Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol.2 [1st ed. 2020] 978-3-030-21008-3, 978-3-030-21009-0

Table of contents :
Front Matter ....Pages i-x
Front Matter ....Pages 1-1
Speech Signal Embedding into Digital Images Using Encryption and Watermarking Techniques (Mourad Talbi)....Pages 3-13
Performance Study of Neural Network Unscented Kalman Filter for Denoising ECG Signal (Sabah Gaamouri, Mounir Bousbia-Salah, Rachid Hamdi)....Pages 14-23
Signal Reconstruction Based on the Relationship Between STFT Magnitude and Phase Spectra (Raja Abdelmalek, Zied Mnasri, Faouzi Benzarti)....Pages 24-36
Real-Time Implementation of an Optimized Speech Compression System in STM32F4 Discovery Board (Souha Bousselmi, Safa Saoud, Adnen Cherif)....Pages 37-48
Degradation Process Analysis and Remaining Useful Life Estimation in a Control System (Nabila Mabrouk, Med Hedi Moulahi, Fayçal Ben Hmida)....Pages 49-65
Data Quality of the Information Collected from GPR on a 3D Structure (Rim Ghozzi, Samer Lahouar, Chokri Souani)....Pages 66-76
DSP Real-Time Implementation of DOST Algorithm Used for Speech Enhancement (Safa Saoud, Souha Bousselmi, Mouhamed Ben Nasr, Adnen Cherif)....Pages 77-88
An Enhanced SDPE Method for Long Delay Multipath Mitigation in GNSS Applications (Wafa Feneniche, Khaled Rouabah, Mustapha Flissi, Salim Atia, Sabrina Meguellati, Salah Eddine Mezaache)....Pages 89-98
Front Matter ....Pages 99-99
A Novel Operational Transconductance Amplifier Based an RLC-Low-Pass Filter (Karima Garradhi, Néjib Hassen, Thouraya Ettaghzouti, Kamel Besbes)....Pages 101-110
Accurate High Level Resources and Power Estimators for FPGAs (Sonia Mami, Younes Lahbib, Yassine Hachaïchi)....Pages 111-123
High Isolation with Neutralization Technique for 5G-MIMO Elliptical Multi-antennas (Marwa Daghari, Chafai Abdelhamid, Hedi Sakli, Kamel Nafkha)....Pages 124-133
Gaussian Process Based Method for Point and Probabilistic Short-Term Wind Power Forecast (Ali Lahouar)....Pages 134-147
UWB-MIMO Array Antennas with DGS Decoupling Structure (Chafai Abdelhamid, Marwa Daghari, Chafaa Hamrouni, Hedi Sakli)....Pages 148-155
Cascade Control Based on TLBO-FOPID for Grid-Connected PV Systems (Afef Badis, Mohamed Habib Boujmil)....Pages 156-166
Embedded Linux Hardware/Software Architecture for Electrical Measurement Acquisition (H. Ben Mansour, L. Chaarabi, K. Jelassi)....Pages 167-176
Simulation and Experimental Study of GaAs Substrate Thermal Conductivity Using 3-Omega Method (A. A. Guermoudi, P. Y. Cresson, A. Ouldabbes, T. Lasri)....Pages 177-184
Drift Effects and Trap Analysis of Power-GaN-HEMT Under Switching Power Cycling (Manuel A. González-Sentís, Patrick Tounsi, Alain Bensoussan, Arnaud Dufour)....Pages 185-193
Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems (Hussein El Ghor, Julia Hage, Nizar Hamadeh, Rafic Hage Chehade)....Pages 194-203
A Novel FPGA-Based Digital Filter Using Fuzzy Logic to Ensure Electromagnetic Compatibility (Yosr Bchir, Soufien Gdaim, Djilali Hamza, Abdellatif Mtibaa)....Pages 204-212
Nonlinear Plasmonic Photoresponse of Field Effect Transistors at Terahertz High Irradiation Intensities (A. Mahi)....Pages 213-219
Simulation and Measurement of a New Circularly Polarized Patch Antenna for WiMAX Applications (El Amjed Hajlaoui)....Pages 220-229
Implementation of Robust Fractional Controller to a Thermal System Using Genetic Algorithm Approach (Aymen Rhouma, Sami Hafsi, Faouzi Bouani)....Pages 230-239
Input-Constrained Controller Design for Nonlinear Systems (Sabrina Aouaouda, Lotfi Moussaoui, Ines Righi)....Pages 240-253
Fast Convergence of 2D DWT-WCIP Method Applied to Study a Complex FSS Structure (S. Bennour, N. Sboui)....Pages 254-260
Absorption Enhancement in an Amorphous Silicon Using a Cluster of Plasmonic Hollow Ring Nano-Antennas (Abdalem A. Rasheed, Khalil H. Sayidmarie, Khalid Khalil Mohammed)....Pages 261-268
Front Matter ....Pages 269-269
Design of a Content-Based Communication Model Using Caching Technique for VANETs (Mohamed Anis Mastouri, Salem Hasnaoui)....Pages 271-280
Monitoring of Greenhouse Based on Internet of Things and Wireless Sensor Network (Achouak Touhami, Khelifa Benahmed, Fateh Bounaama)....Pages 281-289
A Fuzzy Queue Scheduling Controller to Enhance QoS for Terminal Communication (Jamila Bhar)....Pages 290-301
The Performance of RFID Tags in Close Proximity to Human Body (K. Khoder, K. Kaja, A. Choumane, S. Boksmati, H. Amoud)....Pages 302-311
Impact of Phone and Hand Position on SAR Distribution Using Liquid-Based PIFA Antenna (Dina Serhal, Najat Nasser, Mohamed Rammal, Patrick Vaudon)....Pages 312-320
Study and Design of Time Modulated Antenna Array with Low Sides Bands Levels (Alaa Saleh, Mohamad El-Khatib, Mohamad Chakaroun)....Pages 321-326
A Tunable Microwave Bandpass Filter (Mohamed Al Khatib, Alaa Saleh, Mohamad Chakaroun, Mohamed Shehade)....Pages 327-336
Power Saving Approach in LTE Using Switching ON/OFF eNodeB and Power UP/DOWN of Neighbors (Narjes Lassoued, Noureddine Boujnah)....Pages 337-349
Formal-Based Modeling and Analysis of a Network Communication Protocol for IoT: MQTT Protocol (Jamila Hcine, Imene Ben Hafaiedh)....Pages 350-360
Model Based Validation of Real Time QoS for NCDCLA Protocol in Wireless Sensor Networks (Amra Sghaier, Aref Meddeb)....Pages 361-372
SDR-Based Transmitter of Digital Communication System Using USRP and GNU Radio (Nabiha Ben Abid, Chokri Souani)....Pages 373-381
Performance Evaluation of Nonlinear LMMSE-SVR Equalizer for High-Speed Radio Systems (Anis Charrada)....Pages 382-389
A Real-Time Flash-Floods Alerting System Based on WSN and IBM Bluemix Cloud Platform (Hamadi Lirathni, Amira Zrelli, Med Hchemi Jridi, Tahar Ezzedine)....Pages 390-399
Implementation of Web Browser Extension for Mitigating Clickjacking Attack (Amine Benmerzoug, Lalia Saoudi)....Pages 400-411
Integrating Rate-Dependent Transit Time in Dial-A-Ride Problem (Sonia Nasri, Hend Bouziri)....Pages 412-422
Spectral Capacity in Cognitive Networks (Haider Farhi, Abderraouf Messai)....Pages 423-429
Cloud Service for Edge Configuration in Home-Based Healthcare Context (Imen Ben Ida, Abderrazak Jemai)....Pages 430-439
Least Squares Channel Estimation of an OFDM Massive MIMO System for 5G Wireless Communications (Abdelhamid Riadi, Mohamed Boulouird, Moha M’Rabet Hassani)....Pages 440-450
Back Matter ....Pages 451-452

Citation preview

Smart Innovation, Systems and Technologies 147

Med Salim Bouhlel Stefano Rovetta Editors

Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol.2 123

Smart Innovation, Systems and Technologies Volume 147

Series Editors Robert J. Howlett, Bournemouth University and KES International, Shoreham-by-sea, UK Lakhmi C. Jain, Faculty of Engineering and Information Technology, Centre for Artificial Intelligence, University of Technology Sydney, Broadway, NSW, Australia

The Smart Innovation, Systems and Technologies book series encompasses the topics of knowledge, intelligence, innovation and sustainability. The aim of the series is to make available a platform for the publication of books on all aspects of single and multi-disciplinary research on these themes in order to make the latest results available in a readily-accessible form. Volumes on interdisciplinary research combining two or more of these areas is particularly sought. The series covers systems and paradigms that employ knowledge and intelligence in a broad sense. Its scope is systems having embedded knowledge and intelligence, which may be applied to the solution of world problems in industry, the environment and the community. It also focusses on the knowledge-transfer methodologies and innovation strategies employed to make this happen effectively. The combination of intelligent systems tools and a broad range of applications introduces a need for a synergy of disciplines from science, technology, business and the humanities. The series will include conference proceedings, edited collections, monographs, handbooks, reference books, and other relevant types of book in areas of science and technology where smart systems and technologies can offer innovative solutions. High quality content is an essential feature for all book proposals accepted for the series. It is expected that editors of all accepted volumes will ensure that contributions are subjected to an appropriate level of reviewing process and adhere to KES quality principles. ** Indexing: The books of this series are submitted to ISI Proceedings, EI-Compendex, SCOPUS, Google Scholar and Springerlink **

More information about this series at http://www.springer.com/series/8767

Med Salim Bouhlel Stefano Rovetta •

Editors

Proceedings of the 8th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT’18), Vol.2

123

Editors Med Salim Bouhlel SETIT Lab University of Sfax Sfax, Tunisia

Stefano Rovetta DIBRIS - University of Genoa Genoa, Genova, Italy

ISSN 2190-3018 ISSN 2190-3026 (electronic) Smart Innovation, Systems and Technologies ISBN 978-3-030-21008-3 ISBN 978-3-030-21009-0 (eBook) https://doi.org/10.1007/978-3-030-21009-0 © Springer Nature Switzerland AG 2020 This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

Preface

This book collects selected and revised papers that were presented at the international conference SETIT 2018, The Conference on the Sciences of Electronics, Technologies of Information and Telecommunications, covering all topics in the fields of information and communication technologies and related sciences. The conference was held in Hammamet, Tunisia, from December 18 to 20, 2018. The aim of this conference series, a major international event, is to bring together researchers and developers from both academia and industry to report on the latest scientific and theoretical advances in their respective areas, fostering a cross-disciplinary dissemination that would be otherwise made difficult by the extreme specialization of each field. This is a recent trend that characterizes very successful events and publications, and encourages scholars and professionals to overcome disciplinary barriers. In today’s information-centered world, the relevance of hardware, software, telecommunications cannot be overestimated; even fields traditionally involved only marginally, like factory automation and production engineering, are currently discovering the value of data and information, with such trends as the Industry 4.0 or the exponential growth of machine learning and AI, and with all the consequent technological developments that are needed to support them. Both theoretical advances and interesting applications were submitted to the conference, as it gave special emphasis to interdisciplinary works at the intersection of two or more of the covered areas. The papers that are included in this collection have been selected after their presentation at the conference and were carefully revised. But, in addition to this scientific production per se, the event had also another important role, providing an occasion for exchanging experiences and for introducing many young scientists in their training phase to an international scientific community, giving them opportunities for networking and professional growth.

v

vi

Preface

We are therefore grateful to the contributors of this collection, and to all participants, for their cooperation, interest, enthusiasm, and lively interactions, that helped making the conference not only a scientifically stimulating event, but also a memorable experience. March 2019

Med Salim Bouhlel Stefano Rovetta

Contents

Signal Processing Speech Signal Embedding into Digital Images Using Encryption and Watermarking Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mourad Talbi

3

Performance Study of Neural Network Unscented Kalman Filter for Denoising ECG Signal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sabah Gaamouri, Mounir Bousbia-Salah, and Rachid Hamdi

14

Signal Reconstruction Based on the Relationship Between STFT Magnitude and Phase Spectra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Raja Abdelmalek, Zied Mnasri, and Faouzi Benzarti

24

Real-Time Implementation of an Optimized Speech Compression System in STM32F4 Discovery Board . . . . . . . . . . . . . . . . . . . . . . . . . . Souha Bousselmi, Safa Saoud, and Adnen Cherif

37

Degradation Process Analysis and Remaining Useful Life Estimation in a Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Nabila Mabrouk, Med Hedi Moulahi, and Fayçal Ben Hmida

49

Data Quality of the Information Collected from GPR on a 3D Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Rim Ghozzi, Samer Lahouar, and Chokri Souani

66

DSP Real-Time Implementation of DOST Algorithm Used for Speech Enhancement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Safa Saoud, Souha Bousselmi, Mouhamed Ben Nasr, and Adnen Cherif

77

An Enhanced SDPE Method for Long Delay Multipath Mitigation in GNSS Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Wafa Feneniche, Khaled Rouabah, Mustapha Flissi, Salim Atia, Sabrina Meguellati, and Salah Eddine Mezaache

89

vii

viii

Contents

Electronics A Novel Operational Transconductance Amplifier Based an RLC-Low-Pass Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Karima Garradhi, Néjib Hassen, Thouraya Ettaghzouti, and Kamel Besbes Accurate High Level Resources and Power Estimators for FPGAs . . . . 111 Sonia Mami, Younes Lahbib, and Yassine Hachaïchi High Isolation with Neutralization Technique for 5G-MIMO Elliptical Multi-antennas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 Marwa Daghari, Chafai Abdelhamid, Hedi Sakli, and Kamel Nafkha Gaussian Process Based Method for Point and Probabilistic Short-Term Wind Power Forecast . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 Ali Lahouar UWB-MIMO Array Antennas with DGS Decoupling Structure . . . . . . . 148 Chafai Abdelhamid, Marwa Daghari, Chafaa Hamrouni, and Hedi Sakli Cascade Control Based on TLBO-FOPID for Grid-Connected PV Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 Afef Badis and Mohamed Habib Boujmil Embedded Linux Hardware/Software Architecture for Electrical Measurement Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167 H. Ben Mansour, L. Chaarabi, and K. Jelassi Simulation and Experimental Study of GaAs Substrate Thermal Conductivity Using 3-Omega Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 A. A. Guermoudi, P. Y. Cresson, A. Ouldabbes, and T. Lasri Drift Effects and Trap Analysis of Power-GaN-HEMT Under Switching Power Cycling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 Manuel A. González-Sentís, Patrick Tounsi, Alain Bensoussan, and Arnaud Dufour Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194 Hussein El Ghor, Julia Hage, Nizar Hamadeh, and Rafic Hage Chehade A Novel FPGA-Based Digital Filter Using Fuzzy Logic to Ensure Electromagnetic Compatibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204 Yosr Bchir, Soufien Gdaim, Djilali Hamza, and Abdellatif Mtibaa Nonlinear Plasmonic Photoresponse of Field Effect Transistors at Terahertz High Irradiation Intensities . . . . . . . . . . . . . . . . . . . . . . . . 213 A. Mahi

Contents

ix

Simulation and Measurement of a New Circularly Polarized Patch Antenna for WiMAX Applications . . . . . . . . . . . . . . . . . . . . . . . . 220 El Amjed Hajlaoui Implementation of Robust Fractional Controller to a Thermal System Using Genetic Algorithm Approach . . . . . . . . . . . . . . . . . . . . . . 230 Aymen Rhouma, Sami Hafsi, and Faouzi Bouani Input-Constrained Controller Design for Nonlinear Systems . . . . . . . . . 240 Sabrina Aouaouda, Lotfi Moussaoui, and Ines Righi Fast Convergence of 2D DWT-WCIP Method Applied to Study a Complex FSS Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254 S. Bennour and N. Sboui Absorption Enhancement in an Amorphous Silicon Using a Cluster of Plasmonic Hollow Ring Nano-Antennas . . . . . . . . . . 261 Abdalem A. Rasheed, Khalil H. Sayidmarie, and Khalid Khalil Mohammed Telecommunications and Networks Design of a Content-Based Communication Model Using Caching Technique for VANETs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Mohamed Anis Mastouri and Salem Hasnaoui Monitoring of Greenhouse Based on Internet of Things and Wireless Sensor Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 Achouak Touhami, Khelifa Benahmed, and Fateh Bounaama A Fuzzy Queue Scheduling Controller to Enhance QoS for Terminal Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290 Jamila Bhar The Performance of RFID Tags in Close Proximity to Human Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302 K. Khoder, K. Kaja, A. Choumane, S. Boksmati, and H. Amoud Impact of Phone and Hand Position on SAR Distribution Using Liquid-Based PIFA Antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312 Dina Serhal, Najat Nasser, Mohamed Rammal, and Patrick Vaudon Study and Design of Time Modulated Antenna Array with Low Sides Bands Levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321 Alaa Saleh, Mohamad El-Khatib, and Mohamad Chakaroun A Tunable Microwave Bandpass Filter . . . . . . . . . . . . . . . . . . . . . . . . . 327 Mohamed Al Khatib, Alaa Saleh, Mohamad Chakaroun, and Mohamed Shehade

x

Contents

Power Saving Approach in LTE Using Switching ON/OFF eNodeB and Power UP/DOWN of Neighbors . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Narjes Lassoued and Noureddine Boujnah Formal-Based Modeling and Analysis of a Network Communication Protocol for IoT: MQTT Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 Jamila Hcine and Imene Ben Hafaiedh Model Based Validation of Real Time QoS for NCDCLA Protocol in Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361 Amra Sghaier and Aref Meddeb SDR-Based Transmitter of Digital Communication System Using USRP and GNU Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Nabiha Ben Abid and Chokri Souani Performance Evaluation of Nonlinear LMMSE-SVR Equalizer for High-Speed Radio Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382 Anis Charrada A Real-Time Flash-Floods Alerting System Based on WSN and IBM Bluemix Cloud Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390 Hamadi Lirathni, Amira Zrelli, Med Hchemi Jridi, and Tahar Ezzedine Implementation of Web Browser Extension for Mitigating Clickjacking Attack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400 Amine Benmerzoug and Lalia Saoudi Integrating Rate-Dependent Transit Time in Dial-A-Ride Problem . . . . 412 Sonia Nasri and Hend Bouziri Spectral Capacity in Cognitive Networks . . . . . . . . . . . . . . . . . . . . . . . . 423 Haider Farhi and Abderraouf Messai Cloud Service for Edge Configuration in Home-Based Healthcare Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430 Imen Ben Ida and Abderrazak Jemai Least Squares Channel Estimation of an OFDM Massive MIMO System for 5G Wireless Communications . . . . . . . . . . . . . . . . . . . . . . . . 440 Abdelhamid Riadi, Mohamed Boulouird, and Moha M’Rabet Hassani Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451

Signal Processing

Speech Signal Embedding into Digital Images Using Encryption and Watermarking Techniques Mourad Talbi(&) Center of Researches and Technologies of Energy of Borj Cedria (CRTEn), Borj Cedria, Postal Box N°95, 2050 Hammam-Lif, Tunisia [email protected]

Abstract. In the present article, is proposed a novel encryption-watermarking approach. It is applied for embedding speech signals into digital images. It consists at first step in encrypting the watermark. After that, the obtained encrypted watermark is embedded into the host image. This watermark encryption is performed employing Arnold cat map and the embedding is performed employing a secure image watermarking technique based on LWT
SVD. We evaluate the performance of this proposed technique by comparing it to an existing encryption watermarking approach called DSAWM. The latter is also a secure watermarking scheme based on DWT
SVD. It also applies Arnold Cat Map for encrypting the watermark at first step and then inserting the obtained encrypted watermark into the cover image. Evaluation and comparison of the two techniques are performed by computing the SegSNR; PESQ; PSNR; RMSE, pcc, MAE and SSIM. The obtained results from this evaluation show the performance of this proposed encryption-watermarking approach. In fact, the PSNR, RMSE, pcc, MAE and SSIM values show a good perceptual quality of the watermarked images. Moreover, the SegSNR and PESQ values show a very good perceptual quality of the speech signals reconstructed after watermark extraction and decryption. We also tested the robustness of the proposed technique by applying three sorts of attacks on the watermarked images. Those attacks are Median, JPEG Compression and additive White Noise attacks. The results obtained from calculation of the SNR, PESQ; PSNR and SSIM, show the robustness of this proposed approach against those attacks. Keywords: Image watermarking  Encryption  Arnold cat map Lifting wavelet transform  Singular values decomposition



1 Introduction With the digital technology development, communication, computer science and network, online services are extensively hurled. However it’s piracy, unlawful copying, spiteful manipulations and counterfeiting have prompted the necessity of protection of multimedia [1]. An efficient way for solving those problems is the digital watermarking.

© Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 3–13, 2020. https://doi.org/10.1007/978-3-030-21009-0_1

4

M. Talbi

Indeed, it provides copyright protection of data and this by inserting further information (watermark) so it can be extracted for making an assertion concerning the multimedia data [2]. In watermarking, both bit stream and the host file are respectively called the message and the asset. Both cryptography and steganography are other famous techniques. Digital watermarking [3] has numerous applications such as file reconstruction, content authentication, owner identification and broadcasting monitoring. The goal of a good watermarking approach consists in preserving the watermark robustness against spiteful attacks in spectral, spatial and hybrid domains. It has to find a good compromise between imperceptibility and robustness [2]. The image watermarking schemes usually work in the transform or in the spatial domain. In the spatial domain, both watermark embedding and extraction are performed by modifying both intensity and the value of color of some chosen pixels. Numerous spatial domain techniques are proposed and among them we can mention LSB (Least Significant Bit), patchwork technique with streak block mapped coding, approach based on district intersecting. Different schemes in the transform domain, are spread spectrum. Among them, we can mention the DWT (Discrete Wavelet Transform) and the DCT (Discrete Cosine Transform). To guarantee the watermark security, its encryption is performed before embedding [2]. The encryption process consists in encoding information or messages in such manner that its extraction is performed by only authorized parties. Keys may be private or public. Public ones are obtainable by any one through a public accessible directory [2]. Private ones are confidential to its respective proprietors [2]. Public–Private key is an efficient pair since the watermark is encrypted employing a public key and is decrypted employing a private key. Sanpreet et al. [2] have proposed a technique of digital image watermarking employing DWT and SVD and also an encryption based on chaos. The DWT
SVD based approach was employed for the watermark embedding [2]. Chaotic maps own a good potential for designing dynamic permutation map and this owing to sensitivity to initial conditions. Chaotic output signal has random statistical properties employed for operations of confusion and diffusion in a crypto-system. Random statistical property permits to make watermark random in aggressors eye, however its decryption can be performed by authorized individual. One level decomposition of the host image is performed via DWT and SVD. The encryption of the watermark is performed using chaotic map and the obtained encrypted watermark is decomposed employing one level DWT
SVD [2]. The encrypted watermark’s lower frequency singular value is inserted into the cover image lower frequency singular value [2]. The IDWT (DWT inverse) is applied for the watermarked image reconstruction. Then the decryption of the extracted watermark is performed [2]. Al-Janabi et al. [4] proposed a secure technique for hiding one or more images at the same time inside the host image having the same size. This technique employs the genetic algorithm for generating secret key and also for selecting optimum mixing matrix values. This key is shared between only the message sender and receiver in order to guarantee that the data sources are reserved securely and their detection by untrusted third parties is hard. In this work, is proposed a novel encryption-watermarking approach. It consists at first step in encrypting the watermark before inserting the obtained encrypted watermark into the host image. Encryption is performed via Arnold Cat Map [2] and the embedding is

Speech Signal Embedding into Digital Images Using Encryption

5

performed via a secure watermarking scheme based on LWT
SVD [5]. In the rest of this paper we will detail the LWT
SVD based image watermarking approach proposed in [5]. After that, we will detail our proposed encryption-watermarking scheme for embedding speech signals into digital images. Then, we will present results and discussions and the conclusion will be given at the end of this paper.

2 The Watermarking Method Based on LWT-SVD In our previous research works, we have proposed some techniques of speech signal embedding into digital image [5, 6]. Among them, we will detail the technique based on LWT
SVD [5]. It consists at first step in cutting the speech signal into short frames. Each of them is then transposed in order to construct a speech matrix which will be processed as an image (speech image). This speech image is considered as a watermark to be embedded into the host image. The embedding and extraction are performed via a secure image watermarking method based on LWT
SVD [5]. The latter [5] consists at first step in applying the LWT
2D [7, 8] to the original image (I) and four sub-images, LL; HL; LH and HH, are obtained. The sub-band LL designates the lower resolution approximation component. The others sub-images HL; LH and HH are respectively horizontal, vertical and diagonal detail coefficients [6]. The second step of this approach [5] consists in applying the SVD [9] to both HH and the watermark and having the matrixes Uw , Sw and Vw for the watermark and the matrixes Uh , Sh and Vh for HH. After that, the Singular Values (SVs) of HH are replaced with those of the watermark. Then, the signature is generated [5] and it is inserted into LL. After that, the SVD is applied in order to have the modified sub-image HH holding now the SV’s of the watermark. Then, the ILWT (LWT Inverse) is applied to the modified LL, LLM , the modified HH, HHM and the two sub-bands HL and LH, for reconstructing the watermarked image [5]. The different steps of the Watermark Extraction are detailed in the following subsection. 2.1

Watermark Extraction

1. By employing the Haar Mother Wavelet, we decompose the watermarked image into 4 sub-images,HH; HL; LH, and LL, 2. Then, LL is decomposed up to the fourth level in order to obtain at the last level four sub-bands, LLw 4, HLw 4, LHw 4 and HHw 4, 3. Then Applying the Singular Values Decomposition (SVD) to the watermark logo, 4. After that, the signature is generated using the matrixes Vw and Uw , 5. Then, extracting the signature from LLw 4 and HHw_4, employing all coefficients which are in number of 512, 6. Then comparing the embedded and the extracted signatures, 7. Proceed to extract the watermark when the authentication is successful,

6

M. Talbi

8. Then applying the SVD to the modified sub-image HHM , 9. After that, Extracting the SVs from HHM , 10. Then, Reconstructing the watermark employing the SVs and the orthogonal matrixes, Uw and Vw .

3 The Proposed Encryption-Watermarking Technique In this paper, the proposed encryption-watermarking technique can be summarized by the following block diagrams illustrated in Figs. 1 and 2.

Fig. 1. Encryption-watermarking sub-system proposed for speech signal embedding.

As shown in Fig. 1, we first multiply by a factor a, the speech image (Is ) constructed from the original speech signal [5]. Then the modified speech image (IS0 ) is encrypted using the encryption system based on chaotic map and proposed in [2]. After that we multiply the encrypted speech image by the factor 1a in order to obtain the modified encrypted image (IE ). Then, IE is embedded into the host image by using the previous image watermarking technique [5] in order to have finally the watermarked image, Iw . As shown in Fig. 2, the extraction and reconstruction of the speech signal are performed by five steps where the first consists in extracting the watermark, IE from the watermarked image, Iw . This extraction is performed using the previous image watermarking technique [5]. The second step consists in multiplying IE by the factor a in order to obtain the modified image, IE0 . The third step consists in decrypting IE0 . The fourth step consists in multiplying the decrypted image obtained in the third step by the factor 1a in order to have the extracted speech image, IS . Finally, the fifth step consists in the speech signal reconstruction from IS [5].

Speech Signal Embedding into Digital Images Using Encryption

7

Fig. 2. The speech image extraction and speech signal reconstruction.

4 Results and Discussion In this part, is tested the performance of the proposed encryption-watermarking system (based on LWT
SVD [5, 10] and chaotic Map encryption [2, 11]) by comparing it to the image watermarking method proposed in [2]. The latter is a robust and secure technique employing DWT
SVD and Chaotic Map. For this comparative study, we have introduced an encryption block in the image watermarking system proposed in [5]. In this block, is used the chaos based encryption employing Arnold Cat Map [2]. As previously mentioned, the proposed overall watermarking system is illustrated in Fig. 1. For simulation of the proposed technique and the second one proposed in [2], we have chosen in this section, 200 as the value of a. Figures 3, 4, 5 and 6 show an example of Image Watermarking using the technique based on DWT
SVD and Chaotic Map [2]. Figure 5 shows the original, the encrypted and the decrypted speech images. According to Fig. 3, the original and the reconstructed speech signals are very similar that’s why we have passed to the spectral domain and draw their spectrograms (Fig. 6). Those spectrograms show that the reconstructed speech signal presents some distortions compared to the original speech signal. These distortions are especially occurred in unvoiced regions of the signal. These regions are located in high frequencies in the spectrogram ( [ 3 kHz).

8

M. Talbi (a) Original Speech Signal 1 0.5 0 -0.5 -1

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5 4

x 10 (b) Reconstructed Speech Signal 1 0.5 0 -0.5 -1

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5 4

x 10

Fig. 3. An example of image watermarking applying the image watermarking method based on DWT
SVD and Chaotic Map [2]: (a) Original speech signal, (b) Extracted or reconstructed speech signal obtained after Watermark extraction and decryption.

Fig. 4. The same example of image watermarking applying the image watermarking method based on DWT
SVD and Chaotic Map [2]: (a) the cover image, (b) the watermarked image, (c) the watermark logo which is the encrypted speech image, (d) the extracted watermark.

encrypted

input 50 100 150 200 250

50 100 150 200 250 50100150200250

decrypted 50 100 150 200 250

50100150200250

50100150200250

Fig. 5. Input: the original speech image, the encrypted speech image using Chaotic Map, the decrypted speech image.

Speech Signal Embedding into Digital Images Using Encryption

9

Fig. 6. The Spectrograms: (a) the spectrogram of the original speech signal, (b) the spectrogram of the speech signal reconstructed from the extracted speech image (Fig. 3).

In Table 1 are listed the results obtained from computations of SegSNR; PESQ; PSNR; MSE; pcc; MAE and SSIM. Those results are obtained for an example of embedding a speech signal (Fig. 3) into an image of Tower (Fig. 4a). This embedding is performed by using the image encryption-watermarking technique (Figs. 1 and 2) proposed in this work. Table 1. An example of image watermarking applying the two techniques: the obtained values of SegSNR; PESQ; PSNR; MSE; pcc; MAE and SSIM. The watermarking technique The proposed technique The watermarking technique using DWT
SVD and Chaotic Map [2]

SegSNR (dB) 24.0662 24.0662

PESQ

PSNR MSE pcc

MAE

SSIM

4.4935 31.133 7.07 2.5774e+05 4.1250 0.8355 4.4935 31.969 6.42 2.5853e+05 3.8655 0.8631

According to those results (Table 1), the technique using DWT
SVD and Chaotic Map proposed in [2], is slightly better than the proposed technique. However, the main advantage of the proposed technique consists in using the LWT
2D instead of DWT
2D [12]. In fact, the LWT
2D is a powerful tool of image analysis and this thanks to its faster and efficient implementation compared to DWT
2D. The LWT
2D saves times and has a better frequency localization feature and this permits to overcome the shortcomings of DWT
2D [13]. 4.1

Robustness Against Attacks

In this sub-section we are interested in the robustness of the proposed technique against three types of attacks which are JPEG compression, Median and White noise attacks. Each of those attacks is applied on the watermarked image and the PSNR and SSIM are

10

M. Talbi

computed in order to evaluate the perceptual quality of this watermarked image compared to the host image. The SNR and the PESQ are also computed in order to evaluate the perceptual quality of the extracted speech signal compared to the original signal. In Table 2 are listed the values obtained from the calculation of PSNR; SSIM; SNR and PESQ and this for an example of image watermarking applying the proposed technique in the three cases of the mentioned attacks. Table 2. The results obtained from the calculation of PSNR; SSIM; SNR and PESQ for an example of image watermarking applying the proposed encryption-watermarking technique with one of the mentioned attacks. Type of attack

JPEG Compression Attack Median Attack Additive White Noise attack r ¼ 15

The proposed encryption-watermarking technique with a ¼ 1000 SNR (dB) PESQ PSN-R SSIM 13.9876 2.9361 35.5278 0.8234 17.9428 3.0674 34.4563 0.8692 14.6590 3.0313 23.9535 0.3270

(a) Cover image

(b) W atermarked image signed with secret key

(c ) W atermark logo

Fig. 7. An example of image watermarking using the encryption-watermarking approach: (a) the host image, (b) the watermarked image signed with secret key and attacked by JPEG 0 Compression, (c) the watermark logo obtained after encryption of the modified speech image IS (Fig. 1). 1

1

0

0

-1

0

1

2

3

4 x 10

(a)

-1

0

1

2

3

4 4

4

x 10

(b)

Fig. 8. The same example of image watermarking (Case of JPEG Compression Attack): (a) Original speech signal, (b) Speech signal reconstructed after watermark extraction and decryption.

Speech Signal Embedding into Digital Images Using Encryption (a) Cover image

11

(b) W atermarked image signed with secret key

(c ) W atermark logo

Fig. 9. The same example of image watermarking: (a) the host image, (b) the watermarked image signed with secret key and attacked by Median Attack, (c) the watermark logo obtained after encryption of the modified speech image IS0 (Fig. 1).

1

1

0

0

-1

0

1

2

3

4 x 10

(a)

-1

0

1

2

4

3

4 x 10

4

(b)

Fig. 10. The same example of image watermarking (Case of Median Attack): (a) Original speech signal, (b) Speech signal reconstructed after watermark extraction and decryption.

(a) Cover image

(b) W atermarked image signed with secret key

(c ) W atermark logo

Fig. 11. The same example of image watermarking: (a) the host image, (b) the watermarked image signed with secret key attacked by additive White Noise (r ¼ 15), (c) the watermark logo obtained after encryption of the modified speech image IS0 (Fig. 1).

12

M. Talbi

1

1

0

0

-1

0

1

2

-1 4 0

3 x 10

(a)

1

2

3

4

4

x 10

4

(b)

Fig. 12. The same example of image watermarking (Case of Additive White noise Attack ðr ¼ 15Þ): (a) Original speech signal, (b) Speech signal reconstructed after watermark extraction and decryption.

Table 2 and Figs. 7, 8, 9, 10, 11 and 12 show the robustness of the proposed technique against those three attacks. In fact, the reconstructed speech signal has an acceptable perceptual quality and this according to the values of SNR, PESQ. Concerning the perceptual quality of the watermarked image, it is good in cases of the JPEG Compression and Median attacks and it is worse in case of the additive White noise attack and this according to the PSNR and SSIM values.

5 Conclusion In this work, is proposed a novel encryption-watermarking technique. It is applied for embedding speech signals into digital images. This technique consists in encrypting the watermark before the embedding of the obtained encrypted watermark into the cover image. This embedding was performed using a secure watermarking technique based on LWT
SVD. The encryption was performed using Arnold Cat Map. We evaluated the performance of this proposed technique by comparing it to an encryption-watermarking approach called DSAWM. The latter is existing in literature and is also a secure watermarking one based on DWT
SVD. It also applies Arnold Cat Map for encrypting the watermark. After that, is performed the embedding of the obtained encrypted watermark into the cover image. The evaluation and comparison of these two techniques are performed by computing the SegSNR; PESQ; PSNR; pcc; MAE and SSIM and the obtained results show that the second encryption-watermarking approach, DSAWM, is slightly better than the proposed technique. However, the main advantage of the proposed technique consists in using the LWT
2D instead of DWT
2D. In fact, the LWT
2D is a powerful tool of image analysis and this thanks to its faster and efficient implementation compared to DWT
2D. The LWT
2D saves times and has a better frequency localization feature that permits to overcome the shortcomings of DWT
2D. We also tested the robustness of our proposed technique against three types of attacks which are JPEG compression, Median and additive White noise Attacks. The results obtained from calculation of the SSIM; PSNR; SNR and PESQ, show the robustness of the proposed technique against those three attacks. In fact, the reconstructed speech signal

Speech Signal Embedding into Digital Images Using Encryption

13

has an acceptable perceptual quality and this according to the values of SNR and PESQ. Concerning the perceptual quality of the watermarked image, it is good in cases of the JPEG Compression and Median attacks and it is worse in case of the additive White noise attack and this according to the values of PSNR and SSIM.

References 1. Chandra, M., Pandey, S., Chaudhary, R.: Digital watermarking technique for protecting digital images. In: 3rd IEEE International Conference on Computer Science and Information Technology, pp. 226–233. IEEE, Chengdu, China (2010) 2. Sanpreet, S., Savina, B., Sukhjinder, S.: Robust and secure image watermarking using DWTSVD and chaotic map. Int. J. Adv. Res. Comput. Commun. Eng. 4(9), 111–116 (2015) 3. Ingemer, C., Mathew, M., Jeffrey, B., Jessica, F., Ton, K.: Digital Watermarking and Stegnography, 2nd edn. Morgan Kaufmann Publishers, Burlington (2007) 4. Al-Janabi S., Al-Shourbaji, I.: A hybrid Image steganography method based on genetic algorithm. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 398–404. IEEE, Hammamet, Tunisia (2016) 5. Talbi, M., Bouhlel, Med S.: Singular values decomposition and lifting wavelet transform for speech signal embedding into digital image. J. Recent Adv. Electr. Electron. Eng. 12(2), 131–142 (2019) 6. Ben Fatima, S., Talbi, M., Ezzeddine, T.: LWT-SVD secure image watermarking technique. In: International Conference of Electronics, Communication and Aerospace Technology (ICECA), pp. 510–517, IEEE, Coimbatore, India (2017) 7. Daubeches, I., Sweldens, W.: Factoring wavelet transform into lifting steps. J. Fourier Anal. Appl. 4(3), 247–269 (1998) 8. Fan, W., Chen, J., Zhen, J.: SPIHT algorithm based on fast lifting wavelet transform in image compression. In: Hao, Y., et al. (eds.) Computational Intelligence and Security. CIS 2005, Lecture Notes in Computer Science, pp. 838–844. Springer, Berlin, Heidelberg (2005) 9. Lafi, S., Khalfallah, A., Med Salim, B.: A cholesterol lesion detection approach based on SVD decomposition. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 503–508. IEEE, Hammamet, Tunisia (2016) 10. Mourad, T., Med Salim, B.: Secure Image Watermarking Based on LWT and SVD. Int. J. Image Graph. 18(4), 00560-1–00560-25 (2018) 11. Wang, B., Zhou, S., Zheng, X., Zhou, C., Dong, J., Zhao, L.: Image watermarking using chaotic map and DNA coding. Optik – Int. J. Light Electron Optics 126(24), 4846–4851 (2015) 12. Chabchoub, S., Mansouri, S., Ben Salah, R.: Impedance cardiography heartbeat classification using LP, DWT, KNN and SVM. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 53–57. IEEE, Hammamet, Tunisia (2016) 13. Gu, Z.F., Li, G.-F., Yang, Z.-X.: Study on digital image watermark algorithm based on chaos mapping and DWT. In: The International Conference on Advanced Mechatronic Systems, pp. 160–164. IEEE, Tokyo, Japan (2012)

Performance Study of Neural Network Unscented Kalman Filter for Denoising ECG Signal Sabah Gaamouri(&), Mounir Bousbia-Salah(&), and Rachid Hamdi(&) Department of Electronics, BADJI Mokhtar Annaba University, LASA Laboratory, 23000 Annaba, Algeria [email protected], [email protected], [email protected]

Abstract. The aim of this paper is a removal of noise from electrocardiogram (ECG) signals. A dual unscented Kalman filter based on multilayer perceptron (MLP) has been proposed for removing the artificial white, colored Gaussian noises and non-stationary muscle artifact from ECG signals. The (MLP) is used as the nonlinear functional form of the unknown model. Dual Unscented Kalman filter (UKF) is used for the part of the algorithm that estimates the clean state and the weights of the network. The obtained results are compared with other enhancement conventional filters, such as, normalized least mean square (NLMS) and Butterworth filter (BF). The quantitative study of output of the different methods has been presented based on mean squared error (MSE), signal to noise ratio (SNR) and peak signal to noise ratio (PSNR). By considering these parameters, the experimental comparative analysis has shown that the UKF-MLP had optimal performance and capability than conventional filters for denoising ECG signal. Keywords: Electrocardiogram


Unscented Kalman filter
Neural network

1 Introduction The electrocardiogram (ECG) is a medical graphic are produced from translation of electrical activity of the cardiac. It gives a lot of information on the physiology of this organ. This cardiac signal contains various segments, such as, P wave, QRS complex and T wave [1]. The denoising of ECG signal represents one of the main problems of biomedical signal processing. When an ECG is recorded at the medical clinic, it would be contaminated with different noises from biological and environmental resources. For instance, ECG artifacts are caused by movement of electrodes, respiration and power line interferences [1]. Distorted signal makes the clinical interpretation and information retrieval very difficult. Therefore, it is necessary to remove all such disturbances in ECG for correct diagnosis. So in the literature, many methods have been proposed for the elimination of different noise from the ECG signal. For instance, band-pass filtering (BF) [2] cannot eliminate the real non-stationary, muscle artifact, because this type of noise is overleaped with heart components in the frequency domain, especially in the © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 14–23, 2020. https://doi.org/10.1007/978-3-030-21009-0_2

Performance Study of Neural Network

15

range of 0.01 Hz–100 Hz [3]. On the other hand, the least mean square adaptive algorithm (LMS) [4] which is not able to track the rapidly varying non-stationary signal such as the ECG within each heart beat because it needs to adapt the size of their parameter for each new iteration. This causes excessive low pass filter of mean parameters like QRS complex. The recursive least-squares (RLS) [5] is another adaptive algorithm. It is the fastest of all traditional adaptive methods. However, the main disadvantage of this algorithm is a computational cost. Recently, many documents have presented the wavelet denoising (WD). The main reason for using this method is the properties of good representation in the time-frequency of the nonstationary signals, such as, ECG signal [5]. The Kalman filter (KF) has been extensively studied in many applications, particular for ECG enhancement. On the other hand, the nature of most of the systems are nonlinear; in this case, the denoising of ECG signal is reformulated as a nonlinear estimation problem, where the extended Kalman filter (EKF) is the optimal solution popular approach to apply of the KF for a nonlinear signal. However, the performance quality of the EKF depends on the correct advance knowledge of process observation noise and covariance matrices Q and R, respectively. For highly nonlinear systems, the EKF suffers difficulty to obtain these parameters. This caused a significant deterioration of the performance [6]. The drawbacks of the EKF are overcome by the unscented Kalman filter (UKF). This gives a performance equal to KF for linear systems. Also, it applies in a better way to non-linearities without going through the linearized steps of the EKF. The UKF is more accurate in estimating the mean and the covariance up to second order by samples construction with various weights based on deterministic styles. The problem related with UKF is resides in computational cost and numerical instability. This causes the diverging of the algorithm. On the other hand, the multilayer perceptron (MLP) is one of the most implemented neural network topologies. The three layers include one input layer, one hidden layer and one output layer. Each one is composed of several nodes (artificial neurons). It computes one output from multiple real-valued inputs by forming a linear combination according to its input weights. Afterwards, it may put the output through some a nonlinear function [7]. Noise removal and analysis of the ECG signal by KF and MLP have been previously studied. One particular study conducted by Moein [8] applied Kalman filter to eliminate low pass noise. A suitable dataset based on filtered ECG signal is then configured and used to train Multi Layer Perceptron Neural Network (MLPNN). Avendano et al. [9] proposed a technique for estimating time-varying parametric spectrum from ECG sequences by using Kalman filters. Ahrens et al. [10] compared the performance of the EKF and the UKF in order to solve the nonlinear state-space model and for subsequent imaging of the activation/depolarization times of the cardiac muscle. Sameni et al. [11] have proposed a nonlinear Bayesian filtering framework to filter one channel of noisy ECG signals. Su [12] proposed a dynamical model which permits the Bayesian filtering framework to function online. The objective of this study is to enhancement the performance of this filter (UKF) and to reduce measurement uncertainty. We then propose a new method to remove the different types of noise from ECG signal. In this case, we consider the UKF to train a MLP. The proposed method is based on online training, where the weights are

16

S. Gaamouri et al.

updated directly after the presentation of a training model. It does not require the storage of the entire input output record. The UKF algorithm improves the learning speed and reduces the number of tuning parameters, thus achieving numerical stability. Moreover, and in order to adapt the technique to several ECGs, the model parameter of the MLP is automatically selected. To evaluate the proposed approach, different portions of artificial white, colored Gaussian noises and real muscle artifact have been added to clean ECGs. Also, the results are evaluated with MSE, SNR and PSNR [13] of the proposed filter outputs compared with traditional ECG denoising techniques. The results show that this method can tracks this physiological signal with different types of noises.

2 The UKF for ECG Enhancement When the model is highly nonlinear, the UKF is used to solve the problem of the estimation state for any nonlinear stochastic system from noisy observations. UKF is a filter based on the unscented transform (UT). The deterministic sampling technique is used for capture the mean and the covariance estimations with a minimal set of samples [6]. 2.1

State-Space Equations of Signal and Noise Models

The discrete clean signal xk and the observation vector yk at time instant k are modeled as a nonlinear auto-regression as follows:
 xðkÞ ¼ f xðk1Þ ; . . .xðkpÞ ; w þ vðkÞ

ð1Þ

yðkÞ ¼ xðkÞ þ nðkÞ

ð2Þ

Where xk and yk are the clean and the noisy ECG signals. The noise terms vðkÞ and nðkÞ are Gaussian white noise with known zero-mean normally-distributed random variables and covariance matrices Qk and Rk , respectively. f ð:Þ is the evolution of the nonlinear state of past values of xk parameterized by w. In order to estimate the state vector Xk by using the UKF structure, Eqs. (1) and (2) can be written in a state-space form by: XðkÞ ¼ F½ðXðk1Þ Þ þ GvðkÞ

ð3Þ

yðkÞ ¼ HXðkÞ þ nðkÞ

ð4Þ

XðkÞ ¼ ½xðkÞ ; xðk1Þ ; xðk2Þ ; . . .; xðkp þ 1Þ T

ð5Þ

F½ðXðk1Þ Þ ¼ f ðxðkÞ ; . . .; xðkp þ 1Þ ; wÞ; xðkÞ . . .xðkp þ 2Þ T

ð6Þ

The hidden state vector is:

Performance Study of Neural Network

H ¼ ½1 0. . .. . .0 and G ¼ H T

17

ð7Þ

Where p represents the dimensional vectors for the excitation noise and observation respectively. F½ðXðk1Þ Þ can be expressed as AXðkÞ . A is in a canonical form and it can controllable. 2.2

The UKF to Train MLP for ECG Denoising

The UT is a technique for computing the statistics of a random variable which undergoes a nonlinear transformation. The UKF is a straightforward application of the UT for state estimation. The main steps of UKF are explained in [5]. We assume that the ECG is stationary over short segments with each one having a various model. The available measurement is yðkÞ which includes additive noise nðkÞ . During the UKF operation, the noisy ECG yk is segmented into non-overleaped and short frames (25 ms).These frames are extracted by using Hamming window which is defined as follows:  wðnÞ ¼ 0:54  0:46 cos

 2pn N1 N1 n ; N1 2 2

ð8Þ

When ECG is a stationary signal and in order to solve this problem, we use the MLPNN for modeling the nonlinear process f ð:Þ. Afterwards we compute the estimation of ^x based on the UKF approach. The connecting weights of the UKF algorithm for MLP training are organized as a state vector. The UKF states are computed by the UT. The training operation includes, (1) assemble the training data, (2) create a design of the network object, (3) the network is trained and simulated his response with new inputs. Thus, we will easily predict the output [14] (Fig. 1).

Time update UKF

Measurement update UKF

Fig. 1. Block diagram of the proposed method

The state evolution and observation equations are given respectively as: Wk þ 1 ¼ Wk

ð9Þ

dk ¼ y k þ nk ¼ f ð W k ; x k Þ þ nk

ð10Þ

The state evolution is a simple an identity matrix. The neural network f ðWk ; xk Þ acts as a nonlinear observation on w. The training of MLPNN with UKF is presented in the following steps.

18

S. Gaamouri et al.

• Initialization: b 0 ¼ E ½W0  W P0 ¼ E



b0 W0  W



ð11Þ

b0 W0  W

T

ð12Þ

• Calculation of the sigma points: b k1 w0 ¼ v0;k1 ¼ W b k1 þ vi;k1 ¼ W b k1  vi þ L;k1 ¼ W

k ð L þ kÞ

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðL þ kÞPk1 wi¼ i

ð13Þ 0:5 ð L þ kÞ

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ðL þ kÞPk1 wi þ L ¼ i

0:5 ð L þ kÞ

ð14Þ ð15Þ

where ¼ a2 ðL þ K Þ  L, i = 1, 2, …, L L is the state dimension. The parameter k is used to control the covariance matrix. • UKF update time equations: vi;knk1 ¼ vi;k1 b k ¼ W P k ¼

X2L i¼0

wi vi;k1

ð16Þ ð17Þ

h ih iT b k vi;knk1 W b k W w v i i;knk1 i¼0

ð18Þ

  Y i;knk1 ¼ f vi;knk1 ; xk

ð19Þ

X2L

^y k ¼

X2L

wi Y i;knk1

ð20Þ

  T Y i;knk1  ^y wi Y i;knk1  ^y þ Rk k k

ð21Þ

i¼0

• UKF measurements update equations: Sk ¼

X2L i¼0

Gk ¼

h i b k Y i;knk1^y W w v i i;knk1 k i¼0

X2L

ð22Þ

Performance Study of Neural Network

19


 bk ¼ W b k þ Gk S1 W dk  ^y k k T  Kk ¼ P k Hk Sk

ð23Þ

T Pk ¼ P k  Kk Sk Kk

ð24Þ

3 UKF Adapted for Colored Measurement Noise The experiments have shown that in fact, the ECG measurement error is not white noise. There are various methods to generate colored noise as well as realistic ECG artifacts. In the case of colored disturbance, the state-space Eqs. (3) and (4) must be modified before Kalman filtering approaches are applied. The observation noise operation is given by state-space formulation as follows: nðkÞ ¼ An nðk1Þ þ Gn vn ðk Þ

ð25Þ

nðkÞ ¼ Hn nðkÞ

ð26Þ

Where nðkÞ is a vector for lagged values of nðkÞ and vn ðkÞ is a white noise described by the covariance matrix Qk which is dependent on the quality of the ECG measurement. An is a simple state transition matrix in a controllable canonical form. Gn and Hn are of the same formas G and H given in (3) and (4). The white noise has a flat spectral density function over all frequencies. On the other hand, realistic noise sources have non-flat spectral densities that reduce in power at bigger frequencies. The relation of spectral density function is defined by [11]: Sð f Þ/

1 fb

ð27Þ

Where f is the sampling and b is a parameter that measure of the noise color, artificial white noise (b = 0) and artificial pink noise (b = 1). When discrimination between whet is signal and what is noise becomes to be very difficult when the additive noise is colored and highly non stationary. For this reason, we consider that the measurement yðkÞ ¼ xðkÞ þ nðkÞ represents the addition of two signals yðkÞ ¼ X1 ðk Þ þ X2 ðk Þ. We therefore formulate state equations containing X1 ðkÞ and X2 ðk Þ: 

  X1 ðk Þ F1 ½X1 ðk  1Þ G1 ¼ þ X2 ðk Þ F2 ½X2 ðk  1Þ 0 yðkÞ ¼ ½ H 1

0 G2

 X ðk Þ H2  1 X 2 ðk Þ



v1 ð k Þ v2 ð k Þ

ð28Þ ð29Þ

20

S. Gaamouri et al.

Where these equations are similar to the colored noise formulations (25) and (26) where n is replaced by X2 and has a corresponding nonlinear model F2 .

4 Results and Discussion The simulation has been successfully implemented in the MATLAB. This section shows the results obtained and comparative analysis between proposed technique and conventional methods. To test the performance of the different approaches, artificial white, colored Gaussian noise and real non-stationary muscle artifact with various variances were produced and added to ECG records. Then the corrupted ECG was presented to the suggested techniques. For the optimization of the consistency of the filters output, the experience was implemented over both the MIT-BIH arrhythmia [15] and MIT-BIH noise stress test databases [16]. In the MLPNN structure, the input layer has a number of neurons equal to the dimension of the data (13 neurons in this paper). The number of neurons of the output layer is equal to the number of classes to be discriminated (1 neuron). The number of hidden neurons depends on the degree of nonlinearity and the dimensionality of original problem (we use 4 neurons). In WD, the reported results are based on the Coiflets3 mother wavelet with 6 levels of decomposition. We chose the length of window = 15 in MF. The frequency band in BF is [0.5 Hz 20 Hz]. We evaluate the performance of several methods using the MSE which is a common evaluation measure for de-noising process and we considered the PSNR in order to calculate the signal peak values like R wave. This detects the heart disease, such as, cardiac arrhythmia. Figure 2 shows the results obtained after denoising the corrupted signal by white Gaussian noise at input the signal-to-noise ratio (SNR = 5 dB) using the UKFMLPNN. It can be seen from a visually comparison of these results that the proposed method has tracked the original signal. The UKF-MLPNN demonstrates the smoothest result, especially around the sharp turning points of the signal (complexes QRS).

0.4

0.2

(b)

-0.2 -0.4

noisy

UKF-MLPNN 1.4

1.2

0.2

(c)

0 -0.2 -0.4

Am plitude(m v)

0.6

0.4

0

1.6

0.8

Original

0.6

A m plitude(m v)

(a)

Am plitude(m v)

0.8

1

0.8

0.6

0.4

0.2

-0.6

-0.6

-0.8

-0.8

0

-1

-1

-0.2

0

100

200

300

400

500

Time(s)

600

700

800

900

1000

0

100

200

300

400

500

Time(s)

600

700

800

900

1000

-0.4

0

100

200

300

400

500

600

700

800

900

1000

Times(s)

Fig. 2. Typical filtering results for an input signal of 5 dB. (a) Original ECG signal. (b) Noisy signal. (c) UKF-MLPNN output.

Figure 3 shows typical filtering results for an input signal of 5 dB. The original signal represented by (a) is 122 records of normal rhythm from the database. Noisy signal obtained by additive artificial pink Gaussian noise is represented in (b). The result of the proposed method is represented in (c). In order to investigate the performance of the proposed method, the MIT-BIH noise stress test database was considered as shown in Fig. 4. The results indicate that UKF-

Performance Study of Neural Network

UKF-MLPNN 1.4

0.6

1.2

(b)

0 -0.2 -0.4

Am plitude(mv)

0.2

0.2

(c)

0

-0.2 -0.4

Amplitude(mv)

0.4

0.4

Am plitude(m v)

noisy

Original

0.6

(a)

1.6

0.8

0.8

1

0.8

0.6

0.4

0.2

-0.6

-0.6

-0.8

-0.8

0

-1

-1

-0.2

0

100

200

300

400

500

600

700

800

900

0

1000

21

100

200

300

400

500

600

700

800

900

-0.4

1000

0

100

200

300

400

Time(s)

Time(s)

500

600

700

800

900

1000

Times(s)

Fig. 3. Removing Artificial colored Gaussian noise. (a) Original signal. (b) Corrupted signal. (c) UKF-MLPNN output.

MLPNN has removed motion artifacts, while preserving diagnostic morphological information of the signal. This may be due to the accuracy of the mean and the covariance estimates of the UKF. 250

100

UKF-MLPNN

muscle noise 200 0

(b)

-200

Amplitude(mv)

(a)

Amplitude(mv)

150

-100

100

50

0

-50

-300

-100

-150 -400

-200

-500

-250 0

100

200

300

400

500

600

700

800

0

100

200

300

400

500

600

700

800

Times(s)

Time(s)

Fig. 4. Removing real colored noise. (a) Record118e12 from the MIT BIH noise stress test database. (b) UKF-MLPNN output.

Tables 1 and 2 demonstrate the performance comparison based on the MSE and PSNR measures between the proposed method and the conventional filters. We confirm that the UKF-MLPNN gives the smaller MSE values and bigger PSNR values than the BF, MF and WD. Table 1. The complete comparative analysis for denoising ECG signal using BF, WF,WD, and UKF-MLP filter on the basis of MSE, PSNR at the non-stationary white and pink Gaussian noises at SNR = 5 dB. Method BF MF WF UKFMLP

Signal database ECG1 ECG2 ECG1 ECG2 ECG1 ECG2 ECG1 ECG2

MSE for white noise 0.122 0.076 0.012 0.011 0.012 0.009 0.002 0.0006

MSE for Pink noise 0.175 0.156 0.134 0.135 0.113 0.112 0.016 0.009

PSNR for white noise 45.16 47.21 55.23 55.61 55.23 55.23 63.01 68.24

PSNR for pink noise 43.59 44.08 44.75 44.72 45.49 45.53 53.98 56.48

22

S. Gaamouri et al.

Table 2. The complete comparative analysis for denoising ECG signal using BF, WD, MF and EK-MLP filter on the basis of MSE, PSNR at the non-stationary muscle noise at SNR = 12 et 18 dB. Data signal

MSE of BF

118e12 118e18

7.005 6.421

MSE of MF 6.403 5.819

MSE of WF 3.721 3.005

MSE of UKF 1.371 0.983

PSNR of BF

PSNR of MF

PSNR of WF

27.57 27.94

27.96 28.37

30.31 31.24

PSNR of UKF 34.65 36.09

5 Conclusion In this paper an UKF to train MLPNN method have been designed for the filtering of ECG signals. The weight relationship and filter coefficients are updated according to multiple input vectors to obtain a more effective result. It was successful in the online estimation and removal of the different noises for real ECG signal. ECG signal removing experiments that were performed show that the proposed method outperformed conventional filters. We hope that this study will be an effective solution for reducing removal noises problems for heart signal. Future work include the test of the proposed algorithm on extraction of the parameters of ECG signal, such us, peak R for classification of heart diseases related to this wave.

References 1. Oliveira, B.R.D., Duarte, M.A.Q., Abreu, C.C.E.D., Vieira Filho, J.: A wavelet-based method for power-line interference removal in ECG signals. Res. Biomed. Eng. 34(1), 73–86 (2018) 2. Hajri, J.B.R., Ghnimi, S., Sboui, N.: Design of SIW iris-coupled-cavity band-pass filter circuit using wave concept iterative process method. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 209–212. IEEE, Tunisia (2016) 3. Chhatrapal, S., Jaspinder, S.: ECG siganl denoising using digital filter and adaptive filter. Int. Res. J. Eng. Technol. 4(6), 2043–2047 (2017) 4. Berrached, N.: Détection et Classification Automatiques d’Arythmies Cardiaques. In: 5th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 1–12. IEEE, Tunisia (2009) 5. Chabchoub, S., Mansouri, S., Salah, R.B.: Impedance cardiography heartbeat classification using LP, DWT, KNN and SVM. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 53–57. IEEE. Tunisia (2016) 6. Derradji, N., Mounir, B.S., Maamar, B.: Multi-sensor data fusion for wheelchair position estimation with unscented Kalman filter. Int. J. Autom. Comput. 15(2), 217–227 (2017) 7. Makrem, B.J., Imen, J., Kaïs, O.: Study of speaker recognition system based on feed forward deep neural networks exploring text-dependent mode. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 355–360. IEEE, Tunisia (2016)

Performance Study of Neural Network

23

8. Moein, S.: An MLP neural network for ECG noise removal on Kalman filter. In: Advances in Computational Biology, pp. 109–116. Springer, New York (2010) 9. Avendano, L.E., Castellanos, C.G., Ferrero, J.M.: Spectrum estimation and adaptive denoising of electrocardiographic signals using Kalman filters. In: Computers in Cardiology, pp. 952–928. IEEE, Colombia (2006) 10. Ahrens, H., Argin, F., Klinkenbusch, L.: Comparison of the extended Kalman filter and the unscented Kalman filter for magnetocardiography activation time imaging. Adv. Radio Sci. 11(k.1), 341–346 (2013) 11. Sameni, R., Shamsollahi, M.B., Jutten, C., Clifford, G.D.: A nonlinear bayesian filtering framwork for ECg denoising. IEEE Trans. Biomed. Eng. 54(12), 2172–2185 (2007) 12. Su, A.W.H.: ECG noise filtering using online model based bayesian filtering techniques. These de maitrise, University of Waterloo (2013) 13. Tamilselvi, R., Beham, M.P., Merline, A., Roomi, S.M.M., Saravanan, B., Ruba, T.: Detection of fetal stress from maternal abdominal electrocardiogram signal. J. Comput. Sci. Eng. 6(4), 65–70 (2018) 14. Kutucu, H., Almryad, A.: An application of artificial neural networks to assessment of the wind energy potential in Libya. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 405–409. IEEE, Tunisia (2016) 15. MIT-BIH arrhythmia data base. http://physionet.org/physiobank/database/mitdb/. Accessed 24 May 1997 16. MIT-BIH noise stress test database. http://www.physionet.org/physiobank/database/nstdb/. Accessed 13 June 2000

Signal Reconstruction Based on the Relationship Between STFT Magnitude and Phase Spectra Raja Abdelmalek(B) , Zied Mnasri, and Faouzi Benzarti Ecole Nationale d’Ing´enieurs de Tunis, Signal, Image and Technology of Information, University Tunis El Manar, Tunis, Tunisia [email protected], [email protected], [email protected]

Abstract. Signal reconstruction by spectrogram inversion from shorttime Fourier transform (STFT) magnitude spectrum has gained renewed interest since a few years. Actually, recent advances in compressive sensing made it possible to recover high quality signals from partial spectral data. In addition, recent theoretic works have revealed novel relationships between STFT magnitude and phase. Therefore, in this paper, a novel algorithm for signal reconstruction, based on the explicit relationship between STFT magnitude and phase is presented in many variants. Objective evaluation using signal-to-error ratio (SERdB ) shows the advantages and the limits of each variant.

Keywords: Short-time Fourier transform (STFT) Signal reconstruction · STFT Magnitude spectrum · Phase retrieval problem Signal-to-error ratio (SERdB )

1

·

Introduction

Signal reconstruction, mainly audio signals, and also images, from partial spectral information, such as the STFT magnitude spectrum only (or the phase spectrum only) with the estimation of the missing data is a research topic which appeared since 1980 [1]. The relevance of this research topic is related to many applications such as (a) Signal encoding and transmission, especially in a realtime context which requires minimal size of data, (b) Understanding the very role of phase and magnitude in signal perception for people having auditive or visual impairments, since experiments have shown that signals with artificially reconstructed phase or magnitude could be perceived almost natural; and (c) Setting the minimal conditions for signal estimation from partial information, since a minimal amount of data or parameters should be available, especially while conducting subjective measures for intelligibility, recognition or identification. Several speech signal reconstruction algorithms use the magnitude spectrum without using any information of the phase spectrum. The information c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 24–36, 2020. https://doi.org/10.1007/978-3-030-21009-0_3

Smart Innovation, Systems and Technologies

25

contained in the STFT magnitude spectrum is important to reconstruct a signal with a good quality regarding naturalness and intelligibility. In addition, in many research, estimating unknown phase spectrum from known magnitude spectrum contributes to significant intelligibility, thus justifying the researches concerning the relationship between the phase and the magnitude spectrum. In 1982, Portnoff proved that there is an explicit relationship between the firstorder derivatives magnitude and phase spectrum under certain conditions [2]. Recently several works have returned to this relationship. In 2013, a complete evaluation of the first-order, the second-order and mixed derivatives of the phase and log of the magnitude of a given STFT with a specific window was proposed by Auger et al. in 2013 [3]. In 2017, a theoretical basis was presented by [4], where a relationship which directly connects the group delay, the instantaneous frequency, and the phase and amplitude spectrum under a particular STFT structure that uses a Gaussian window was determined. These works are in the continuation of a series of signal reconstruction algorithms that consider both phase and magnitude spectrum. Among the iterative algorithms, the first (and the most influential) algorithm for signal reconstruction from (modified) STFT magnitude is the Griffn and Lim algorithm (GL algorithm) [5]. A realtime version of GLA was introduced by Beauregard et al. in [6] (Real Time Spectrogram Inversion Algorithm RTISI). This algorithm is still iterative, but the signal is reconstructed frame by frame such that only a few iterations are necessary in order to get a good result. The second version of this algorithm is Real Time Spectrogram Inversion Algorithm with look ahead (RTISI-LA) [7]. Recently, a real-time algorithm was introduced in [8], called Single Pass Spectrogram Inversion (SPSI). This algorithm is based on a non-iterative method [8]. In this algorithm, the instantaneous frequency is estimated in each frame by peak picking and quadratic interpolation. Another non-iterative algorithm termed Real-Time Phase Gradient Heap Integration (RTPGHI) was proposed in 2017 [9]. RTPGHI is based on the phase-magnitude relationship introduced in [9], which allows estimating the phase increments between neighboring STFT coefficients only from the magnitude spectrum. The aim of the work presented in this paper consists in implementing signal reconstruction using the theoretic relationship between the STFT magnitude and phase recently established in [4]. Then many variants of the implemented algorithm are evaluated (iterative vs. non-iterative). The rest of this paper is organized as follows: Sect. 2 presents the state of the art of signal reconstruction algorithms is reviewed; Sect. 3 explains the theoretic relationship between STFT magnitude and phase is presented; then in Sect. 4, the developed algorithms are described with the yielding results, and finally, in Sect. 5, the findings are thoroughly discussed and commented.

2

Signal Reconstruction Algorithms

The signal reconstruction algorithms problem has started since 1980 in MIT labs by Hayes et al. [1]. Their aim was to determine theoretically the minimal conditions required to reconstruct a signal from partial spectra. Then two theorems appeared:

26

R. Abdelmalek et al.

Theorem of Hayes et al. (1980). The theorem of Hayes et al. developed in 1980 proves the possibility to reconstruct signal from phase spectrum only under the following conditions: A sequence which is known to be zero outside the interval [0, N−1] is uniquely specified to within a scale factor by (N −1) distinct samples of its phase spectrum in the interval 0 < ω < p if it has a z-transform with no zeros on the unit circle or in conjugate reciprocal pairs [1]. Theorem of Van Hove et al. (1983). The theorem of Van Hove et al. appeared in 1983 proves the possibility to reconstruct signal from magnitude spectrum only under the following conditions: Let x(n) and y(n) be two real, causal, and finite extent sequences with ztransforms which have no zeros on the unit circle. If Ax (x : x0 ) = Ay (x : x0 ) for all x then x(n) = y(n). Where  |Ss (ω)| if −ω0 < Φs (ω) < ω0 + π (1) As (ω : ω0 ) = −|Ss (ω)| otherwise and Ss (ω) et Φs (ω) are the magnitude and the phase spectra of the signal s(n), ω = 2πf where f is the frequency. As (ω : ω0 ) is then called the signed spectrum of s(n) [12]. 2.1

Iterative Signal Reconstruction Algorithms

The Griffin and Lim Algorithm (GLA). Griffin and Lim proposed in 1984 the first and probably the most generic signal reconstruction algorithm [5,10]. The problem was formulated such that a time-domain real signal has to be estimated from only a known STFT magnitude spectrum, while the unknown STFT phase spectrum can be initialized to zero or to a random value, following the theorem of Hayes [1]. To solve this problem, the Griffin and Lim algorithm (GLA) consists in minimizing the distance measure between the STFT of the original signal |Xω (mS, ω)| and of the reconstructed signal |Yω (mS, ω)| defined in (2), is minimized at each iteration. +∞ 

1 D= 2π =−∞



m

π

(|Xω (mS, ω)| − |Yω (mS, ω)|)2 dω

(2)

π

where S is the synthesis step size, m is the index of the frame and ω is the Fourier angular frequency. The estimate x(i+1) (n) is computed by inverting the STFT of xi (n) with a modified magnitude to find then a signal with STFT spectrum as close as possible to the original magnitude. The estimated frame x(i+1) (n) is updating using the following function +∞ i+1

x

(n) =

n=−∞

W (ms − n) +∞

π

n=−∞

Xˆωi (mS, n)e−jnω dω ω=−π W 2 (mS − n)

(3)

Smart Innovation, Systems and Technologies

where

X i (mS, ω) Xˆωi (mS, n) = |Yω (mS, ω)| ωi |Xω (mS, ω)|

27

(4)

GLA algorithm generally reaches high quality reconstruction after a large number of iterations, often more than 10. To cope with this problem, Slaney proposed an approach to reconstruct audio signal from cochleagrams and correlograms based on the GLA technique [11]. Nevertheless. RTISI and RTISI-LA Algorithms. In Griffin and Lim algorithm, the process consists in reconstructing the whole signal for each iteration, which is computationally very expensive [5]. Beauregard et al. proposed two real-time implementations based on the idea of GLA algorithm: The real time iterative spectrogram inversion (RTISI) in 2005 [6] and the real time iterative spectrogram inversion with look-ahead RTISI-LA in 2007 [7]. First the baseline algorithm RTISI is based on reconstructing the signal according to the time sequential signal (i.e. frame by frame) in order to keep the number of iteration as minimum as possible. This technique allows reconstructing a frame at each iteration, following these steps: (a) Consider that the first (m−1) frames are generated and denoted ym−1 (n), the mth frame is reconstructed from the overlap-added results of the estimated frames of y(n): (m − 1), (m − 2) and (m − 3), while the fourth quarter of the partial frame m is initialized all zero using a synthesis window overlap equal to 3/4. (b) The iteration functions cf. (3) and (4) are updated as in Griffin and Lim algorithm which is restrained to the current frame m only at each step, instead of updating it for the whole signal. The first frame of the constructed signal is estimated by combining a zero initial phase and the target magnitude spectrum. Then, the RTISI algorithm continues with the successive frames by combining the frame m with the partial frame ym−1 (n)w(n − mS) until reconstructing the whole signal [6]. Therefore, a second process RTISI with Look-ahead [7] was developed in 2007. In the previous method (RTISI), only the previous frame are considered to estimate the frame m. In contrast, the RTISI-LA method performs phase estimation of RTISI algorithm on k frames after the current frame so the estimation of the frame m is influenced only by the futures frames m to m+k−1 [7]. To overcome the delay due to the iterative implementations, which is not appropriate to real-time applications, some non-iterative signal reconstruction algorithms were developed. 2.2

Non Iterative Signal Reconstruction Algorithms

Single Pass Spectrogram Inversion Algorithm (SPSI). Beauregard and al. proposed an algorithm which reconstructs a signal in a single iteration in 2015

28

R. Abdelmalek et al.

[8]. The SPSI algorithm employs peak picking followed by quadratic interpolation of the magnitude spectrum in order to identify the instantaneous frequency [8]. SPSI algorithm proceeds as follows for each frame: (a) The successive peaks of the STFT magnitude spectrum |X(mS, ωj−1 )|, |X(mS, ωj )| and |X(mS, ωj+1 )| are considered, where m is the frame index, S is the synthesis step size, wj = 2π.j N represents the peak’s angular frequency and N is the number of points of the discrete Fourier transform. (b) The true values of the magnitude spectrum bins are identified using a quadratic interpolation defined in (5) α−γ 2(α − 2β + γ)

(5)

α = |X(mS, ωj−1 )|

(6)

β = |X(mS, ωj )|

(7)

γ = |X(mS, ωj+1 )|

(8)

p= where

The value of p is the deviation between the true and the estimated peaks as a proportion of the bin size. it should be within the range [−0.5, 0.5]. (c) The true value of the frequency of the peak is calculated using (9) ωj =

2π(j + p) N

(9)

where j is the position of the peak bin. (d) The adjusted phase of the peak is calculated and the accumulated phase corresponding to the bin j is given in (10) φm,j = φm−1,j + Sω,j .

(10)

(e) The remaining bins phase values are identified according to the sign of p. (f) The reconstruction algorithm continues with the successive frame by combining the phase obtained so forth with the original magnitude until reconstructing the whole signal [8]. Real-Time Spectrogram Inversion Using Phase Gradient Heap Integration Algorithm (RTPGHI). The goal of this algorithm is to estimate the unknown STFT phase spectrum from the known STFT magintude spectrum. The algorithm process is performed in two steps [9]. (a) using a numerical differentiation to obtain the phase gradient. Then, an approximation of the STFT phase gradient ∇φ = (φω (m, n), φt (m, n) is calculated using a centered difference scheme defined in (11) and (12). δ (Alog (m, n + 1) − Alog (m, n − 1)) φω (m, n) = − 2aM

(11)

Smart Innovation, Systems and Technologies

29

aM 2π.am (Alog (m + 1, n) − Alog (m − 1, n)) + (12) φt (m, n) = 2δ M Where Alog (m, n) = log(A(m, n)), A(m, n) is the STFT magnitude spectrum, δ is the window’s time frequency ratio, a is the time step samples and M is the number of frequency channels. (b) According to the coefficients magnitude, in this step, the numerical gradient integration is done using a cumulative sum. First, the process consists in marking coefficients from the frame n as unknown. Secondly, it continues by inserting into the heap the coefficients from the frame n − 1 which was making all potential initial points. Then, to start the integration, the biggest coefficient is removed from the heap in order to be used to spread the phase to the only neighbor in the frame n. Finally, the process carry on until no coefficients having unknown phase are left in the heap. The ˆ signal sˆ is reconstructed by combining the estimate phase φ(m, n) with the origˆ φ(m,n) ˆ using the overlap-and-add inal magnitude such that S(m, n) = A(m, n)e method with a dual window such us gabor window [9].

3

Explicit Relationship Between Magnitude and Phase of the STFT

Since the start of using STFT as an analysis tool, many research have tried to investigate how the magnitude and phase of the STFT are linked together. However, a perfect estimation of one spectrum requires an exact knowledge of the other one, which presents several difficulties in most cases, as will be mentioned afterward. In [2], the STFT could be considered as the output of a filter bank spectrum analyser (cf. (13)). 

+∞

X(t, ω) = −∞

w(t − u)x(u)e−juω du

(13)

where X(t, ω) can be described by X(t, ω) = A(t, ω)ejφ(t,ω)

(14)

where A(t, ω) and φ(t, ω) present respectively the magnitude and phase spectra using a Gaussian analysis window defined by t2

w(t, ω) = e− 2σ2

(15)

where σ is the standard deviation. Then, the first order derivatives of STFT magnitude and phase of X(t, ω) along time and angular frequency can be obtained by the system of Eqs. (15) and (16). ∂log(A(t, ω)) 1 ∂(φ(t, ω)) T + 2 =− 2 (16) ∂t σ ∂ω σ

30

R. Abdelmalek et al.

1 ∂log(A(t, ω) ∂φ(t, ω) − 2 =ω (17) ∂t σ ∂ω Using a definition of the group delay (GD) and the instantaneous frequency (IF) given by ∂φ(t, w) GD(t, ω) = − (18) ∂ω 1 ∂φ(t, w) IF (t, ω) = (19) 2π ∂t The Eqs. (15) and (16) were reformulated in [4] to obtain the Eqs. (20) and (21). ∂A(t, ω) T GD(t, ω) = σ 2 + (20) ∂t 2 1 ∂A(t, ω) ω IF (t, ω) = + (21) 2 2πσ ∂ω 2π Equations (19) and (20) show that group delay and instantaneous frequency, which are both related to the phase by definition, are also linked to the magnitude [4]. So, the STFT phase φ(t, ω) can be obtained from the known STFT magnitude using either the group delay related equation i.e (19) or the instantaneous frequency related one, i.e (20), respectively.  Tω ∂log(A(t, ω)) dω − + C1 (22) φ(t, w) = −σ 2 ∂t 2  ∂log(A(t, ω)) 1 dt + ωt + C2 (23) φ(t, w) = 2 σ ∂ω To confirm the above relationships in practise, since our foal is to reconstruct signal by estimate the phase spectrum from the magnitude A(t, ω) using STFT, the discrete time version of (21) and (22) are necessary. Then, the discrete time delay (DGD) and the discrete instantaneous frequency (DIF) can be calculated by (23) and (24) respectively, as mentioned in [4]. An+1,k + δ N σ 2 fs log + 2 An−1,k + δ 2fs

(24)

kfs An,k+1 + δ N + log 8π 2 σ 2 fs An,k−1 + δ N

(25)

DGD(n, k) = DIF (n, k) =

ω.N Where n = t.fs is the discrete time index, k = 2pif is the frequency bin, s δ is a small positive constant used for numerical stability and N = T.fs is the length of the analysis window. Then a recursive solution of (21) and (22) can be described by (25) and (26)

An+1,k An+1,k−1 + δ πσ 2 fs2 log −π φˆn,k = φˆn,k−1 − 2N An−1,k An−1,k−1 + δ

(26)

2πk N An,k+1 An−1,k+1 + δ + (27) log 8π 2 σ 2 fs An,k−1 An−1,k−1 + δ N ˆ 0) = 0 or π and Where the initial phase values in case of real signals are φ(n, ˆ φ(0, k) = 0 [4]. φˆn,k = φˆn−1,k +

Smart Innovation, Systems and Technologies

4

31

Experiments

In this section, we propose an algorithm based on the relationship between phase and magnitude spectra, in different variants. First, the proposed algorithm is described and then evaluated for different values of number of iterations to compare the four variants of our algorithm in order to determine which one give the best quality. 4.1

Proposed Algorithm

In this section, the different variants of our signal reconstruction algorithm which estimates phase from the original STFT magnitude spectrum are presented. The relationship between STFT magnitude and phase spectra were already published in [2–4] in slightly different forms obtained using different techniques. Actually, in the above mentioned references, the authors use different STFT phase conventions so that the explicit equations relating phase and magnitude spectra differ. In this paper, the phase spectrum is estimated using (22) from the original magnitude spectrum in many variants. The goal of this algorithm is to estimate the phase spectrum from the STFT magnitude spectrum only, and then to reconstruct each frame by the inverse STFT. Four variants of the implemented algorithm are described below and presented in Fig. 1. Variant V1. The first frame is reconstruct by combining the original phase with a phase equal to zero.  An,1 = Aoriginal k=1 (28) φˆn,1 = 0 At the first iteration, the estimated phase φˆn,k is calculated using (24), then, (i) at each iteration, the phase estimate φˆn,k is updated using the same RTISI algorithm technique [7]. Variant V2. Assuming that the phase spectrum until the frame (n − 1) is reconstructed, the frame S(n − 1, :) is shifted, with an appropriate overlap rate. Initially the new samples introduced by the overlap are set to zero. Then the initial (i) phase φˆn,k is estimated using (24). Finally φˆn,k is updated similarly to variant V1. Variant V3. In this variant, the initial phase is calculated like in variant V1, i.e. (i) without overlap. However, the phase estimate φˆn,k is updated at each iteration using (24), in opposition to variants V1 and V2 where (24) is used only once. Variant V4. This variant is similar to variant V2, as it starts by shifting the (i) current frame, and to variant V3 by updating φˆn,k by using (24) at each iteration. 4.2

Objective Evaluation

Test Databases. To assess the quality of the reconstruction algorithm and the efficiency of each variant, audio signals consisting in speech (monophonic),

32

R. Abdelmalek et al.

Fig. 1. The proposed phase estimation algorithm with different variants

Smart Innovation, Systems and Technologies

33

and music (polyphonic) were used. To configure the test sample set, 90 audio segments are chosen from three different databases to cover different aspects of music and speech signal such as duration, sampling frequency, intonation and encoding rate. The test sample set contains, as mentioned in Table 1, 60 Arabic speech from [13] and [14], and 30 classical music segments. Error Measure. Generally, in reconstruction signal problems, the objective evaluation consists in calculating the signal-to-error ratio in dB (SERdB ) [5], as defined by (26) between the original and the reconstructed signals. π +∞ 1 2 m=−∞ 2π ω=−π |X(mS, ω)| dω SERdB = 10log10 +∞ (29)  π 1 ˆ (|X(mS, ω)| − |X(mS, ω)|)2 dω m=−∞ 2π

ω=−π

ˆ where |X(mS, ω)| and |X(mS, ω)| are the STFT magnitude spectra of the original signal and the reconstructed signal, respectively. Table 1. Test databases characteristics Database

Type

Sampling frequency Mean duration Number of samples

Database1 Speech 16 KHz

3s

30

Database2 Speech 48 KHz

15 s

30

Database3 Music

15 s

30

44.1 KHz

Evaluation Results. In this section, the performance of our different implementation versions of the proposed algorithm based on relationship between phase and magnitude is evaluated for different values of number of iterations. Many tests were conducted, before choosing the following optimal parameters values: Gaussian window with σ = 2.5 ms, frame length N = 512 and shift rate equal to 1/2. The SERdB results are shown in Table 2. In Table 2, it looks obvious that variants V2 and V4 perform much better than variants V1 and V3. Actually both rely on starting by overlapping the previous frame, before estimating the phase. Figure 2 confirms these results and shows clearly that both variants reach their best performance after 5 iterations, which is relatively a small number in comparison to other iterative signal reconstruction algorithms, such as RTISI and RTISI-LA [7]. However, updating the phase estimate φˆn,k at each iteration using (24), as in variant V4 gives almost the same results as calculating it once, as in variant V2. This means that the first estimate of φˆn,k using (24) is enough to reach the best performance, leading to a smaller computational load.

34

R. Abdelmalek et al. Table 2. Signal-to-error ratio (SERdB ) measures

Database

Variant 1 iterations 2 iterations 3 iterations 5 iterations 10 iterations

Database1 V1 V2 V3 V4

17.56 30.49 17.27 22.38

18.35 33.35 19.29 29.44

19.65 41.19 20.16 36.04

20.82 52.90 21.06 51.86

21.08 52.78 21.92 52.75

Database2 V1 V2 V3 V4

17.65 30.55 16.43 23.42

18.65 33.83 18.03 28.68

20.39 40.67 19.25 37.2

22.54 44.99 21.19 40.29

24.04 44.45 23.80 43.6

Database3 V1 V2 V3 V4

17.49 31.04 14.17 25.21

18.32 35.68 17.95 28.55

18.68 40.88 18.56 37.27

19.16 43.94 19.04 43.05

21.20 41.79 20.20 41.56

Fig. 2. Signal-to-error ratio (SERdB ) measures

5

Conclusion

Signal reconstruction using phase estimation from the STFT magnitude spectrum was studied in this paper. Phase estimation was achieved based on the

Smart Innovation, Systems and Technologies

35

theoretic relationships between STFT magnitude and phase spectra, revealed by the works of [2] and [4]. Then a novel algorithm was proposed, in four variants. The variants differ in (a) whether using an initial estimate of the current frame by overlapping the previous one or not, and (b) whether updating the phase estimate using the explicit relationship between phase and magnitude at each iteration or only at the first one. Three audio signals databases, covering monophonic speech and polyphonic music were used for objective evaluation, where the signal-to-error ratio (SERdB ) was calculated. The collected results show that to reach a high quality reconstruction (SERdB > 40dB), overlap is required but phase updating using the relationship with magnitude spectrum is required only at the beginning. As an outlook, first subjective tests (MOS and DMOS) will be carried out to assess the quality of the reconstructed signals, and secondly non-iterative variants will be investigated.

References 1. Hayes, M.H., Lim, J.S., Oppenheim, A.V.: Signal reconstruction from phase or magnitude. IEEE Trans. Acoust. Speech Signal Process. 28(6), 672–680 (1980) 2. Portnoff, M.R.: Magnitude-phase relationships for short-time Fourier transforms based on Gaussian analysis windows. In: Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP), vol. 4, pp. 186–189 (1979) 3. Auger, F., Chassande-Mottin, E., Flandrin, P.: On phase magnitude relationships in the short-time Fourier transform. IEEE Signal Process. Lett. 19(5), 267–270 (2012) 4. Shimauchi, S., Kudo, S., Koizumi, Y., Furuya, K.: On relationships between amplitude and phase of short-time Fourier transform. In: Proceedings of the IEEE Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 676–680 (2017) 5. Griffin, D.W., Lim, J.S.: Signal estimation from modified short-time Fourier transform. IEEE Trans. Acoust. Speech Signal Process. 32(2), 236–243 (1984) 6. Beauregard, G.T., Zhu, X., Wyse, L.: An efficient algorithm for real-time spectrogram inversion. In: Proceedings of the 8th International Conference on Digital Audio Effects (DAFx), pp. 116–118 (2005) 7. Zhu, X., Beauregard, G.T., Wyse, L.: Real-time signal estimation from modified short-time Fourier transform magnitude spectra. IEEE Trans. Audio Speech Lang. Process. 15(5), 1645–1653 (2007) 8. Beauregard, G.T., Harish, M., Wyse, L.: Single pass spectrogram inversion. In: Proceedings of the IEEE International Conference on Digital Signal Processing (DSP), pp. 427–431 (2015) 9. Zdenek, P., Sondergaard, P.L.: Real-time spectrogram inversion using phase gradient heap integration. In: Proceedings of the International Conference on Digital Audio Effects (DAFx), pp. 17–21 (2016) 10. Nawab, S.H., Quatieri, T.F., Lim, J.S.: Signal reconstruction from short-time Fourier transform magnitude. IEEE Trans. Acoust. Speech Signal Process. 31(4), 986–998 (1983) 11. Slaney, M., Naar, D., Lyon, R.F.: Auditory model inversion for sound separation. In: IEEE International Conference on Acoustics, Speech, and Signal Processing, vol. 2, pp. 77–80 (1994)

36

R. Abdelmalek et al.

12. Van Hove, P.L., Hayes, M.H., Lim, J.S., Oppenheim, A.V.: Signal reconstruction from signed Fourier transform magnitude. IEEE Trans. Acoust. Speech Signal Process. 31(5), 1286–1293 (1983) 13. Boudraa, M., Boudraa, B., Guerin, B.: Elaboration d’une base de donn´ees Arabes phon´etiquement ´equilibr´ee. In: Actes du colloque Langue Arabe et technologies in-formatiques avanc´ees, Casablanca, pp. 171–187 (1993) 14. Halabi, N., Wald, W.: Phonetic inventory for an Arabic speech corpus. In: Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC), Slovenia, pp. 734–738 (2016)

Real-Time Implementation of an Optimized Speech Compression System in STM32F4 Discovery Board Souha Bousselmi(B) , Safa Saoud, and Adnen Cherif Laboratory of Analysis and Processing of Signals, and Electric and Energy System, Science Faculty of Tunis, University of Tunis El-Manar, 1060 Tunis, Tunisia [email protected]

Abstract. This paper presents an optimization, and a real-time implementation of a wavelet based speech compression system in STM32F4 discovery card. The optimization is done on the one hand by considering a Voice Activity Detection (VAD) to reduce the complexity and on the other hand by using a new quantization approach that codes each sample with fewer bits. The performance of the embedded audio codec is evaluated with a test technique called Processor-in-the-Loop (PIL) and using objective measures that can predict the perceived quality of the signal, namely SNR, PSNR and MSE. The compression efficiency is measured with the compression factor (CR). This research highlights the importance of the proposed optimizations. Indeed, they increase the CR without damaging the voice quality. The practical study shows that the proposed system meets the temporal and material requirements. Voice clarity is assessed with the Mean Opinion Score (MOS). Keywords: Discrete wavelet transform (DWT) · Speech compression Voice Activity Detection (VAD) · STM32F4 Discovery board · Processor in the loop (PIL)

1

·

Introduction

Audio compression is a growing field of research. The goal of any compression technique is to reduce transmission and storage costs by reducing signal redundancy and eliminating spectral and temporal characteristics that are perceptually insignificant. A compression system is always the result of a compromise between four main criteria of compression: quality of the reconstructed signal, compression ratio, complexity and delay. A good compression system permits to obtain the best quality of the reconstructed signal with the best compression ratio, the least complexity and the lowest delay. Signal compression occurs in several applications such as fixed and mobile digital telephony transmission, packet network (internet) transmission, video conferencing, radio and television, as well as various storage applications. In the literature, audio compression methods are c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 37–48, 2020. https://doi.org/10.1007/978-3-030-21009-0_4

38

S. Bousselmi et al.

classified according to the quality of the reconstituted signal: lossless compression and lossy compression. The first one restores the signal perfectly; however, it can not achieve a high compression rate such as RLE coding, Huffman coding. The other one loses some information, but it provides a height compression ratio according to the desired quality [1]. Usually, the most commonly used audio compression algorithms fuse these two kinds to increase the compression ratio as much as possible. There are a wide variety of audio compression techniques. The majority of compression techniques use time-frequency representations such as wavelet transform, wavelet packet transform. The discrete wavelet transform is a robust signal processing tool that allows multi-resolution analysis of speech signals. The importance of this transformation in the design of an audio compression system lies in the fact that it concentrates the energy in a few coefficients. This transformation is defined by a pair of quadrature mirror filters. The choice of these filters and their orders is of major importance in audio compression. Wavelet-based compression provides greater analytical finesse and adapts to the properties of speech signals. It reduces the transmission rate by eliminating redundancy with an orthogonal wavelet. The decomposition of a signal with wavelets is important, however, the application of DWT for silence frames is not necessary. So, we need to differentiate between active frame and silent frames by using the voice activity detection (VAD). This last is among the methods used in speech signal codecs. It is intended to reduce the amount of data to be transmitted and to detect periods of silence. Indeed, in a telephone conversation the speaker speaks only 1/3 of the time on average. That makes 1/3 of the time of a conversation consist of silence easily reproducible and therefore not encoded by the codec. The principle of detecting the speech activity of an audio or speech signal consists in differentiating the frames that contain voice to those that do not contain. In this work we have used the voice activity detection defined in G729. the size of an audio file relies upon the number of samples and the number of bits used to code each sample, so reducing the number of bits to encode each sample reduces the size of the audio signal. This led us to think of a new quantization approach that encodes each sample with 8 bits instead of 16. Many works deals with the implementation of audio codec based on wavelets transform in DSP board [7,8]. However, the use of DSP boards makes the application a bit expensive. This leads us to look for other cheaper solutions based on microcontrollers such as the STM32F4 card developed by ST Microelectronics. This development board is effective for audio applications as it is proven in the following works [2–4]. Indeed, it combines several features such as performance, real-time, low voltage and low power consumption as well as full integration of necessary audio devices such as microphone and audio output. In this context, we tested the feasibility and evaluated the performance of the proposed audio codec in a STM32F4 Discovery development board marketed under the name “STM32f407G-DISC”. Following this introduction the paper is organized as follows; the next section provides a description of each step leading to the development of the proposed algorithm. It is also presents the criteria considered to assess the quality of this algorithm. Section 4 exposes and discusses the

Real-Time Implementation of an Optimized Speech Compression System

39

experimental results. The different steps leading to the real-time implementation are explained and evaluated in the Sect. 5. Finally, Sect. 6 provides a conclusion of this work.

2

Optimized Speech Compression Algorithm Using DWT

The optimized audio compression system is illustrated by the following flowchart, the different steps are detailed in the following paragraphs (Fig. 1).

Fig. 1. Flow chart of the proposed algorithm

2.1

Voice Activity Detection

Various algorithms of voice activity detection where evoked in literature such as the algorithm of Y. Ephram and D. Mallah. The main objective of VAD is to determinate whether an audio frame contains speech (Active frame)or no, in order to reduce the transmission rate during silents frames. In this work, we have considered the VAD algorithm described under the G729 recommendation [6]. Figure 2 presents an example of VAD for sx22.wav extracted from TIMIT database. 2.2

Discrete Wavelets Transform

The second step of the proposed approach is to decompose the active frame using DWT. This later converts the temporel representation of an input signal into a time-frequency. The advantage of this transformation is to reduce redundancy. The DWT concentrates the signal energy into a few coefficients. Consequently, many coefficients will be either zero or have negligible magnitudes. this work is yield with wavelets ‘db10’ and we have decomposed the signal at level 5.

40

S. Bousselmi et al.

Fig. 2. VAD of sx22.wav according to G729

2.3

Thresholding

Thresholding consists of rejecting the DWT coefficients below a given threshold. This step is fundamental in transform-based compression; in fact, it makes the signal parsimonious. The most well known thresholding methods were presented by Donoho [36]. These are soft and hard thresholding. In this work, we have exploited the hard thresholding presented in this equation:  CRe if |CRe | ≥ T (1) CRe = 0 otherwise 2.4

New Quantization Approach

The principle of this approach is to quantify each sample with 8 bits instead of 16 bits in order to reduce the compression ratio. Indeed, the size of a file depends on the number of samples and the number of bits used to quantify each sample. For our simulations, all the signals are extracted from the TIMIT database, so the original samples are quantized to 16 bits. This strategy consists in converting each number to an integer that can be encoded on 8bit so it must be between −128 and 127. For example, x = 0.2350 the value will be 23 using the following formula: (2) Xquant = f ix(x(n) ∗ 102 ) With Fix: Function that rounds the elements of X to nearest integers to zero. The reconstitution of each value is performed by applying the following equation: Xrec = Xquant ∗ 10−2 2.5

(3)

ZRLE Coding

ZRLE is a variant of RLE used when the most frequent symbol is zero. This technique encodes a series of zeros using two values. The first one indicates the beginning of the sequence while the second one specify the number of occurrences of zeros. For example, the series 1, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0 would be coded as follows: 1, 0, 4, 2, 0, 6.

Real-Time Implementation of an Optimized Speech Compression System

3

41

Evaluation Criteria

To evaluate the effectiveness of the proposed algorithm, we used objective and subjective metrics. Objective metrics are purely mathematical parameters that have not need much material and time. All the standard metrics have been used are defined below. – Compression ratio (CR)

C=

length(x(n)) length(c(n))

(4)

c(n), presents the compressed signal. – Signal to noise ratio (SNR) SN R = 10 log10 (

δx2 ) δe2

(5)

where: δx2 is the mean square of the original signal. δe2 is the mean square difference between the original and reconstructed audio signals. – Peak Signal to noise ratio (PSNR) P SN R = 10 log10

N X2

(6)

2

|x − r|

N , is the number of samples of the reconstructed signal. X, is the maximum 2 absolute square value of the signal. |x − r| is the energy of error between the reconstructed and original signal. – Normalized root mean square error (NRMSE)  N RM SE =

2

(x(n) − r(n)) (x(n) − μx (n))2

(7)

x(n), is the original signal, r(n) is the reconstructed signal, and μx (n))2 is the mean of the original signal. – Absolute Category Rating Numerous subjective tests are used in the literature. The most used is the Absolute Category Ranting (ACR), in which a number of listeners judge the quality of the reconstructed signal according to a rating scale. ACR is often used in ITU-T applications such as G711, G728. The average numeric score over the different experiments gives a mean Opinion score (MOS). The correspondence among the scores and the distinctive judgments are provided in the table below.

42

S. Bousselmi et al.

Table 1. Rating scale and descriptions for the absolute category-rating (ACR) test Rating ACR description-MOS

4

5

Excellent

4

Good

3

Fair

2

poor

1

Bad

Test and Results

In this phase, a MATLAB software was developed to implement the optimized audio compression algorithm based on DWT. In order to examine the performance of this algorithm, a comparative study between the wavelet-based compression and the proposed system is done. This comparison is based on objective criteria that we have previously introduced. In all experiments, we have examined signals obtained from the TIMIT database [5]. Throughout the Table 2, it is clear that the proposed algorithm (ODWT) gives a better rate than those obtained by DWT without optimisation. The obtained results affirm the superiority of the proposed algorithm. Indeed, the compression factor increased from one to four without sacrificing the speech quality. The comparison of the quality obtained by the two algorithms confirmed that the criteria (SNR, PSNR and NRMSE) are maintained. This is explained by the fact that the proposed system increases the compression ratio without affecting the quality of the signal. Table 2. Performance evaluation of the optimized algorithm Trame Algorithm sx19.wav sx29.wav Cr SNR PSNR NRMSE Cr SNR PSNR NRMSE 64

DWT ODWT

0,92 18,49 39,59 2,02 19,81 40,90

0,11 0,10

0,86 4,79

18,49 45,55 18,35 45,41

0,11 0,12

128

DWT ODWT

1,59 18,67 39,77 3,58 19,01 40,10

0,11 0,11

1,36 6,61

19,67 46,73 19,49 46,55

0,10 0,10

512

DWT ODWT

2,37 18,80 39,89 6,33 19,39 40,48

0,11 0,10

2,34 8,59

20,08 47,13 19,83 46,88

0,09 0,10

1024

DWT ODWT

4,18 18,61 39,66 7,05 20,11 41,16

0,11 0,09

2,62 20,90 47,95 10,90 18,86 45,91

0,09 0,11

2048

DWT ODWT

4,53 19,76 40,82 8,18 20,91 41,96

0,10 0,09

6,11 20,36 47,41 13,19 20,71 47,76

0,09 0,09

Real-Time Implementation of an Optimized Speech Compression System

5

43

Real-Time Implementation in STM32f4

Real-time assessments have an exceptional importance, particularly for audio applications which have time constraints such as streaming audio and Voice over IP (VoIP). In these applications both input signal and output signal are processed simultaneously, this is explained by the fact that the average processing time per sample plus the overhead is less than the sampling period. In this context, we have examined our algorithm on the STM32F4 microcontroller. 5.1

STM32F4 Discovery Board Overview

The STM32F4 Discovery board includes an STM32F407 Microcontroller based on the ARM Cortex-M4F 32 bits core which operates with a clock frequency up to 168 MHz. It has a DSP and it features a floating point instructions. It includes high-speed memories (Flash memory up to one Mbyte, as much as 192 Kbytes of SRAM). This development board is appropriate for audio programs. In fact, it is equipped with a microphone ST MEMS MP45DT02 and a digitalto-analog conversion (DAC) with integrated class D speaker driver. As indicated in the figure bellow this board includes LEDs, push buttons and a USB OTG micro-AB connector. The Fig. 3 provides an overview of the STM32F4 Discovery board.

Fig. 3. STM32F4 Discovery board

5.2

Rapid Prototyping Technology

The STM32F4 board is supported by several development environments such as IAR, Keil, Coocox which are reliable and satisfying to make embedded applications. However, they require prior knowledge and take a relatively long time to implement such an application. In order to avoid all this and to facilitate the implementation of the proposed algorithm we have used an interface

44

S. Bousselmi et al.

between Simulink and the STM32F4 board. To make this connection we have installed two toolboxes “Embedded Coder Support Package for STMicroelectronics STM32F4-Discovery Board” and “ARM-Cortex M4”. There are three steps to implementing a Simulink model on the board: first, the Embedded Coder Support Package for STMicroelectronics STM32F4-Discovery Board Toolbox is used to interact with the ports and devices on the board; Then the Embedded Coder automatically converts the Simulink model into a C/C++ code that will be compiled by “GNU TOOLCHAIN” to result a binary code. The latter will be flashed in the memory of the microcontroller through the tool “ST LINK Utility”. 5.3

Real-Time Implementation of a Wavelet Codec on the STM32F4 Board

In this section, we propose the implementation of an optimized audio compression algorithm on the STM32F4 board. To do this, we exploited the rapid prototyping described in the previous paragraph. First, we designed our algorithm under Simulink and configured its code generation parameters to use the ‘ert.tlc’ file and the STM32F4-Discovery as a target board. This configuration is illustrated in Fig. 4. Subsequently, the embedded coder automatically generates a C language project that includes all necessary codes and libraries. This strategy greatly reduces the difficulty of implementing audio codecs under the STM32F4 board. Figure 5 presents the Simulink model of the optimised audio codec based on DWT in STM32F4.

Fig. 4. Configuration parameters for Simulink model

Real-Time Implementation of an Optimized Speech Compression System

45

Fig. 5. Simulink model of audio compression system in STM32F4

Initially, the analog signal captured by the microphone is converted into a digital signal through the analog-to-digital converter (ADC). After the compression and decompression processes, the resulting digital signal is converted into an analog signal via digital-to-analog converter (DAC), finally sent to the audio output. The “Mic IN” block is used to acquire the audio signal of the MEMS microphone integrated in the STM32F4Discovery card. The MEMS microphone produces a pulse modulated signal (PCM) of type int 16. The Mic IN block uses double buffering to read the audio data from the MEMS microphone. The double buffering implements two buffers: one is used by the DMA to read the audio signal from the MEMS microphone while the other is used to send audio frames to the audio processing algorithm. This prevents the buffer from being overwritten while it is being used by the audio algorithm. The “audio.out” block writes the processed audio data to the audio device connected to the processor on the STM32F4 board. This block also uses the double-buffering technique to avoid overwriting when sending data to the device. To design the Simulink model of the algorithm presented in Fig. 5, we have used predefined blocks in the “STM32F4 Embedded Encoder Support Package for STMelectronics STM32F4” dedicated to the STM32F4 development board and the Simulink model based on block “embedded Matlab function”. Figure 6 shows some predefined bolcks for the STM32F4 development board. For a purpose of justifying the performances of the proposed algorithm, we have used a very interesting test technique proposed by Mathworks called “Processor in the loop”. This later tests the code executed by the real target board in the virtual environment Simulink (Co-Simulation). The PIL model of the audio codec is shown in Fig. 7. The model has the following blocks: - Subsystem: contains the different steps of the Simulink model

46

S. Bousselmi et al.

Fig. 6. Simulink library dedicated to the STM32F4 card

Fig. 7. Simulink model for PIL simulation of codec based on DWT.

of the DWT compression algorithm. - Subsystem 1: contains the PIL model. Scope 1: displays the temporal variation of the original signal before any treatment. - Scope: displays the reconstructed signal generated by Simulink. - scope 2: displays the reconstructed signal generated by the STM32F4 board. Figure 8 shows on the one hand that the signal reconstructed by the Simulink model is identical to that generated by the STM32F4 board and on the other hand that the compression using the proposed codec on the STM32F4 board did not visually degrade the reconstructed signal.

Real-Time Implementation of an Optimized Speech Compression System

47

Fig. 8. Time variation before and after compression.

For an accurate evaluation of the reconstructed signals after compression/decompression on the STM32F4 board, a listening test is necessary. For this we have done the ACR test wherein 10 listeners listens to audio sequences and judges the quality perceived according to a rating scale defined in the Table 1. The average of the listening test received is 3.8.

6

Conclusion

In this paper, an optimized speech compression algorithm using DWT with VAD and a new quantization approach has been yield. The proposed algorithm proves its reliability to increase the compression ratio without affecting the voice quality according to the performance assessments released using the objective criteria. In this respect, a comparative study with the wavelet based compression demonstrate the ability and the effectiveness of our algorithm. In fact, it increases the compression factor from 1 to four without losing neither the speech intelligibility nor the quality. Finally, the real-time test of speech compression codec has been efficiently implemented in STM32F4.

References 1. Sayood, K.: Introduction to Data Compression, 3rd edn. Morgan Kaufmann, Burlington (2012) 2. Saggese, A., et al.: Time-frequency analysis for audio event detection in real scenarios. In: 2016 13th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). IEEE (2016)

48

S. Bousselmi et al.

3. Zhou, J., Zhang, X., Dong, J.: Design of multichannel audio power amplifier monitoring and protection system. Inf. Technol. 9, 037 (2016) 4. Bu´s, S., J¸edrzejewski, K.: Real-time pitch shifting using a general purpose microcontroller. In: Photonics Applications in Astronomy, Communications, Industry, and High Energy Physics Experiments 2017, vol. 10445. International Society for Optics and Photonics (2017) 5. Zue, V., Seneff, S., Glass, J.: Speech database development at MIT: TIMIT and beyond. Speech Commun. 9(4), 351–356 (1990) 6. Benyassine, A., et al.: ITU-T Recommendation G. 729 Annex B: a silence compression scheme for use with G. 729 optimized for V. 70 digital simultaneous voice and data applications. IEEE Commun. Mag. 35(9), 64–73 (1997) 7. Aloui, N., Bousselmi, S., Cherif, A.: Optimized speech compression algorithm based on wavelets techniques and its real time implementation on DSP. Int. J. Inf. Technol. Comput. Sci. (IJITCS) 7(3), 33–41 (2015) 8. Wang, F., Zhou, Q., Cai, K., et al.: Implementation of G. 729 ADPCM on DSP. J.-Sichuan Univ. Nat. Sci. Ed. 44(4), 785 (2007)

Degradation Process Analysis and Remaining Useful Life Estimation in a Control System Nabila Mabrouk1,4(&), Med Hedi Moulahi2,4, and Fayçal Ben Hmida3,4 1

2

Institut Superieur des Etudes Technologiques de Charguia II, 2035 Ariana, Tunisia [email protected] Institut Superieur des Etudes Technologiques de Nabeul, 8000 Nabeul, Tunisia [email protected] 3 Ecole Nationale Supérieure d’Ingénieurs de Tunis, Tunis, Tunisia [email protected] 4 Université de Tunis – Laboratoire, LISIER, ENSIT, Tunis, Tunisia

Abstract. In this paper, we develop the stochastic degradation model in a control system, which deteriorates over time. The degradation process depends on different operational condition. An actuator can be worked under multiple stress levels. The aim is to improve the reliability of the actuator. We propose to model the degradation in an actuator with a gamma process in order to predict the Remaining Useful Lifetime. Obviously, a maintenance activity can be planned. To illustrate the performances of the approach, we consider a simulated double-tank level control system. Keywords: Degradation process RUL estimation


Reliability in control systems


1 Introduction The objective of this paper is to apply reliability theory in double-tank level control system in order to improve the remaining useful life prediction. The reliability of the system is difficult to model because of the interactions between predictive maintenance via actuators and the control system [1, 2]. Only some work [2, 6, 7] has investigated this difficulty and already give partial answers. The aim is to find a relationship between the control law, the actuator’s loss of effectiveness and its degradation. In the first step, we investigate the stochastic degradation models in control system. The reduction of safety margins for the good functioning of the system is the result of the degradation in the actuator that disturbs the progressive evolution of its state [6]. The modeling of the degradation is important for the prognostic in systems failure. The optimization of maintenance planning actions needs the estimation of the remaining useful life of actuator. Degradation evolution in actuator is Modeled by a stochastic process with random effects, These processes belongs to the Levy family [4]. Wiener process and Gamma process are the most frequently models used in literature. Gamma process, is used for a monotonic degradation evolutions while in the Wiener model, the © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 49–65, 2020. https://doi.org/10.1007/978-3-030-21009-0_5

50

N. Mabrouk et al.

increments follow a normal distribution, this process represent non-monotonic degradation evolutions [8]. This paper is organized as follows. In Sect. 2, we outline the general framework of Degradation in an actuator. In Sect. 3, we present the models of discrete degradation process and models of continuous degradation process in an actuator. In Sect. 4 we study the prognostic and Remaining Useful Life. In Sect. 5, an example is considered (motor pump system) to illustrate and verify the validity of the approach. Finally, we provide some final conclusions.

2 Actuator Degradation Type An actuator deterioration process is considered as a source of performance deterioration in physical system. If dðtÞ describes the actuator deterioration according to time t, the actuator capacity can be given by: Ca ðtÞ ¼ Ca0  dðtÞ

ð1Þ

Where Ca0 is the initial actuator capacity. In the following, we describe a different wear in an actuator: • Actuator wear Wear refers to the degradation of a surface leading to debris production. The most commonly wear is the combination of different degradation processes evolutionary and irreversible: adhesive friction, corrosion, erosion, fatigue. The phenomenon of wear appears during the relative sliding of two surfaces in contact on which a load is applied [1]. • Actuator choc Actuator is submitted to randomly shock that affects a random quantity of damage to the actuator. • Actuator thermal degradation Aging depends strongly on the temperature; we speak thermochemical process and thermal degradation. The deterioration rate d(t) of the winding insulation of an actuator heated to temperature T given by a deterministic model with the equation as follow [2]: d ðtÞ ¼ d0 ðt0 Þec0 ðtt0 Þ

ð2Þ

Where d ðt0 Þ the initial degradation measured at t ¼ t0 and c0 is the parameter related to conduction thermal of the material.

3 Actuator Degradation Process Modeling According to the type of the system, we can find two degradation classes: models of discrete degradation process and models of continuous degradation process [3].

Degradation Process Analysis and Remaining Useful Life

3.1

51

Discrete Models of Degradation Process

Discrete degradation models allow to shock-type [4] deteriorations. The most used model for modeling shocks is the Poisson process with intensity k. The level of damage wj produced by the j-th shock are identically distributed and independent. If an actuator is subjected to a shock-like degradation, it is assumed that the degradation occurs according to a compound Poisson process. This modeling hypothesis applied to an actuator, let Nt the cumulative number of shocks, the cumulative damage in an actuator at t is given by the curve see Fig. 1 dt ¼

XN t j¼1

wj

ð3Þ

Fig. 1. Discrete degradation model

This case shows that an actuator is subject of 39 shocks only at discrete times until t = 3500 time unit. 3.2

Continuous Models of Degradation Process

Continuous degradation models assume knowledge of the laws of increments of degradation between two consecutives instants. For continuous degradation, the Gamma process and the Wiener process are the Levy process models and they are the most used in prognostic.

52

3.3

N. Mabrouk et al.

The Gamma Process

Gamma process was successfully applied to model the monotonic and gradual deterioration for example a fatigue, crack growth or corrosion. This stochastic process was frequently used to describe the degradation of systems. The stochastic process modeling cumulative degradation according to a gamma process with independent, non-negative increments is given by dG ðtÞ ¼ dG ðt0 Þ þ Gða; bÞ [ 0

ð4Þ

Where dG ðt0 Þ the initial degradation (a, b) the shape and the scale parameter and G(a, b) the gamma process defined by the following properties [8]. • The increments of the process G are independent and stationary • G is a continuous stochastic process, 8t2 ; t1 ðt2  t1  0Þ the increment Gðt2 Þ  Gðt1 Þ follows a gamma distribution whose density function is: f G ð xÞ ¼

baðt2 t1 Þ xaðt2 t1 Þ1 expðbxÞ Cðaðt2  t1 ÞÞ

ð5Þ

Where ðaðt2  t1 Þ; bÞ the shape and the scale parameter; C the Gamma function is defined by: Z

þ1

Cðaðt2  t1 ÞÞ ¼

xaðt2 t1 Þ1 expðxÞdx

ð6Þ

0

The gamma process is characterized by positive increments. It is therefore well adapted to model of the Monotonic degradations (See Fig. 2) such as wear, corrosion. The mean and the variance of a process gamma are respectively worth EðdG ðtÞÞ ¼ dG ðt0 Þ þ

3.4

a a t ; Var ðdG ðtÞÞ ¼ 2 t b b

ð7Þ

The Wiener Process

The stochastic Wiener process (See Fig. 3) is modeling by a cumulative degradation dw ðtÞ with a linear drift parameter l and shape parameter r: dw ðtÞ ¼ dw0 ðt0 Þ þ lt + rBðtÞ

ð8Þ

Where dw0 ðt0 Þ the initial degradation, BðtÞ is a continuous stochastic process, 8t2; t1 ðt2  t1  0Þ the increment Bðt2 Þ  Bðt1 Þ follows a normal distribution with zero pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi average and a standard deviation ðt2  t1 Þ

Degradation Process Analysis and Remaining Useful Life

53

Fig. 2. Continuous degradation model: Gamma process path

Fig. 3. Continuous degradation model: Wiener process path

The drift and diffusion of a Wiener process are respectively (Fig. 3): E ðdw ðtÞÞ ¼ dw0 ðt0 Þ þ lt

ð9Þ

V ð dw ð t Þ Þ ¼ r 2 t

ð10Þ

54

N. Mabrouk et al.

4 Prognostic and Remaining Useful Life The prognosis of the failures actuator occupies an important place in operational safety. It is a process of predicting the future evolution of the capacity of the actuator. We noted tprog the moment at which the prognosis is considered and RUL the Remaining Useful Life actuator. Let us denote T the time failure in an actuator, the Remaining Useful Life (RUL) at tprog is a random variable defined for tprog \T: RUL ¼ T  tprog

ð11Þ

In the following, we will like to determine the pdf (probability density function) of RUL when the result of observations made since the initial moment. In the prognostic, there are two different assumption for the monitoring information [2]. • The first assumption: A model can describe the evolution of the degradation • The second assumption: The evolution of degradation can’t be described by a model. 4.1

Degradation Process with Model

The modeling of degradation actuators is a very important step to evaluate the reliability in actuator. Loss of the actuator effectiveness is considered to result from the dynamic evolution of the deterioration process. Thus, the deterioration in the actuator can be a stochastic or determinist process (see Fig. 4).

Fig. 4. Monitoring information with model

Degradation Process Analysis and Remaining Useful Life

55

In case where an actuator has a monotone gradual stochastic deterioration behavior, other degradation processes should be considered [5], for example, the Gamma process or Poisson process. In the following we use Gamma process to describe the degradation in actuator, this stochastic process was successfully applied to model the monotonic, progressive and slow degradation [8, 9]. It is used to model increasing degradations such as wear, corrosion and crack propagation. A homogeneous gamma process Cða; bÞ where a and b are respectively the shape and the scale parameter with the properties: • X(0) = 0, for t = 0, • For t  0 X(t) is independent and stationary. • 8t; sðt  s  0Þ, the increment X(t) − X(s) follows a gamma distribution, the shape parameter a(t−s) and the scale parameter b as in [10] X ðtÞ  X ðsÞ  Cðaðt  sÞ; bÞ ¼ faðtsÞ;b ðxÞ FaðtsÞ;b ð xÞ ¼

1 baðtsÞ xaðtsÞ1 ebx for x  0 Cðaðt  sÞÞ

ð12Þ ð13Þ

5 Maximum Likelihood Estimator of Parameters (MLE) To estimate the scale parameters a and the shape parameters b we use the maximum likelihood method. Consider the degradation increment Dxj ; j 2 f1. . .mg, from the density function, the likelihood function of the increment Dxj is [11]: LðaDt; bÞ ¼

Ym

f ðDxj Þ ¼ j¼1 aDt;b

Ym

baDt ðDxj ÞaDt1 ebDxj j¼1 CðaDtÞ

ð14Þ

Then LnðLðaDt; bÞÞ ¼

Xm j¼1

aDtlnðbÞ  lnðCðaDtÞÞ þ

Xm j¼1

   ððaDt  1Þ ln Dxj  bDxj ð15Þ

The expression of the partial derivative of Ln(LðaDt; bÞÞ compared to a   @lnðLðaDt; bÞÞ Xm ¼ ðDtlnðbÞ  Dt:wðaDtÞ þ Dt ln Dxj j¼1 @a

ð16Þ

56

N. Mabrouk et al.

The expression of the partial derivative of Ln(LðaDt; bÞÞ compared to b @ lnðLðaDt; bÞÞ Xm aDt ¼  Dxj Þ ð j¼1 @b b

ð17Þ

Or 0

C ðxÞ w¼ CðxÞ

ð18Þ

The estimators ^a and ^b are obtained by solving (19): Xm j¼1

  ðDtlnðbÞ  Dt:wðaDtÞ þ Dt ln Dxj ¼ 0 Xm j¼1

ð

aDt  Dxj Þ ¼ 0 b

ð19Þ ð20Þ

Which gives: Xm mDt ^a Pm þ Dt lnðlnðDxj ÞÞ  wð^ aDtÞ ¼ 0 j¼1 j¼1 ðDxj Þ

ð21Þ

^b ¼ ^a PmmDt j¼1 ðDxj Þ

ð22Þ

To solve Eqs. (20) and (21) we can use numerical method such as Newton Raphson method.

Fig. 5. Monitoring information without model

Degradation Process Analysis and Remaining Useful Life

5.1

57

Degradation Process Without Model

Generally, performance degradation is the result of environment factors and the operating conditions. The foundation of modeling can be collected from the history degradation data of actuators (see Fig. 5). However, the degradation process can be modeled with statistic inference method, and it will be used to describe the prior information of the Bayesian method. Therefor, the data base is collected to estimate the PDF of posterior information. The parameters of the model process are real time updated and we can get the real time reliability by the specific failure of actuator. Bayesian inference is an efficient approach to evaluate the unknown parameters of a given model. When it is difficult to obtain the analytical posterior distribution, Markov Chain Monte Carlo method can be used. 5.2

Prognostic and RUL Estimation

To illustrate the actuator breakdown, a failure threshold L and degradation paths of many components can be considered. We describe the degradation in an actuator at time t a Gamma process model is used. The failure time for a component i is TLi which its evolution path i first hits the failure threshold L. If we consider one component reaches a failure threshold L. The distribution function of failure time is defined below: TL ¼ inf ft [ 0; Xt  Lg

ð23Þ

According to Eq. (22), the first hitting time TL for one component satisfies Eq. (23). F ðx; a; bÞ ¼ PðTL  tÞ ¼ PðX ðTL Þ  LÞ ¼ 1  PðX ðTL Þ  LÞ Z

L

F ðx; a; bÞ ¼ 1  0

bat at1 bx x e dx CðatÞ

ð24Þ ð25Þ

b Where fat;b ð xÞ ¼ CðatÞ xat1 ebx : at

6 Case Study: Two-Tank Level Control System To verify the validity of the approach, we consider a motor pump system and we assume that the stochastic degradation process is applied in the actuator. This case is presented in many work such as in [6, 7]. 6.1

System Description

Water is pumped into tank1 by a motor pump drives, then, the out flow from tank1 feeds tank2 (see Fig. 6). A PID controller calculates the motor pump control input. The water level of tank1 and tank2 measured with a level measurement sensor. The cross sectional area of the tank1 and tank2 is respectively S1 and S2 (Table 1).

58

N. Mabrouk et al.

Fig. 6. A double-tank levels control system Table 1. Physical parameters Parameters S1 ; S2 h1 ; h2 g K v1 ; K v2 sa Ka K a0 qin u uM

Description Tank cross-sectional area of tank1 and tank2 Fluid level of tank1 and tank2 Acceleration of gravity (m2/s) Parameter of valve of V 1 and V 2 Time constant (motor pump) Servo amplifier gain Initial Servo amplifier gain Flow rate (m3/s) Command signal Maximum input

The flow rate qin and the input u can be written in the form of a first order system. dqin 1 Ka ¼ qin þ u sa dt sa

ð26Þ

Regardless of the time t considered during the operation of the pump uðtÞ 2 ½0; uM ; the water flows out at the bottom of tank1 and tank2 through valves V1 andV2 at flow rates according to Torricelli’s law: q1;out ¼ Kv1

pffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffi 2gh1 ; q2;out ¼ Kv2 2gh2

ð27Þ

Degradation Process Analysis and Remaining Useful Life

59

The equations that describe the system can be described by using the mass balance equation dh1 ðtÞ 1 Kv pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ qin ðtÞ  1 2gh1 ðtÞ dt S1 S1

ð28Þ

dh2 ðtÞ Kv1 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Kv2 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ 2gh1 ðtÞ  2gh2 ðtÞ dt S1 S2

ð29Þ

The control objective of the actuator is to adjust the level of tank 2 according to the set-point evolution. 6.2

Actuator Degradation Process

The actuator deterioration process is considered as a source of performance deterioration in physical system. Control systems can deteriorate via the actuators [6]. Indeed, the loss of capacity of an actuator can cause a loss of performance of the system. We assume that the capacity of the actuator decreases from the nominal value according to the Gamma process without the covariates. If d ðtÞ  Ga ða  t; bÞ describes the actuator accumulated deterioration according to time t, the actuator capacity can be given by: Ka ðtÞ ¼ Ka0  dðtÞ

ð30Þ

Table 2. Numerical value of physical parameters Parameters S1 ¼ 25 Kv1 ¼ 8 sa ¼ 1 S2 ¼ 20 Kv2 ¼ 6 g ¼ 9:82 umax ¼ 10 Ka,min= 5 L = 15 Initial condition h1 ð0Þ ¼ 5 h2 ð0Þ ¼ 0 Ka0 ¼ 20 PID controller parameters Kp ¼ 4 TI ¼ 19 TD ¼ 1:5

To illustrate the proposed approach a simulation of the two-tank level control system was used. The following Table 2 gives the numerical value of different parameters used in simulation. The simulation result (Figs. 7, 8, 9, and 10). We assume the law of growth degradation process d ðt þ DtÞ  dðtÞ is a gamma distribution Ga ðaðt þ DtÞ  aðtÞ; bÞ, for all t [ 0 and Dt [ 0. The expression of function density is given by [7]:

60

N. Mabrouk et al.

Fig. 7. Set point

Fig. 8. Flow rate

f ðd Þ ¼

bðaðt þ DtÞaðtÞÞ d ðaðt þ DtÞaðtÞÞ1 expðbdÞ Cðaðt þ DtÞ  aðtÞÞ

Where C : Gamma function.

ð31Þ

Degradation Process Analysis and Remaining Useful Life

61

Fig. 9. Actuator capacity

Fig. 10. Water level of tank1

The probability distribution function of RUL An actuator is considered in a good state at the time of prognostic tpro , we are interested to know the RUL of the actuator.

62

N. Mabrouk et al.

  • If the current degradation level d tpro is unknown, the conditional reliability is defined by:

Fig. 11. Deterioration process

Ri ðtjTi [ tpro Þ ¼ PðTi  tjTi [ tpro Þ ¼ Pðdi ðtÞ  LjTi [ tpro Þ

ð32Þ

  Pðdi ðtÞ  di ð0ÞÞ  L  Ri tjTi [ tpro ¼    P di tj  di ð0Þ  L

ð33Þ

  • If we know the law of d tpro ; we have more precise predictions of remaining useful life than Eq. (32). Figure 11 illustrates the effect of knowing the degradation of the actuator at t ¼ tprog : The probability distribution function of RUL of actuator is less than h is [12]: PðRULX ðtÞ  hÞ ¼ 1  PðRULX ðtÞ [ hÞ ¼ 1  PðYðt þ hÞ  L=XðtÞ ¼ xðtÞÞ PðY ðt þ hÞ  xðtÞ  L  xðtÞÞ ¼1 PðXðtÞ  L PðRULX ðtÞ  hÞ ¼ 1  Where h is a predefined period.

R LxðtÞ

0

fah;b ðuÞdu 0 fat;b ðvÞdv

RL

ð34Þ

Degradation Process Analysis and Remaining Useful Life

63

Fig. 12. FHT by Monte Carlo simulations, a = 0.02 and b = 0.6

Fig. 13. Probability density of the remaining useful life

The prognostic degradation path is illustrated in Fig. 12. Ka ðtÞ ¼ Ka0  d ðtÞ, L = 15 (Failure Threshold Level) etKa0 ¼ 20: The estimated value of the shape parameter a ¼ 0:02 and the scale parameter b ¼ 0:06. We use the Monte Carlo simulation to obtain the distribution of First Hitting Time.

64

N. Mabrouk et al.

We use softwear Matlab to evaluate the statistical proprieties of pdf RUL see Fig. 13 for more details [13] The mean of Tmean ¼ 878:2380 and the time of prognostic Tpog ¼ 450, given Eq. 11, we calculate RUL ¼ 428:2380. RUL is obtain with Matlab simulation, Tmean ¼ 878:2380 and the time of prognostic Tpog ¼ 450, given by Eq. 10, we calculate RUL ¼ 513.2. However we use the RUL to improve plant production availability and reduce down time cost.

7 Conclusion This work is a proposal for a positioning of the problem of the RUL evaluation of a dynamic control system with a stochastically deteriorating actuator. We aim to model the degradation process in an actuator. At the first time we study the discrete models of degradation process. At the second time we analyse and study continuous models of degradation process. Then we combine the dynamic and the stochastic part system modeling using only the output of the system. We show the ability to combine the deterministic behavior of a feedback control system with the stochastic deterioration process for the actuator. In this framework, the loss of effectiveness of actuator is modeled by the random gaps which intersect the deterministic trajectory of closed-loop system only at random discrete times. At the end a case of study is applied to illustrate our approach.

References 1. Warburton, D., Strutt, J.E., Allsopp, K.: Reliability prediction procedures for mechanical components at the design stage. In: Proceedings of the Institution of Mechanical (1998) 2. Langeron, Y.: Modélisation stochastique pour la sûreté de fonctionnement des systèmes commandés. Ph.D. thesis, Université Technologique de Troyes (2015) 3. Castanier, B.: Modélisation stochastique et optimisation de la maintenance conditionnelle des systèmes à dégradation graduelle. Ph.D. thesis, Université Technologique de Troyes (2001) 4. Nakagawa, T.: Shock and Damage Models in Reliability Theory. Springer Science and Business Media (2007) 5. Ngoc Nguyen, D., Dieulle, L., Grall, A.: Remaining useful lifetime prognosis of controlled systems: a case of stochastically deteriorating actuator. Math. Probl. Eng. 2015, 1–16 (2015) 6. Ngoc Nguyen, D., Dieulle, L., Grall, A.: Feedback control system with stochastically deteriorating actuator: remaining useful life assessment. In: World Congress. Cape Town, South Africa (2014) 7. Aggab, T., Kratz, F., Vrignat, P., Avila, M.: Remaining useful life prediction method using an observer and statistical inference estimation methods. In: The 10th International Conference on Mathematical Methods in Reliability (2017) 8. Van Noortwijk, J.M.: A survey of the application of gamma processes in maintenance. Reliab. Eng. Syst. Saf. 94, 2–21 (2009) 9. Abdel-Hameed, M.: Lévy Processes and Their Applications in Reliability and Storage. Springer, New York (2014)

Degradation Process Analysis and Remaining Useful Life

65

10. Lawless, J., Crowder, M.: Covariates and random effects in a gamma process model with application to degradation and failure. Lifetime Data Anal. 10(3), 213–227 (2004) 11. Zhang, Y.: Wiener and gamma processes overview for degradation modelling and prognostic. Norwegian University of Science and Technology (2015) 12. Park, C., Padjet, W.J.: Acceleration degradation models for failure based on geometric Brownian motion and gamma process. Lifetime Data Anal. 11, 511–527 (2005) 13. Moulahi, M.H., Ben Hmida, F.: gamma process based degradation growth model on reliability of bearing in motor pump. In: Euro-Mediterranean Conference on Mathematical Reliability (ECMR), pp. 57–68 (2018)

Data Quality of the Information Collected from GPR on a 3D Structure Rim Ghozzi1,2(&), Samer Lahouar1,3, and Chokri Souani1,4 1

4

Laboratoire de Microélectronique et Instrumentation, Faculté des Sciences de Monastir, Université de Monastir, 5000 Monastir, Tunisia [email protected], [email protected], [email protected] 2 Ecole Nationale D’Ingénieurs de Sousse, Université de Sousse, 4023 Sousse, Tunisia 3 Center for Research in Microelectronics and Nanotechnology, CRMN, Technopole of Sousse, 4054 Sousse, Tunisia Institut Supérieur des Sciences Appliquées et de Technologie de Sousse, Université de Sousse, 4003 Sousse, Tunisia

Abstract. Ground Penetrating Radar (GPR) is a non-destructive geophysical survey technique based on the analysis of electromagnetic waves propagation phenomena (refraction, reflection and diffraction) in the subsoil. By analyzing the received signals, it is possible to obtain a representation of the different layers of the subsoil. The data quality of the GPR signal becomes more important to get quantitative information about the underground layers, such as their thickness. For the evaluation of the quality of the received signal, air-launched and groundcoupled antenna systems are used. The effects of an incident field on the received signal may have to be considered. This paper examines the effects of the GPR antennae orientation on the received signal for the air-launched and ground-coupled antenna systems. Also, the different types of incident field sources are tested in order to obtain a better quality received signal, by comparing the quantitative information after the data analysis. The effects of these changes in types of incident field sources are measured by changes in arrival time. To see these effects, a 3D numerical modeling has been done using the open source software gprMax. From the results it appears that the Ricker and Gaussiandotnorm waveforms are the best incident fields in ground-coupled antenna with parallel and perpendicular configuration, respectively. Keywords: Ground Penetrating Radar (GPR) Thickness estimation
3D structure




Non-destructive inspection




1 Introduction Ground Penetrating Radar (GPR) is a non-destructive prospecting technique used for the inspection of civil engineering structures [1–3]. GPR system is based on sending an electromagnetic pulse in the environment to be studied, using a transmitting antenna © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 66–76, 2020. https://doi.org/10.1007/978-3-030-21009-0_6

Data Quality of the Information Collected from GPR

67

(Tx). This pulse is reflected on the interfaces between layers of different dielectric properties (see Fig. 1). A receiving antenna (Rx) measures the amplitude of the reflected electromagnetic signal over time, which is called an A-scan. The latter was used to determine the thicknesses of the underground layers by measuring the time intervals between the emission of the pulse and the recording of its echo. In Fig. 1, the A-Scan GPR echo signal (S) includes the Direct Wave (DW) and interfaces echo signals (Si) of the ith layer and eri is the relative permittivity of the ith layer. The simplest method for studying the reflected signal based on a trace is to measure the peak-to-peak time of the reflection [4, 5]. That is, the time difference between these peaks.

x y

A-scan GPR echo signal (S)

Tx Rx Air Background

Amplitude

Surface

Direct Wave (DW) Layer 1 (εr1)

Interface 1 echo signal (S1)

Layer 3 (εr3)

(a)

Time

Layer 2 (εr2)

Interface 2 echo signal (S2)

(b)

Fig. 1. (a) Structure of three-layers, (b) GPR echo in the multilayer structure [6].

2 Methodology This paper aims to select the best incident waveforms (Gaussian waveform, Mexican hat waveform…). The electromagnetic simulation software (gprMax) is used to generate GPR data. The simulator was developed by Dr. Giannopoulos using the FiniteDifference Time-Domain (FDTD) method [7, 8]. This simulator allows specifying the technical characteristics of the antenna (incident waveforms of the sources (Tx antenna), frequency, amplitude, time window) and the geometry models (model size, space discretization, model material, multilayer structure). In order to prove this, the first step is to create two 3D geometric models containing underground multilayer structure. Figure 2 shows the 3D geometric models of the underground geometry. The model domain 1  2  2 m, has a spatial discretization of Dx = Dy = Dz = 0.01 m and the boundary conditions of the numerical simulation model are set to be the Perfectly Matched Layer (PML). The two models represent two layers with different permittivity. There are two types of antenna which are ground-coupled and air-launched [6, 9]. The two types of antenna used in this research. The receiver drawn by a red triangle and the transmitter depicted by a blue triangle antennae. The two antennae are separated from each other with a distance of 20 cm. The distance between ground- and air-coupled antennae and the ground surface is 2 cm and 40 cm, respectively. The time window is 12 ns. Table 1 provides the subsurface geometry and the parameters of the antennae.

68

R. Ghozzi et al.

z

Tx Rx Y X

40 cm 2 cm

Surface

d1

Layer 1 (εr1)

Interface 1

Layer 2 (εr2)

Layer 1 (εr1)

Layer 2 (εr2)

Model 2 (b)

Model 1 (a)

Fig. 2. 3D geometric models for two types of antennae: (a) ground-coupled, (b) air-launched [10]. Table 1. GPR model description. Geometry models

Antenna (Tx/Rx)

Domain (m) Box (m) Relative permittivity of layer 1 (er1) Relative permittivity of layer 2 (er2) d1 (cm) Polarization Type of scan Position between Tx/Rx (cm) Type of incident field source (Tx) Amplitude of incident field source Frequency of incident field source (MHz) Configuration

(2 2 1) (1 2 1.85) 4 {2, 6} 20 x A-Scan 20 {Gaussian, Gaussiandotnorm, Ricker} 1 500 { Parallel, Perpendicular}

At the interface between two mediums, the electromagnetic wave (EM) will be reflected and transmitted. At a normal incidence, the reflection coefficient ðRÞ for an EM wave is the following one [11, 12]: pffiffiffiffiffiffi pffiffiffiffiffiffi er1  er2 R ¼ pffiffiffiffiffiffi pffiffiffiffiffiffi er1 þ er2

ð1Þ

The amplitude of the echo signals is proportional to the contrast in dielectric properties between the layers [6, 13]. If medium 2 has a higher relative permittivity

Data Quality of the Information Collected from GPR

69

than medium 1 ðer2 [ er1 Þ, then R is a negative number, which means that the polarization of the wave is reversed and vise-versa. In the two models the first layer whose relative permittivity is 4 and thickness is 20 cm. The second having a value of permittivity equal to six or two. Transmitter and receiver antennae can also be used in different configurations, as they are illustrated in Fig. 3. The parallel and perpendicular configuration is when the axis of polarization collinear and perpendicular to the antenna axis, respectively. Changing the orientation of antennae can significantly change the response of GPR Because the radiation patterns of typical GPR antennae are not omnidirectional, antennae radiate or receive electromagnetic waves more effectively in specific directions than others. Tx Rx

(a)

(b)

Fig. 3. (a) Perpendicular configuration, (b) Parallel configuration.

3 Data Analysis 3.1

Type of Incident Waveforms

The different waveforms of the source used in this paper are: Gaussian f1(t), the normalized first derivative of Gaussian (gaussiandotnorm f2(t)) and Ricker waveform ((Mexican Hat) which is the negative, normalized second derivative of a Gaussian waveform f3(t)) [14]. The equations of waveforms and plots (as depicted in Fig. 4) are shown below using the following parameters: the amplitude of 1, centre frequency of 500 MHz, the time window of 12 ns and a time step of 19.25 ps. f1 ðtÞ ¼ enðtvÞ

2

rffiffiffiffiffi 2 e nðt  nÞenðtvÞ f2 ðtÞ ¼ 2 2n   2 f3 ðtÞ ¼  2nðt  nÞ2 1 enðtvÞ where n ¼ 2p2 f 2 , v ¼ 1f and is the frequency.

ð2Þ ð3Þ ð4Þ

70

R. Ghozzi et al.

Fig. 4. (a) Gaussian waveform - time domain, (b) Gaussiandotnorm waveform - time domain, (c) Ricker waveform - time domain.

3.2

Numerical Simulation

Signal processing on the collected time domain signal plots is executed by MATLAB software. In time domain, the peak-to-peak time of the reflected signal was measured to determine the thickness of the underground first layer. For the air-launched antenna, the thickness is estimated using peaks of Reflected Wave (RW) by interfaces between two mediums. While for the ground-coupled antenna, the thickness is estimated using peaks of direct and reflected waves (DW and RW). The different scenarios (in Table 2) are tested using different incident waveforms. The results of reflected signals when the incident fields are Gaussian, Gaussiandotnorm and Ricker shown in Figs. 5, 6 and 7, respectively. Table 2. Different scenarios.

Structure 1 er1 ¼ 4; er2 ¼ 6 Structure 2 er1 ¼ 4; er2 ¼ 2

Air-launched Parallel configuration Scenario 1

Perpendicular configuration Scenario 2

Ground-coupled Parallel Perpendicular configuration configuration Scenario 5 Scenario 6

Scenario 3

Scenario 4

Scenario 7

Scenario 8

Data Quality of the Information Collected from GPR

t1

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

71

t2

Fig. 5. Numerical simulation of scenarios with Gaussian incident field, (a)–(h): Senario 1– Senario 8.

72

R. Ghozzi et al.

Fig. 6. Numerical simulation of scenarios with Gaussiandotnorm incident field, (a)–(h): Senario 1–Senario 8.

Data Quality of the Information Collected from GPR

73

Fig. 7. Numerical simulation of scenarios with Ricker incident field, (a)–(h): Senario 1–Senario 8.

74

R. Ghozzi et al.

The travel-time is determined as the distance between the maximums (positive or negative) of the reflected pulses at the interfaces. Therefore, the travel time ð c DtÞ is the difference between the transmission times of peak n°1 and peak n°2 (see Eq. 5). c Dt ¼ t2  t1

ð5Þ

In scenario 1, when the source is a Gaussian waveform (see Fig. 5), the transmission time of the first peak is set to 4.45 ns and the transmission time of the second peak is set to 7.11 ns. As a consequence, the travel time is equal to 2.66 ns. With the same method of calculation, when the source is Gaussiandotnorm and Ricker, the travel times are 2.68 ns and 2.66 ns, respectively (see Figs. 6 and 7). The travel times for all scenarios are reported in Table 3. The performance of measuring the travel time, at different source types in parallel and perpendicular antennae, is evaluated by computing the relative error (RE) (see Table 4), which is defined as follows:   Dt  c Dt i   RE ð%Þ ¼  ð6Þ   100  Dt  where c Dt i is the estimated value, Dt is the true value, and i is the number of scenario. Actually, the thickness of the 1st layer (d1 ¼ 0:2 m), the dielectric constant of the medium (er1 ¼ 4) and the true travel-time (Δt = 2.67 ns), are calculated using the equation of the EM wave, which is given by the following equation [15]: Dt ¼

pffiffiffiffiffiffi 2 er1 d1 c

ð7Þ

where ‘c’ is the speed of light in vacuum (c = 0.3 m/ns). Table 3. Results of efficient incident waveforms to estimate travel time. c Dt i ðnsÞ

Scenario 1

Gaussian 2.66 Gaussiandotnorm 2.68 Ricker 2.66

Scenario 2

Scenario 3

Scenario 4

Scenario 5

Scenario 6

Scenario 7

Scenario 8

2.68 2.68 2.66

2.66 2.68 2.66

2.66 2.68 2.66

1.37 2.64 2.28

2.20 2.18 2.85

2.03 2.66 2.33

2.01 2.04 2.84

Table 4. Relative travel time estimation error. RE (%)

Scenario 1

Gaussian 0.37 Gaussiandotnorm 0.37 Ricker 0.37

Scenario 2

Scenario 3

Scenario 4

Scenario 5

Scenario 6

Scenario 7

Scenario 8

0.37 0.37 0.37

0.37 0.37 0.37

0.37 0.37 0.37

48.69 1.12 14.61

17.60 18.35 6.74

23.97 0.37 12.73

24.72 23.60 6.37

For the air-launched antenna (in scenarios 1–4) the relative error equal to 0.37% at different source types. For the ground-coupled antenna with parallel configuration (in

Data Quality of the Information Collected from GPR

75

RE (%)

scenarios 5, 7) the Ricker waveform is the best incident field. Whereas for the groundcoupled antenna with perpendicular configuration (in scenarios 6, 8) the Gaussiandotnorm waveform is the best incident field. Figure 8 depicts the relative travel time estimation error for ground-coupled antenna.

50

Gaussian

40

Gaussiandotnorm

30

Ricker

20 10 0 Scenario Scenario Scenario Scenario 5 6 7 8

Fig. 8. Relative travel time estimation error for ground-coupled antenna.

4 Conclusion In this paper, different types of incident fields of sources have been tested in order to obtain better GPR data reflected by underground layer. FDTD simulations of various types of GPR signals transmitted in different scenarios. A-Scans have been also acquired by considering multiple antenna orientations to see how important the configuration factor may be for selecting the best incident waveform: with the axis of the transmitter/receiver antennae perpendicular and parallel to the polarization. To estimate the thickness of first underground layer, the results have demonstrated that the Ricker and Gaussiandotnorm waveforms are the best incident fields in ground-coupled antenna with parallel and perpendicular configuration, respectively.

References 1. Lai, W.W.L., Derobert, X., Annan, P.: A review of Ground Penetrating Radar application in civil engineering: a 30-year journey from locating and testing to imaging and Diagnosis. NDT & E Int. 96, 58–78 (2018). https://doi.org/10.1016/j.ndteint.2017.04.002 2. Xie, P., Wen, H.J., Xiao, P., Zhang, Y.Y.: Evaluation of ground-penetrating radar (GPR) and geology survey for slope stability study in mantled karst region. Environ. Earth Sci. 77, 122 (2018). https://doi.org/10.1007/s12665-018-7306-9 3. Ghozzi, R., Lahouar, S., Souani, C.: The estimation of buried empty cylindrical tubes characteristics using GPR. In: Advanced Technologies for Signal and Image Processing (ATSIP’2018), Sousse, Tunisia (2018). https://doi.org/10.1109/atsip.2018.8364486

76

R. Ghozzi et al.

4. De Coster, A., Van der Wielen, A., Gregoire, C., Lambot, S.: Evaluation of pavement layer thicknesses using GPR: a comparison between full-wave inversion and the straight-ray method. Constr. Build. Mater. 168, 91–104 (2018). https://doi.org/10.1016/j.conbuildmat. 2018.02.100 5. Ghozzi, R., Lahouar, S., Souani, C., Besbes, K.: Peak detection of GPR data with lifting wavelet transform (LWT). In: Advanced Systems and Electric Technologies (IC_ASET). IEEE, Hammamet, Tunisia (2017). https://doi.org/10.1109/aset.2017.7983663 6. Baili, J., Lahouar, S., Hergli, M., Al-Qadi, I.L., Besbes, K.: GPR signal de-noising by discrete wavelet transform. NDT & E Int. 42, 696–703 (2009). https://doi.org/10.1016/j. ndteint.2009.06.003 7. Giannopoulos, A.: Modelling ground penetrating radar by gprMax. Constr. Build. Mater. 19, 755–762 (2005). https://doi.org/10.1016/j.conbuildmat.2005.06.007 8. Warren, C., Giannopoulos, A., Giannakis, I.: gprMax: open source software to simulate electromagnetic wave propagation for Ground Penetrating Radar. Comput. Phys. Commun. 209, 163–170 (2016). https://doi.org/10.1016/j.cpc.2016.08.020 9. Khamzin, A.K., Varnavina, A.V., Torgashov, E.V., Anderson, N.L., Sneed, L.H.: Utilization of air-launched ground penetrating radar (GPR) for pavement condition assessment. Constr. Build. Mater. 141, 130–139 (2017). https://doi.org/10.1016/j.conbuildmat.2017.02.105 10. Diamanti, N., Annan, A.P.: IEEE: air-launched and ground-coupled GPR. In: Data 11th European Conference on Antennas and Propagation (EUCAP), Paris, France (2017). https:// doi.org/10.23919/eucap.2017.7928409 11. Iswandy, A., Serma, A., Setan, H.: Ground penetrating radar (GPR) for subsurface mapping: preliminary result. Geoinf. Sci. J. 9, 45–62 (2009) 12. Syambas, N.R.: An approach for predicting the shape and size of a buried basic object on surface ground penetrating radar system. Int. J. Antennas Propag. (2012). https://doi.org/10. 1155/2012/919741 13. Loizos, A., Plati, C.: Accuracy of pavement thicknesses estimation using different ground penetrating radar analysis approaches. NDT & E Int. 40, 147–157 (2007). https://doi.org/10. 1016/j.ndteint.2006.09.001 14. Warren, C., Giannopoulos, A.: gprMax User Guide Release 3.1.12017. www.gprmax.com 15. Lahouar, S., Al-Qadi, I.L.: Automatic detection of multiple pavement layers from GPR data. NDT & E Int. 41, 69–81 (2008). https://doi.org/10.1016/j.ndteint.2007.09.001

DSP Real-Time Implementation of DOST Algorithm Used for Speech Enhancement Safa Saoud(&), Souha Bousselmi, Mouhamed Ben Nasr, and Adnen Cherif ATSEE Laboratory, Sciences Faculty of Tunis, University of Tunis El-Manar, Tunis, Tunisia [email protected]

Abstract. The real-time implementation of speech enhancement is a vital tool destined to ameliorate the speech quality and intelligibility for auditors. In this paper, a speech denoising hardware implementation is developed in order to be used in recognition, synthesis, and coding applications. So, we propose a realtime implementation of speech enhancement approach for single channel in a noisy environment on the basis of Discrete Orthonormal Stockwell Transform (DOST) at the aim to ameliorate the speech quality and intelligibility. The speech enhancement system was tested on DSP TMS320C6416 processor and the obtained results have shown that it has met the real-time requirements in terms of memory consumption (Ko) and number of cycles (MCPS). For a subjective criterion, we have used the Mean Opinion Score (MOS) to evaluate the perceptual quality. Keywords: Speech Enhancement (SE)
DOST Speech quality and intelligibility
MOS


DSP


1 Introduction We are surrounded by a noisy environment, such as in the street (car traffic, construction works), in the cars (the sound of the engine and wind) in the restaurants (the sound of people chattering). For this reason, speech can be easily contaminated by acoustic background noise. Hence, speech enhancement (SE) has become a fundamental means for speech signal processing and it is widely applied in a rising number of audio applications, such as cellular telephone and automatic speech recognition (ASR). The main problem in these systems consists of improving their performances in terms of quality and intelligibility in noisy conditions. In the literature, there are many speech enhancement algorithms which have been proposed, such as Discrete Fourier Transform (DFT), Kalman [1] and Wiener [2] filtering, Spectral Subtraction [3], Wavelet Transform (WT) [4, 5], Recurrent Neural Network [26],…etc. All the above mentioned approaches have their own favors and drawbacks. The spectral subtraction (SS) used to enhance the speech corrupted by additive stationary background noise and it has two major inconvenient representing by its influence by musical noise and its disability to reduce noise in silence periods [6]. Concerning Wiener filter, it is destined to decrease the Mean Square Error © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 77–88, 2020. https://doi.org/10.1007/978-3-030-21009-0_7

78

S. Saoud et al.

(MSE) among the clean and the estimated signal. In fact, the characteristics of clean speech are necessary to enhance speech for spectral and Wiener approaches [2]. However, clean speech is not always available in real-time. To enhance speech recovered form noisy speech, sinusoidal model has been proposed in [7] instead of harmonic structure of speech signal [8]. Minimum Mean Square Error (MMSE) estimator has been also adopted for the same task [9]. In [10], it has been assumed that using SS and MMSE as speech enhancement methods, the noise affects uniformly the speech signal in the whole frequency range. However, this signal is affected differently by colored noises at its different frequency ranges in real life conditions. To benefit from the advantages of Kalman filter, an estimation of signal parameters from clean speech has been proposed by Paliwal and Basu in [11] before the corruption of speech signal by white noise. In [1], an extension of this work has been introduced with colored and random noises. In these aforementioned methods, finding such compromise between SNR and intelligibility should be preserved. In time domain, the reconstruction of enhanced signal is carried out using the phase of the noisy signal. Indeed, it has been assumed that this phase occurs due to noise. In time-frequency domain, Short-Time Fourier Transform (STFT) [12] and Wavelet Transform (WT) are considered as well-known methods. In signal and image processing, discrete wavelet transform (DWT) is an emerged and a powerful tool that relies on a selective smoothing at each scale of the time frequency plot. Thus, only the scale information is provided in DWT which makes its application limited as the frequency information and the absolutely-referenced phase are required [13]. In fact, the phase estimation of the processed signal can be performed using Phase Aware (PA) speech enhancement techniques by applying conjugate symmetry of the short-time Fourier spectrum [15–17]. Stockwell Transform (ST) [13] has been considered as a new approach for timefrequency analysis. In multi-resolution analysis domain, ST method is defined as an exceptional case of wavelet transform (WT) and it is the time-frequency resolution of STFT. Therefore, ST method can be established by hybridizing WT with STFT [18]. Using ST approach, a more specific relation can be offered between time and frequency distribution of speech signal. Nevertheless, the dimension of the original data set is redundantly doubled by ST which makes it unfeasible and computationally costly. This is the case when using a big size of data sets. To enhance its effectiveness, DOST method has been integrated in [18] at the aim to decrease the redundancy of ST approach and makes it useful and more suitable for real-life applications. Indeed, a set of orthonormal basis functions which focus the Fourier spectrum of a signal represents the basic element of DOST approach. Also, this later samples the time-frequency representation given by ST approach without information redundancy and preserves the favor phase properties. Although, the DOST is fairly recent with comparison by other transforms, it has been confirmed to its practicability in certain domains, such as image restoration [14], image compression [13, 18, 19] and image texture analysis [18]. Furthermore, it has been effectively employed in signal analysis by canalizing the instantaneous frequencies. In this context, we have proposed a new approach based on DOST algorithm for real-time speech enhancement in noisy conditions. This approach has been

DSP Real-Time Implementation of DOST Algorithm

79

implemented on a fixed point digital signal processing (DSP) board and compared to DWT algorithm via objective and subjective tests. The remaining sections of this paper are structured as follows: The proposed speech enhancement method is given in Sect. 2. In Sect. 3, a real-time implementation of the suggested model is presented. Conclusion and perspectives are exhibited in Sect. 4.

2 The Proposed Speech Denoising Method An illustration of the block diagram of the proposed speech denoising system is given in Fig. 1.

DOST

Thresholding

IDOST

Enhanced speech

Noisy speech

Fig. 1. DOST approach based speech denoising.

The expression of the noisy speech signal is given as: Y ð nÞ ¼ Sð nÞ þ W ð nÞ

ð1Þ

Where, s (n), w (n), and y (n) represent clean speech, noise and noisy speech signal, respectively. In this paper, the proposed denoising system is composed of three stages: DOST transform thresholding, and IDOST transform. 2.1

The Discrete Orthonormal Stockwell Transform (DOST)

The initial step of the proposed approach is to decompose the signal (speech) using DOST transform. This later is presented as an orthonormal version of S-Transform. Also, it is expressed as an inner product between the basis function d[k] and temporal series h[k]. Indeed, the kth basis vector is given as: 1 XN1 Xv þ b=21 i2p k f i2pbs f ips e Ne e h½ k  Sðv;b;sÞ ¼ \d ½k½v;b;s ; h½k ¼ pffiffiffi k¼0 f vb=2 b

ð2Þ

Where: m, b and s represent the center of each frequency band, the bandwidth and the location in time respectively. The discrete ST can produce N2 coefficients for a given signal of length N, whereas the DOST represents the similar signal with only N coefficients. Thus, DOST is a non superfluous version of ST.

80

S. Saoud et al.

By introducing FFT into DOST, we can calculate DOST more rapidly such as: Sðv;b;sÞ ¼ \d½k½v;b;s ; h½k

    b=21 N1 v þX 1 X k s ¼ pffiffiffi exp i2p f exp i2p f expðipsÞHðf Þ N b b k¼0 fvb=2

2.2

ð3Þ

Thresholding

The thresholding is a main step for a transform-based enhancement. Indeed, it consists to reject the DOST coefficients according to a given threshold. In general, the soft and hard thresholding are usually used [25]. In this study, the soft one is used which is given by the following equation:  Xsoft ¼

signðxÞðjXj  jsjÞ; 0

if jXj [ 0 if jXj  0

ð4Þ

Where, X and s represent the DOST coefficients and a threshold value, respectively. 2.3

Inverse Discrete Orthonormal Stockwell Transform (IDOST)

At the end, we have applied the inverse DOST (IDOST) in order to get the enhanced speech signal.

3 Real-Time DSP Implementation For audio applications, real-time experiences are still an increasing search axis. Indeed, the input and the generated output signal are processed incessantly which can explain clearly that the average processing time per sample is inferior than the sampling period In fact, our algorithm has been tested on DSP-TMS320C6416 using a developed starter kit DSK board and a software tool (code composer studio). Moreover, a rapid prototyping (RTW) tool from Math-works has been employed. 3.1

DSK C6416 Overview

In general, the TMS320C6416 is a fixed point digital signal processor which operates with a 1 GHz clock represents the main component of DSK C6416 board. Indeed, this later includes the following components: • A stereo codec AIC23 for interfacing audio applications contains ADC converters for signal taking (LINE IN or MIC IN) and DAC converters for signal export (LINE OUT, HP OUT). • Flash memory is of capacity 512 KB • The SDRAM is of 16 MB capacity

DSP Real-Time Implementation of DOST Algorithm

81

• A JTAG emulation allowing communication with an external host through a USB connection. • 4 LEDs and 4 switches (DIP) allowing a simple dialogue between the user and the DSP. • Configurable initialization options (SW3). For numerical intensive algorithms, the TMS320C6416 board based on Very Long Instruction Word (VLIW) architecture is a good choice. Indeed, the material architecture of DSK board is shown in Fig. 2.

Fig. 2. DSK board architecture.

3.2

Rapid Prototyping Technology

In order to speed up the prototyping of the proposed algorithm on DSP processor and to avoid the development of complex mathematical functions in C/C++, we have explored the Embedded Target library from C6000 DSP Platform plus Real-time Workshop (RTW) [20]. For RTW, it has been dedicated to generate automatically a project in C language from a Simulink model. Also, a MATLAB Link has been used to interface MATLAB with code composer studio (CCS) Development Tools in order to provide an executable file (.out) which can be loaded into DSK-C6416 board. The flow diagram illustrating the connection between MATLAB-Simulink and RTW with DSK C6416 board is shown in Fig. 3. In this section, we explain the implementation principle on the DSK C6416 board. At first, we have modeled the proposed approach by MATLAB-SIMULINK, and then a C code has been generated by SIMULINK Coder interface. After that, this code has been automatically sent to CCS software which has made its compilation and execution to get finally an executable file ‘.out’ which has been loaded into the DSK C6416 board (Fig. 4). The Simulink modeling of DOST denoising system on TMS320C6416 board is shown in Fig. 5. The DSP Starter Kit (DSK) block, has a primordial importance by offering the access to diverse settings of processor hardware. The extraction of this block can be performed from C6000 Target Preference library.

82

S. Saoud et al.

Fig. 3. The flow diagram illustrating the connection between MATLAB-Simulink and RTW with DSK C6416 board.

Modeling of the proposed approach on SIMULINK

Compilation and execution of the proposed approach by CCS File.c

Loading Code on DSKC641 File.out

Fig. 4. Implementation steps on DSK C6416.

Figure 6 below depicts the different blocks of DOST system using Simulink that contain different functions which are extracted from Embedded MATLAB Function and C6000 library [21, 22]. The proposed model has been created using Embedded MATLAB Function in order to design DOST transform and Thresholding function. Also, elementary block sets from Simulink libraries has been used at the aim to decompose and reconstruct the speech signal entered from AIC23 codec, and then truncating the respective DOST coefficients. After that, a project is built and compiled in the Integrated Development Environment (IDE) of CCS via RTW. The obtained executable file (.out) is loaded in C6000-based target.

DSP Real-Time Implementation of DOST Algorithm

83

Fig. 5. Simulink modeling of DOST denoising approach on DSP board.

Fig. 6. Simulink modeling of DOST denoising approach.

4 Results of Real-Time Speech Enhancement Algorithm To evaluate the real-time performances of the proposed system, the number of cycles in Million Instructions per second (MCPS) and the memory consumption in Kilobytes (Ko) have been calculated as it is shown in Table 1. From the following Table 1, we observe that the overall required number of cycles (MCPS) for running the DOST algorithm is better using DSP/BIOS (9.21 MCPS) compared to other algorithms without DSP/BIOS. However, the obtained results are much better using DWT with DSP/BIOS, compared to other approaches (MCPS = 7.82 and memory consumption = 215.92). Thus, the CPU speed (1 GHz) and the memory consumption (the flash memory is of 512 KB and the SRAM is of 16 MB) of TMS320C6416 board meet up the real-time requirements.

84

S. Saoud et al. Table 1. Real-time performances of the proposed algorithm

Methods DOST denoising with DSP/BIOS DOST denoising without DSP/BIOS DWT denoising with DSP/BIOS DWT denoising without DSP/BIOS

Number of cycles (MCPS) 9.21 21.74

Memory consumption (Ko) 275 244

7.82 30.96

215.92 215.25

3 2.5 2 1.5 1 0.5 0

DWT

Airport

Babble SNR(dB)

10dB

5dB

0dB

10dB

0dB 5dB

DOST

0dB 5dB 10dB

PESQ Score(dB)

To examine the performances of our implemented method, we have conducted 2 tests using PESQ (Perceptual evaluation speech quality) [23] and the time domain waveforms as objective criteria test. For subjective evaluation, we have used the Mean Opinion Score (MOS). Also, in all tests, we have used the speech signal extracted from noisy speech corpus (NOIZEUS) database [24] and a various additives noises (Airport, babble, car) at 0 dB, 5 dB and 10 dB was used to evaluate the denoising algorithm. The results of the proposed method shown in Figs. 7, 8 and 9 are compared with the DWT denoising using db10 at 3 level and soft thresholding. It’s clear from Fig. 7 below, that our real time DOST denoising improves the PESQ score more than real time DWT denoising.

Car

Fig. 7. PESQ score evolution.

Figure 8 shows the time domain waveforms of the clean signal, the noisy signal and the enhanced signal respectively when the speech is corrupted by airport noise. We observe that the proposed algorithm reduces noise from the speech signal. A subjective test in terms of MOS of the speech enhancement systems implemented in the DSK C6416 board is presented in this section.

DSP Real-Time Implementation of DOST Algorithm

85

3.5 3 2.5 2 1.5 1 0.5 0

DOST

Airport

Babble

Car

Fig. 9. MOS score evolution.

10 dB

5 dB

0 dB

10 dB

5 dB

0 dB

5dB

10dB

DWT 0dB

MOS

Fig. 8. (a) Original speech signal. (b) Speech signal degraded by airport noise at 5 dB. (c) Enhanced signal by DOST algorithm.

86

S. Saoud et al.

For an accurate evaluation, the MOS (Mean Opinion Score) is considered the most used evaluation method from which 7 volunteers listen to each audio file and give their note concerning the perceived quality of the enhanced signals on the basis of a rating scale. The following Table 2 gives a rating scale for MOS test. Table 2. Rating scale for MOS subjective test. Rating 5 4 3 2 1

Label Excellent Good Fair Poor Bad

For this test, we have evaluated two implemented algorithms in noisy conditions (speech is corrupted by different types of noise such as: car, airport, babble) with different SNR levels), and then calculating their MOS scores as it is shown in figure below. We observe that the DOST denoising approach has the best intelligibility in terms of MOS in different noisy environments (car, airport, babble) and at different SNR levels (0 dB, 5 dB and 10 dB) compared to DWT algorithm.

5 Conclusion In this paper, the real time implementation for single channel speech denoising in a noisy environment has been presented. We have exploited the DOST to optimize the performances of speech enhancement in terms of quality and intelligibility. Indeed, a comparative study with DWT approach has been carried out and has shown that the proposed algorithm has outperformed DWT algorithm in terms of speech intelligibility with preserving the speech quality. In real-time test, the DOST approach has been implemented in TMS320C6416 platform and has led to reduce significantly the system complexity especially when DSP/BIOS is utilized. In order to investigate from the performances of the proposed approach, we suggest testing our system on other electronic architectures, such as FPGA, and Raspberry Pi3. Also, we tend to extend the domain application of the proposed approach into diverse field applications such as image denoising.

References 1. Chabane, B., Daoued, B.: On the use of Kalman filter for enhancing speech corrupted by colored noise. WSEAS Trans. Sig. Process. 4, 657–666 (2008) 2. Sreenivas, T.V., Kirnapure, P.: Codebook constrained Wiener filtering for speech enhancement. IEEE Trans Speech Audio Process. 4, 383–389 (1996)

DSP Real-Time Implementation of DOST Algorithm

87

3. Boll, S.: Suppression of acoustic noise in speech using spectral subtraction. IEEE Sig. Process. 27(2), 113–120 (1979) 4. Hassen, F.S.: Performance of discrete wavelet transform (DWT) based speech denoising in impulsive and Gaussian noise. J. Eng Sustain. Dev. 10(2), 175–193 (2018) 5. Nasr, M.B., Talbi, M., Cherif, A.: Arabic speech recognition by bionic wavelet transform and MFCC using a multi layer perceptron. In: Proceedings of the SETIT’12, pp. 803–808 (2012) 6. Zhang, Y., Zhao, Y.: Real and imaginary modulation spectral subtraction for speech enhancement. J. Speech Commun. 55, 509–522 (2012) 7. Jensen, J., Hansen, J.H.L.: Speech enhancement using a constrained iterative sinusoidal model. IEEE Trans. Speech Audio Process. 9, 731–740 (2001) 8. Anderson, D.V., Clements, M.A.: Audio signal noise reduction using harmonic modeling. In: Proceedings of the IEEE International Conference on Acoustics. ICASSP (1999) 9. Epharaim, Y.: A minimum mean square error approach for speech enhancement. In: Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (1990) 10. Dash, T.K., Solanki, S.S.: Comparative study of speech enhancement algorithms and their effect on speech intelligibility. In: 2nd International Conference on Communication and Electronics Systems (ICCES). IEEE (2017) 11. Paliwal, K.K., Basu, A.: A speech enhancement method based on Kalman filtering. In: Proceedings of ICASSP’87, pp. 177–180, Dallas, TX, USA (1987) 12. Parchami, M., Zhu, W.P., Champagne, B., Plourde, E.: Recent developments in speech enhancement in the short-time Fourier transform domain. IEEE Circ. Syst. Mag. 16(3), 45– 77 (2016) 13. Wang, Y., Orchard, J.: On the use of the Stockwell transform for image compression. In: SPIE Electronic Imaging Algorithms System VII, p. 7245 (2009) 14. Wójcicki, K., Milacic, M., Stark, A., Lyons, J., Paliwal, K.: Exploiting conjugate symmetry of the short-time Fourier spectrum for speech enhancement. IEEE Sig. Process. Lett. 15, 461–464 (2008) 15. Stark, A.P., Wójcicki, K.K., Lyons, J.G., Paliwal, K.K.: Noise driven short-time phase spectrum compensation procedure for speech enhancement. In: Inter Speech, pp. 549–552, September 2008 16. Samui, S., Chakrabarti, I., Ghosh, S.K.: Improved single channel phase-aware speech enhancement technique for low signal to- noise ratio signal. IET Sig. Process. 10(6), 641– 650 (2016) 17. Stockwell, R.G.: A basis for efficient representation of the S-transform. Digital Sig. Process. 17(1), 371–393 (2007) 18. Yan, Y., Zhu, H.: The generalization of discrete Stockwell transforms. In: EURASIP, pp. 1209–1213 (2011) 19. Huang, H., Sun, F., Babyn, P., Zhou, Z., Wang, L.: Medical-image denoising and compressing using discrete orthonormal S transform. In: 2nd International Conference on Electrical, computer Engineering and Electronics (ICECEE 2015), vol. 291, pp. 291–296. ICECEE (2015) 20. Texas instruments: TMS320 DSP/BIOS v5. 42 users guide. -01-20(2010) 21. Math Works: Real-time workshop for use with SIMULINK, user’s guide. Version 6, June 2004 22. Texas instruments: TMS320 DSP/BIOS. v5.42, User Guide, spru423I, Août (2012) 23. Hu, Y., Loizou, F.C.: Perceptual evaluation of speech quality (PESQ), and objective method for end-to-end of speech quality assessment of narrowband telephone network and speech codecs. ITUT Recommendation, p. 862. ITU (2000)

88

S. Saoud et al.

24. Hu, Y., Loizou, P.: NOIZEUS: a noisy speech corpus for evaluation of speech enhancement algorithms (2005) 25. Issaoui, H., Bouzid, A., Elloouze, N.: Comparison between soft and hard thresholding on selected intrinsic mode selection. In: Proceedings of SETIT’12, pp. 712–715 (2012) 26. Talbi, M., et al.: Speech enhancement with bionic wavelet transform and recurrent neural network. In: 5th International Conference: Sciences of Electronic, Technologies of Information and Telecommunications SETIT (2009)

An Enhanced SDPE Method for Long Delay Multipath Mitigation in GNSS Applications Wafa Feneniche(&), Khaled Rouabah, Mustapha Flissi, Salim Atia, Sabrina Meguellati, and Salah Eddine Mezaache Electronics Department, ETA Laboratory, Mohamed El Bachir El-Ibrahimi University of Bordj Bou Arreridj, 34031 Bordj Bou Arreridj, Algeria [email protected]

Abstract. In Global Navigation Satellite System (GNSS) applications, the Binary Offset Carrier (BOC) modulation provides superior benefits vis-a-vis the classical Binary Phase Shift Keying (BPSK) one. Nevertheless, this modulation type presents a major drawback due to the presence of secondary peaks in envelope of the received signal Autocorrelation Function (ACF). This is due to the fact that the ACF side peaks may create false lock points in the Discrimination Function (DF) envelope of the Delay Locked Loop (DLL) structure, producing ambiguity in both acquisition and tracking processes. In the present paper, the Strobe Double Phase Estimator (SDPE) technique for multipath (MP) effect reduction is used in BOC modulated signals acquisition and its performance is analyzed and then enhanced by applying a proposed scheme that is efficient and valid for all BOC modulated signals. The suggested method is based on the use of the SDPE approach for coherent configuration in conjunction with the same approach for non-coherent configuration. As result, the ACF resulting from this combination is unambiguous. Besides, a large evaluative comparison between the new algorithm and the SDPE technique is achieved together with a performance evaluation of the proposed method in MP environment based on the Root Mean Square Error (RMSE) criterion. The obtained results have shown that our proposed scheme presents better performances in the presence of MP. Keywords: GNSS
BOC
BPSK
ACF
DF
Ambiguity
SDPE
RMSE Notations and Abbreviations

SDPE: MP: BOC: ACF: RMSE: GPS: GNSS: DF: DLL: SPLL: DPE:

Strobe Double Phase Estimator Multipath Binary Offset Carrier Autocorrelation Function Root Mean Square Error American Global Positioning System Global Navigation Satellite Systems Discrimination Function Delay Locked Loop Subcarrier Phase Locked Loop Double Phase Estimator

© Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 89–98, 2020. https://doi.org/10.1007/978-3-030-21009-0_8

90

CF: PLL: N: HRC: LOS:

W. Feneniche et al.

Correlation Function Phase Locked Loop Modulation order High Resolution Correlator Line of Sight

1 Introduction Positioning systems have become part of the history of telecommunications’ technology since the appearance of the Global Positioning System (GPS) representing the American GNSS. The GPS satellites transmit the BPSK modulated signals with spread spectrum, specifically designed for the litter measurement [1, 2]. These signals are continually being improved in order to allow a better accuracy in measurements and a more efficient use of the transmission channel. Europeans aim to develop their own navigation system, the Galileo system [2], to be used with their private satellites. At first, this system was confronted with the challenging issue of interoperability with other GNSS systems (GPS, Glonass, …) [2]. The solution to this problem emerged from the development of the BOC modulated signals [2], which would permit efficient transmission channel and a rigorous sharing of the allowed frequency band with other navigation systems [2]. This bandwidth efficiency is achieved by displacing the energy of the signal away from the middle of the frequency band, thus providing a higher spectral separation degree between BOC modulated signals and other GNSS signals [3, 4]. However, these new signals of this type suffer from the presence of secondary peaks in the ACF and thus the appearance of false zero crossings points in the DF [5, 6]. Lots of solutions have been proposed by researchers attempting to address these problems in the most unfavorable transmission cases, especially in the presence of noise [1] and MPs [1–8]. Some of them can be easily affected by the MPs that have a long delay (between 0.5 and 1 chips) [9, 10], some others can manage to stay valid as long as the MP signals are absent and the rest of these techniques have a low elimination capability of secondary peaks and false lock points [11]. Therefore, the ambiguity is always present because of the lack of robustness of phase estimation process in degraded environment [11–14]. In order to avoid the negative impact of the MPs phenomenon, a recent method, called SDPE, is proposed in [15]. This method effectively exploits the subcarrier as the actual ingredient that distinguishes BOC modulated signals from BPSK modulated signals [15]. Its principle is to generate two strobe reference waveforms that replace the usual subcarrier that is tracked using the Subcarrier Phase Locked Loop (SPLL) in the Double Phase Estimator (DPE) technique [15]. The SDPE technique mitigates the biased phase error induced by those MPs whose delays exceed the strobe gate size [15]. Nevertheless, the appearance of some secondary peaks in the SDPE CF as well as the appearance of some false zero crossing points in the SDPE DF, trigger new challenges for the acquisition and tracking GNSS receiver processes [15]. Herein, we suggest a new unambiguous method valid for all BOC signals. It is based on the combination of the coherent and non-coherent configurations of the SDPE

An Enhanced SDPE Method for Long Delay Multipath Mitigation

91

structure CFs. The proposed technique can completely remove any side-peaks and any false lock points from the CF and the DF of BOC modulated signals, respectively. Besides, it keeps the sharp main-peak of the SDPE CF and the same slope level of the SDPE DF, which maintains, thus, the MPs rejection property of the SDPE. Furthermore, the simulation results, based on Matlab, show that the proposed method presents a better performance than the conventional SDPE technique and the classical one regarding MPs. The rest of this paper is structured as follows. In Sect. 2, a short description of BOC signals is introduced. In Sect. 3, the basic concept of SDPE is presented. Section 4 describes the principle of the proposed technique. The performance evaluation results of the proposed method and the SDPE method are compared and discussed in Sect. 5. In Sect. 6, we end up with conclusions.

2 BOC Signals In GNSS, the basic continuous time varying BOC modulated signal “SðtÞ” is defined by [16]: SðtÞ ¼ DðtÞ:CðtÞ:Sp ðtÞ:PðtÞ Where: DðtÞ: is CðtÞ: is Sp ðtÞ: is PðtÞ: is

ð1Þ

the data signal of frequency f d ; the spreading code of frequency f c ; the subcarrier signal of frequency f sc ; carrier signal of frequency f p

The BOC signal is usually noted by BOCðf sc ; f c Þ. The “f sc ” and “f c ” frequencies are multiples of the reference frequency “fr = 1.023 MHz”. They are given by f sc ¼ m:f r and f c ¼ n:f r , respectively, where “m” and “n” are positive integers [5]. 2.1

Correlation and Discrimination Functions

The number of secondary peaks and false crossing points presented in the BOC ACF and the BOC DF are in particular dependent on the modulation order denoted by NðN ¼ 2f sc =f c Þ [2]. The normalized BOC ACF and DF for ideal BOC (f sc ; f c ) signals with different values of N are presented respectively in the left and right of Fig. 1. This figure shows that the increase of N allows to obtain a sharp main-peak and a high-level slope in the BOC ACF and the BOC DF, respectively, which maintains good performances. However, it produces N − 1 different alternated couples of symmetric secondary peaks [2], which produces, as shown in the same figure, a significant number of zero crossing points in the BOC DF. Hence, in GNSS receiver, the DLL locking can be done on any secondary peak, which causes strong ambiguities nearby the central peak of the BOC ACF.

92

W. Feneniche et al. BOC(1,1) BOC(14,2)

1

0.5

0.5

Normalized DFs

Normalized ACFs

1

0

-0.5

0

-0.5

-1

-1 -1

-0.5 0 0.5 Time Delay in "Chips"

1

-1

-0.5 0 0.5 Time Delay in "Chips"

1

Fig. 1. Normalized ACFs and DFs for BOC (f sc ; f c ) modulated signals (N = 2,14).

3 Strobe Double Phase Estimator The SDPE technique is a new design of the BOC subcarrier that mitigates the MPs effects [15]. In the SDPE, the code and the subcarrier are tracked separately using three loops, which are the standard DLL, the standard Phased Locked Loop (PLL) and the SPLL. In the GNSS receiver, both of the proposed subcarriers WG1 ðtÞ and WG2 ðtÞ are formed, respectively, by the principal strobe waveforms G1 ðtÞ or G2 ðtÞ; which can be expressed by the following equations [15]: G1 ðtÞ ¼ G2 ðtÞ ¼

X1 i¼0

X1 i¼0

gðt  iTc Þci ðtÞ

ð2Þ

gðt  iTc ÞcðtÞ

ð3Þ

Where: gð t Þ ¼

Xm þ n j¼1

xj pðt  jlÞ

ð4Þ

Here, xj represents the sequence value of the jth strobe. l is the width of the gate ðl ¼ Tc =4Þ. The waveforms G1 ðtÞ and G2 ðtÞ are found near each chip of the code [15]. ci ðtÞ represents the ith PRN code. The two reference waveforms WG1 ðtÞ and WG2 ðtÞ used within the SPLL correlation can be described as follows [15]: WG1 ðtÞ ¼ G1 ðtÞ

ð5Þ

1 WG2 ðtÞ ¼ G2 ðtÞ þ G2 ðt  Tc Þ 2

ð6Þ

An Enhanced SDPE Method for Long Delay Multipath Mitigation

3.1

93

SDPE Correlation and Discrimination Functions

The simulation results of the SDPE CF and DF for both reference waveforms WG1 ðtÞ and WG2 ðtÞ are presented, respectively, in the left and the right of Fig. 2, where BOC (14,2) signal is considered as example. 5

5

x 10

x 10 WG 1 WG

2

2 DFs Amplitudes

CFs Amplitudes

2

1 0

-1 -2

1 0 -1 -2

-0.5

0 0.5 Time Delay in "Chips"

-0.5

0 0.5 Time Delay in "Chips"

Fig. 2. SDPE CFs and DFs for BOC (14,2) signal (WG1 ðtÞ and WG2 ðtÞ).

This figure shows that both SDPE CFs present a sharp main peak and both SDPE DFs have a high slope. Furthermore, it can be seen that a side peak and a false lock point appear both at -Tc time delay in each of the SDPE CFs and each of the SDPE DFs, respectively, which may cause an additional MP error with a same delay Tc [15]. However, it can be noticed that the resulting CFs are similar to those of HighResolution Correlator (HRC) structures.

4 Principle of the Proposed Method To overcome the limitation of all BOC signals in terms of side peaks and false lock points effects, we propose an unambiguous method based on the combination of coherent and non-coherent SDPE approach and characterized by the suggested unambiguous CF, which is expressed by the following equation: RP ðsÞ ¼ RSDPE ðsÞ þ ðRSDPE ðsÞÞ2

ð7Þ

Where, RP ðsÞ and RSDPE ðsÞ are the proposed and the traditional SDPE CFs, respectively. Note that the principle of our proposed method, shown in Fig. 3, is valid for both classical SDPE waveforms. The new obtained CFs and DFs are illustrated, respectively, in the left and the right of Fig. 4.

94

W. Feneniche et al.

NC-SDPE True CFs Amplitudes

SDPE Enhanced SDPE

False

Fig. 3. Principle of the proposed method. x 105

x 105 5

W G1 W G2 DFs Amplitudes

CFs Amplitudes

5

0

-5

0

-5 -0.5 0 0.5 Time Delay in "Chips"

-0.5 0 0.5 Time Delay in "Chips"

Fig. 4. Proposed CFs and DFs for BOC (14,2) signal (WG1 ðtÞ and WG2 ðtÞ).

As can be seen in Fig. 4, the proposed scheme completely removes all side peaks and all false lock points of the SDPE CFs and DFs, respectively. Furthermore, the proposed correlations keep the sharpness of the main peak of the SDPE CFs, which insures a high loop gain for the discriminator.

An Enhanced SDPE Method for Long Delay Multipath Mitigation

95

5 Simulation Results In order to illustrate the performance of the enhanced SDPE method in MP environment, both SDPE and enhanced SDPE schemes were simulated. It is interesting to describe the bias on the characterization of the zero-crossing point corresponding to the traditional SDPE scheme and that corresponding to our proposed method. For this purpose, the first scenario is established considering two noiseless components of the received signal, namely: the Line Of Sight (LOS) signal and an MP signal of amplitude 0.5 with respect to the LOS. Here, we consider four different values of the MP delay with respect to the LOS: 0.1, 0.3, 0.6 and 0.9. The DLL DF outputs for both structures WG1 ðtÞ and WG2 ðtÞ are shown in Figs. 5 and 6 respectively. As illustrated in these two figures, while our proposed method suffers only from the presence of an additional triangular curve characterizing the MP effect, the traditional scheme, in addition to this same problem, it exhibits other curves that provoke the apparition of other side zero crossings, which causes more ambiguity in the receiver positioning. SDPE Enhanced SDPE

1

1

0.5

0.5 0

Normalzed DF

0

-0.5

-0.5 -1 -1

-1 -0.5

0

0.5

1

-1.5 -1

1

1

0.5

0.5

0

0

-0.5

-0.5

-0.5

0

0.5

1

-1

-1 -1

-0.5

0

0.5

1

1.5 -1 0 Time Delay in "Chips"

1

2

Fig. 5. Enhanced SDPE DF (WG1 ðtÞ) in presence of MP.

In the second scenario, the tests are realized for a noisy signal corresponding to WG2 ðtÞ waveform. Here, a precorrelation bandwidth of 24 MHz is used to estimate the Root Mean Square Error (RMSE) of the MP errors for both structures. The errors values are found by defining the zero crossing points belonging respectively to the SDPE and its enhanced SDPE version. The MP signal is chosen with an amplitude equal to 0.5 and a delay, with respect to the LOS, in the interval from 0 to 300 m [17, 18]. When the MP signal is in phase (i.e. 0°) with respect to the LOS, the considered MP errors are those at the maximum points. The RMSEs of both structures versus MP delay are given in Fig. 7. As shown in this figure, our method, although sensitive to short delay MP, shows the best

96

W. Feneniche et al.

1

1

0.5

0.5 0

0

-0.5

-0.5 Normalzed DF

SDPE Enhanced SDPE

-1

-1

-1.5 -1

-0.5

0

0.5

-1

1

1

1

0.5

0.5

0

0

-0.5

-0.5

-1

-1

-1.5 -1 -0.5

0

0.5

1

-0.5

1.5 -1 Time Delay in "Chips"

0

0

0.5

1

1

1.5

2

Fig. 6. Enhanced SDPE DF (WG2 ðtÞ) in presence of MP. 16 SDPE Enhanced SDPE

14

RMSE in "Meters"

12 10 8 6 4 2 0

0

50

100 150 200 RelaƟve MP delay in "Meters"

250

300

Fig. 7. RMSE versus relative MP delay for SDPE and enhanced SDPE.

overall performance in the presence of MP. Besides, for long delay MPs, the code error envelope decays to zero. Therefore, it is clear that our proposed method outperforms the SDPE. Indeed, the error and its variation band are considerably attenuated. This is mainly due to the total absence of secondary peaks in the proposed combination.

An Enhanced SDPE Method for Long Delay Multipath Mitigation

97

6 Conclusion In this paper, we have suggested an efficient scheme for the reduction of MPs effect in GNSS applications. This method is based on combining the coherent and the noncoherent versions of the SDPE structure. The original SDPE, which offers a solution to the MP constraint, is unbalanced by its higher sensitivity to long delay MP signals due to the presence of secondary peaks. The obtained simulation results have shown that the enhanced SDPE scheme performance is better than those of the existing SDPE. This is mainly due to the suppression of the lateral peaks and the false look points in the proposed method CF and DF respectively.

References 1. Kaplan, E.D., Hegarty, C.J.: Understanding GPS: Principles and Applications, 2nd edn. Artech House, London (2006) 2. Betz, J.W.: Binary offset carrier modulations for radionavigation. Navigation 48(4), 227–246 (2001) 3. Feneniche, W., Rouabah, K., Attia, S., Flissi, M., Djendel, S., Saoudi, A., Chikouche, D.: Receiver Precorrelation bandwidth effects on BOCs/c-PRN structures. In: International Conference on Automatic Control, Telecommunications and Signals, pp. 1–6. Badji Mokhtar Annaba University, Algeria (2017) 4. Avila-Rodriguez, J.A.: On generalized signal waveforms for satellite navigation. Ph.D. Dissertation. University FAF Munich (2008) 5. Rouabah, K., Atia, S., Flissi, M., Bouhlel, M.S., Mezaache, S.: Efficient technique for DLL S-curve side zero-crossings cancellation in global positioning system/Galileo receiver. IET Sig. Process. 13(3), 338–347 (2019) 6. Nunes, F.D., Sousa, F.M.G., Leitao, J.M.N.: Gating functions for multipath mitigation in GNSS BOC signals. IEEE Trans. Aerosp. Electron. Syst. 43(3), 951–964 (2007) 7. Attia, S., Rouabah, K., Chikouche, D., et. al: Side peak cancellation method for sine-BOC (m,n) modulated GNSS signals. EURASIP J. Wireless Commun. Netw. 1(34), 1–14 (2014) 8. Parkinson, B.W., Spilker, J.J., Axelrad, P., Enge, P.: Global Positioning System: Theory and Applications, 1st edn. American Institute of Aeronautics and Astronautics, USA (1996) 9. Rouabah, K., Chikouche, D., Attia, S.: Evaluation des Erreurs de Poursuite de Code dans les Récepteurs C/A – GPS et BOC (1,1) – GALILEO en Présence des Multitrajets. In: Fourth International Conference: Sciences of Electronic, Technologies of Information and Telecommunications, p. 7. IEEE, Hammamet, Tunisia (2007) 10. Chung, L.C., Juang, J.C.: An adaptive multipath mitigation filter for GNSS applications. J. Adv. Sig. Process. 2008, 1–10 (2008). (Open Access Journal) 11. Rouabah, K., Flissi, M., Attia, S., Chikouche, D.: Unambiguous multipath mitigation technique for BOC(n,n) and MBOC-modulated GNSS signals. Int. J. Antennas Propag. 2012, 1–13 (2012). (Open Access Journal) 12. Rouabah, K., Attia, S., Flissi, M., medjdoub, I., Chikouche, D.: GNSS multipath mitigation using finite difference derivatives with five-point stencil. In: 6th International Conference on Sciences of Electronic, Technologies of Information and Telecommunications, pp. 690. IEEE, Hammamet, Tunisia (2012)

98

W. Feneniche et al.

13. Kheloufi, N., Lounis, M.: Errors sources elimination for accurate evaluation of displacement using global positioning system. In: 6th International Conference on Sciences of Electronic, Technologies of Information and Telecommunications, pp. 220–225. IEEE, Hammamet, Tunisia (2012) 14. Titouni, S., Rouabah, K., Flissi, M., Atia, S., Feneniche, W., Chikouche, D.: General analytical models characterizing multipath running average error for C/A-GPS and BOC (n, n) Galileo signals. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications, pp. 520–526. IEEE, Hammamet, Tunisia (2016) 15. Chengtao, X., Zhe, L., Xiaomei, T., Feixue, W.: Strobe double phase estimator: a multipath mitigating technique for BOC signal in GNSS based on double phase estimator: a multipath mitigating technique for GNSS BOC signal. Int. J. Satell. Commun. Netw. 35(3), 249–261 (2017) 16. Chebir, S., Aidel, S., Rouabah, K., Attia, S., Flissi, M.: GNSS signals acquisition and tracking in unfavorable environment. Radioengineering 27(2), 557–571 (2018) 17. Salem, M., Bouallègue, A., Jarboui, S.: Vector parameter estimation over flat Rayleigh fading channel: Cramer-Rao lower bound generalized expressions. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications, pp. 87–91. IEEE, Hammamet, Tunisia (2016) 18. Attalah, M.A., Laroussi, T., Aouane, A., Mehanaoui, A.: Adaptive filters for direct path and multipath interference cancellation: application to FM-RTL-SDR based Passive Bistatic Radar. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications, pp. 461–465. IEEE, Hammamet, Tunisia (2016)

Electronics

A Novel Operational Transconductance Amplifier Based an RLC-Low-Pass Filter Karima Garradhi1(&), Néjib Hassen1, Thouraya Ettaghzouti1, and Kamel Besbes1,2 1

Micro-electronics and Instrumentation Laboratory, University of Monastir, Monastir, Tunisia [email protected], {nejib.hassen, Kamel.besbes}@fsm.rnu.tn, [email protected] 2 Centre for Research on Microelectronics and Nanotechnology of Sousse, Technopole of Sousse, Sousse, Tunisia

Abstract. This paper presents a new OTA circuit based on bulk-driven technique. It is worked with ±0.6 V supply voltages and power consumption is 40.14 lW. In order to assess the proposed OTA pre-fabrication performances, post-layout have been performed with Cadence Virtuoso in TS18SL 0.18 µm technology Tower Jazz. The PLS of proposed OTA showed a gain of 54 dB, a wide linearity and it occupies a small effective area, where its dimensions are occupied only 38.34 µm
27.75 µm. Based on the proposed OTA, we analyzed an RLC Low-Pass filter which has a good perfect. Keywords: OTA

 Bulk-driven  RLC filter

1 Introduction An OTA is the most considerable analog block which become known in several applications such as multipliers [1, 2], current conveyor [3, 4], instrumentation amplifier [5] and universal filter [6, 7]. The expression of OTA is given by: Iout ¼ I2  I1 ¼ ðVin þ  Vin ÞGm

ð1Þ

However, operational transconductance amplifier requires constant gm circuits and good linearity. In this case, many research works are carried out to achieve these requirements. Therefore, there are several techniques have been developed, such as crossing-coupling [8], floating gate [9], source degeneration [10] and class-AB [11]. In this work, we used the bulk-driven technique in OTA circuit. A diversity of RLC filter circuits using active elements such that current conveyor (CC) [12, 13], an operational amplifier, operational transconductance amplifier (OTA), have affected important attention. This paper is organized as follows: new general operational transconductance amplifier using bulk driven technique is presented in Sect. 2. Post-layout simulations are proposed and discussed in Sect. 3. The implementation of floating inductance and floating resistance based on OTA is addressed in Sect. 4. RLC Low-Pass filter using as © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 101–110, 2020. https://doi.org/10.1007/978-3-030-21009-0_9

102

K. Garradhi et al.

anactive element floating inductance and floating resistance is implemented in Sect. 5. At the end of this paper, the conclusion is presented in Sect. 6.

2 Proposed OTA 2.1

Description of Circuit Operation

The Fig. 1 presented the proposed OTA circuit using bulk driven [14].

Fig. 1. Proposed OTA circuit using bulk driven

The expression of the threshold voltage is given by: pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi VTH ¼ VTH0 þ cn ð j2/F j  VBS  j2/F j

ð2Þ

The currents I1 and I2 are given as follows: h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffii2   2Id1 ¼ ISS  Iout ¼ bn VG  VS  VTH0  cn ð3Þ j2/F j  Vcm þ VS þ Vid =2   2/F  h pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffii2 2Id2 ¼ ISS  Iout ¼ bn VG  VS  VTH0  cn j2/F j  Vcm þ VS  Vid =2  j2/F j ð4Þ

Hence, Iout is given by

A Novel Operational Transconductance Amplifier

Iout

Iout

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi !2 u pffiffiffiffiffiffiffiffiffiffi u cn bn ISS bn Vid2 c n pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Vid t1  4ISS 2 j2/F j  Vcm þ VS 2 j2/F j  Vcm þ VS

pffiffiffiffiffiffiffiffiffiffi cn bn ISS 1 pffiffiffiffiffiffiffiffiffiffi bn p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi bn ISS ffi Vid  8 ISS 2 j2/F j  Vcm þ VS

cn pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 j2/F j  Vcm þ VS

103

ð5Þ

!3 Vid3

ð6Þ

Based Eq. 5, the Gm expressed as: pffiffiffiffiffiffiffiffiffiffi cn bn ISS Gm ffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2 j2/F j  Vcm þ VS

ð7Þ

3 Post-layout Simulation Results of OTA Post-Layout simulations of proposed OTA circuit are verified by Cadence Tower Jazz technology 0.18 µm TS18SL. The transistors aspect ratios are presented in Table 1. Table 1. Transistor aspect ratios of the proposed OTAs Transistors M1, M2 M3, M4, M5, M6 M3a, M4a, M5a, M6a M9, M10, M9a, M10a Mc

W (lm)/L(lm) 10/5 10/1 20/5

Figure 2 showed the layout of the proposed OTA. It occupies a small effective area, where its dimensions are (38, 34 µm
27, 75 µm). Figure 3 permit to compare the simulation and post-layout simulation DC transfer characteristic according to the input voltage from ±1.2 V. Hence, it showed a good linearity about ±1 V. Figure 4 compared the schematic simulation and PLS results of transconductance Gm according to the differential input voltage. As a result, it has a large linearity in ±0.8 V and presents a maximum value of 5.33 µS. Figures 5 and 6 respectively described the simulation and PLS of frequency response and phase margin of OTA. It achieved a gain at 57 dB, a phase margin of 63° and GBW at 4 MHz. It can be noted in these figures that post-layout simulations and schematics simulations of the proposed OTA are nearly the same. From Table 2, we summarized the simulation characteristics of OTA using the bulk driven technique with certain of the late works.

104

K. Garradhi et al.

Fig. 2. A layout of the proposed OTA circuit using bulk driven

Fig. 3. Simulation and post-layout simulation of DC transfer characteristics

Fig. 4. Simulation and post-layout simulation of transconductance

A Novel Operational Transconductance Amplifier

105

Fig. 5. Simulation and post-layout simulation of open loop frequency response

Fig. 6. Simulation and post-layout simulation of phase margin

4 Implementing Floating Inductance and Resistance Using one grounded capacitor and four proposed OTAs, floating active inductor can be implemented as shown in Fig. 7. The voltage across active floating inductance is: VAB ¼ ðVA  VB Þ

ð8Þ

The expressions of output currents across OTA1 and OTA3 are given by: I1 ¼ gm1 VA and I3 ¼ gm3 VB

ð9Þ

The output voltage V of OTA1 and OTA3 is connected by the grounded capacitor C1 is:

106

K. Garradhi et al.

Table 2. Comparison table between proposed OTA and others reported in the literature Performance parameters Technology CMOS (µm) Supply voltage (V) Power consumption (µW) DC gain (dB) Phase margin (deg) GBW (MHz) Transconductance (lS) Linear range (V) PSRR+ (dB) PSRR− (dB) CMRR (dB) Input Noise Density (µV/√Hz)@frequency (MHz) Positive slew rate (V/lS) Negative slew rate (V/lS) Cap. load (pF)

Simulation results Post-layout simulation [18] [19] 0.18 0.18 0.18 0.18 ±0.6 ± 0.6 0.6 0.7 40.14 38.2 0.4 130 57 54 82 56 63 63 60 60 4 4 0.01 3.2 5.33 4.99 ±1.2 ±1.2 158 154 140 139 144 141 114 80 0.8 0.7 0.15 1 1 0.1 15.24 14.95 5.6 −17.16 −16.88 5.6 1 1 15

[20] 0.18 0.5 0.06 70.4 54 0.09

70 70 106

0.96 0.96 30

Fig. 7. Realization of floating inductance using four OTAs

V¼

I1 þ I3 sC1

ð10Þ

The expressions of output currents across OTA2 and OTA4 are given by: I2 ¼ gm2 V and I4 ¼ gm4 V

ð11Þ

For superior performance, the transconductance of the OTA1, OTA2, OTA3 and OTA4 are identical (gm1 = gm2 = gm3 = gm4 = gm). After substituting V from Eq. (15) in Eq. (16), we found: I2 ¼

g2m ðVB  VA Þ g2 ðVA  VB Þ and I4 ¼ m sC1 sC1

ð12Þ

A Novel Operational Transconductance Amplifier

107

Therefore, the impedance ZAB is expressed as: ZAB ¼

ðVA  VB Þ ðVA  VB Þ sC1 ¼ ¼ 2 ¼ sL IA IB gm

ð13Þ

Consequently, the expression for inductance can be expressed as: L¼

C1 g2m

ð14Þ

Fig. 8. Floating resistor using proposed OTA

The implementation of floating resistance based two OTA is shown in Fig. 8. The equivalent resistance is given by the inverse of transconductance (gm1 = gm2 = gm) and can be calculated as follows: R¼

1 gm

ð15Þ

5 An RLC Filter The proposed RLC low-pass filter is presented in Fig. 9. It is composed of one floating resistance, one floating inductor and one grounded capacitor. The implementations of resistance and inductor are realized by OTA which has been explained in the previous paragraphs. The transfer function of the RLC filter is given as follows: 1 Vout ¼ 2 RLC Vin s þ Lsþ

1 LC

ð16Þ

108

K. Garradhi et al.

Fig. 9. RLC low-pass filter: (a) Passive elements (b) Active elements

Consequently, the pole frequency (x0) and the pole quality factor (Q0) of this transfer function can be given as follows: 1 gm 1 x0 ¼ pffiffiffiffiffiffi ¼ pffiffiffiffiffiffiffiffiffi and Q ¼ R CC1 LC

rffiffiffiffi rffiffiffiffiffiffi L C1 ¼ C C

ð17Þ

Fig. 10. Frequency response of RLC low-pass filter

The workability of proposed RLC low-pass filter has been demonstrated by Cadence Tower Jazz technology 0.18 µm TS18SL. To test the functioning of our filter, we set C = C1 = 1pF (gm1 = gm2 = gm3 = gm5 = gm6 = 5. 33µS, R = 0.18 MΩ and L = 35mH). The result of the frequency response is demonstrated in Fig. 10. The cut-off frequency is 600.4 kHz. The result of Monte-Carlo analysis for the bandwidth of this filter is depicted in Fig. 11. The variation incidence of this filter is 9.49%.

A Novel Operational Transconductance Amplifier

109

Fig. 11. Simulation results of Monte Carlo analysis for RLC low-pass filter

6 Conclusion A new proposed OTA using bulk driven is introduced. It achieved a high linearity, a gain of 54 dB. Based on this later, a floating resistor, floating inductance, and low pass RLC filter are implemented. This filter showed a good acceptance.

References 1. Nandini, A.S., Madhavan, S., Sharma, C.: Design and implementation of the analog multiplier with improved linearity. Int. J. VLSI Design Commun. Syst. (VLSICS) 3(5), 93 (2012) 2. Kaewdang, K., Surakampontorn, W.: On the realization of electronically current-tunable CMOS OTA. Int. J. Electron. Commun. (AEU) 61, 300–306 (2007) 3. Rani, R., Rai, P., Athiya, J.: Design and analysis of second generation current conveyor based low power operational transconductance amplifier. Int. J. Electr. Eng. Technol. (IJEET) 6(4), 09–15 (2015) 4. Senani, R.: Novel circuit implementation of current conveyers using an OA an OTA. Electron. Lett. 16, 2–3 (1980) 5. Jain, S.: Design high CMRR, high slew rate instrumentation amplifier using OTA and CDTA for biomedical application. Int. J. Eng. Res. 2(5), 332–336 (2013) 6. Garradhi, K., Hassen, N., Besbes, K.: Low-voltage and Low-power OTA using source degeneration technique and its application in Gm-C filter. In: Conference, IDT, pp. 221–226. IEEE (2016) 7. Garradhi, K., Hassen, N., Ettaghzouti, T., Besbes, K.: Realization of current mode biquadratic filter employing multiple output OTAs and MO-CCII. Int. J. Electron. Commun. (AEÜ) 83, 168–179 (2018) 8. Chen, J., Sanchez-Sinencio, E.: Frequency-dependent harmonic-distortion analysis of a linearized cross coupled CMOS OTA and its application to OTA-C filters. IEEE Trans. Circuit Syst. 53, 499–510 (2006) 9. Alsibai, Z.: Floating-gate operational transconductance amplifier. Int. J. Inf. Electron. Eng. 3, 4 (2013) 10. Baruqui, F.A.P., Petraglia, A.: Linearly tunable CMOS OTA with constant dynamic range using source degenerated current mirrors. IEEE Trans. Circuits Syst. 53, 791–801 (2006)

110

K. Garradhi et al.

11. López-Martín, A.J., Baswa, S., Ramirez-Angulo, J.: Low-voltage super class AB CMOS OTA cells with very high slew rate and power efficiency. IEEE J. Solid-State Circuits 40(5), 1068–1077 (2005) 12. Ettaghzouti, T., Hassen, N., Besbes, K.: High performance low voltage low power voltage mode analog multiplier circuit. In: Conference 2016, SETIT, pp. 527–531. IEEE (2016) 13. Ettaghzouti, T., Hassen, N., Besbes, K.: SIMO type mixed mode biquadratic filter using second generation current conveyor circuits. In: Conference 2016, SETIT, pp. 539–543. IEEE (2016) 14. Garradhi, K., Hassen, N., Ettaghzouti, T., Besbes, K.: A low voltage and low power OTA using bulk-driven technique and its application in Gm-C filter. In: Conference 2018, SSD, IEEE (2018) 15. Akbari, M., Hashemipour, O.: A 0.6 V, 0.4uW bulk driven operational amplifier with rail to rail input/output swing. Analog Integr. Circ. Sig. Process 86, 341–351 (2016) 16. Raikos, G., Vlassis, S.: 0.8 V bulk-driven operational amplifier. Analog Integr. Circ. Sig. Process 63, 425–432 (2010) 17. Sharan, T., Bhadauria, V.: Fully differential bulk driven class AB sub-threshold OTA with enhanced slew rates and gain. J. Circuits Syst. Comput. 26(1), 1750001 (2017)

Accurate High Level Resources and Power Estimators for FPGAs Sonia Mami1,2 , Younes Lahbib2,3 , and Yassine Hacha¨ıchi3,4(B) 1

4

Facult´e des Sciences de Tunis, Universit´e de Tunis El Manar, Tunis, Tunisia [email protected] 2 Research Laboratory LAPER UR-17-ES11, Universit´e de Tunis El Manar, Tunis, Tunisia {younes.lahbib,Yassine.Hachaichi}@enicarthage.rnu.tn 3 ENICarthage, Universit´e de Carthage, Tunis, Tunisia Research Laboratory Smart Electricity and ICT, SEICT, LR18ES44, National Engineering School of Carthage, University of Carthage, Tunis, Tunisia

Abstract. In this paper, we propose a rapid and high level resource estimator for Field-Programmable Gate Arrays (FPGAs). The design is represented by its data flow graph. The proposed approach provides an estimation of the number of slices and Digital Signal Processors (DSPs). We also propose a power estimation based on the aformentioned resource estimation and the spreadsheet of Xilinx Xpower Estimator (XPE). The method is tested on a set of benchmarks. The results are very satisfactory. The average error reached respectively 0%, 2.64% and 9.14% for the DSP, the slice and the power estimation.

Keywords: High level estimation

1

· DSP · DFG · FPGA · Inference

Introduction

In the recent years, the need for high level resource and power estimation increased significantly. In fact, designers have to know, early in the design process, how the described design is going to perform in order to choose the best way to implement it. Moreover, with the evolution of high level code generators [1] and the constant need to optimize its flow, the area and power estimation becomes a key milestone in the design space exploration (DSE). Many estimation tools have been proposed to respond to this need. In [2], authors presented a statistical modeling approach able to generate hardware prediction models targeting different application domains. The proposed tool takes as input a HLL (High Level Language) namely the C language. Authors based their model on neural network and linear regression. The reported results show a low accuracy with an average error reaching 23.8%. In [3], an empirical area model is proposed. This model targets several components such as shifters, add/Sub operators and multipliers. The tool takes as input a HLL namely Matlab and generates estimation models relative to specific c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 111–123, 2020. https://doi.org/10.1007/978-3-030-21009-0_10

112

S. Mami et al.

IP cores. Authors based their work on linear regression. The estimation error was about 8.2% according to [2]. The Achilles heel of this approach is that it doesn’t take into consideration the optimizations performed by the synthesis tool. In [4], an area and time estimation model for the control part of the design was proposed. The reported error was about 10.3%. In [5], authors proposed a prediction model based on linear regression. The input of their tool is the C language. The reported error in area estimation reached 31%. In [6], a method of resource estimation in Field-Programmable Gate Array (FPGA) is proposed. The proposed approach takes as input the RTL representation of the application and generates an estimation which doesn’t take into consideration the optimizations performed by the synthesis tool. It only models the steps that the synthesis tool is expected to perform. This work targets only some families of Xilinx FPGAs namely Spartan3, Virtex4 and Virtex5. The reported error is about 20%. In [7], authors proposed a rapid estimation of DSP utilization for efficient high level synthesis. They state that high level synthesis tools don’t take into consideration the impact of optimizations done by the backend synthesis tool. Therefore, they present a more accurate model to estimate DSP numbers, however, they didn’t generalize their studies and didn’t apply their method on any benchmark, they just estimate the number of DSP blocks while varying the bitwidth of multiplication’s inputs. In this work, we propose an accurate high level resources estimation method which takes into account the optimizations done by the synthesis tool. The proposed approach takes as input a high level specification of the considered design namely its DFG (Data Flow Graph). Concerning the high level power estimation, the two main axes existing in the literature are the analytical models and the spreadsheet. The analytical models are created on the basis of the results obtained through the synthesis and the implementation of several configurations of the considered model. These results are then analysed using mathematical approaches such as the curve fitting and the regression analysis [3], the machine learning with automatic feature selection [8] and the geometric programming [9]. The spreadsheet is provided by the FPGA constructor. This alternative is interesting since all the features of the target are taken into consideration. Consequently, the tool provides a better estimation. However, the accuracy of the results depends essentially on the resources estimation accuracy. In our case, we adopt the second alternative since we proposed an accurate resources estimation which allows a better and faster power estimation process with lower error. This work is based on a combination of the analysis of the synthesis inference strategies and a low level exploration of the studied operators in the DFG. This ensures the generation of accurate estimation since the proposed approach takes into consideration the optimizations performed by the synthesis tool. Moreover, all the interpretations derived from the exploration of the studied operators which allows a generalization of the results.

High Level Estimators for FPGAs

113

This paper is organized as follow, the first section focuses on the theoritical background. The second one shows the main results and analysis of the inference strategy of the synthesis tool. The third and the fourth sections present respectively the area and the power estimation flows. Section 5 summarizes the experimental results. Section 6 concludes this work.

2

Theoritical Background

In this section, we present the data flow graph as it is the input of the proposed high level estimation approach. 2.1

Data Flow Graph

The Data Flow Graph (DFG) is a graphical representation composed by nodes and directed arcs. In mathematical terms, we can say that a DFG is a graph defined by Eq. 1 G = (V, E), V = (v1 , ..., vn ), E = (e1 , ..., em ) ⊂ V × V

(1)

where V is the set of nodes (or vertices) and E the set of directed edges ei = (vj , vk ). In digital signal processing, nodes of the data flow graph are operators, essentially adders, subtractors and multipliers. 2.2

Defined Blocks DFG (DBDFG)

The DBDFG is a combination between the conventional DFG and the Bloc Definition Diagram (BDD). The BDD is a diagram which shows blocks with their attributes (types). It is an appropriate method to define the components of a system. Therefore, the DBDFG is a graph G = (V,E,A), where V = (v1 , ..., vn ) is the set of nodes (or vertices), E = (e1 , ..., em )⊂ VxV the set of directed edges ei = (vj , vk ) and A = (a1 , ..., an ) is the set of attributes where ap is the attribute (operation type) of the node vp . Each operation node is described by the structure shown in Eq. 2: node =

(2)

Where ID is the identifier of the node, type its attribute and bitin is the number of bits of the nodes’ operands. bitin is a vector of b values where b is the number of operands of the considered node. (b = 2 for multiplications, additions and subtractions, b = 1 for shifters). bitin is determined by considering the type and bitin of the node’s predecessors as shown in the pseudo code 1. Later, the aformentioned representation will be considered to provide an accurate area estimation. But before, the synthesis inference strategy of the multipliers is analysed in order to extract the optimizations performed by the sythesis tool.

114

S. Mami et al.

Algorithm 1. Bitin determination 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14:

3

Input : bitin (pred node), type(pred node) NOTE Pred Node if type(pred node)=addition or subtraction then if bitin (pred node1) = bitin (pred node2) then bitin (node) = bitin (pred node) + 1 ; else bitin (node) = M ax(bitin (pred node1), bitin (pred node2)) ; end if else if type(pred node)=multiplication then bitin (node) = bitin (pred node1) + bitin (pred node2) else bitin (node) = bitin (pred node) end if

Analysis of the Synthesis Inference of Multipliers

When a multiplication is inferred, two cases are considered by the synthesis tool, the Two Variables Multiplication (TVM) and the Single Constant Multiplication (SCM). Figure 1 shows a summary of the results reported by the synthesis tool for a TVM. It is clear that multiplications are almost systematically implemented using DSP blocks. This is perfectly coherent since multiplication is the major operation in a DSP block. When the bitwidth is lower than 4, the tool adopts a LUT based implementation since the DSP blocks are precious and limited resources.

Fig. 1. Synthesis inference for the TVM

For the SCM case, the synthesis tool operates differently according to the value of the constant.

High Level Estimators for FPGAs

115

– C1: if there is an integer “i” such as the constant “C” can be written as follows C = 2i then a shift based implementation is applied. – C2: if the constant is odd and lower than 7, then a LUT (LookUp Table) implementation is adopted. – C3: if the constant is even and lower than 7, then a shift and LUT implementation is considered. – In the remained cases, a DSP is inferred. This analysis is perfectly coherent. In fact, as it is shown in Fig. 2, the multiplication is nothing more than the sum of all of the powers of 2 in any given number. For optimization reasons, this definition is directly applied by the synthesis tool in the mentioned cases.

Fig. 2. The multiplication operation

To summarize the results of the SCM analysis, we can say that if the number of bits is lower or equal to 4, the multiplication is inferred as the sum of the shifted variable (Eq. 3). var × cst =

3 

var × (2i × ji )

i=0

=



shif t(var, kZ )

(3)

Z

Where ji is the value of the ith bit of the constant, Z, the number of ones in the constant and kZ the position of the Z th one. According to this analysis, we class multiplications into 2 categories: multDSP et multslice . This classification allows to distinguish the multiplications mapped in DSP blocks from the multiplications mapped in LUTs. The two classes will be studied separately.

4

Area Estimation

In this paper, we measure the area in terms of slices and DSPs. It is notable that in digital signal processing, the principal operations are multiplication, addition/subtraction and logic operations. Concerning the add/sub and logic operators, it is clear that they are almost always implemented in LUT slices. Multipliers are in most cases mapped into DSP blocks. However, as mentioned

116

S. Mami et al.

in the study of the synthesis inference, this operation is sometimes implemented in LUTs. The area estimation flow is summarized in Fig. 3. As it is shown, each block takes as input the number of bits of its inputs (bitin ). – For the add/Sub operators, bitin = (n, m) – For the mult operators, two cases will be treated: • the two variables multiplication: In this case the estimator takes as input n and m. • the single constant multiplication: In this case bitin = (nvar , k, Z), we will see later that k and Z are features extracted from the constant. – For the shift operators, bitin = (n), k is the amount of shift and σ its direction.

Fig. 3. The area estimation

4.1

Estimation of DSP Blocks

In order to generalize the estimation, we work on large multiplication. The mapping of such operation into DSP blocks requires the partionning of the operands into 17 bits sub-words. We note M = [MH , ML ] and N = [NH , NL ] the operands of the multiplication where ML and NL are respectively the 17 least significant bits of M and N and MH and NH are respectively the most significant bit of M and N. m is the bitwidth of M et n the bitwidth of N. The equation of the large multiplication sub-products is shown Eq. 4   M × N = (MH × NL ) + (ML × NH ) >> n   (4) + (MH × NH ) >> 2n + (ML × NL )

High Level Estimators for FPGAs

117

Now according to the values of m and n, the operation will be mapped in one DSP or more with some LUTs or without. Figure 4 show the mapping optimisation for m and n ∈ [0, 37].

Fig. 4. Mapping of multiplication

As it is shown in Fig. 4, 7 areas are proposed. For reasons of symmetry, we focus in this analysis on x-axis only. 1. In this case, m and n ≤ 18, therefore MH = NH = 0. We have a unique multiplication to perform (ML × NL ), we need one 18 × 18 DSP. 2. In this case 19 ≤ n ≤ 20 and 19 ≤ m ≤ 20, therefore MH =0, We have 2 multiplications to perform ML × NL and ML × NH . The first one is mapped in a DSP block, the second one is mapped using LUTs. 3. In this case 20 < n ≤ 35 and m ≤ 18, We have 2 multiplications to perform ML × NL and ML × NH . Both multiplication are mapped in DSP blocks. 4. In this case 35 < n ≤ 37 and m ≤ 18, We have 3 multiplications to perform ML × NL , ML × NH and ML × N[36..37] Where N[36..37] is the 36th and the 37th bits of N. The first 2 multiplication are mapped into DSP blocks, the third in LUTs. 5. In this case 18 ≤ n ≤ 35 and 18 ≤ m ≤ 35, we have 4 multiplications to perform ML × NL , ML × NH , MH × NL , MH × NH mapped into 4 DSP blocks. 6. In this case 36 ≤ n ≤ 37 and 18 ≤ m ≤ 35, we have 6 multiplications to perform ML × NL , ML × NH mapped into 4 DSP blocks and ML × N[36..37] and MH × N[36..37] mapped in LUTs. 7. In this case 36 ≤ n ≤ 37 and 36 ≤ m ≤ 37, we have 9 multiplications to perform. They are mapped into k DSP and LUTs. where k = 6 and 8 respectively when n = 36 and 37.

118

S. Mami et al.

Algorithm 2. DSP Estimation Algorithm 1: 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: 18: 19: 20: 21: 22: 23: 24: 25: 26: 27: 28: 29: 30: 31: 32:

Input : n,m  n: width of N, m: width of M Output : nDSP, nLUT  ×  m−1  nmult= n−1 17 17 if n≤ 18 then  nH is the highest sub-word. nH =nL =n; else nHtemp =n; while nHtemp > 17 && nHtemp >0 do nHtemp =nHtemp -17; end while nH =nHtemp end if if m≤ 18 then  mH is the highest sub-word. mH =mL =m; else mHtemp =m; while mHtemp >17 && mHtemp >0 do mHtemp =mHtemp -17; end while mH =mHtemp end if if nH ∈ [2, 3] && mH ≤13 then  ndsp=nmult- m−1 17 nLUT=(nH − 1) × mH else if mH ∈ [2, 3] && nH ≤13 then  ndsp=nmult- n−1 17 nLUT=(mH − 1) × nH else if nH ∈ [2, 3] && nH ∈ [2, 3] then  ndsp=nmult- n−1 17 nLUT=(nH − 1) × mH else ndsp=nmult end if

The generalized algorithm which provides the closest values to the synthesis tools is given in Algorithm 2. As it is shown in Algorithm 2, the estimator takes as input the bitwidth of the operands and provides the number of DSP blocks and eventually the number of LUTs used. nmult is the number of 18 × 18 multiplication required to perform the whole multiplication without optimization. From line4 to line18, we determine the number of bits of the highest subword for each operand. In fact, according to their values, an optimization will be applied or not. If the bitwidth of one of the operands is lower than 3 and the bitwidth of the other is lower than 13, the use of DSP blocks is rationalized and some of them will be replaced by LUTs.

High Level Estimators for FPGAs

4.2

119

Estimation of Slices

According to the operation’s types and the number of inputs bits, the number of slices noted S is evaluated. The proposed equations are shown in Eqs. 5 and 6 respectively for the Add/Sub and the Shift operations. S(addmaxin ) = maximum(n, m)

(5)

Where n and m are the operands’ bitwidth. S(shif t) = n − σk

(6)

Where n is the number of the shifter’s input bits and k the amount of shift and σ the direction of shift (σ = −1 if the shift is right and 1 if it is left). According to the analysis done in Sect. 3, the number of slices for a single constant multiplication is given by Eq. 7. S(multSlice ) =

 Z−1 

S(add) +

i=1

=

 Z−1 

Z 

 S(shif t)

i=1

(nvar ) +

i=1

Z 

 (nvar + k)

(7)

i=1

We remember that Z is the number of ones in the constant. k is the position of the Zth one (see Sect. 3). The total number of slices in a given DFG is shown in Eq. 8.   Slice(add) + Slice(shif t) TSlice = ( Na

+



Ns

Slice(multSlice )) × α

(8)

Nm

where Na , Ns and Nm are respectively the number of add/sub, shifters and SCM multipliers in the DFG and α is a correction scaling factor proposed in [2].

5

Power Estimation Flow

The power estimator introduced in this work, see Fig. 5, consists of the aformentioned resource estimator and the spreadsheet Xilinx Power Estimator (XPE) [10]. The equation (Eq. 9) giving power consumption is: P = Ps + P d

(9)

Ps (static power) comprises the user design dependent static power and the device static power.

120

S. Mami et al.

Fig. 5. The power estimation flow

The design static power is the amount of extra power consumed when no switching activity is programmed. It relies essentially on the clock frequency and the input/output. The device static is the transistor leakage power when the device is powered and not configured. The device static depends principally on the features of the device: target, ambient temperature and voltage. Pd (dynamic power) mainly depends on the resource utilization and the switching activity. Pd is defined as demonstrated in Eqs. 10 and 11. Pd = Presources + PI/O + Pclk

(10)

Presources = Plogic + PDSP

(11)

where Several power estimation approaches are presented in literature. In our case, we opt for the approach based on spreadsheet [10]. In fact, this approach has the advantage of taking into consideration all the FPGA’s features. The operating frequency and an estimation of the logic resources and their corresponding toggle rate are inserted by the user. Hence, a high level power estimation is provided. The toggle rate aforementioned is the preliminary switching information, pre-estimated in [10].

6

Experiments

We applied the proposed estimations on a set of well-known benchmarks including the Cordic based Loeffler DCT with different precision extracted from [11], the conventional Loeffler based DCT, a filter named mibench2 and the color

High Level Estimators for FPGAs

121

Table 1. The benchmarks’ features Benchmarks

Number of I/O Nodes’ number and types

DCT-CORDIC-P1 [11] 16

34Add/Sub, 12Shift, 0Mult

DCT-CORDIC-P2 [11] 16

40Add/Sub, 18Shift, 0Mult

DCT-CORDIC-P3 [11] 16

46Add/Sub, 24Shift, 0Mult

DCT-Loeffler

16

29Add/Sub, 0Shift, 11Mult

Mibench2

4

8Add/Sub, 0Shift, 8Mult

Color conv.

6

8Add/Sub, 0Shift, 9Mult

conversion algorithm. The features of the designs are summarized in Table 1. We consider that the input size is 8 bits. The designs are implemented on Spartan6 XC6SLX16-3CSG324 using Xilinx System Generator. The results in terms of slices and DSPs are summarized respectively in Tables 2 and 3. The values between brackets corresponds to the error () measured as shown in Eq. 12, when comparing the estimation (est.) with the experimental (exp.) results. (%) =

exp. − est. est.

(12)

We remark that the predicted in theory, and the experimental, resources used (DSPs and slices) are almost identical. The average estimation error is about 2.64% for the number of slices and 0% for the DSP. Table 2. Estimated and experimental slices Benchmark architectures Estimated Tslice Experimental Tslice (%) DCT-Cordic-Loeffler-P1

900

858(4.8%)

DCT-Cordic-Loeffler-P2

1110

1086(2.2%)

DCT-Cordic-Loeffler-P3

1334

1322(0.9%)

DCT-Loeffler

446

438(1.79%)

Mibench2

191

193(1%)

Color conv.

151

159(5.2%)

Table 3. Estimated and experimental DSPs Benchmark architectures Estimated nDSP Experimental nDSP (%) DCT-Cordic-Loeffler-P1

0

0(0%)

DCT-Cordic-Loeffler-P2

0

0(0%)

DCT-Cordic-Loeffler-P3

0

0(0%)

DCT-Loeffler

13

13(0%)

Mibench2

5

5(0%)

Color conv.

5

5(0%)

122

S. Mami et al.

The power consumption is measured with XPA (Xpower Analyzer). The chosen clock frequency is 100 Mhz and the supply power 1V. The results are presented in Table 4. Table 4. Estimated and experimental Plogic Benchmark

Estimated Plogic Experimental Plogic (%)

DCT-Cordic-Loeffler-P1 21

20(4.7%)

DCT-Cordic-Loeffler-P2 26

23(11.5%)

DCT-Cordic-Loeffler-P3 31

28(9.67%)

DCT-Loeffler

15

13(13%)

Mibench2

6

5(16%)

Color conv.

3

3(0%)

The average error in estimating the power consumption is 9.14%.

7

Conclusion

The high level area and power estimation are considered as a key step in the design space exploration performed by high level synthesis tools. The earlier in the design process the resources and power are predicted, the greater savings can be achieved. In this paper, we propose a rapid and high level area and power estimation for FPGA which provides accurate results without even first going through the synthesis and the implementation phases. The proposed area estimator takes as input the data flow graph of the considered design, scans it, classes the nodes according to their types and predicts the number of slices and DSPs. It is important to notice that the proposed method takes into consideration the optimization done by the synthesis tool. The power estimator takes as input the results provided by the aformentioned estimator, the clock frequency and the inputs/outputs number and size and provides an accurate estimation of the power consumption. The reported results are satisfactory. In fact, the estimation error barely attained 2.64% in terms of slice utilization and 9.14% in terms of power consumption. The estimated number of DSP blocks is perfectly correct in all the cases.

References 1. Gasmi, K., Rebaya, A., Amari, I., Hasnaoui, S.: Workflow for multi-core architecture: from Matlab/Simulink models to hardware mapping/scheduling. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 142–148 (2016)

High Level Estimators for FPGAs

123

2. Meeuws, R., Ostadzadeh, S.A., Galuzzi, C., Sima, V.M., Nane, R., Bertels, K.: Quipu: a statistical model for predicting hardware resources. ACM Trans. Reconfigurable Technol. Syst. Article No. 3 (2013) 3. Deng, L., Sobti, K., Chakrabarti, C., Zhang, Y.: Accurate area, time and power models for FPGA-based implementations. J. Signal Process. Syst. 63, 39–50 (2011) 4. Chuong, L.M., Lam, S.-K., Srikanthan, T.: Area-time estimation of controller for porting C-based functions onto FPGA. In: Proceedings of the RSP 2009, pp. 145– 151 (2009) 5. Cilardo, A., Durante, P., Lofiego, C., Mazzeo, A.: Early prediction of hardware complexity in HLL-to-HDL translation. In: Proceedings of the International Conference on Field Programmable Logic and Applications (FPL), pp. 483–488 (2010) 6. Schumacher, P., Jha, P.: Fast and accurate resource estimation of RTL-based designs targeting FPGAs. In: International Conference on Field Programmable Logic and Applications (FPL), pp. 59–64 (2008) 7. Aung, Y.L., Lam, S.-K., Srikanthan, T.: Rapid estimation of DSPs utilization for efficient high level synthesis. In: IEEE International Conference on Digital Signal Processing (DSP), pp. 1261–1265 (2015) 8. Yu, Y., He, L.: FPGA power estimation using automatic feature selection (abstract only). In: Proceedings of the 2016 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, pp. 282–282 (2016) 9. Mehri, H., Alizadeh, B.: Analytical performance model for FPGA based reconfigurable computing. Microprocess. Microsyst. 39, 796–806 (2015) 10. Xilinx Power Estimator User Guide, UG440 (v2014.1) Xilinx Homepage (2018) 11. Mami, S., Saad, I.B., Lahbib, Y., Hachaichi, Y.: Enhanced configurable DCT cordic loeffler architectures for optimal power-PSNR trade-off. J. Signal Process. Syst. 90, 371–393 (2017)

High Isolation with Neutralization Technique for 5G-MIMO Elliptical Multi-antennas Marwa Daghari1(B) , Chafai Abdelhamid1(B) , Hedi Sakli1(B) , and Kamel Nafkha2(B) 1

2

Research Laboratory of Modeling Analysis and Control of Systems, National Engineering School of Gabes, 6029 Gabes, Tunisia [email protected], [email protected], [email protected] Research Laboratory in High Frequency Electronic Circuits and Systems, Faculty of Sciences of Tunis, El Manar I, 2092 Tunis, Tunisia [email protected]

Abstract. In this paper, an elliptical multi-antennas working at 3.5 GHz for possible future fifth-generation (5G) MIMO (Multiple-Input Multiple-Output) mobile communication system is presented. Firstly, the antenna unit was designed and simulated to be operating at 3.5 GHz. Then, a neutralization line technique was used in designing multi antennas to allow good isolation and reduce the mutual coupling between this antennas. Simulated results show that, over the entire frequency band of interest, the proposed configuration could attain good isolation and low correlations between antennas with only neutralization line technique. Compared to multi-antennas without neutralization line, the proposed structure reduces the coupling with 24 dB at 3.5 GHz. Keywords: Elliptical multi antennas · High isolation · MIMO antenna · Mutual coupling · Neutralization line technique Applications (WIFI, UWB applications)

1

· 5G

Introduction

Over recent years, due to the exceptional growth in the use of wireless communication technologies, a huge commercial opportunity has opened for the mobile industry [1]. Wireless service applications including internet of things, realtime video streaming, live HDTV broadcast and online games, mobile electronic devices should be able to handling these high data rate applications demand [2]. 5G is a future mobile communication technology which is designed to achieve a maximum of channel capacity and delivering high data rate traffic for wireless applications [3]. In order to increase the data transmission rate without increasing the transmitted signal power or the bandwidth of the system, one solution is to take advantage of the diversity through the use of several antennas at the c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 124–133, 2020. https://doi.org/10.1007/978-3-030-21009-0_11

High Isolation with Neutralization Technique

125

transmission and the reception: this is the MIMO (Multiple Inputs Multiple Outputs) technique [4]. This technique is well-known as one of the core technologies in the fifth generation mobile communication. In fact, it can be used to guarantee multiplexing and also diversity gain which can improve link quality and capacity of wireless systems [5]. Wireless systems based on MIMO technique, use multiple antennas both in the transmitted and received sides in order to achieve important capacity gain compared with the SISO (single-input-single-output) systems [1]. So, increasing the number of antennas on mobile terminal devices is an option to raise the channel capacity [6]. In this case, the separation distance between the different antennas must be sufficient to guarantee the independence of the signals. However, this separation distance is limited by the scares space available for an antenna in mobile phone. So, antennas will be closely spaced which results to an inevitable strong mutual coupling among antenna elements. Due to this mutual coupling, multi antennas system efficiency is decreased and performance of the global system is affected. In order to solve the problem of mutual coupling, isolation between antennas in the proposed design should be very high [7]. But, it becomes even difficult to succeed in providing high isolation, particularly when designing small antennas for hand held devices with MIMO elements. Recently, with this major challenge with MIMO antennas to reduce mutual coupling, several techniques have been studied and used [8]. They are known as isolation techniques. In this paper, we propose several ways to decrease the mutual coupling and correlations between two elliptical antennas element ports with neutralization line technique. The rest of the paper is structured as follows. Section two presents the proposed MIMO elliptical antenna element design and simulations. Then, Section three is considered for the design and simulated results of MIMO elliptical multi-antennas system with and without isolation technique. Finally, Section fourth concludes the paper.

2 2.1

Geometrical of the Proposed Elliptical Antenna Element Design Antenna Element Design

Printed antennas are characterized by low cost, light weight and also, easy construction. These features are so suitable and desirable for 5G handheld antenna devices. For this raison, we propose, in this paper, an elliptical patch antenna which is designed for 5G Wireless applications (like Wifi and UWB applications). The geometry of the proposed printed elliptical antenna is shown in Fig. 1, and the design parameters were calculated and optimized to obtain a wellcomportment of the antenna through the UWB (ultra wide band) frequency and specially at 3.5 GHz operating frequency [9,15–18]. The elliptic patch antenna is printed on an FR-4 substrate of thickness h = 1.6 mm, relative permittivity r = 4.4 and loss tangent tgδ = 0.02. The substrate size of this proposed antenna is Ws × Ls . The two axes of antenna are 2a and 2b and the size of the ground plane is chosen to be rectangular with dimensions Wg ×Lg . There is p distance as

126

M. Daghari et al.

a gap between the radiating element and ground plane. The antenna is excited via 50 Ω microstrip lines of Wf = 3 mm width.

Fig. 1. Geometry of elliptical MIMO antenna.

2.2

Simulations Results

In this section, the principal objective is to select the best ground plane length Lg and the elliptical axe a length, which give more performance at 3.5 GHz. The selecting criteria of the best length of Lg and a is the S-parameter return loss value (S1,1 ). Our choice abides with the requirements of antenna design and adaptation. So, to design and optimize the proposed MIMO elliptical antenna, simulations with CST Microwave Studio (MWS) is carried out to complete S1,1 characteristics, maximum gain, efficiency and radiation pattern. A parametric analysis is so performed to get optimized antenna parameters. The overall optimized parameters at the lower resonant frequency of 3.5 GHz are listed in Table 1. Results presented in Fig. 2 shows the simulated antenna reflection coefficient (S1,1 ) curves against the frequency for the designed antenna with optimized parameters value involved in Table 1. It can be noticed that simulated reflection coefficient curves are already lower then −10 dB level through the UWB frequency band (3.1–10.6 GHz). So, the proposed antenna exhibits good performances in UWB. The simulated gain peak of this UWB elliptical antenna after optimization is shown in Fig. 3. As can be noticed in this figure gain is maximum with a value of 2.74 dB at 3.5 GHz operating frequency. The 2-D simulated radiation patterns of the antenna at the frequency of 3.5 GHz in the H-plane (x–z

High Isolation with Neutralization Technique

127

Table 1. Elliptical antenna optimized parameters(mm) Patch layer Ground layer Substrate layer a: 14.5

Lg: 19

Ls: 44

b: 10.0

Wg: 46

Ws: 46

p: 0.40

plane), and E-plane (y–z plane) are shown respectively, in Figs. 4 and 5. Elliptical antenna has omnidirectional radiation patterns, so we can conclude that the MIMO elliptical antenna was optimized to radiate at 3.5 GHz frequency.

Fig. 2. Simulated reflection coefficient curves of the proposed antenna.

3 3.1

MIMO Elliptical Multi-antennas Design Elliptical Multi Antennas Configurations and Simulations

In the previous section, antenna element was first designed and optimized using CST Microwave Studio. Now, we placed two elliptical antenna elements with identical size and configuration, symmetrically on the left and right edges. d represents the Gap distance between the two excited ports P1 and P2 and it is optimized to 44 mm. So, the e distance separates the edge of the two ellipses is taken e = λ6 = 15 mm. The geometries of the proposed MIMO elliptical multi-antennas are shown in Fig. 6. The multi-antennas were designed using CST Microwave Studio. During the simulations, Port 1 (P1 ) and Port 2 (P2 ) were excited with 50 Ω microstrip line. So, Port 2 and Port 1 have the same impedance response. The simulated S parameters of the proposed configuration are given in Fig. 7. From the obtained S parameters, each element resonated at 3.6 GHz with −46.87 dB of S1,1 loss value. S2,1 is taken as the parameter characterizing the isolation between the two power ports of the antennas which was −11.66 dB

128

M. Daghari et al.

Fig. 3. 2D-antenna gain peak.

Fig. 4. 2D-Antenna radiation pattern in the H-plane at 3.5 GHz.

Fig. 5. 2D-Antenna radiation pattern in the E-plane at 3.5 GHz.

of value. Due to enough distance among each element, mutual coupling is high between two antennas which causes a 100 MHz offset of the resonance frequency from 3.5 GHz to 3.6 GHz. For MIMO applications, it is fundamental for the closely-spaced multi-antennas to offer low mutual coupling. In the next section we use an isolation technique with neutralization to reduce correlations between the two elliptical antennas.

High Isolation with Neutralization Technique

129

Fig. 6. Geometry of MIMO elliptical multi antennas.

Fig. 7. Simulated S-parameters for elliptical multi-antennas.

3.2

Elliptical Multi-antennas Configuration with Neutralization Technique

Currently, divers methods are available to reduce the strong coupling among multi-antennas elements. Generally, it can be classified into: polarization decoupling method, port decoupling method, and also ground decoupling method [10]. As a very simple and typical port decoupling method, the neutralization technique. It should be noted the originality of this method because it consists of connecting the radiating elements to better decouple their power port. In fact, to maximize the energy radiated by a powered antenna, it must be ensured that all the energy transmitted to it, is not lost in the second antenna which is at the same time charged by 50 Ω. We must therefore minimize the S2,1 which will be taken as the parameter characterizing the isolation between the two power ports of the antennas [11–14]. It is then a question of judiciously inserting a self between the two antenna elements to obtain an overall behavior of the rejection filter. So, a microstrip line, being fine and short, naturally has very strong characteristic impedance can thus be seen as an inductance [12,13]. This method will

130

M. Daghari et al.

be adopted to improve isolation between the two elliptical antenna elements. In this section, the same geometry of the two elliptical multi-antennas shown in Fig. 6 was adopted excepting that a neutralization line is inserted between the two radiated elements. The configuration shown in Fig. 8 represents the geometry of this structure. Simulations are carried out to verify the effectiveness of the neutralization line. A analysis was used to select the best neutralization line width which gives high isolation between the two antenna. Due to enough distance among elements, S2,1 value of less than −10 dB is exhibited in simulations results. As a comparison, the S-parameters of the two antennas with and without the neutralization line are provided in Fig. 9. As we can see, isolation between elements is efficiently improved to −35.19 dB at 3.5 GHz. It is mainly due to the higher mutual coupling in the case without neutralization. Simulations results demonstrate, so the effectiveness of the neutralization isolation technique to effectively, reduce mutual coupling between multi-antennas elements. The simulated maximum gain radiation patterns of the multi-antennas at 3.5 GHz resonant frequency in the H-plane (x–z plane), and E-plane (y–z plane), with

Fig. 8. Geometry of MIMO elliptical multi-antennas with neutralization.

Fig. 9. Reflection loss values of antenna 1.

High Isolation with Neutralization Technique

131

and without neutralization are shown in Figs. 10, 11 and 12, respectively. It is shown that the gain does not vary much. Also, at 3.5 GHz, we notice an increase of gain by 1 dBi to reach 5 dBi with the use of neutralization technique.

Fig. 10. Gain peak of multi-antennas at 3.5 GHz.

Fig. 11. 2-D radiation pattern of multi-antennas in the H-plane at 3.5 GHz.

Fig. 12. 2-D radiation pattern of multi-antennas in the E-plane at 3.5 GHz.

132

4

M. Daghari et al.

Conclusion

In this paper, a two MIMO Elliptical multi-antennas operating at 3.5 GHz for 5G mobile application is designed. Neutralization line technique is used to reduce correlations between antennas and also to simplify the design process. Simulation results of S-parameters (S12, and S2,1 ) ensure that the magnitude of transmission is less than −24 dB between the two antennas ports around 3.5 GHz. This simulated value demonstrates the performance of the isolation method and also indicates that the multi-antennas applies well to MIMO 5G systems. Moreover, the proposed two elliptical antennas is a simple and low cost design which is a good choice for future 5G mobile applications (like Wifi and UWB applications) and also may provide a reference for the design of antenna in MIMO-5G system.

References 1. Jamil, A., Yusoff, M.Z., Yahya, N.: Current issues and challenges of MIMO antenna designs. In: Proceeding of IEEE the 3rd International Conference on Intelligent and Advanced Systems (ICIAS), Philippines, pp. 1–5 (2010) 2. Roy, S., Foerster, J.R., Somayazulu, V.S., Leeper, D.G.: Ultrawideband radio design: the promise of high-speed, short-range wireless connectivity. In: Proceeding of IEEE, vol. 92, pp. 295–311 (2004) 3. Wallace, J.W., et al.: Experimental characterization of the MIMO wireless channel: data acquisition and analysis. IEEE Trans. Wirel. Commun. 2, 335–343 (2003) 4. Cho, Y.S., Kim, J., Yang, W.Y., Kang, C.G.: MIMO-OFDM Wireless Communications with MATLAB. Wiley, Hoboken (2010) 5. Jusoh, M., Jamlos, M.F.B., Kamarudin, M.R., Malek, M.F.B.A.: A MIMO antenna design challenges for UWB application. Progress Electromagnet. Res. B 36, 357– 371 (2012) 6. Zheng, L., Tse, D.N.C.: Diversity and multiplexing: a fundamental tradeoff in multiple-antenna channels. IEEE Trans. Inf. Theory 49, 1073–1096 (2003) 7. Malathi, A.C.J., Thiripurasundari, D.: Review on isolation techniques in MIMO antenna systems. Indian J. Sci. Technol. 9(35) (2016). https://doi.org/10.17485/ ijst/2016/v9i35/96704 8. Zhang, Y., Niu, B.: Compact ultrawideband (UWB) slot antenna with wideband and high isolation for MIMO applications. Progress Electromagnet. Res. C 54, 9–16 (2014) 9. Sayidmarie, K.H.: Design aspects of UWB printed elliptical monopole antenna with impedance matching. In: Loughborough Antennas & Propagation Conference, UK (2012) 10. Diallo, A., Luxey, C., Thuc, P.L., Staraj, R., Kossiavas, G.: Study and reduction of the mutual coupling between two mobile phone PIFAs operating in the DCS1800 and UMTS bands. IEEE Trans. Antennas Propag. 54, 3063–3074 (2006) 11. Zhang, S., Pedersen, G.F.: Mutual coupling reduction for UWB MIMO antennas with a wideband neutralization line. IEEE Antennas Wirel. Propag. Lett. 15, 166– 169 (2016) 12. Su, S., Lee, C., Chang, F.: Printed MIMO-antenna system using neutralization-line technique for wireless USB-dongle applications. IEEE Trans. Antennas Propag. 60, 456–463 (2012)

High Isolation with Neutralization Technique

133

13. Saleh, A.M., Sayidmarie, K.H., Abd-Alhameed, R.A., Jones, S.M.R., Noras, J.M., Excell, P.S.: Compact tri-band MIMO antenna with high port isolation for WLAN and WiMAX applications. In: Antennas and Propagation Conference (LAPC), Loughborough (2016) 14. Moharram, M.A., Kishk, A.A.: General decoupling network design between two coupled antennas for MIMO applications. Progress Electromagnet. Res. Lett. 37(2013), 134–142 (2013) 15. Ben Mbarek, S., Ben Hassen, M., Choubani, F.: 2-D FDTD analysis of CPW antenna for electromagnetic near-field applications. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT 2016), pp. 62–65, March 2016 16. Jebali, N., Beldi, S., Gharsallah, A.: RFID antennas implanted for pervasive healthcare applications. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT 2016), pp. 149–152, March 2016 17. Zoubiri, B., Mayouf, A., Mayouf, F., Abdelkebir, S., Devers, T.: Rectangular microstrip antenna gain enhancement using elliptical EBG structure. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT 2016), pp. 386–388, March 2016 18. Sethi, W., Vettikalladi, H., Fathallah, H., Hindi, M.: 1x2 equilateral triangular dielectric resonator nantenna array for optical communication. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT 2016), pp. 7–9, March 2016

Gaussian Process Based Method for Point and Probabilistic Short-Term Wind Power Forecast Ali Lahouar(B) Laboratory of Advanced Technology and Intelligent Systems (LATIS), National Engineering School of Sousse (ENISO), University of Sousse, Sousse, Tunisia [email protected]

Abstract. Wind power is becoming one of the most promising renewable energy sources. With a total capacity exceeding 486 gigawatts worldwide in 2016, wind power optimization, forecast and control become more challenging than ever before. Forecasting wind turbines output for a period of time in advance is beneficial for grid managers, since it allows them to optimize their generation plans and to control the production of conventional thermal or nuclear plants. This paper proposes then a Gaussian process based method for predicting the production of a wind farm for one and two hours in advance. Both point and probabilistic forecasts are performed through customizable prediction intervals with different confidence levels. The model is tested using real data from Sidi Daoud wind farm in northeast Tunisia. Results are analyzed and compared to similar methods in terms of various assessment metrics. Keywords: Wind power forecast · Short term · Wind speed and direction · Gaussian process · Probabilistic forecast Prediction intervals

1

·

Introduction

Efficient exploitation of energy resources becomes more challenging in modern grids, with the increasing penetration levels of renewable generation worldwide. Optimized control of wind farms in particular is difficult due to wind intermittence and complexity of weather patterns. Accurate forecast of future wind power is therefore mandatory to ensure continuous and stable electricity supply. Indeed, intermittent generation can inject disturbances into the grid, and may cause frequency regulation problems. Probabilistic forecast is able to generate different possible outcomes of future power, in order to take into account several possible scenarios. The transition from deterministic to probabilistic predictions is mandatory because of uncertainties associated with electricity generation and trade. In fact, probabilistic forecast is a powerful tool to manage power reserves and to optimize bidding strategies. c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 134–147, 2020. https://doi.org/10.1007/978-3-030-21009-0_12

Gaussian Process Based Method

135

New schemes for security evaluation are proposed in the literature to take into account the disturbances induced by intermittent generation [1]. Among these disturbances, previous research papers cite the effects on voltage stability [2]. The impact of wind forecast uncertainty on power systems is likewise evaluated [3]. Under these circumstances, probabilistic forecast shows a high flexibility. Indeed, it is proven that probabilistic wind power prediction can contribute to efficient operation of electricity markets with high penetration levels [4]. Probabilistic wind power forecast facilitates several tasks in modern power grids, such as optimal setting of operating reserves [5], predictive control of battery energy storage system [6], and unit commitment [7]. The state of the art of probabilistic wind power forecasting is already well developed. A review of different methods can be found easily in the literature [8]. Most methods make use of either physical or statistical models. Physical models are generally based on numerical weather prediction, while statistical methods focus mainly on time series. Artificial intelligence, machine learning, and heuristic optimization are common approaches to deal with wind power and speed forecast. For example, a framework based on the particle filter algorithm is applied to predict the output of each wind farm apart [9]. Fuzzy k-means clustering algorithm, support vector regression and quantile regression are utilized in a probabilistic forecast framework [10]. Quantile regression is also combined with extreme learning machine to generate nonparametric probabilistic forecasting, where quantiles are produced using a linear programming optimization model [11]. A method based on Gaussian process, making use of a local moving window, is also proposed for this purpose [12]. Variants of this method are also proposed, such as warped Gaussian process [13]. The k-nearest neighbors algorithm is likewise utilized in probabilistic forecasting, either alone [14] or combined with kernel density estimator [15]. Quantile regression for probabilistic prediction is performed using a reproducing kernel Hilbert space framework [16]. Double seasonal Holt Winters and conditional density kernel estimation are developed in order to estimate the probabilistic density of wind power [17]. Sparse Bayesian learning and discrete wavelet transform are carried out in order to solve the problem of wind power forecast [18]. Several other methods are also developed, such as gradient boosting machine [19], radial basis function neural networks [20], kernel density estimator with logarithmic transformation [21], Markov chain models [22], and sparse vector autoregression [23]. The abundance of sophisticated methods proves the importance of probabilistic power prediction and justifies the need for more advanced approaches to handle the increasing uncertainty. This paper proposes a short-term probabilistic wind power forecast based on Gaussian process (GP). The method itself has already been used before. The novelty, however, lies in the selection of inputs, commonly called features. In order to take into consideration the spatial correlation between wind turbines, the wind speed is averaged spatially and used as input. The wind direction is also appended to the input vector. It will be proved that wind direction and spatially averaged speed affect the overall farm production. The lead time is set to 1 and

136

A. Lahouar

2 h in advance. Nine prediction intervals are provided by the GP model according to confidence levels ranging from 10% to 90%. Various evaluation metrics are suggested to assess the forecast accuracy. To our best knowledge, the effects of spatially averaged wind speed and direction in a Gaussian process forecaster were not assessed before this work. The remainder of the paper is organized as follows. Section 2 defines the forecast methodology, Sect. 3 presents and discusses results, and Sect. 4 concludes the paper.

2 2.1

Forecast Methodology Data Preprocessing

Renewable energy and smart grid technologies in Tunisia are getting more and more attention by researchers [24,25], since they are in continuous development. Therefore, the data used in this work will be taken from Sidi Daoud wind farm in northeast Tunisia. The total installed capacity is 53.5 MW. The rated power of the 70 wind turbines in the farm ranges from 330 kW to 1320 kW. Turbines are aligned from northeast to southwest, perpendicularly to the wind dominant direction. The wind speed is metered above the nacelle of each individual wind turbine apart. The wind direction is measured from the meteorological mast inside the farm, in addition to temperature, relative humidity and atmospheric pressure. The power output of each turbine, in addition to the aforementioned meteorological factors, are given in the form of time series sampled at 10 min for the years 2010 and 2011. The huge amount of data includes necessarily some errors. The preprocessing is therefore mandatory. Major corrections are: filling gaps (missing samples) with previous values, removing redundant and repeated samples, and avoiding sensors failure periods (large periods with immovable or inconsistent values). In addition, the overall produced power output of the farm is calculated, by summing the power outputs of all turbines. Time series averaging is also proposed in order to get new quantities. The averaging is either spatial, or temporal. Spatial averaging is obtained by calculating the mean of 70 wind speed measures, in order to benefit from spatial smoothing effect. It is proven in Fig. 1 that the spatially averaged speed is more correlated to the farm power output than the speed metered by the meteorological tower. Temporal averaging is applied to the power output, where each consecutive six values are replaced by their mean. This operation provides smoother curves, and thus easier to predict. 2.2

Feature Selection

The feature selection is the process of choosing the appropriate inputs of the forecaster. Let X be the input vector, and Yˆ the provided output distribution, intended to estimate the real output Y . In this paper, the lead time, or forecast horizon, is set to 1 and 2 h in advance. Therefore, all features will be selected from hour h in order to predict power at hour h + 1 or h + 2. Let S be the wind

Gaussian Process Based Method

137

1 January 2011

12

Wind speed (m/s)

Wind averaged speed Wind speed Wind power 9

6

3

0

0

6

12

18

24

Hours Fig. 1. Wind spatially averaged speed (solid curve), wind speed from the weather mast (dotted curve), and overall power output (solid curve with dots) on January 1st , 2011

speed, D the wind direction, and P the wind power. S¯ will be used to denote the spatially averaged wind speed (mean of 70 measures), and P˜ to denote the temporal average of power output (mean of six consecutive values). Six values ¯ D} at hour h (from h : 00 to h : 50), will be selected of each quantity in {P, S, to predict the average power P˜ at hour h + 1 or h + 2. The input vector X, is then expressed as follows: X = [Ph:00 , . . ., Ph:50 , S¯h:00 , ...S¯h:50 , Dh:00 , . . ., Dh:50 ]T

(1)

Where Ph:00 is the power measured at h : 00, Dh:30 is the wind direction metered at h : 30, and so on. The total number of inputs is therefore 18. The average power P˜h at hour h is naturally given by: 1 P˜h = Ph:i0 6 i=0 5

(2)

˜ of P˜ The output Yˆ provided by the GP forecaster, should be an estimation Pˆ at hour h + 1 or h + 2, according to the required forecast horizon. ˜h+2 Yˆ = Pˆ˜h+1 or Yˆ = Pˆ

(3)

Past values of power are utilized according to the autocorrelation plot of Fig. 2, which is extended to include 24 past hours. The autocorrelation reveals the similarities between the power time series and its lagged version. Lags with maximum similarity are those close to zero. Even with a lag of three hours, the correlation is still higher than 0.8. Therefore, it is judicious to utilize the six

138

A. Lahouar Autocorrelation function of wind power

1 0.9 0.8

Magnitude

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

0

3

6

9

12

15

18

21

24

Lag (hours) Fig. 2. Autocorrelation plot of wind power time series, sampled at 10 min

Fig. 3. Histogram of wind direction samples in 2011

power measurements at the current hour h in order to predict the average power of the next hours h + 1 and h + 2. The spatially averaged wind speed S¯ is the second category of inputs. It is selected according to Fig. 1, where its correlation with power output is obvious. Consequently, it has certainly a great impact on the generated power. The third family of inputs, which is the wind direction D, affects also the power output. The distribution of wind direction occurrences in 2011 are given in Fig. 3, recorded

Gaussian Process Based Method

139

Surface fit in 2011

Wind power (MW)

60

40

20

0 North West South East North

Wind direction

0

5

10

15

20

25

Wind averaged speed (m/s)

Fig. 4. Surface plot of wind power fitted to averaged speed and direction in 2011

each 10 min throughout the year. Major directions are northwest and southeast, which are perpendicular to the turbines alignment in the farm. It is therefore expected that power is maximum for these directions in particular, and minimum elsewhere. The surface plot of Fig. 4 asserts this assumption, where the generated power is fitted to wind speed and direction. Obviously, the average wind speed is the most influential factor that drives the power output. However, at the rated speed of 15 m/s, some peaks and valleys appear on the surface according to wind direction. Indeed, peaks of power arise for northwest and southeast directions, while power valleys dominate the remainder. 2.3

Mapping Function

This section describes the input/output mapping function of the Gaussian process (GP). The GP has several advantages. It is probabilistic; different prediction intervals can be constructed according to different confidence levels. Furthermore, it is versatile. Different kernels may be specified according to a covariance function. The GP is utilized in this paper for supervised regression. Let Sn be a training set containing n observations Sn = {(X1, Y1 ), . . ., (Xi , Yi ), ..., (Xn , Yn )}, where Xi is a vector of dimension d (d = 18 in this paper), and Yi is the corresponding scalar target. The set may be written Sn = {X, Y }, where X is a d × n matrix. Let X ∗ be an unseen input vector, i.e. does not belong to Sn . The aim of the prediction is to determine the output scalar Y ∗ that corresponds to this new input. According to the GP, the target Y ∗ follows a normal distribution [26]:

Where:

Y ∗ ∼ N (μ, σ 2 )

(4)

μ = K(X, X ∗ )K(X, X)−1 Y

(5)

140

A. Lahouar

σ = K(X ∗ , X ∗ ) − K(X, X ∗ )K(X, X)−1 K(X ∗ , X) K is a covariance function, for example the squared exponential:   1 ∗ ∗ 2 K(X, X ) = exp − |X − X | 2 N is the normal distribution, whose probability density f is given by:   1 (x − μ)2 f (x) = √ exp − 2σ 2 2πσ 2

(6)

(7)

(8)

Several prediction intervals may be obtained from Y ∗ through quantiles. Let F be the cumulative distribution function of the random variable Y ∗ . The αth quantile of Y ∗ is given by [27]: QY * (α) = FY−1 ∗ (α) = inf {y : FY * (y) ≥ α}; y ∈ R

(9)

Where 0 ≤ α ≤ 1. For example, the QY ∗ (0.6) is the quantile below which 60% of Y ∗ observations are expected. Prediction intervals (PI) are constructed from quantiles according to confidence levels. For instance, the PI associated to the confidence level 80% is defined by: PI 80% (Y ∗ ) = [QY ∗ (0.10), QY ∗ (0.90)]

(10)

The median value QY ∗ (0.50) will be utilized for point forecast and for comparison with similar methods. Nine PI are constructed for probabilistic forecast, ranging from 10% to 90%.

3 3.1

Case Study Assessment Metrics

In order to evaluate quantitatively the prediction accuracy, several metrics are already in use by researchers. For point forecast provided by the median value QY ∗ (0.50), four assessment criteria are proposed in this paper. They are the mean absolute error (MAE ), the root mean squared error (RMSE ), the mean absolute percentage error (MAPE ) and the mean absolute scaled error (MASE ), defined by: s 1 ˆ |Pi − Pi | (11) MAE = s i=1   s 1  RMSE =  (Pˆi − Pi )2 (12) s i=1 MAPE =

s 1  |Pˆi − Pi | × 100 s i=1 Ps

(13)

Gaussian Process Based Method

⎛

⎞

s ⎜ 1 ⎜ ⎜ MASE = s i=1 ⎜ ⎝

⎟ ⎟ |Pˆi − Pi | ⎟ s ⎟  ⎠ |Pj − Pj−1 |

1 s−1

141

(14)

j=2

Where s is the number of hours in the testing period. For each hour h, the ˜h+1 or Pˆ ˜h+2 provided by quantity Pˆi stands for the predicted average power Pˆ ∗ the GP median value QY (0.50), according the prediction horizon. Naturally, Pi denotes the actual measured average power P˜h+1 or P˜h+2 , also according to the forecast s horizon. Ps is the actual power averaged over the whole test period, Ps = 1s i=1 Pi . Two additional metrics are suggested for probabilistic forecast, which are the Pinball loss function (PLF ), and the Winkler score (WS ). The PLF is defined for each quantile QY ∗ (α) by [28]: PLF (QY ∗ (α)) =  (1 − α)(QY ∗ (α) − Pi ) if Pi < QY ∗ (α) if Pi ≥ QY ∗ (α) α(Pi − QY ∗ (α))

(15)

Where α ∈ {0.1, 0.2, ..., 0.9}. In addition, let PI c% (Y ∗ ) be the prediction interval associated to the confidence level c% = (1 − 2α) × 100: PI c% (Y ∗ ) = [QY ∗ (α), QY ∗ (1 − α)]

(16)

Where α ∈ {0.05, 0.10, ..., 0.45}. The width of PI c% (Y ∗ ) is denoted δc% , with δc% = QY ∗ (1 − α) − QY ∗ (α). The WS is defined by [28]: WS (PI c% (Y ∗ )) = ⎧ δc% for Pi ∈ PI c% (Y ∗ ) ⎪ ⎪ ⎨ 1 for Pi < QY ∗ (α) δc% + (QY ∗ (α) − Pi ) α ⎪ ⎪ 1 ⎩δ + (P − Q ∗ (1 − α)) for P > Q ∗ (1 − α) i Y i Y c% α

(17)

All proposed metrics are proposed to quantify the distance between real and predicted samples, but from different points of view. 3.2

Results and Discussion

The aforementioned GP forecaster is tested for a lead time of 1 h in advance, on four different weeks in 2011. The seven first days, 1st to 7th day of February, May, August and November, are selected for this purpose. Test periods are selected intentionally across different seasons in order to verify the prediction accuracy under different circumstances. Forecast results are given in Fig. 5 in the form of fan plot, for August and November only. Only four days of each period are shown for sake of clarity. Actual power is represented by the dashed curve, whereas point forecast is represented by the solid curve. The nine prediction intervals, ranging

142

A. Lahouar

Overall produced power (MW)

60

3 to 6 August 2011, hour-ahead prediction 10% 20% 30% 40% 50% 60% 70% 80% 90% Actual Predicted

50 40 30 20 10 0 48

72

96

120

144

Hours

Overall produced power (MW)

60 50 40 30

3 to 6 November 2011, hour-ahead prediction 10% 20% 30% 40% 50% 60% 70% 80% 90% Actual Predicted

20 10 0 48

72

96

120

144

Hours Fig. 5. Actual curve and prediction intervals of one hour ahead wind power forecast

from 10% to 90%, are given in the form of color gradient. The darkest color stands for the narrower PI (10%), and the lighter color stands for the wider PI (90%). For this short lead time of 1 h in advance, results of point forecast are already satisfactory. With the exception of some sudden and very sharp variations, the predicted curve succeeds to follow the major trends of the actual curve. Forecast accuracy is verified under different scenarios, even with very spiky curves like those of November, or abnormally low production like that of August. The GP is compared to some conventional prediction methods, namely persistence (PER),

Gaussian Process Based Method

143

artificial neural networks (ANN) and support vector machines (SVM). Comparison is performed in terms of the four aforementioned metrics, MAE , RMSE , MAPE , and MASE . Results are given in Table 1, where all metrics are computed on the first 7 days of each test month. With the exception of November, the GP gives always the lower error. In fact, the special spiky curve in November makes the prediction more difficult. Table 1. Evaluation metrics of one hour ahead point forecast using four different methods Period

1 to 7 February

Criterion PER

ANN

1 to 7 May SVM

GP

PER

ANN

SVM

GP

M AE

2.9533

2.7317

4.2068

2.3509

3.8553

2.9549

3.9594

2.8284

RM SE

4.2109

4.3042

7.0569

3.6503

5.7793

4.3870

5.8416

4.3202

M AP E

14.1811 13.1170 20.2002 11.2883 17.0541 13.0710 17.5148 12.5114

M ASE Period

1.0036

0.9283

1.4296

0.7989

1 to 7 August

Criterion PER

ANN

0.9956

0.7631

1.0225

0.7304

1 to 7 November SVM

GP

PER

ANN

SVM

GP

M AE

1.2273

1.1110

1.7525

0.9724

3.0585

2.6699

4.2139

3.0737

RM SE

2.0932

1.6665

2.6286

1.5260

4.5172

4.2176

7.6371

4.8488

M AP E

25.6834 23.2499 36.6736 20.3498 19.7480 17.2389 27.2086 19.8463

M ASE

1.0116

0.9158

1.4445

0.8015

1.0024

0.8750

1.3811

1.0074

Table 2. Evaluation metrics of one hour ahead probabilistic forecast P LF February May

August November W S February May

0.10

0.6411

0.7327 0.4172

0.7069

90% 11.1283

0.20

0.9153

1.0901 0.5379

1.0374

80%

0.30

1.0766

1.2905 0.5334

1.2514

0.40

1.1879

1.3849 0.4675

0.50

1.2417

0.60 0.70

August November

10.8624 9.3174

8.6480

8.6704

8.4632 7.2595

6.7379

70%

7.0120

6.8445 5.8710

5.4492

1.3902

60%

5.6940

5.5580 4.7674

4.4249

1.4050 0.4837

1.4681

50%

4.5633

4.4543 3.8207

3.5462

1.2650

1.3898 0.5780

1.4841

40%

3.5478

3.4631 2.9705

2.7414

1.2307

1.3045 0.6063

1.4136

30%

2.6069

2.5446 2.1827

1.7314

0.80

1.0798

1.1272 0.5596

1.2384

20%

1.7140

1.6731 1.4351

1.1207

0.90

0.7788

0.8144 0.4012

0.9287

10%

0.8502

0.8299 0.7118

0.6237

The probabilistic forecast is useful in the presence of some shifts between actual and predicted curves. In many cases, this shift is compensated by prediction intervals. This compensation, albeit not very clear, can be observed in

144

A. Lahouar

February and May in Fig. 5. Indeed, prediction intervals for a lead time of one hour in advance are narrow, and therefore their usefulness does not show up properly. The PLF and WS are computed in order to evaluate the prediction accuracy. Results appear in Table 2. The PLF is evaluated for each quantile, where α ∈ {0.1, 0.2, ..., 0.9}, and averaged over the whole test period of 7 days in each month. In a similar manner, the WS is calculated for each prediction interval, with confidence levels ranging from 10% to 90%. It is likewise averaged over the whole test period. For both criteria, lower values reflect more accurate prediction. The PLF , for all quantiles, has lower values in August. Since the PLF characterizes only the forecast sharpness, it is expected that August results will be the best. Indeed, the actual power curve in August is much less spiky than other months, and it is consequently more predictable. However, the WS is minimum in November. In fact, the WS depends on PI widths. Since November PI are very narrow, the November WS is naturally lower than that of other months. Table 3. Evaluation metrics of two hours ahead probabilistic forecast P LF February May

W S February May

0.10

1.0860

1.4228 90% 19.8559

19.3694

0.20

1.6272

2.0687 80% 15.4703

15.0913

0.30

1.9315

2.4245 70% 12.5113

12.2048

0.40

2.1278

2.5777 60% 10.1597

9.9107

0.50

2.2303

2.6351 50%

8.1421

7.9426

0.60

2.2167

2.6351 40%

6.3303

6.1752

0.70

2.1191

2.4954 30%

4.6514

4.5374

0.80

1.8663

2.1713 20%

3.0583

2.9834

0.90

1.3594

1.5628 10%

1.5169

1.4798

Figure 6 shows the forecast results for a lead time of two hours in advance, from the 1st to the 7th day of February and May 2011. In this case, the point prediction encounters some difficulties. In many cases, a shift of one or two hours appear between actual and predicted curves. In this case, prediction intervals reveal their advantages. With larger bounds than those of hour-ahead forecast, PI are able to cover the actual curve in most cases. In February for example, the dashed curve is almost always covered by at least one PI, with the exception of the spectacular increase at the first hours of 3th February. The same phenomenon is observed in May. The inconvenient that may be cited here, is that very spiky curves cannot be followed easily even in the presence of different PI. However, the forecast accuracy is pretty satisfactory, and it is far away from point prediction in terms of sharpness. The PLF and WS are computed in case of two hours ahead prediction for the seven first days of February and May. The results, for each quantile and PI, are given in Table 3. Values of PLF and WS are a bit

Gaussian Process Based Method

Overall produced power (MW)

60

145

3 to 6 May 2011, 2 hours ahead prediction 10% 20% 30% 40% 50% 60% 70% 80% 90% Actual Predicted

50 40 30 20 10 0 48

72

96

120

144

Hours

Overall produced power (MW)

60

3 to 6 February 2011, 2 hours ahead prediction 10% 20% 30% 40% 50% 60% 70% 80% 90% Actual Predicted

50 40 30 20 10 0 48

72

96

120

144

Hours Fig. 6. Actual curve and prediction intervals of two hours ahead wind power forecast

higher now, which is expected for longer forecast horizons. In terms of PLF , the prediction in February, with respect to May, is somehow better. Nevertheless, in terms of WS , both months are almost equal.

4

Conclusion

This paper proposed a short-term wind power forecasting approach based on Gaussian process (GP). The GP is a probabilistic and versatile method, able to provide several prediction intervals with different confidence levels. The forecast

146

A. Lahouar

methodology was organized as follows. First, a set of wind power, speed and direction from a wind farm in Tunisia is considered and processed. Then, an analysis is carried out in order to select the best candidates to the forecaster input. Finally, an input/output mapping function is built using the GP specific equations. The output of the model was utilized for one hour and two hours of prediction in advance. In both cases, the GP forecaster shows very satisfying results in terms of several evaluation metrics. The future work involves the optimization of GP parameters, and the refinement of inputs with the intention of improving the forecast accuracy.

References 1. Le, D., Berizzi, A., Bovo, C.: A probabilistic security assessment approach to power systems with integrated wind resources. Renew. Energy 85, 114–123 (2016) 2. Bian, X., Geng, Y., Yuan, F., Lo, K.L., Fu, Y.: Identification and improvement of probabilistic voltage instability modes of power system with wind power integration. Electric Power Syst. Res. 140, 162–172 (2016) 3. Xydas, E., Qadrdan, M., Marmaras, C., Cipcigan, L., Jenkins, N., Ameli, H.: Probabilistic wind power forecasting and its application in the scheduling of gas-fired generators. Appl. Energy 192, 382–394 (2017) 4. Botterud, A., Zhou, Z., Wang, J., Sumaili, J., Keko, H., Mendes, J., Bessa, R.J., Miranda, V.: Demand dispatch and probabilistic wind power forecasting in unit commitment and economic dispatch: a case study of illinois. IEEE Trans. Sustain. Energy 4(1), 250–261 (2013) 5. Matos, M.A., Bessa, R.J.: Setting the operating reserve using probabilistic wind power forecasts. IEEE Trans. Power Syst. 26(2), 594–603 (2011) 6. Kou, P., Gao, F., Guan, X.: Stochastic predictive control of battery energy storage for wind farm dispatching: using probabilistic wind power forecasts. Renew. Energy 80, 286–300 (2015) 7. Botterud, A., Zhou, Z., Wang, J., Valenzuela, J., Sumaili, J., Bessa, R.J., Keko, H., Miranda, V.: Unit commitment and operating reserves with probabilistic wind power forecasts. In: 2011 IEEE Trondheim PowerTech, pp. 1–7, June 2011 8. Zhang, Y., Wang, J., Wang, X.: Review on probabilistic forecasting of wind power generation. Renew. Sustain. Energy Rev. 32, 255–270 (2014) 9. Li, P., Guan, X., Wu, J.: Aggregated wind power generation probabilistic forecasting based on particle filter. Energy Convers. Manag. 96, 579–587 (2015) 10. Huang, C.M., Huang, Y.C., Huang, K.Y., Chen, S.J., Yang, S.P.: Deterministic and probabilistic wind power forecasting using a hybrid method. In: 2017 IEEE International Conference on Industrial Technology (ICIT), pp. 400–405, March 2017 11. Wan, C., Lin, J., Wang, J., Song, Y., Dong, Z.Y.: Direct quantile regression for nonparametric probabilistic forecasting of wind power generation. IEEE Trans. Power Syst. 32(4), 2767–2778 (2017) 12. Yan, J., Li, K., Bai, E.W., Deng, J., Foley, A.M.: Hybrid probabilistic wind power forecasting using temporally local gaussian process. IEEE Trans. Sustain. Energy 7(1), 87–95 (2016) 13. Kou, P., Liang, D., Gao, F., Gao, L.: Probabilistic wind power forecasting with online model selection and warped Gaussian process. Energy Convers. Manag. 84, 649–663 (2014)

Gaussian Process Based Method

147

14. Mangalova, E., Shesterneva, O.: K-nearest neighbors for GEFCom2014 probabilistic wind power forecasting. Int. J. Forecasting 32(3), 1067–1073 (2016) 15. Zhang, Y., Wang, J.: K-nearest neighbors and a kernel density estimator for GEFCom2014 probabilistic wind power forecasting. Int. J. Forecasting 32(3), 1074–1080 (2016) 16. Gallego-Castillo, C., Bessa, R., Cavalcante, L., Lopez-Garcia, O.: On-line quantile regression in the RKHS (reproducing kernel hilbert space) for operational probabilistic forecasting of wind power. Energy 113, 355–365 (2016) 17. Aguilar, S., Souza, R.C., Pensanha, J.F.: Predicting probabilistic wind power generation using nonparametric techniques. In: 2014 International Conference on Renewable Energy Research and Application (ICRERA), pp. 709–712, October 2014 18. Wei, Z., Liu, S.M., Wei, D., Wang, Z.J., Yang, M.L., Li, Y.: Probabilistic wind power forecast using sparse Bayesian learning of unified kernel function. In: 2014 IEEE Conference and Expo Transportation Electrification Asia-Pacific (ITEC Asia-Pacific), pp. 1–4, August 2014 19. Lee, D., Baldick, R.: Probabilistic wind power forecasting based on the Laplace distribution and golden search. In: 2016 IEEE/PES Transmission and Distribution Conference and Exposition (TD), pp. 1–5, May 2016 20. Sideratos, G., Hatziargyriou, N.D.: Probabilistic wind power forecasting using radial basis function neural networks. IEEE Trans. Power Syst. 27(4), 1788–1796 (2012) 21. Zhang, Y., Wang, J., Luo, X.: Probabilistic wind power forecasting based on logarithmic transformation and boundary kernel. Energy Convers. Manag. 96, 440–451 (2015) 22. Carpinone, A., Langella, R., Testa, A., Giorgio, M.: Very short-term probabilistic wind power forecasting based on Markov chain models. In: 2010 IEEE 11th International Conference on Probabilistic Methods Applied to Power Systems, pp. 107–112, June 2010 23. Dowell, J., Pinson, P.: Very-short-term probabilistic wind power forecasts by sparse vector autoregression. IEEE Trans. Smart Grid 7(2), 763–770 (2016) 24. Nasraoui, K., Lakhoua, N., Amraoui, L.E.: Study and analysis of micro smart grid using the modeling language SysML. In: 2017 International Conference on Green Energy Conversion Systems (GECS), pp. 1–8, March 2017 25. Lakhoua, M.: Systemic analysis of a wind power station in Tunisia. J. Electr. Electron. Eng. 4, 83–88 (2011) 26. Do, C.B.: Gaussian Processes. Stanford University, Stanford (2007) 27. Meinshausen, N.: Quantile regression forests. J. Mach. Learn. Res. 7, 983–999 (2006) 28. Weron, R.: Electricity price forecasting: a review of the state-of-the-art with a look into the future. Int. J. Forecasting 30(4), 1030–1081 (2014)

UWB-MIMO Array Antennas with DGS Decoupling Structure Chafai Abdelhamid1(B) , Marwa Daghari1(B) , Chafaa Hamrouni1(B) , and Hedi Sakli2(B) 1

2

Research Laboratory of Modeling Analysis and Control of Systems, National Engineering School of Gabes, 6029 Gabes, Tunisia [email protected], [email protected], [email protected] SYSCOM Research Laboratory, National Engineering School of Tunis, Tunis El Manar University, Le Belvedere, 1002 Tunis, Tunisia [email protected]

Abstract. In this paper, we present a new Ultra Wide Band (UWB) rectangular patch antenna conception method. Antenna impedance and pass band are studied depending on the dimensions of the notches introduction on developed patch and its presented ground. Obtained bandwidth is between 3.3 GHz and 17.8 GHz. The maximum gain is under 3.5 dBi and 8.7 dBi depending on operating frequency. The suggested UWB antenna is characterized by simplicity of design and power supply, a very low cost and a relatively stable radiation pattern over a very wide range. In addition, a network of two UWB-MIMO antennas with a DGS (Defected Ground Structure) decoupling structure, which consists in creating a slot in the ground plane of two antennas, is proposed. This isolation structure represents a notch filter to minimize the effect between adjacent antennas. Compared to the UWB-MIMO network without DGS, the proposed antenna array not only maintains the performance of each antenna, that reduces the coupling of 25 dB at 5.4 GHz and 14 dB at 9.8 GHz and provides total isolation more than 15 dB in the bandwidth spectrum [3.3–17.8] GHz. Therefore, it is suitable for satellite communications in the C, X, and Ku bands. Keywords: UWB-MIMO antenna · Isolation of antennas DGS Technique · MIMO satellite communications in C, X, and Ku bands

1

·

Introduction

New radio telecommunications systems development presents a very important evolution of data transmission point of view in several fields of military and commercial application. Consequently, this evolution requires a new antenna architecture capable of adapting to these evolutions [1], which leads to the design of antennas satisfying numerous constraints such as: low manufacturing cost, c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 148–155, 2020. https://doi.org/10.1007/978-3-030-21009-0_13

UWB-MIMO Array Antennas with DGS Decoupling Structure

149

compactness and UWB operation. In general, the major drawback affecting the limitation of a patch antenna in microstrip technology is its narrow bandwidth, which is intrinsically linked to its resonant nature [2]. UWB antennas applications are various in long range and new generations of wireless networks like X-band (10.7–12.75 GHz), and Ku (12.75–13.25 GHz, 13.75) −14.5 GHz, 17.3– 18.4 GHz, 17.3–17.7 GHz, 17.7–19.7 GHz) [3–20]. Several techniques have already been applied to broadband antennas design. By way of example, the addition of the different shapes of the slots at the level of the radiating element for the production of broad strips [4,5,14]. In this paper, we propose a rectangular patch antenna on FR-4 substrate of very wide bandwidth, using a notch at the level of the ground plane with a change of shape for the radiating element. Compared with traditional microstrip antennas, the bandwidth is around 14.5 GHz assuming a reflection factor limit of −10 dBi. A DGS isolation technique based on the introduction of the slits into the ground plane is then used to decouple the two UWB-MIMO antennas placed side by side. Decoupling of 15 dB in the band [3.3–17.8] GHz is obtained.

2

UWB Antenna Structure Design

By tuning ultra-wide band technology into mobile terminals. The substrate used is the FR4-epoxy permittivity 4.4, loss tangent 0.02 and thickness 1.6 mm. Initially starting from a square patch fed by a microstrip line on substrate FR-4 containing in the underside a partial ground plane, this construction can not cover a very wide band. As a result, we transform the profile in front of the patch as shown in Fig. 1 by introducing circular notches, because the latter is focused on the antenna’s sides. Figure 1 displays the shape of the given UWB antenna in front view and bottom view. The optimized geometrical parameters of this antenna are given in Table 1. The curve in Fig. 2 shows the simulation under HFSS of the reflection coefficient of this antenna. It is noted that the antenna has a very wide band of width 14.5 GHz which varies from 3.3 GHz to 17.8 GHz. From Fig. 3 it can be seen that the VSWR stationary wave ratio values of the proposed patch antenna as a function of the frequency are less than

Fig. 1. Proposed UWB antenna configuration. (a) Front view and (b) Back view.

150

C. Abdelhamid et al.

2 in the band indicated above. The antenna is well adapted. Simulation results of the antenna with the optimum dimensions show that the maximum gain is greater than 3.5 dBi as a function of the frequency in the band studied and which reaches it´s maximum of 8 dBi at 11.3 GHz and 15 GHz. The suggested UWB antenna radiation pattern is described in Fig. 4 for both 4 GHz, 11.3 GHz and 15 GHz frequencies. The number of lobes increases with increasing frequency due to the existence of higher order modes. Table 1. Optimized parameters of proposed antenna UWB. Parameters Values Parameters Values R1

1.5

W3

7.8

R2

1.5

W4

1.66

R3

1

W5

0.3

R4

1

L1

7.5

R5

1

L2

1.5

R6

1

WL

3

W1

10

LL

0.76

W2

4

LC

3.3

Lsub

35

Wsub

30

Lg

12

Wg

30

Wp

15

Lp

14.5

Lf

13.6

Wf

2.8

Fig. 2. Antenna return loss.

Figure 5 demonstrates that the gain curve of this antenna in the plane E (y–z plane) and in the plane H (x–z plane) for the frequencies 4 GHz, 8 GHz, 11.3 GHz and 15 GHz. The suggested UWB antenna delivers a moderately constant radiation pattern over a very wide band. It can be seen that the distribution of the surface currents is very important at the edges of the supply line is maximum

UWB-MIMO Array Antennas with DGS Decoupling Structure

151

Fig. 3. VSWR simulation of the proposed antenna UWB.

Fig. 4. Maximal gain of the UWB antenna for the proposed antenna; f = 4 GHz, (b) f = 11.3 GHz, (c) f = 15 GHz.

on the edges of the line and the patch for low frequencies. As the frequency increases, this distribution becomes more and more important on the edges of the patch as well as on the line-patch transition (in Fig. 6).

Fig. 5. The Simulated radiation pattern, (a) E-plane and (b) H-plane, for the suggested antenna for f = 4 GHz, f = 8 GHz, f = 11.3 GHz and f = 15 GHz.

It can be seen that the distribution of the surface currents is very important at the edges of the supply line is maximum on the edges of the line and the patch for low frequencies. As the frequency increases, this distribution becomes more and more important on the edges of the patch as well as on the line-patch transition (in Fig. 6).

152

C. Abdelhamid et al.

Fig. 6. E field of the suggested antenna for f = 4 GHz, 8 GHz, 11.3 GHz and 15 GHz.

3

Decoupling Structure Design

The DGS main method is based on changes in the ground plane to change the present distribution on this plane. The DGS method can provide both slow wave and band cut filter characteristics. In antenna design, this method has been used to improve the isolation of a dual polarization patch antenna. In multiantenna systems, the effect of the band-cut filter is used to decrease the common coupling among antennal components. A common use of the DGS method is to insert slots on the ground plane. The efficiency of the DGS decoupling method makes it applied to different kinds of antennas. The application of the DGS structure to different multi-antenna systems does not require much modification, as it´s operation depends on the resonant frequency rather than the types of antennas. The simulation results show good transmission quality with low power loss over the entire band by reading slits at the S12 100 mA (the saturation of the measuring instrument).

Fig. 5. Measured gate leakage current during the power cycling test. Note that DUTs 2 to 5 are driven at VG = 5 V and DUTs 6 to 8 are driven at VG = 4 V.

Drift Effects and Trap Analysis of Power-GaN-HEMT

191

In contrast, the DUTs driven at VG = 4 V present a gate leakage current below the max given by datasheet. We observed that there is an initial degradation after S1, and then the gate leakage current remains stable during the rest of the power cycling (Fig. 6).

Fig. 6. Measured gate threshold voltage during the power cycling. The curve shows that the VTH shift after the power cycling is more significant in DUTs driven at VG = 5 V.

Figures 7 and 8 shows the evolution of the CG(V) curve of a DUT driven at VG = 5 V and VG = 4 V during the power cycling.

Fig. 7. CG(V) characteristic a GaN HEMT driven at VG = 5 V. In blue, before the power cycling, in green after stress S1 and in red after S2.

192

M. A. González-Sentís et al.

Cg(Vg) at S0

3.7E-10

Cg(Vg) at S1

Capacitance (F)

Cg(Vg) at S2 center of S0

3.3E-10

center of S1 center of S2

2.9E-10

2.5E-10

0

1

VG (V)

2

3

4

Fig. 8. CG(V) characteristic a GaN HEMT driven at VG = 4 V

We can observe that the hysteresis surface increases in devices driven at VG = 5 V more than in devices driven at VG = 4 V. By observing the center of each surface, note that the shift of the CG(V) characteristic is similar that the shift on the threshold voltage shown in Fig. 6.

5 Conclusion and Future Work We have introduced a power cycling test environment for a discrete GaN power transistors tacking into account the specific failure mechanisms in power GaN HEMT and thermo-mechanical issues. Failure progresses are observed with measuring diverse electrical parameters. We suggested CG(V) curves to investigate the evolution of traps during the test, and the CG(V) surface hysteresis as an interesting degradation indicator to track the evolution of trapping effects. We propose that the new traps created during the power cycling may be originated by overshoots on the gate voltage during the power cycling on DUTs driven at VG = 5 V. Therefore, future research should be directed in order to approve or deny certain hypotheses and to clarify the origin of the traps by realizing C-DLTS measurements, and RDSON measurements to make a link between CG(V) hysteresis and Dynamic On-Resistance degradation. Acknowledgments. This work is conducted in the frame of the IRT Saint-Exupery Robustness Electronic project sponsored by Airbus Operations, Airbus Group Innovations, Continental Automotive France, Hirex Engineering, Nexio, Safran Electrical & Power, Thales Alenia Space France, Thales Avionics and the French National Agency for Research (ANR).

Drift Effects and Trap Analysis of Power-GaN-HEMT

193

References 1. Lutz, J., Franke, J.: Reliability and reliability investigation of wide-bandgap power devices, invited paper ESREF (2018) 2. Fu, J.Z., Fouquet, F., et al.: Evolution of C-V and I-V characteristics for a commercial 600 V GaN GIT power device under repetitive short-circuit tests. ESREF (2018) 3. Meneghini, M., et al.: Power GaN Devices, pp. 299–300. Springer, Berlin (2017) 4. Li, K., et al.: GaN-HEMT RDSON characterisation and modelling. COMPEL IEEE (2016) 5. EPC Corporation: EPC2019 Enhancement Mode Power Transistor, datasheet, Sep. 2015 6. Lidow, A., et al.: eGaN FET drivers and layout considerations. EPC Corporation (2016) 7. Song, S., et al.: Failure mechanism analysis of a discrete 650 V enhancement mode GaN-onSi power device with reverse conduction accelerated power cycling test. IEE (2017) 8. Meneghini, M., et al.: Reliability and failure analysis in power GaNHEMTs: an overview. IEEE, June 2017 9. Dobresc, L., Petrov, M., et al.: Threshold voltage extraction methods for MOS transistors. CAS 2000 Proceedings IEEE, Oct 2000 10. Meneghini, M., et al.: Buffer traps in Fe-Doped AlGaN/GaN HEMTs. IEEE Dec 2014 11. Saadaoui, S., Ben Salem, M.M., Gassoumi, M., Maaref, H., Gaquière, C.: J. Appl. Phys. 110 (2011), 013701

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems Hussein El Ghor1(B) , Julia Hage2 , Nizar Hamadeh1 , and Rafic Hage Chehade1 1

2

LENS Laboratory, Faculty of Technology, Lebanese University, B.P. 813, Saida, Lebanon [email protected] Faculty of Technology, Lebanese University, B.P. 813, Saida, Lebanon

Abstract. In this paper, we investigated the problem of developing scheduling techniques for uniprocessor real-time systems that enhances energy saving while still tolerating up to k transient faults to preserve the system’s reliability. Two scheduling algorithms are proposed: The first scheduler is an extension of an optimal fault-free energy-efficient scheduling algorithm, named ES-DVFS. The second algorithm aims to decrease the consumption of energy by using the slack time for the recovery operation when faults occur. The experimental results show that the proposed approach significantly reduces the consumption of energy when compared to the previous schedulers. Keywords: Real-time scheduling · Fault-tolerant Energy management · Energy harvesting

1

· Checkpointing ·

Introduction

Many embedded real-time systems usually operate in harsh environments. To function correctly, they have to respect the timing constraints and at the same time decrease energy consumption even in the presence of faults. Therefore, besides their timing and energy constraints, these systems usually have serious fault-tolerant limitations. Energy management is achieved by the most popular solution, namely dynamic voltage and frequency scaling (DVFS) [1,2], with the aim to reduce energy consumption during system operation and to prolong the battery lifetime by dynamically scaling down the processor supply voltage as much as possible and without violating the tasks deadlines. In reality, processor faults can be categorized as: transient and permanent faults [3]. We focus in this paper on the transient fault since, in most computing systems, the majority of errors are due to transient faults [4]. In the case of an energy-efficient system, reliability also means ensuring that the system will never be short of energy to ensure its treatment. Anticipation of possible cases of energy can, again, be implemented on the basis of the flexibility offered by the system at the level of execution of tasks. c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 194–203, 2020. https://doi.org/10.1007/978-3-030-21009-0_18

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems

195

In this work, we focused on the problem of real-time scheduling under reliability and energy constraints. Its about considering real-time jobs that have needs which are expressed, on the one hand, in terms of processing time and energy consumed by the processor and, on the other hand, in terms of the number of tolerated faults. A job configuration is energy overloaded, this means that the amount of energy consumed is greater than the amount of available energy. In addition, the amount of execution time requested is smaller than the available capacity, the system will therefore typically be able to meet all its deadlines or else catastrophic consequences will occur. A major question that needs to be answered is: how to schedule real-time jobs with energy constraints where the system keeps reliable and able to tolerate up to k faults. To answer this question, we first propose a uniprocessor Earliest Deadline First (EDF) scheduling algorithm and we then derive an exact and efficient feasibility condition by considering energy management and fault-tolerance. Second, the proposed algorithm is derived to achieve energy autonomous utilization of the processor while respecting deadlines of each task in the task set. The rest of the paper is organized as follows. In the next section, we summarize the related work. In Sect. 3, we present the model and terminology. The fault tolerant speed schedule was then presented in Sect. 4. Section 5 shows through experimental results the energy savings of the proposed algorithms and Sect. 6 concludes the paper.

2

Related Work

In both industry and academia fields, researchers have found some techniques to enhance energy saving in embedded systems. Among these, DVFS has risen as one of the best framework level methods for energy consumption. DVFS scheduling reduces the supply voltage and frequency when conceivable for preserving energy consumption. Subsequently, a large number of procedures considering the issue of limiting the consumption of the needed energy without violating the timing constraints on uniprocessor systems are widely presented in literature for different task models. Many of the previous work that studied the problem of energy efficient frameworks for real-time embedded systems apply the DVFS technique to reduce the processor energy consumption [5–8]. In [6], authors proposed a DVFS scheme under EDF scheduling policy to decrease dynamic power consumption for real-time systems. In [8], we settle the hypothesis for energy consumption in real-time systems, we proposed an energy efficient real-time scheduling algorithm of aperiodic tasks for wireless sensors. Specifically, we applied the concept of DVFS technique to the process of real-time scheduling. Further, we proposed in [9] an energy guarantee real-time scheduling algorithm that applies the DVFS technique targeting energy harvesting systems. We show that our scheduler achieves capacity savings when compared to other schedulers. On the other side, fault tolerance objectives are of uppermost importance for embedded systems [10]: system failures can occur in real-time computing

196

H. El Ghor et al.

systems and can result in hardware errors and/or deadline misses. It was found that the most common errors in computing systems are soft errors, and hence most researches target their work on soft errors to present fault tolerant systems. Such research efforts produced scheduling algorithms with the joint consideration of energy and timing constraints in fault tolerant systems. More recently, Zhao et al. [11] presented the Generalized Shared Recovery (GSHR) technique to reserve computing resources, which can be used by other tasks to enhance the energy efficiency. Later, this work was extended to be applied to a real-time periodic task model [10]. The proposed algorithms aim to determine the processor scaling factor and the reserved resources for every task to enhance the minimization of energy while still guaranteeing the reliability requirement at the task-level. The advantage of the GSHR scheduler comes from the fact that the reliability of the system can be increased when applying the DVFS technique. Recently, Han et al. developed effective scheduling algorithms that can save energy when considering that the proposed real-time system can tolerate up to k failures when scheduling a set of aperiodic tasks on a single processor under the EDF policy [12]. For this sake, authors proposed three algorithms: The first two algorithms are based on the previous work performed in [6]. The third algorithm extends the first two by considering that the computing resources are no longer reserved and hence better energy saving performance can be achieved. The main drawback of this work is that the problem of improving the system reliability in presence of failures cannot be solved by a simple modification to the work done in [6].

3

Model and Terminology

We consider the system model and their corresponding notations. Then, we present the problem formulation. 3.1

Task Model

We consider a set of n independent aperiodic real-time jobs J = {J1 , J2 , · · · , Jn }, where Ji denotes the ith job in J and is characterized by a three tuple (ai , ci , di ). The definition of these parameters are as follows: – ai is referred to the arrival time, this means that the time when job Ji is ready for execution. – ci is referred to the worst case execution time (WCET) under the maximum available speed Smax of the processor. – di is considered as the absolute deadline of job Ji . We denote the laxity of the job Ji by di − (ai − ci ). We consider that the job set J is said to be feasible in the real-time manner and under fault-free scenario. In other words, there exists a feasible schedule for J in abscence of energy considerations, where all deadlines in are respected.

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems

3.2

197

Power and Energy Model

We assume the speed/frequency of the processor is equipped with a DVFSenabled with N discrete frequencies f ranging from fmin = f1 ≤ f2 ≤ · · · ≤ fN = fmax . We consider the notation processor speed SN , or slowdown factor, as the ratio of the computed speed to the highest processor speed, this means that SN = fN /fmax . The CPU speed can be changed continuously in [Smin , Smax ]. Consequently, when a job Ji is executed under speed Si , the worst case execution time of Ji becomes equal to ci /Si . For embedded systems, the processor and off-chip devices such as memory, I/O interfaces and underlying circuits mainly consume the major part of the energy [13]. In this paper, we distinguish between frequency-dependent and frequency-independent components of the consumed power. Specifically, we adopt the overall power consumption (P ) at a slowdown factor S as follows: P = Pind + Pdep = Pind + Cef S α

(1)

Where Pind stands for the frequency-independent power that includes the constant leakage power and the power consumed by off-chip devices [12], which is independent of the system frequency and supply voltage. Cef is denoted as the effective switching capacitance. α is the dynamic power exponent, which is a constant usually larger than or equal to 2. Pdep is considered to be the frequency-dependent active power, which includes not only the processor power, but also any power that depends on the processing speed S. Consequently, the energy consumption of a job Ji that runs at the speed Si , denoted as Ei (Si ), can be expressed as: Ei (Si ) = (Pind + Cef Siα ). 3.3

ci Si

(2)

Energy Storage Model

The used system relies on an energy storage unit (battery or supercapacitor) with an ideal capacity, namely C, that corresponds to a maximum stored energy. The energy level of the battery must remain between two predefined boundaries, namely Cmin and Cmax , where C = Cmax − Cmin . We consider that C(t) stands for the level of energy in the battery at time t. We state that the stored energy at any time is less than the ideal storage capacity, this means C(t) ≤ C 3.4

∀t

(3)

Fault Model

During the execution of any operation on a computing system, both transient and permanent faults may affect the system due to various reasons, like hardware defects or system errors. In this paper, we consider only transient faults since it has been shown to be dominant over permanent faults especially with scaled technology sizes [14].

198

H. El Ghor et al.

The proposed system can afford a maximum of k transient faults. The used system is usually able to detect faults when a job ends its execution. We assume that the energy and time overhead caused by fault detection, denoted as EOi and T Oi respectively, are not negligible and are independent of the variations in the processor frequency. Generally, there is not restriction on the occurrence of faults during the execution of jobs and multiple faults may occur when executing a single job [12]. The fault recovery scheme in this paper is based on re-executing the affected job. Consequently, Ri stands for the maximum recovery overhead for executing a job Ji under the maximum speed Smax , which is equal to ci , or Ri = ci . When a fault occurs during any job execution, say Ji , a recovery job of the same deadline di is released, which is subject to preemption as well.

4 4.1

Fault Tolerant Speed Schedule Overview of the Scheduling Scheme

In this section, we present a fault-tolerant DVFS scheduling approach for a dynamic-priority real-time job set on uniprocessor systems to enhance energy saving while still guaranteeing the timing constraints. The proposed algorithm is based on the Energy Saving - Dynamic Voltage and Frequency Scaling (ESDVFS) algorithm that we previously proposed in [8]. To better understand tthe proposed approach and before proceeding, we first state some basic definitions and then briefly reiterate the general concept of ESDVFS. Definition 1. Given a real-time job set J of n independent aperiodic jobs such that J = {J1 , J2 , · · · , Jn }. – J (ts , tf ) denotes the job set contained in the time interval φ = [ts , tf ], i.e jobs that are ready to be executed at time ts and with deadlines smaller than or equal to tf . J (φ) = {Ji | ts ≤ ai < di ≤ tf }. – W (φ) denotes the overall amount of the jobs’ workload in J (φ) in the time interval [ts , tf ], that means that the total worst case processing time of jobs completely embedded in the time interval,  W (φ) = ci (4) Ji J (φ)

– The processor load h(φ) over an interval φ = [ts , tf ] is defined as h(φ) =

W (φ) tf − ts

(5)

– The intensity of jobs in the time interval φ = [ts , tf ], denoted as I(φ), is defined as    ci I(φ) = max

Jj J (φ)

di ≤dj

dj − (tf − ts )

(6)

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems

199

– We consider that the fault-related overhead of a time interval φ = [ts , tf ], denoted as Wk (φ) is Wk (φ) = Wr (φ) + WT O (φ)

(7)

Where Wr (φ) stands for the worst-case workload that is reserved to be used in case of recovery, i.e. Wr (φ) = k × (Rl + T Ol ) and l represents the index of the job with the maximum recovery time in J (φ). Jl = {Ji | max(Ri + T Oi ), Ji J (tφ)} and WT O (φ) is considered as the overhead due to fault detection from regular jobs, i.e.  WT O (φ) = T Oi (8) Ji J (φ)

Further, Wk (φ) ≥ Wk−1 (φ) for k ≥ 1, since all recovery of jobs have positive execution times. For this sake, we restrict our work to a k-fault tolerant system that can exactly tolerate k faults when investigating the worst-case reserved recovery of fault scenarios. – The energy demand of a job set J in the interval φ = [ts , tf ] is  Ek (Sk ) (9) g(φ) = ts ≤rk ,dk ≤tf

Given a real-time job set J , ES-DVFS was provably optimal in minimizing energy consumption in on-line energy-constrained setting by providing sound dynamic speed reduction mechanisms [8]. The ES-DVFS approach can provide a feasible energy efficient technique, which is function of the processor frequency where the time constraints of all jobs in J are still respected. Under this assumption, the ready jobs in the used interval are not executed with fixed speed as the previous work in [6], but are dynamically adjusted on the fly. 4.2

Concepts for the EMES-DVFS Model

ES-DVFS is optimal in case of fault-free conditions. Hence, To make the above ES-DVFS fault-tolerant, we adopt a scheduler (we call it MES-DVFS) is to take into consideration the fault recovery when calculating the effective processor load and intensity in any interval φ = [ts , tf ], i.e. to replace h(φ) and I(φ) with hm (φ) and Im (φ) respectively, such that  ci + k × Rl hm (φ) =

Ji J (φ)

dmax − WT O (φ) − k × T Ol

(10)

Where dmax is the longest deadline in J (φ) and WT O (φ) stands for the overall overheads due to fault-detection for original jobs as defined in Definition 1. In addition, the intensity of the jobs in J (φ) at current time t is    ci + k × Rl Im (t) = max

Jj J (φ)

di ≤dj

dj − t − WT O (φ) − k × T Ol )

(11)

200

H. El Ghor et al.

When a fault is detected, and for the sake of reducing the total energy consumption for the regular jobs and their recovery copies, MES-DVFS runs the copy of the recovered job using a defined processor speed (Si ≤ Smax ). However, this may not be energy efficient since, in practice, the fault rate is considered to be very low. An extended approach for MES-DVFS (we call it EMES-DVFS), is to execute the recovery copies under the highest possible processor speed, usually at Smax . Hence, the intensity calculation of the jobs in J (φ) can be modified correspondingly, as Eq. 12    ci Ie (t) = max

di ≤dj

Jj J (φ)

dj − t − Wk (φ)

(12)

Further, the effective processor load of the jobs in J (φ) can also be modified correspondingly, as Eq. 13  ci he (φ) = 4.3

Ji J (φ)

dmax − Wk (φ)

(13)

Description of the EMES-DVFS Scheduler

We consider a job set J of n jobs J = {J1 , J2 , · · · , Jn } that can tolerate up to k faults. Let Q(φ) be the list of ready but uncompleted jobs for execution in the time interval φ = [ts , tf ]. We can formulate our EMLPEDF algorithm to obey the following rules: Rule 1: The EDF scheduler is used to select the future running jobs in Q(φ). Rule 2: The processor is imperatively idle in [ts , ts + 1) if Q(φ) is empty. Rule 3: The processor is imperatively busy in [ts , ts + 1) if Q(φ) is not empty and 0 < C(ts ) ≤ C. Hence, the following steps must be performed: 1. Select the job, say Ji with the highest priority. 2. Calculate the effective processor load he (φ) and intensity Ie (φ) using Eqs. 13 and 12 respectively. 3. Set the speed Sei of job Ji to the maximum between he (φ) and Ie (φ). Rule 4: If Se i < Smin , then Se i = Smin ∀ Ji J (φ). Rule 5: If job, say Jj is released with dj < di , then update Sei by Rule 3. Rule 6: If job, say Jk is released with dk > di , then complete the execution of Ji . Rule 7: If job, say Jk is released with dk > di , and ck > dk − di then update Sei by Rule 3. Rule 8: Calculate the energy consumption Ei (Se i ) according to Eq. (2). Rule 9: Calculate the energy level in the battery when the job ends its execution. Rule 10: Remove job Ji from the queue Q(φ). Rule 11: Repeat step (1)–(8) until the queue Q is empty.

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems

5

201

Simulation Results

We compare the performance of four scheduling algorithms: EMES-DVFS, MESDVFS, NPM and LPSSR proposed in [12]. NPM scheme executes jobs with maximum frequency and does not scale down the voltage/frequency. We developed a discrete event-driven simulator in C that generates a job set J where the number of jobs varies from 10 to 50. The simulation is repeated 100 times for the same number of jobs. For the sake of clarity, we use NPM as a reference schedule that represents the schedule of given set of jobs J without incorporating DVFS. We consider that all the plotted energy consumptions are normalized to NPM. We consider that α = 2, Cef = 1, Pind = 0.05, and Smin is equal to 0.25. We compute the simulation results by using a discrete DVFS processor that operates on 8 frequency levels {1.00, 0.86, 0.76, 0.67, 0.57, 0.47, 0.38, 0.28} as in the PentiumM processor. 5.1

Experiment 1: Energy Consumption by Varying the Number of Jobs

First, we take interest in how energy consumption of the processor changes when we vary the number of jobs. We report here the results of four simulation studies where the fault rate is set to 10−5 and the number of jobs varies from 10 to 50. Further, we consider that the number of faults is strictly equal to 1 (k = 1). In each of the four schedulers, we compute the normalized energy consumption metric of the used speed. Figure 1 shows the expected energy consumption of EMES-DVFS and MES-DVFS versus previous schedulers like NPM and LPSSR. From Fig. 1, we find that the energy consumption of the four schedulers increases as the number of jobs becomes larger. This is reasonable since the likelihood of having large slack time that can be used for DVFS is diminishing as we increase the number of jobs. Further, a significant amount of energy saving is gained by EMES-DVFS and MES-DVFS schemes since it can benefit from the significant amount of slack time that decreases the expected consumption of energy. In other words, EMES-DVFS and MES-DVFS can effectively assign the speeds for every job such that the job set becomes feasible at a speed closest to the critical speed. When the number of jobs is low, we find that the reduction in energy consumption that is achieved by the tested algorithms are approximately the same. This is because most jobs are executed at the lowest speed. With the increasing number of jobs, our scheduler demonstrates its great advantage in achieving a high energy saving. As an average, EMES-DVFS can achieve an additional 51% and 20% of energy saving when we compare it with NPM and LPSSR, respectively. Further, the difference in energy consumption is around 12% between EMES-DVFS can MES-DVFS, since in the EMES-DVFS scenario, the recovery from one faulty job is performed at maximum processor speed and subsequent slack is left for DVFS.

202

H. El Ghor et al. 1.2 LPSSR

MES−DVFS

EMES−DVFS

NPM

Normalized Energy Consumption

1

0.8

0.6

0.4

0.2

0

5

10

15

20

25

30 35 Number of Jobs

40

45

50

55

Fig. 1. Energy savings by varying the numbers of jobs, k = 1

We conclude that our approach gains more energy savings in such a way that it can benefit from the slacks generated during execution and hence it can use all the available slack time.

6

Conclusions

In this work, we proposed and evaluated a new novel approach, which aims to enhance energy savings when scheduling a real-time job set that can tolerate up to k transient faults while still respecting time and energy constraints. We benefit from the slacks generated during run-time to the maximum extent in such a way that all the available slack time is used for energy reduction, which is carried out using dynamic voltage and frequency scaling (DVFS). Under this notion, we propose an algorithm that estimates an optimal speed reduction mechanism which maintains feasibility within predefined timing constraints when no more than k faults occur. Our scheduler dynamically adjusts the jobs’ slowdown factors by using the run-time slacks which may be increased for recovery demands of the system. It differs from the previous approaches where the assignments of job frequencies are predetermined, and hence it is more flexible and adaptive in minimizing energy consumption while still keeping the systems reliability at a desired level. Simulation results proved that the presented scheduler can significantly reduce energy consumption when compared with the existing works.

Energy-Aware Fault-Tolerant Real-Time Scheduling for Embedded Systems

203

References 1. Shin, Y., Choi, K., Sakurai, T.: Power optimization of real-time embedded systems on variable speed processors. In: Proceedings of the International Conference on Computer-Aided Design, pp. 365–368 (2000). https://doi.org/10.1109/ICCAD. 2000.896499 2. Quan, G., Hu, X.: Energy efficient fixed-priority scheduling for real-time systems on variable voltage processors. In: Proceedings of the Design Automation Conference, pp. 828–833 (2001). https://doi.org/10.1109/DAC.2001.156251 3. Srinivasan, J., Adve, S.V., Bose, P., Rivers, J., Hu, C.K.: Ramp: a model for reliability aware microprocessor design. IBM Research Report, RC23048 (2003) 4. Castillo, X., McConnel, S.R., Siewiorek, D.P.: Derivation and calibration of a transient error reliability model. IEEE Trans. Comput. 31, 658–671 (1982). https:// doi.org/10.1109/TC.1982.1676063 5. Aydin, H., Melhem, R., Mosse, D., Mejia-Alvarez, P.: Power-aware scheduling for periodic real-time tasks. IEEE Trans. Comput. 53(5), 584–600 (2004) 6. Yao, F., Demers, A., Shenker, S.: A scheduling model for reduced CPU energy. In: Proceedings of the 36th Annual Symposium on Foundations of Computer Science, pp. 374–382, October 1995 7. Zhang, Y., Chakrabarty, K., Swaminathan, V.: Energy-aware fault tolerance in fixed-priority real-time embedded systems. In: Proceedings of the 2003 IEEE/ACM International Conference on Computer-Aided Design, ICCAD 2003 (2003) 8. El Ghor, H., Aggoune, E.M.: Energy efficient scheduler of aperiodic jobs for realtime embedded systems. Int. J. Autom. Comput. 1–11 (2016) 9. EL Ghor, H., Chetto, M.: Energy guarantee scheme for real-time systems with energy harvesting constraints. Int. J. Autom. Comput. (to appear) 10. Zhao, B., Aydin, H., Zhu, D.: Energy management under general task-level reliability constraints. In: IEEE 18th Real Time and Embedded Technology and Applications Symposium (2012) 11. Zhao, B., Aydin, H., Zhu, D.: Generalized reliability-oriented energy management for real-time embedded applications. In: 48th ACM/EDAC/IEEE Design Automation Conference (DAC), pp. 381–386, June 2011 12. Han, Q., Niu, L., Quan, G., Ren, S., Ren, S.: Energy efficient fault-tolerant earliest deadline first scheduling for hard real-time systems. Real-Time Syst. 50, 592–619 (2014) 13. Burd, T.D., Brodersen, R.W.: Energy efficient CMOS microprocessor design. In: Proceedings of the HICSS Conference, January 1995 14. Hazucha, P., Svensson, C.: Impact of CMOS technology scaling on the atmospheric neutron soft error rate. IEEE Trans. Nuclear Sci. 47(6), 2586–2594 (2000). https:// doi.org/10.1109/23.903813

A Novel FPGA-Based Digital Filter Using Fuzzy Logic to Ensure Electromagnetic Compatibility Yosr Bchir1(&), Soufien Gdaim1, Djilali Hamza2, and Abdellatif Mtibaa1

2

1 Laboratory of Electronics and Microelectronics of the FSM, National Engineering School of Monastir, University of Monastir, Avenue Ibn al-Jazzar, 5019 Monastir, Tunisia [email protected], [email protected], [email protected] Electrical and Computer Engineering Department, Queen’s University, Kingston, ON, Canada [email protected]

Abstract. Instead of analog Electromagnetic Interference (EMI) filters, digital filters have been the alternate EMI mitigation solution in power converters to overcome the passive and analog active filters ‘limitations such as size, cost, size and high power consumption. In this paper, Common Mode (CM) EMI digital filter based on the use of fuzzy logic technique, is proposed. The filter is modeled and integrated in a field programmable gate Array (FPGA) device. Matlab/Simulink environment is used for the co-simulation PSpice/Simulink and the block sets of Xilinx System Generator (XSG) has been used for a rapid prototyping. The figure-of-merit of the proposed EMI filter is presented. The simulation results using MATLAB/Simulink have been discussed. Keywords: Power inverter


EMI filter
XSG
Fuzzy logic
FPGA

1 Introduction Currently, Power converters are designed to operate at higher switching frequencies with higher voltages and lower losses that enable the reduction of the size and weight of passive components for energy storage. with the emerging switching device technologies such as Silicon Carbide (SiC) and Gallium Nitride (GaN), both conducted and radiated noise emissions have become more significant compared to Silicon (Si) devices based switching converters. Hence, the development of high speed power converters semiconductor devices has aggravated the EMI problems [1, 2]. This fact is mainly due to the increase of the switching frequency of the emerging semiconductor devices, which in turn have an impact on the switching transitions dv/dt and di/dt. The rapid variation in terms of voltage/current induces an EMI noise, and it has been proved that the switching transitions dv/dt well as the common mode voltage generated in ac drives are the root cause of EMI problems in power inverters [3]. The continuous development of power converters implies more efficient and expensive power lines © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 204–212, 2020. https://doi.org/10.1007/978-3-030-21009-0_19

A Novel FPGA-Based Digital Filter Using Fuzzy Logic

205

filters. Complying with Electromagnetic Compatibility (EMC) standards is compulsory in industrial and domestic fields in order to commercialize different products; hence, meeting EMI regulations becomes a challenge for manufacturers. EMI filters have been widely used with power inverters. In fact, passive filtering is the most conventional used technique in reducing EMI noise. Many works was interested in passive filtering thanks to its simplicity of design and performance [4–7], but the disadvantage of high cost and size are discussed by the authors of [8–11] to improve its efficiency and deal with its drawbacks. In [12] active filtering is considered as an alternative to passive one; but it proves an efficiency ups to 1 MHz and this does not cover the range defined by the standard CISPR 22 which considers a frequency range of 150 kHz to 30 MHz [13]. Thanks to the progress of digital processing techniques as well as high speed/precision processors (such ASICs, FPGAs), integrated digital EMI filter has been emerged. In literature, there are some works that deal with the incorporation of digital filtering technique in EMC issues. In [14–17], the authors developed an active digital EMI filter (ADF) to be embedded with the inverter digital controller algorithm in an FPGA chip. The effectiveness of the proposed applications has been proved compared to passive filter. The authors of [18] give an accurate mathematical design of the EMI ADF and specify separately the common mode (CM) and differential mode (DM) noise. In this paper, we propose a design of a novel FPGA-based digital EMI filter in order to overcome conventional filters ‘disadvantages in terms of size, weight and cost, as well as to improve the efficiency of the whole system. The hardware implementation of the established EMI digital active fuzzy filter (DAFF) is developed to be implemented on FPGA using XSG. This paper is organized as follows, in the Sect. 2, the principle of both conventional and fuzzy digital active filters and the hardware design of the developed DAFF using XSG have been presented. Section 3 contains the simulations of the EV’s drive train system with and without integrating the developed DAFF to prove the established concept as well as the interpretation of the simulation results. Finally a conclusion about the paper’s contribution is given.

2 Proposed Digital Active Fuzzy Filter Recently, FPGAs have been frequently used in a wide range of embedded systems that require real-time hardware implementation of their algorithms. FPGA approach allows the implementation of systems in digital signal processing (DSP) with similar features than those using other hardware implementation. However, parallelization and pipelining capability as well as higher sampling rates with FPGA are better than DSP, and low cost and flexibility are advantages as compared to ASICs. XSG has been used to develop hardware-based EMI filter from a system level approach. Matlab/Simulink is employed as a co-design and co-simulation platform for rapid prototyping of an EMI DAFF.

206

2.1

Y. Bchir et al.

Presentation of Conventional Digital EMI Filter

Active noise cancellation technique is realized by using active devices that generate and inject an EMI noise which is the opposite of the EMI system noise. The classical active digital filtering process is based on the same principle of analog active filtering. It contains three elements [14]: – Sensing part: This stage is devoted to sense the noise signal. The used circuit is a RC high-pass filter with a cutoff frequency of 150 kHz. – Control part: The role of this part is to generate a 180° shift signal. – Injection part: This step is about injecting with high fidelity the produced opposed shift noise signal for nullification of the EMI noise signal. The used circuit at this stage is a RC low-pass filter with a cutoff frequency of 30 MHz. The chosen cutoff frequencies aim to comply with the frequency spectrum in accordance with CISPR 22 EMC standards (150 kHz to 30 MHz). 2.2

The DAFF Principle

In DAFF, the control part is substituted by a fuzzy controller that aims to nullify the measured EMI conducted noise. The developed DAFF is a Mamdani-type fuzzy logic controller. Its principle is based on the same principle of active filtering technique. It calculates the difference between the opposite of sensed signal and the injected signal and tends to nullify it. The adjustment of the injected signal depends on the error (e) and its variation in function of time (De). The inputs of the DAFF are the error, (difference between the opposite of the sensed signal and the injected signal) and the derivative of the error signal. The output is the rectified injected signal (see Fig. 1).

Sensed signal

e(k)

ADC

RC high-pass filter

y(k)

DAFF

Comparator e(k-1)

DAC

RC low-pass filter Injected signal

FPGA chip

Fig. 1. The principle of the DAFF

The drive train scheme is designed in Pspice; it includes a battery, a line impedance stability network (LISN) used for EMI measurement, the power inverter and the induction motor (IM). The purpose of this work is to evaluate the effects of switching elements of inverter defined by CM EMI conducted noise. The established design in Pspice is included in Simulink system simulator allowing a single prototype to cosimulate both the established drive train system and the developed DAFF using XSG.

A Novel FPGA-Based Digital Filter Using Fuzzy Logic

207

This co-simulation is executed via SLPS interface: it combines Pspice and Matlab/Simulink for hardware software co-simulation (see Fig. 2).

Sensing circuit

injecting circuit

LISN DAFF

Fig. 2. Schematic of the drive train simulated system with EMI DAFF in MATLAB/Simulink using SLPS interface

Triangular and Trapezoidal membership functions are used for the inputs and the output of the fuzzy logic controller part of the DAFF EMI filter. There are three membership functions of the “error” (e) and the “derivative function of error” (De); five membership functions for the output “rectified signal” (s). The different membership functions of the inputs and the outputs are given by Figs. 3 and 4. Table 1 illustrates the basis of rules of the developed fuzzy system. The defuzzification method for the DAFF is centroid method. It is described by Eq. 1 (where we design by u: the real output, xi: the linguistic variable, and µ: the membership function). P

lðxi Þxi i u¼ P lðxi Þ i

ð1Þ

208

Y. Bchir et al.

Fig. 3. Membership functions of the inputs s

VN

Z

VP

1

0

Fig. 4. Membership functions of the output Table 1. Table of DAFF rules e de P Z N

2.3

P

Z N

Z N VN P Z N VP P Z

FPGA Implementation of the DAFF

Hardware block of the DAFF is established in XSG to be synthesized and downloaded into the target FPGA chip Viretex-5. XSG is a toolbox developed for MATLAB/ Simulink that enables an abstraction algorithm level while keeping traditional Simulink block sets. It generates a synthesizable code VHDL to be directly used for Xilinx FPGA chip implementation. The determination of the fuzzy controller’s characteristics (such as membership functions, defuzzification method, etc.) is empirical; in fact, it is defined by simulations that enable the choice of adequate parameters (see Fig. 2). The design of the DAFF using XSG is given by Fig. 5.

A Novel FPGA-Based Digital Filter Using Fuzzy Logic

209

Fig. 5. Design of the DAFF using XSG

3 Simulation Results and Interpretation The simulation of the system based DAFF is performed using a co-simulation between PSpice and Matlab-Simulink as shown in Fig. 2. The conducted noise is displayed in frequency domain by using the FFT (Fast Fourier Transform) function to evaluate the performance of the proposed approach.

210

Y. Bchir et al.

EMI CM conducted noise (dBμV)

Figures 6 and 7 illustrate respectively the frequency spectrum of EMI CM conducted noise without EMI filter and with integrating the EMI DAFF.

Frequency (Hz)

EMI CM conducted noise (dBμV)

Fig. 6. CM Conducted noise without integrating an EMI filter

Frequency (Hz)

Fig. 7. CM Conducted noise with integrating the EMI DAFF

The first simulation is performed without integrating the EMI filter. As illustrated in Fig. 6, it can be observed that in the first range of frequency from 150 kHz to 0.5 MHz, the CM Conducted EMI noise is attenuated by more than 20dBs: its variation is from 140 dBµV to 126.4 dBµV, the second range of frequency is from 0.5 MHz to 5 MHz and its characterized by some peaks of unwanted EMI noise such as the points (8 MHz, 103.3 dBµV), (10 MHz, 93.22 dBµV), and the third interval of frequency range which is from 5 MHz to 30 MHz is characterized by an average peak 85 dBµV of conducted EMI noise.

A Novel FPGA-Based Digital Filter Using Fuzzy Logic

211

The second simulation is performed with integrating the DAFF as illustrated in Fig. 7. Compared with Fig. 6 it can be seen that a significant attenuation of the CM conducted noise across the frequency range [150 kHz, 30 MHz]. The highest value of CM conducted noise is pointed at the point (150 kHz, 76.83 dBµV). The EMI CM conducted noise is almost invariable with an average value of 67 dBµV. This results are illustrated in Table 2. Table 2. The performance of the novel EMI filtering approach CM conducted EMI noise (dBµV) Frequency range Without filter (MHz) 0.15–0.5 0.5–5 5–30

From 136.8 to 101.3 From 101.3 to 77.8 From 77.8 to 54.28

With integrating the DAFF An average value of 70 An average value of 67 An average value of 67

The standard EN61800-3/ class A 79 73 73

It can be concluded that the DAFF has shown its efficiency in reducing CM conducted EMI noise and complying with the standard EN61800-3/class A.

4 Conclusion In this paper design of novel EMI filtering technique to be involved with a power inverter has been presented. The DAFF is designed using XSG to be embedded into FPGA chip. This approach has the advantages of smaller size, low cost and less power consumption as compared to analog techniques. It has shown its effectiveness in terms of a significant reduction of the CM conducted noise. For future work, experimental results will be conducted, to validate the proposed simulation approach.

References 1. Akagi, H., Shimizu, T.: Attenuation of conducted EMI emissions from an inverter-driven motor. IEEE Trans. Power Electron. 23(1), 282–290 (2008) 2. Lei, X., Feng, F., Jian, S.: Behavioral modeling methods for motor drive system EMI design optimization. In: IEEE Energy Conversion Congress and Exposition, USA, pp. 947– 954 (2010) 3. Aizawa, N., Kubota, K.: Analysis of common-mode voltage elimination of PWM inverter with auxiliary inverter. In: International Conference on Electrical Machines and Systems, Incheon, pp. 889–893 (2010)

212

Y. Bchir et al.

4. Xue, J., Wang, F., Zhang, X.: Design of output passive EMI filter in DC-fed motor drive. Twenty-Seventh Annual IEEE Applied Power Electronics Conference and Exposition (APEC), pp. 634–640 (2012) 5. Jettanasen, C., Thongsuk, S.: Attenuation of high-frequency electromagnetic noise in a single-phase AC motor drive. In: International Conference on Electrical Machines and Systems (ICEMS), Busan, Korea (2013) 6. Wu, W., Jiang, Y., Liu, Y.: A new passive filter design method for overvoltage suppression and bearing currents mitigation in a long cable based PWM inverter-fed motor drive system. In: IEEE International Power Electronics and Motion Control Conference (IPEMCECCE Asia), China (2016) 7. Kalaiselvi, J., Srinivas, S.: Passive common mode filter for reducing shaft voltage, ground current, bearing current in dual two level inverter fed open end winding induction motor. In: International Conference on Optimization of Electrical and Electronic Equipment (OPTIM), Romania (2014) 8. Zhenyang, Y., Shishan, W., Zheng, S.: The reviews of integrated EMI filters applied in power electronic system. In: 2015 Asia-Pacific Symposium on Electromagnetic Compatibility (APEMC), Taiwan (2015) 9. Yitao, L., See, K.Y., Simanjorang, R.: Evaluation on filter topologies in high power density converter design for power quality and EMI control. In: 2015 Asia-Pacific Symposium on Electromagnetic Compatibility (APEMC), Taiwan (2015) 10. Morris, C.T., Han, D., Sarlioglu, B.: Comparison and evaluation of common mode EMI filter topologies for GaN-based motor drive systems. In: 2016 IEEE Applied Power Electronics Conference and Exposition (APEC), USA (2016) 11. Ian, L., Xibo, Y., Neville, M.: A holistic approach to optimize the power density of a silicon carbide (SiC) MOSFET based three-phase inverter. In: 2015 IEEE 11th International Conference on Power Electronics and Drive Systems (PEDS), Sydney, Australia (2015) 12. Chen, W., Yang, X., Wang, Z.: An active EMI filtering technique for improving passive filter low-frequency performance. IEEE Trans. Electromagn. Compat. 48(1), 172–177 (2006) 13. Hamza, D., Jain, P.K.: Conducted EMI noise mitigation in DC-DC converters using active filtering method. In: IEEE Power Electronics Specialists Conference, pp. 188–194 (2008) 14. Hamza, D., Pahlevaninezhad, M., Jain, P.K.: Implementation of a novel digital active EMI technique in a DSP-based DC–DC digital controller used in electric vehicle (EV). IEEE Trans. Power Electron. 28(7), 3126–3137 (2013) 15. Hamza, D., Qiu, M., Jain, P.K.: Application and stability analysis of a novel digital active EMI filter used in a grid-tied PV micro inverter module. IEEE Trans. Power Electron. 28(6), 2867–2874 (2013) 16. Hamza, D., Qiu, M.: Digital active EMI control technique for switch mode power converters. IEEE Trans. Electromagn. Compat. 55(1), 81–88 (2013) 17. Yosr, B., Soufien, G., Hamza, D., Abdellatif, M.: Hardware software co-simulation of a digital EMI Filter using Xilinx system generator. Arch. Electr. Eng. 67(3), 515–527 (2018) 18. Ji, J., Chen, W., Yang, X.: Design and precise modeling of a novel digital active EMI filter. In: IEEE Applied Power Electronics Conference and Exposition (APEC), USA, pp. 3115– 3120 (2016)

Nonlinear Plasmonic Photoresponse of Field Effect Transistors at Terahertz High Irradiation Intensities A. Mahi(B) University Center Nour Bachir, B.P. 900, 32000 El Bayadh, Algeria [email protected]

Abstract. By numerical simulations we investigate the dependence of the current response on the THz radiation intensity for different geometric parameter of InGaAs high-electron mobility transistors (HEMT), we show that the increased current response followed by saturated and then decreased. The results of the rectification current response to a Terahertz single intensity are in good agreement with other measured data. The influence of geometric parameters of the HEMT on the saturation curve of current response is studied in this work. Keywords: Terahertz single intensity · Detection · Modeling Simulation · InGaAs high-electron mobility transistor

1

·

Introduction

Recently, the high electron mobility transistor (HEMT) has exhibited his potentiality as an efficient Terahertz (THz) detector [1,2,5]. It started at 1993, when Dyakonov and Shur (DS) have demonstrated that the incoming THz radiation is able to excite the 2-dimensional electron plasma in the transistor channel [3,4]. In this context, many experimental studies validated DS model, using InAlAs/InGaAs HEMT as THz detector [6–8]. Moreover, interesting properties of the InAlAs/InGaAs HEMT detector were experimentally discovered [9,10]. This experiment brings up the relationship of the photoresponse to the incoming THz radiation intensity, they also showed a saturation of photoresponse when the intensity of THz radiation is higher than some KW/cm2 [9,10]. However, there have been very few theoretical studies of the HEMT detectors at high radiation intensities. In this work, we propose a study of the dependance of the photoresponse THz detector, based on a InAlAs/InGaAs HEMT, to incident radiation power density. The impact of different geometric parameters of the HEMT to the relation-ship between photoresponse and intensity of THz radiation will take place.

2

Theoretical Part

To study the HEMT photoresponse to incident THz radiation, we propose that the THz wave was detected by the gate of the HEMT and it’s modeled by c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 213–219, 2020. https://doi.org/10.1007/978-3-030-21009-0_20

214

A. Mahi

an oscillating voltage source at the same frequency than the incident radiation analogously to what has been already used in this book chapter [11]. In this case, the THz excitation collected through the gate contact is described by the harmonic component of the gate potential, δVg = ΔVg cos(2πf t), where ΔVg is the THz signal amplitude and f its frequency. So, the harmonic component due to THz detection (δVg ) will be added to the voltage gate polarization (VGS ). We utilize our HydroDynamic model (HD) coupled with 2D-Poisson equation (the model is detailed in ref. [13]). For aim to calculate time-dependent drain current which can be expressed : Id (t) = Id0 + ΔId (f ) + ΔId (f ) cos(2πf t + ϕI ), where Id0 is the drain dark current (without THz excitation), ΔId (f ) cos(2πf t + ϕI ) the harmonic drain courant and ΔId (f ) the average response, also its the additional DC current is due to the excitation of the electronic plasma in the transistor. The average response represent the conversion of the THz signal collected by the gate to the drain courant.

Average response (μA)

102 10

VDS= 40 mV VGS= -1 mV

1

100 10-1

ΔVg = 1 mV

f 2D

10-2

= 10 mV = 20 mV

f 3D

10-3 0

2

4

6 8 10 12 Frequency (THz)

14

Fig. 1. Average response of the drain current as functions of the radiation frequency for an excitation applied to the gate of the HEMT, for different THz signal amplitude (ΔVg ), under conditions of polarization drain-source and gate source are VDS = 40 mV and VGS = −1 mV respectively.

In this paper, we studied an InAlAs/InGaAs/InP transistor structure similar to that cited in [12], that is composed of : the gated part Lg = 100 nm with a concentration of electrons ND = 1×1018 cm−3 ; the two ungated access regions + = 5×1018 cm−3 ; the part Lc = 50 nm with a concentration of electrons ND channel thickness δ = 10 nm. The applied voltage to the gate and drain are VGS = −1 mV and VDS = 40 mV, respectively. Figure 1 shows the average current response at room temperature as functions of the THz excitation frequency, for different amplitude THz signal (ΔVg ). Each spectrum displays three resonances peaks, the first peak correspond to the first

THz Photoresponse Saturation Phenomena

215

2D-plasma resonance (f 2D  4 T Hz) and the peak of high value represents the 3D-plasma resonance (f 3D  12 T Hz), as it is demonstrated in ref. [13]. We show also that the average response is variable, depending on the THz signal amplitude. Why we found is useful to determine the dependence of the DC courant on the THz signal amplitude. To achieve this goal, we fixed the frequency of the THz signal radiation and we study the variation of the DC current to the amplitude of this THz signal detected. I (W/cm2)

10

100

1K

10K

VDS = 40 mV VGS = -1 mV

10-3 10 f 2D = 4.5 THz f

3D

10-4

Photoresponse (A)

Average response ( μ A)

100K 10-4

100

= 12 THz

Measured data

1 5

10-6 10 15 20 25 30 35 40 45 50 ΔV g (mV)

Fig. 2. Average response as functions of the amplitude of THz signal (ΔVg ), for THz signal frequency : (Continuous line) f 2D  4 THz and (dashed line) f 3D  12 THz, under condition of polarization drain-source and gate source are VDS = 40 mV and VGS = −1 mV respectively. Compared with (Full circles) photoresponse experimental data measured in ref [10]

Figure 2 shows the dependence of the photoresponse of the HEMT to the THz signal amplitude for different frequencies : Continuous line; f 2D = 4.5 THz and dashed line ; f 3D = 12 THz, correspond to the frequencies of the first 2D-plamsa and 3D-plamsa, respectively (See Fig. 1). Calculate with our HD 2DPoisson model (for more detail about this model see ref [11]), and we compare this analytical result with experimental data published in ref. [10]. For each frequency spectrum in Fig. 2, the average response of current (for f 2D = 4.5 THz or f 3D = 12 THz) increase quasi-linearly as function of the THz signal amplitude followed by square shape, for intense THz signal amplitude value, we can observe a saturation regime followed by a small decreasing of the average response of current. We can also note that the saturation behavior of the DC current is due to electron heating activation, induce when the incoming amplitude signal is very intense [12]. If we compare the theoretical result obtained

216

A. Mahi

in the framework of this work (lined and dashed curve) with the experimental measured data (Full circles) obtained in the ref [10], we can revealed that there is a good concordance in the shape between experimental and our results. using our HD model coupled with the 2D Poisson equation validated by the experimental results, we can go on to study the influence of the geometric parameters on the saturation of photoresponse.

3

Results and Discussion

Let we study the influence of the gate-source voltage VGS on the saturation phenomena. To this purpose, we report in Fig. 3 the average response of the drain current as functions of the amplitude of THz signal detected (ΔVg ), for different gate-source voltage VGS , with signal frequency f 2D  4 THz and drainsource polarization VDS = 40 mV.

Average response (μA)

20 ( f 2D = 4.5 THz)

10 5 VGS = - 1 mV = - 20 mV = - 40 mV = - 80 mV = - 120 mV

1 0

10

20 30 ΔVg (mV)

40

50

Fig. 3. Average response as functions of the amplitude of THz signal (ΔVg ), for different gate-source voltage VGS with signal frequency f 2D  4 THz and drain-source polarization VDS = 40 mV.

According to the Dyakonov-Shur work [4], the quality factors and resonant frequencies are proportional to the square root of the gate voltage VGS . In other words, more the gate voltage is important more the current response is low. For this case, we observed in Fig. 3 that saturation regime decreases when the gate voltage VGS increases. If the gate voltage VGS increases the dissipated energy in the channel becomes important. Therefore, the saturation regime will be unstable on account of the electron heating phenomenon.

THz Photoresponse Saturation Phenomena

217

Average response (μA)

16 14

( f 2D )

12 10

Lg = 70 nm 100 nm 150 nm 200 nm 250 nm

8 6 4 2 0 0

10

20

30 40 ΔVg (mV)

50

60

Fig. 4. Average response as functions of the amplitude of THz signal (ΔVg ), for different gate length Lg with signal frequency f 2D  4 THz and drain-source polarization VDS = 40 mV.

The drain current response as functions of the amplitude of THz signal (ΔVg ), for different gate length Lg , is reported in Fig. 4. As the length of the gate increases, the 2D plasma resonance frequency decreases. However, as the channel is longer, imposed wavelengths are larger and as a result of the lower 2D plasma resonance frequency and the transport in a longer channel become less ballistic. for this reason, when the gate length decreases more, the current response saturation is more visible for higher amplitudes THz signal (see Fig. 4). In the final phase, we show the average response as functions of the amplitude of THz signal (ΔVg ), for different concentration of electrons in the access regions, in Fig. 5. We show that when the different of concentration between the drain and the gate regions is important the saturation of the current appears for low amplitude of THz signal. This could be interpreted as follows : On one hand, increasing carrier concentration in access regions implies an increase in 2D plasma resonant frequency and on the other hand, the Coulomb interaction at long range leading to mutual oscillations between electrons of different region which increases the energy of the electrons. So, when the difference in concentration between regions is important, the saturation phenomenon is more considerable.

218

A. Mahi

Average response (μA)

16 14 12 ( f 2D )

10

N+ = 2 N

8

N+ = 5 N

6

N+ = 8 N

4 2 0 0

10

20

30 40 ΔVg (mV)

50

60

Fig. 5. Average response as functions of the amplitude of THz signal (ΔVg ), for different concentration of electrons in the access regions N + with signal frequency f 2D  4 THz and drain-source polarization VDS = 40 mV.

4

Conclusion

We have studied the current response of a HEMT for high THz signal amplitude, we show that an increasing region is followed by a saturation regime of the detection. This saturation is due to the phenomenon of hot electrons. The influence of geometric parameters of the HEMT to the saturation phenomena of current response is studied too.We have found that for short lengths of the channel and for low gate voltage of the HEMT, the saturation phenomena of current response take place for THz amplitudes more considerable. Finally, we show that for low different of concentration between the drain and the gate regions, the saturation of the current appear for high amplitude of THz signal.

References 1. Gargouri, N., Sakka, Z., Ben Issa, D., Kachouri, A., Samet, M.: A 4GHz.: temperature compensated CMOS ring oscillator for impulse radio UWB. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, pp. 71–75 (2016) 2. Lu, J.-Q., Shur, M.S., Hesler, J.L., Sun, L., Weikle, R.: Terahertz detector utilizing two-dimensional electronic fluid. Electron Device Lett. 19(10), 373–375 (1998) 3. Dyakonov, M., Shur, M.: Detection, mixing, and frequency multiplication of terahertz radiation by two-dimensional electronic fluid. IEEE Trans. Electron Devices 43(3), 380–387 (1996)

THz Photoresponse Saturation Phenomena

219

4. Dyakonov, M., Shur, M.: Shallow water analogy for a ballistic field effect transistor: new mechanism of plasma wave generation by dc current. Phys. Rev. Lett. 71(15), 2465–2468 (1993) 5. Knap, W., Deng, Y., Rumyantsev, S., Shur, M.S.: Resonant detection of subterahertz and terahertz radiation by plasma waves in submicron field-effect transistors. Appl. Phys. Lett. 81(24), 4637–4639 (2002) 6. Teppe, F., Orlov, M., El Fatimy, A., Tiberj, A., Knap, W., Torres, J., Gavrilenko, V., Shchepetov, A., Roelens, Y., Bollaert, S.: Room temperature tunable detection of subterahertz radiation by plasma waves in nanometer InGaAs transistors. Appl. Phys. Lett. 89, 222109 (2006) 7. Nouvel, P., Marinchio, H., Torres, J., Palermo, C., Gasquet, D., Chusseau, L., Varani, L., Shiktorov, P., Starikov, E., Gruˇzinskis, V.: Terahertz spectroscopy of plasma waves in high electron mobility transistors. J. Appl. Phys. 106, 013717 (2009) 8. Hassen, M.B., Mbarek , S.B., Choubani, F.: 2-D FDTD analysis of CPW antenna for electromagnetic near-field applications. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, pp. 62–65 (2016) 9. But, D.B., Drexler, C., Dyakonova, N., Drachenko, O., Ganichev, S.D., Knap, W.: Nonlinear photoresponse of FET THz broadband detectors at high power irradiation. In: 38th International Conference on Infrared, Millimeter, and Terahertz Waves (IRMMW-THz), pp. 1–2 (2013) 10. But, D.B., Drexler, C., Sakhno, M.V., Dyakonova, N., Drachenko, O., Sizov, F.F., Gutin, A., Ganichev, S.D., Knap, W.: Nonlinear photoresponse of field effect transistors terahertz detectors at high irradiation intensities. J. Appl. Phys. 115(6), 164514 (2014) 11. Mahi, A., Palermo, C., Marinchio, H., Sabatini, G., Belgachi, A., Varani, L.: Plasma wave excitation by terahertz electric signal in high electron mobility transistors. In: Adorno, D.P., Pokutnyi, S. (eds.) Advances in Semiconductor Research: Physics of Nanosystems, Spintronics and Technological Applications, chap. 11. Nova Science Publishers (2014) 12. Mahi, A., Palermo, C., Marinchio, H., Belgachi, A., Varani, L.: Saturation of THz detection in InGaAs-based HEMTs: a numerical analysis. Physica B 500, 1–3 (2016) 13. Mahi, A., Marinchio, H., Palermo, C., Belghachi, A., Varani, L.: Enhanced THz detection through phase-controlled current response in field-effect transistors. IEEE Electron. Device Lett. 34(6), 795–797 (2013)

Simulation and Measurement of a New Circularly Polarized Patch Antenna for WiMAX Applications El Amjed Hajlaoui1,2(&) 1

College of Engineering, Qassim University, Buraydah 53452, Saudi Arabia 2 Electronic LAB, El Manar University, 2092 Tunis, Tunisia [email protected]

Abstract. This paper reveals the impact of the insertion of electromagnetic band gap (EBG) structures on the performance of circularly polarized (CP) patch-slot antenna with offset slot. An optimization is necessary to precise physical parameters in the aim to fix the resonance frequency at 3.2 GHz useful for weather radar, surface ship radar, and some communications satellites bands. The proposed antenna possesses lightweight, simplicity, low cost. The circular polarization is ensured by right-hand and left-hand circular polarization process (RHCP and LHCP) due to the two exciting sources. Our investigation will confirm the simulation and experimental results of the EBG antenna involving new EBG structures. Keywords: Circular microstrip patch-slot antenna
RHCP and LHCP polarization
Circular polarization
Microstrip antenna with an offset circular slot
Electromagnetic band gap (EBG) resonator

1 Introduction Antennas analysis and conception have been the core of researches during the last decades [1, 2]. Great investigations were piercing to circularly polarized (CP) antennas, recognized as master key for various wireless and mobile communications systems, sensors in the aim to enhance various output parameters (gain, bandwidth…) [3]. In addition, electromagnetic band gap (EBG) structures are attractive features to permit new functionality of these kind of antennas [4], and will provide solutions to numerous engineering problems [5]. A new concept of circularly polarized EBG antenna will be examined. In the final attained antenna design, two orthogonal polarizations [6] with a low cross-polarization level < Aij ¼ Ai þ k¼1 qk A Pn rkj  ij B ¼ Bi þ k¼1 qk B > : ij qðtÞ ¼ qj ðtÞ; j ¼ 1


n

ð7Þ

The indices rkj equivalent to min or max, indicate which partition of the kth parameter involved in the jth submodel (see [14] for more details). 2.2

Saturated Time-Varying Nonlinear System

The T-S system (6) under actuator saturation and subject to unknown inputs is defined by: 8 < :

x_ ðtÞ ¼

r P 2n P i¼1 j¼1

yðtÞ ¼ CxðtÞ

  li ðnðtÞÞgj ðqðtÞÞ Aij xðtÞ þ Bij satðuðtÞÞ þ Bw wðtÞ

ð8Þ

Input-Constrained Controller Design for Nonlinear Systems

243

where, the saturation function sat : Rnu ! Rnu is defined as: 2

8 > > > > > >
> > satnuðunu Þ > > > : sat u  ¼ signu min u ; j j j j

3 7 7 7 5

ð9Þ umax j



with umax [ 0 denote the saturation levels. The unknown disturbance wðtÞ is defined by j the following set:  Qd ¼

wðtÞ : R

þ

nd þ p

!R

Z ;

1

T

wðsÞ wðsÞds  d

ð10Þ

0

Remark 1. Recently, the nonlinear behavior of the input saturation has been investigated as a convex combination of 2m linear models in [10, 11], and as an alternative polytopic structure using Takagi-Sugeno modeling in [14]. In the proposed work we consider only adjustment on the controller input as a reminiscent of the classical antiwindup configuration.

3 Saturated SOF Control Law 3.1

Control Law

Our objective is to synthesize a robust controller law to guarantee the stability of the proposed class of systems and ensure the desired closed-loop performance, especially the disturbance attenuation by taking into account the input control saturation levels. To achieve this goal, a static output feedback (SOF) controller law is designed as follows: ( uðtÞ ¼

r X

  lj ðnðtÞÞ Kj ðqðtÞÞyðtÞ þ Fj ðqðtÞÞwðtÞ

ð11Þ

j¼1

Pn Pn where the gains K j ðqðtÞÞ¼ 2k¼1 gk ðqðtÞÞK j , and Fj ðqðtÞÞ¼ 2k¼1 gk ðqðtÞÞFj are to be determined. Generally, we can simply set Fj ¼ 0; j¼ 1;


; r [8]. Using the following definition: wð:Þ : Rnu ! Rnu as wðuðtÞÞ¼uðtÞsatðuðtÞÞ and based on the controller law (11), the closed-loop dynamic system can be described as an augmented descriptor form: Ex_ ðtÞ ¼

r X 2n X i¼1 j¼1

   w wðtÞ gi ðnðtÞÞgj ðqðtÞÞ AijxðtÞ þ Bij wðuðtÞÞ þ B

ð12Þ

244

S. Aouaouda et al.

where  x¼ xT

T T; E

u

¼

I 0

 0 Aij þ Bij Ki ðqðtÞÞC ; Aij ¼ Ki ðqðtÞÞC 0

  w ¼ Bw þ Bij Fi ðqðtÞÞ B Fi ðqðtÞÞ



 0 Bij ; Bij ¼ ; I 0 ð13Þ

The following Lemma will be used in the sequel of the paper Lemma 2. [18]: Consider symmetric matrices Fij ; i; j 2 Nr of appropriate dimenP P sions. The following inequality ri¼1 rj¼1 Fij \0 is verified if  Fii \0 2 F i; j 2 Nr ; i 6¼ j r1 ii þ Fij þ Fji 3.2

Control Problem Definition

We propose a new LMI conditions for the design of a SOF controller (11) to preserve stability performance with disturbance attenuation. Specifically, the following properties will have to be satisfied: State constraints: Given vectors Nk 2 Rn ; k 2 Nq 8xð0Þ 2 EðP; cÞ with wðtÞ 2 Qd the closed loop system remains inside the admissible polyhedral set defined by: ð14Þ assuring the Local stability convergence: There exists an ellipsoidal set exponential convergence of undisturbed system responses ði:e:w ðtÞ ¼ 0Þ with initial states remains in this ellipsoid. The convergence decay rate is assumed to be less then a predefined scalar a. L2 -gain performance: The input signal wðtÞ belong to Qd set. In fact, for x(0Þ ¼ 0 and pffiffiffi all allowable values w(t) 2 Qd , this imply that kxðtÞk2 \ k kwðtÞk2 .

4 Main Results The solution of the design stabilizing SOF control problem defined in the previous section is obtained for the T-S system (8) by solving a set of LMI conditions. The following lemma is needed for the theoretical development: Lemma 3. Given two matrices MðqÞ 2 Rnu nx ; and SðqÞ 2 Rnx nx ; i 2 Nr and let be the polyhedral set related with this matrices and expressed by:

where

Input-Constrained Controller Design for Nonlinear Systems

245

 n  o   LðMðqÞ  SðqÞÞ ¼ x 2 Rn :  MðqÞðlÞ  SðqÞðlÞ x  umaxðlÞ ; l 2 Nm Then if

, the inequality wðuÞT QðqÞ1 ðwðuÞ  SðqÞxÞ  0

Pn is satisfied, for QðqÞ ¼ 2j¼1 gj ðqÞQj and any positive diagonal matrix SðqÞ ¼ P2n n j¼1 gj ðqÞS j where the family of functions gj ; j 2 N2 verify the convex sum propriety. The controller gains are derived from the solution of the following theorem. Theorem 1. Given positive scalar a, assume there exist positive definite matrices X1 ðqÞ, positive definite matrices QðqÞ, diagonal positive matrices XðqÞ, matrices X2 ðqÞ, X3 ðqÞ, weighing matrices MðqÞ, N ðqÞ, Ki ðqÞ; Fi ðqÞ, and positive scalars k such that

X1 ðqÞ MðqÞðlÞ  N ðqÞðlÞ

X1 ðqÞ Nk X1 ðqÞ



 i 2 Nr ;

u2maxðlÞ  0;    0; 1

/ii \0;

l 2 Nnu

ð15Þ ð16Þ

k 2 Nq

i 2 Nr

2 / þ /ij þ /ji \0; r  1 ii

ð17Þ

i; j 2 Nr ; i 6¼ j

ð18Þ

where /ij is defined by: 2

/11 ij 6 T ðqÞ K X 6 2 i ðqÞC 6 X1 ðqÞ 6 6 Bij Ki ðqÞC 6 6 /ij ¼ 6 QðqÞT BTij þ XðqÞ 6 6 BT þ F T ðqÞBT w i ij 6 6 X1 ðqÞ 6 4 Ki ðqÞC X3 ðqÞT

 HðX2 ðqÞÞ 0 0 0 FiT ðqÞ 0 0 0

  I 0 0 0 0 0 0

   I 0 0 0 0 0

    2QðqÞ 0 0 0 0

     kI 0 0 0

      I 0 0

       I 0

3  7  7 7  7 7  7 7  7 7  7 7  7 7  5 I ð19Þ

  T with /11 ij ¼ X1 ðqÞAij þ 2aX1 ðqÞ . Then the SOFC(11) with the derived feedback controller gains offer a feasible solution for the discussed problem. Proof. From the inequality (15) (respectively (16)) we can prove the inclusion (respectively ). Now, consider the Lyapunov function V defined as:

246

S. Aouaouda et al.

V ðxðtÞ; ðqÞÞ ¼ xðtÞT EP ðqÞxðtÞ

ð20Þ

with EPðqÞ ¼ PðqÞT E  0

ð21Þ

Pn and PðqÞ¼ 2j¼1 gj ðqÞP j . Let V_ ðxðtÞÞ be the derivative of V ð xðtÞÞ. For closed loop system converge asymptotically to zero if: _ xÞ þ 2aVðxÞ þ xT Qw xkwT w2wðuÞT QðqÞ1 ðwðuÞSðqÞxÞ\0 Vð

, the

ð22Þ

With (12) and (20), the inequality (22) is fulfilled if: 2 6 6 4

 H ATij PðqÞ þ 2aEPðqÞ þ Qw



BTij PðqÞ þ QðqÞ1 S  ðqÞ

2QðqÞ1

T Bw PðqÞ



0

3

7  7 5\0 kI

ð23Þ

where S  ðqÞ ¼ ½SðqÞ 0 ; Qw ¼ diagðI; 0Þ. Let XðqÞ¼ PðqÞ1 . Congruence transformation of (23) with diagðXi ; Qi ; I Þ yields: 2 6 6 4

 H ATij XðqÞ þ 2aXðqÞT E þ XðqÞT Qw XðqÞ



QðqÞT BTij þ S  ðqÞXðqÞ

2QðqÞ

T Bw

0



3

7  7 5\0 kI

ð24Þ

In order to derive easily LMI conditions, the structure of the matrix XðqÞ is considered as follows:

X1 ðqÞ 0 XðqÞ ¼ X3 ðqÞ X2 ðqÞ

 ð25Þ

According to (21), it follows that X1 ðqÞ ¼ X1 ðqÞT , and X3 ðqÞ, X2 ðqÞ are free slack matrices. Let XðqÞ ¼ SðqÞX1 ðqÞ. Using system matrices defined by (13) we can prove that the condition (24) is developed to: 2

 wij 6 X ðqÞT K ðqÞC HðX 2 ðqÞÞ 6 2 i 6 X1 ðqÞ 0 6 6 QðqÞT BT þ XðqÞ 0 4 ij BTw þ FTi ðqÞBTij FTi ðqÞ

  I 0 0

   2QðqÞ 0

3   7 7  7 7\0  7 5 kI

ð26Þ

   with wij ¼ H X 1 ðqÞAij þ X1 ðqÞBij K i ðqÞC þ H X 3 ðqÞT K i ðqÞC þ 2aX 1 ðqÞT . Now, basing on Lemma 1 and Schur complement the inequality (26) is expressed as:

Input-Constrained Controller Design for Nonlinear Systems

2

/11 ij 6 X ðqÞT K ðqÞC 2 i 6 6 X1 ðqÞ 6 6 Bij K i ðqÞC 6 6 6 QðqÞT BTij þ XðqÞ 6 6 BT þ FT ðqÞBT ij w i 6 6 X1 ðqÞ 6 4 K i ðqÞC X 3 ðqÞT

 HðX 2 ðqÞÞ 0 0 0 FTi ðqÞ 0 0 0

  I 0 0 0 0 0 0

   e1 0 0 0 0 0

    2QðqÞ 0 0 0 0

     kI 0 0 0

      e1 1 0 0

       e2 0

247

3   7 7  7 7  7 7 7  7\0 7  7 7  7 7  5 e1 2

ð27Þ We assume that positive scalars e1 ¼e2 ¼I, then the derived inequality is equivalent to (19).This ends the proof. Remark 2. The solution given by Theorem 1 for the control problem defined in Sect. 2 is en reality not practical for control design. In fact this is related to the dependence of inequality (19) on parameters weighting functions. A more tractable solution is then derived by Theorem 2 Theorem 2. Given positive scalar a, assume there exist positive definite matrices X1i , positive definite matrices Qi , diagonal positive matrices Xi , matrices X2i , X3i , weighing matrices Mi , N i , Kij ; Fij and positive scalars k such that

X1i MiðlÞ  N iðlÞ



 i 2 Nr ;

u2maxðlÞ  0;

X1i Nk X1i

   0; 1

 ii \0;

l 2 Nnu

ð28Þ ð29Þ

k 2 Nq

i 2 Nr

2  ii þ  ij þ  ji \0; r1

ð30Þ

i; j 2 Nr ; i 6¼ j

ð31Þ

where  ij is defined by:   H X1i Aij þ 2aX1iT 6 X2iT Kij C 6 6 X1i 6 6 B Kij C ij 6 T T Q B Yij ¼ 6 i ij þ Xi 6 6 T T T B þ F 6 w ij Bij 6 X1i 6 4 Ki ðqÞC X3iT 2

  HðX2i Þ  0 I 0 0 0 0 FijT 0 0 0 0 0 0 0

   I 0 0 0 0 0

    2Qi 0 0 0 0

     kI 0 0 0

      I 0 0

       I 0

3   7 7  7 7  7 7  7 7  7 7  7 7  5 I ð32Þ

248

S. Aouaouda et al.

Then the SOFC(11) with the derived feedback controller gains offer a feasible solution for the discussed problem. Proof. The derived conditions can be easily demonstrated following the same proof of Theorem 1. Thus the proof is omitted here for brevity.

5 Illustrative Results A nonlinear model (33) representing a vehicle system behavior is used to illustrate the efficiency of the proposed controller. A simplified model lateral dynamics of the so-called bicycle model is considered for simulation [19]. A T-S representation is considered with  T the state vector xðtÞ ¼ bT ðtÞr T ðtÞ corresponding to the dynamics of the yaw rate rðtÞ and the vehicle sideslip angle bðtÞ. The vehicle T-S model structure is derived as: 8 < :

x_ ðtÞ ¼

2    P li af  Ai ðvÞxðtÞ þ Bfi ðvÞdf ðtÞ þ Bi Mz ðtÞ

i¼1

ð33Þ

yðtÞ ¼ CxðtÞ

where " Ai ðvÞ ¼ " Bfi ðvÞ ¼

Sfi þ Sri mv Sfi af Sri ar 2 J

2

2Sfi mv 2af Sfi J

Sfi af Sri ar  mv2 Sfi a2f þ Sri a2r 2 Jv

2

# ; Bi ¼ B ¼

 0 1 J

;C ¼ ½0

1

#

1 :

uðtÞ ¼ df ðtÞ is the control input which represent the front steering angle while the yaw moment Mz ðtÞ is considered as disturbance input. yðtÞ is the measured output. The membership functions li ; i 2 f1; 2g are defined as follows:    af      # 1 i li af  ¼ P2   ; #i af  ¼       jaf jci  2bi i¼1 #i af   1þ  ai

ð34Þ

with the following parameter values obtained for a dry road, [19]: a1 ¼ 0:5077; a2 ¼ 0:4748; b1 ¼ 3:1893; b2 ¼ 5:3907, c1 ¼ 0:4356; c2 ¼ 0:5622. the vehicle parameters are as follows: J = 3000 kg.m2, la masse m = 1500 kg, ar ¼ 1:3 m; af ¼ 1:2 m; Sf10 ¼ 60712 N=rad; Sf20 ¼ 4812N=rad; Sr10 ¼ 60088 N=rad, Sr20 ¼ 3455 N=rad. In practice the steering  Hence   the  system  has a physical  limitation.   saturated steering angle is defined by sat df ðtÞ ¼ sign df ðtÞ min df ðtÞ ; dfmax , dfmax is the limitation of the control input. Furthermore, the dynamics of the system (33) depend nonlinearly on the vehicle measured and bounded speed parameter vðtÞ: vmin  vðtÞ  vmax . Where vmin ¼ 5½m=s ,

Input-Constrained Controller Design for Nonlinear Systems

249

and vmax ¼ 30½m=s [8]. Using the first order Taylor approximation 1=v and 1=v2 are written as [8]: 1 1 1 ¼ þ qðtÞ; v v0 v1

1 1 1 ¼ þ2 qðtÞ v2 v20 v0 v1

qðtÞ 2 ½ 1 1

ð35Þ

Where v0 and v1 in (35) are defined by [8]: v0 ¼

2vmin vmax 2vmin vmax ; v1 ¼ min min max v þv v  vmax

ð36Þ

Using Eqs. (35) and based on the polytopic representation described in Sect. 2, the nonlinear vehicle model (33) can be equivalently written by: 8 < :

x_ ðtÞ ¼

2 P 2     P li af  gi ðqÞ Aij xðtÞ þ Bfij df ðtÞ þ Bi Mz ðtÞ

ð37Þ

j¼1 i¼1

yðtÞ ¼ CxðtÞ

where 2 Ai1 ¼ 4 2 Ai2 ¼ 4

2

Sfi þ Sri m



2 2

Sfi þ Sri m

2

1 v0

Bfi1 ¼

fi

1 v1

qmax



Sfi af Sri ar J



1 v0

þ

1 v1

qmin



Sfi af Sri ar J

" 2S  m

þ

1 v0

þ

1 v1 2af Sfi J

qmax





Sfi af Sri ar 1 þ 2 v01v1 qmax  m v20 2 2  Sfi a þ Sri a 2 f J r v10 þ v11 qmax

2





Sfi af Sri ar 1 þ 2 v01v1 qmin  m v20  Sfi a2 þ Sri a2 2 f J r v10 þ v11 qmin

2

#

" 2S  ; Bfi2 ¼

fi

m

1

1 v0

þ

1 v1 2af Sfi J

qmin

1

3 5 3 5

#

and g1 ðqÞ ¼

1  qðtÞ ; 2

g2 ðqÞ ¼ 1  g1 ðqÞ

ð38Þ

Now the obtained vehicle T-S model (37) is used to test the effectiveness of the designed approach. The SOF controller gains Kij , and Fij are calculated from the solution of the optimization problem defined in Theorem 2. For brevity, the solution data details are omitted in this paper. Figure 1 represents the disturbance signal. While Figs. 2 and 3 evaluate the vehicle state responses xn ðtÞ of the nominal system (i.e. without varying parameter) with a comparison to the T-S saturated system state (37). Figure 4 shows the different control inputs applied to the proposed vehicle T-S system and the vehicle membership functions are drawn in Fig. 5. As depicted by simulation results, the designed approach satisfies the objectives of the defined problem.

250

S. Aouaouda et al.

0.05

z

yaw moment (M (t))

0.1

0

-0.05

-0.1

0

1

2

3

4

5

Time(s)

Fig. 1. Vehicle yaw moment 2 x 1n

sideslip angle [deg]

0

x 1v

-2

-4

-6

-8

0

0.5

1

1.5

2

2.5 Time(s)

3

3.5

4

4.5

5

Fig. 2. Sideslip angle time evolution-with and without qðtÞ 40

yaw rate

30 20 x 2n

10

x 2v 0

0

1

2

3

4

Time(s)

Fig. 3. Yaw rate time evolution - with and without qðtÞ

5

Input-Constrained Controller Design for Nonlinear Systems

251

40 u(t) sat(u(t)) umax

control input [deg]

30 20 10 0 -10

0

1

3

2

4

5

4

5

Time(s)

Fig. 4. System steering angle evolution

membership functions

1 0.8 0.6 0.4 0.2 0

0

1

2

3 Time(s)

Fig. 5. Vehicle membership functions

6 Conclusion In the present paper, a new control design solution is presented for constrained nonlinear system represented by T-S model with time-varying parameter. To deal with the input saturation limits and time varying LPV representation is investigated. For the control implementation, the synthesized approach is based on a SOF control law. The designed controller gains are then obtained by solving an optimization problem under LMI constraints.. The effectiveness of the proposed approach has been clearly illustrated on a vehicle lateral dynamics system.

252

S. Aouaouda et al.

References 1. Tanaka, K., Wang, H.: Fuzzy Control Systems Design and Analysis: A Linear Matrix Inequality Approach. Wiley-Interscience, New York (2004) 2. Fang, G.: A survey on analysis and design of model-based fuzzy control systems. IEEE Trans. Fuzzy Syst. 14(5), 676–697 (2006) 3. Aouaouda, S., Chadli, M., Shi, P., Karimi, H.R.: Discrete-time H_/H∞ sensor fault detection observer design for nonlinear systems with parameter uncertainty. Int. J. Robust Nonlinear Control. 25, 339–361 (2015) 4. Sala, A., Arino, C.: Relaxed stability and performance conditions for Takagi-Sugeno fuzzy systems with knowledge on membership function overlap. IEEE Trans. Syst. Men Cybern. 37(3), 727–732 (2007) 5. Hoffmann, C., Werner, H.: A survey of linear parameter-varying control applications validated by experiments or high-fidelity simulations. IEEE Trans. Cont Syst Tech. 23(2), 416–433 (2015) 6. Fergani, S., Menhour, L., Sename, O., Dugard, L., D’Andrea-Novel, B.: Integrated vehicle control through the coordination of longitudinal/lateral and vertical dynamics controllers: Flatness and LPV/ H∞ based design. Int. J. Robust Nonlinear Control 27(18), 4992–5007 (2017) 7. Ichalal, D., Mammar, S.: On unknown input observers for LPV systems. IEEE Trans. Indus. Electron. 62(9), 5870–5880 (2015) 8. Nguyen, A.-T., Zang, H., Sentouh, C., Popieul, J.-C.: Input-constrained LPV Output feedback Control for path following of autonomous Ground vehicles. In: Annual American Conference (ACC) (2018) 9. Tabatabeipour, M., Stroustrup, J., Bak, T.: Fault tolerant control of discrete-time LPV systems using virtual actuators and sensors. Inter. J. Robust Nonlinear Control. 25(5), 707– 734 (2015) 10. Tarbouriech, S., Garcia, G., Gomes da Silva, J., Queinnec, I.: Stability and Stabilization of Linear Systems with Saturating Actuators. Springer-Verlag, London (2011) 11. Saifia, D., Chadli, M., Labiod, S., Guerra, T.M.: Robust H∞ static output feedback stabilization of T-S fuzzy systems subject to actuator saturation. Inter. J. Cont, Aut and Syst. 10(3), 613–622 (2012) 12. Nguyen, A.-T., Dambrine, M., Lauber, J.: Simultaneous design parallel distributed output feedback and anti-windup compensators for constrained Takagi-Sugeno fuzzy systems. Asian. J. Control 18(5), 1641–1654 (2016) 13. Dang, Q.V., Vermeiren, L., Dequidt, A., Dambrine, M.: Robust stabilizing controller design for Takagi-Sugeno fuzzy descriptor systems under state constraints and actuator saturation. Fuzzy Sets Syst. 329, 77–90 (2017) 14. Bezzaoucha, S., Marks, B., Maquin, D., Ragot, J.: Stabilization and output feedback control for Takagi-Sugeno with saturated actuators. Inter. J. Adapt. Control. Signal Process. 30(6), 888–905 (2016) 15. Gao, Y.-Y., Lin, Z., Shamash, Y.: Set invariance analysis and gain-scheduling control for LPV systems subject to actuator saturation. Syst. Control Lett. 46(2), 137–151 (2002) 16. Nguyen, A.-T., Laurain, T., Palhares, R., Lauber, J., Sentouh, C., Popieul, J.-C.: LMI-based control synthesis of constrained Takagi-Sugeno fuzzy systems subject to L2 or L∞ disturbances. Neurocomputing, 207©, 793–804 (2016) 17. Aouaouda, S., Bouarar, T., Bouhali, O.: Fault tolerant tracking control using unmeasurable premise variables for vehicle dynamics subject to time varying faults. J. Frankl. Inst. 351, 4514–4537 (2014)

Input-Constrained Controller Design for Nonlinear Systems

253

18. Tuan, H., Apkarian, P., Narikiyo, T., Yamamoto, Y.: Parameterized linear matrix inequality techniques in fuzzy control design. IEEE Trans. Fuzzy Syst. 9, 324–332 (2001) 19. Aouaouda, S., Chadli, M., Boukhnifer, M., Karimi, H.K.: Robust fault tracking controller design for vehicle dynamics: a descriptor approach. Mechatronics 30, 316–326 (2015)

Fast Convergence of 2D DWT-WCIP Method Applied to Study a Complex FSS Structure S. Bennour1,2(&) and N. Sboui1,3 1

Faculté des Sciences de Tunis, Université de Tunis El Manar, Tunis, Tunisia [email protected], [email protected] 2 Ecole Supérieure des Sciences Appliquées et de Management SESAME, Tunis, Tunisia 3 Riyadh College of Technology, Riyadh, Kingdom of Saudi Arabia

Abstract. The wave concept iterative procedure (WCIP) is a powerful tool dedicated to the numerical analysis of RF circuits. The big problem of this method is that the required calculation time becomes higher when the circuit’s complexity increases, that’s to say for circuits with huge mesh surface. In this paper, we use a fast WCIP method based on 2D Discrete Wavelet Transform (2D DWT) to study a frequency selective surface, FSS, having a U-geometry. This 2D DWT technique is used to optimize the amount of information manipulated by the iterative process in order to get a fast convergence time. Keywords: WCIP


2D DWT
FSS
U-shaped geometry

1 Introduction The engineers in the field of microwave circuits and antennas are constantly invited to create new methodologies and improve those already used. In light of this request, the application of the FSS structure, has been the subject of extensive research to reduce unwanted signals, reuse the same antenna in two or more configurations, or increase antenna performance [1–5]. The choice of the geometry of the elements is one of the main points in the design of the Frequency Selective Surface. Various parameters such as dimensions, polarization, and multiband operation must be adjusted in a well-studied way. Periodic structures are characterized by their large sizes. The description of these structures in the spatial domain requires a very large number of pixels. The WCIP is a numerical method that has been used for the analysis and the study of the frequency response of FSS structures [6, 7] and periodic circuits [8]. This approach is characterized by its convergence towards good results and its stability. This method is also stable when it comes to complex structures, but it takes a long time to achieve the optimal result. For this, a large number of iterations is necessary to analyze the complex structures which require a tight mesh such us FSS structure. This reinforces the problem of numerical complexity since the number of elements representing the circuit is high. This explains why the “WCIP” method takes longer time to converge to the optimal values. In order to improve this method, we have used an image © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 254–260, 2020. https://doi.org/10.1007/978-3-030-21009-0_24

Fast Convergence of 2D DWT-WCIP Method

255

processing technique that converges quickly to the right result. This method uses the 2D discrete wavelet transform (2D DWT) [9, 10]. In fact, we use this transformation to focus on the important part of the structure studied in which electromagnetic fields are important. The 2D DWT separates the lowfrequency components from the high-frequency components of the image being studied and can locate the fast-changing position of an image. In our case, the high frequency content forms the most important part and represents the useful information that correspond to high values of the electromagnetic fields. The separation between the two types of components allows us to focus only on important information while the rest will be ignored. So that the speed of the “WCIP” method can be improved by removing operations on input values that are not necessary. Using this new wavelet-based method minimizes the number of operations by rejecting unnecessary operations and then we can speed up the iterative process. The improved WCIP method denoted 2D DWTWCIP showed its robustness and its ability to improve the performance of WCIP in terms of convergence and computing time for planar circuits, in previous work [9, 10]. In this work we will show the improvement in convergence time for circuits with complex geometry type FSS. In Sect. 2, we present a summary of the iterative method and a brief description of the principle of the wavelet transform. In Sect. 3, the numerical results of convergence and computation time are given. Finally, Sect. 4, deals with conclusions and perspectives.

2 Theory 2.1

Brief Review of the WCIP

The detailed development of the WCIP is given in the following researches [6, 11–17]. We can, briefly, define it as an integral method which uses the wave manipulation to solve problems of electromagnetic modeling. This method deals with the reflected and the incident waves rather than electromagnetic field. These waves are defined by (1) and (2): 1 Ai ¼ pffiffiffiffiffiffi ðEi þ Z0i Ji Þ 2 Z0i

ð1Þ

1 Bi ¼ pffiffiffiffiffiffi ðEi  Z0i Ji Þ 2 Z0i

ð2Þ

With the current density is defined by (3): ~ ~i ^ ~ Ji ¼ H ni

ð3Þ

~ ni is oriented to the area i 2 f1; 2g. ~i and Ei indicate respectively the tangential magnetic and electric field at the H surface.

256

S. Bennour and N. Sboui

Z0i is the impedance of the medium (i). Because it is based on a recurrence relation between incident and reflected waves in two different domains, the method is called iterative. This transition from one area to another is held by the Fast Fourier Mode Transformation FMT and the reverse transformation FMT−1. 2.2

2D DWT WCIP

Wavelets play a significant role in many image processing applications [18–22]. The Discrete Wavelet Transform (2D DWT), is a powerful tool used in image processing such as image analysis, de-noise, image segmentation and other applications. In, this paper we present the use of the 2D DWT in the performance optimization of the WCIP method, when this methods is employed to analyze a complex structure (FSS). Since, the surface of the studied structure is considered as an image. We use this technique to locate the area where the information is important. The 2D DWT decomposes an image into four parts of reduced dimension; each of these parts contains different information from the original image. In fact, detail coefficients represent useful information in the image because they are formed by the points of rapid variation of an image while the approximation coefficients are considered to be noise, so that they will be ignored. A good modification of the approximation coefficients is the simplest way to optimize the amount of data manipulated by the iterative process and to minimize computation time. The simplest method is based on setting all approximation coefficients to zero. This modification removes the low frequencies of the image that are supposed to be noise and represent unnecessary information. For the detail coefficients, we only retained those with a high value. Coefficients with low values, compared to a threshold set in advance, are rejected, since more the value of the wavelet coefficients is high, more than they are significant. The image is then reconstructed using only the remaining wavelet coefficients using the 2D IDWT. In fact, by using this method, the most important points, in which the values of the electromagnetic fields are important, are taken into account, while the negligible values are rejected. The wavelet optimization algorithm of the iterative process consists of the following steps: Wavelet transformation of initial values, cancellation of approximation coefficients, and elimination of image detail values below a threshold set in advance, and calculation only on the remaining values. The 2D DWT-WCIP algorithm is given in Fig. 1.

Fast Convergence of 2D DWT-WCIP Method

257

Begin Definition of a uniform grid N*M

Initial values

Extracting useful information

Iterative process for ‘ni’ iterations on the original matrices

Decomposition of the original matrices by the 2D DWT

Set the approximation coefficients to zero

Iterative process for « Ni-ni» iterations

Keep the detail coefficients of high values

No

Convergence Yes Reconstruction of the original matrices by the 2D IDWT

End

Fig. 1. 2D DWT- WCIP algorithm

Fig. 2. FSS structure with U-shaped geometry

3 Simulation Results The studied circuit is a FSS with a U-shaped geometry. The elementary cell of this structure is given in Fig. 2. The physical parameters of this structure are: er = 4.4, d = 0.02 and h = 0.9 mm.

258

S. Bennour and N. Sboui

The sizes of the elementary cell are Wx = Wy = 20 mm and the total sizes of the FSS are 20 cm  20 cm, which corresponds to 10  10 elementary blocks. dx = dy = 5 mm, Sy = 2 mm and ly = 10 mm. Lx1 = Lx2 = 2 mm [23]. We note from Fig. 3 that the two methods “WCIP” and “2D DWT-WCIP” converge to the same results after 500 iterations.

Fig. 3. Evolution of the parameters [S] according to the iterations

An iteration number of 500 is relatively high for a complex FSS structure that requires a tight mesh of 512 * 512 pixels. The advantage of the “2D DWT-WCIP” method is that only 20 iterations are realized by the conventional iterative process on matrices of size 512 * 512pixels. The remaining iterations up to 500 are completed on small size matrices obtained from the wavelet decomposition of the original matrices. This ensures a fairly fast convergence towards the good results. These results are obtained by Matlab simulation. In Table 1, we note a significant improvement in convergence time for the calculation of S11 and S21 values after 500 iterations by both methods. This improvement is due to the use of the 2D DWT-WCIP algorithm for Ni = 500 iterations and ni = 20 iterations. The mesh used in this structure is 512  512 cells. Table 1. Time comparison between the two methods ni 20 iterations « WCIP » for 500 iterations 5, 2 mn « 2D DWT-WCIP » for 500 iterations 0, 73mn Gain of time 86%

Fast Convergence of 2D DWT-WCIP Method

259

4 Conclusion In this work, we applied the 2D DWT-WCIP to the study of a complex FSS circuit requiring fine mesh and described by large matrices. From the results found and validated previously, we can deduce that the 2D DWTWCIP method is a fairly powerful method that has proved its validity and its ability to optimize the performance of the “WCIP” method when analyzing periodic planar structures. In future works we will show the performance of our method if the number of pixels exceeds 512*512 pixels.

References 1. Arezou, E., Tayeb, A.: Frequency selective surfaces for beam-switching applications. IEEE Trans. Antennas Propag. 61(1), 195–200 (2013) 2. Taylor, P.S., Bathelor, J.C., Parker, E.A.: A passively switched dual-band circular FSS slot array. In: Proceedings of 2011 IEEE-APS Topical Conference on Antennas and Propagation in Wireless Communications (APWC), pp. 648–651 (2011) 3. Valdez, A.: Aplicação de Superfícies Seletivas em Frequência para Melhoria de Resposta de Arranjos de Antenas Planares, in Portuguese, PhD. Thesis, UFRN, Natal, RN (2014) 4. Niroo-Jazi, M., Denidni, T.A.: Electronically sweeping-beam antenna using a new cylindrical frequency-selective surface. IEEE Trans. Antennas Propag. 61(2), 666–676 (2013) 5. Jebali, N., Beldi, S., Gharsallah, A.: RFID antennas implanted for pervasive healthcare applications. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 149–152, Hammamet (2016) 6. Latrach, L., Sboui, N., Gharsallah, A., Baudrand, H., Gharbi, A.: Analysis and design of planar multilayered FSS with arbitrary incidence. Appl. Comput. Electromagn. Soc. J. 23(2), 149–154 (2008) 7. Sboui, N., Salouha, A., Latrach, L., Gharsallah, A., Gharbi, A., Baudrand, H.: Efficient analysis of switchable FSS structure using the WCIP method, ACES JOURNAL 27(3) (2012) 8. Elbellili, T., Azizi, M.K., Latrach, L., Trabelsi, H., Gharsallah, A., Baudrand, H.: Analyzing of one dimensional quasi periodic circuit by using auxiliary sources in a WCIP method. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 34–39. Hammamet (2016) 9. Bennour, S., Sboui, N.: A new WCIP method based on 2D discrete wavelet transform. In: 2nd International Conference on Automation, Control, Engineering and Computer Science (ACECS-2015). Sousse, Tunisia (2015) 10. Bennour, S., Sboui, N.: An Optimized WCIP Method Based on A 2D discrete wavelet transform. Int. J. Appl. Eng. Res. 10(7) (2015) 11. Sboui, N., Gharsallah, A., Gharbi, A., Baudrand, H.: Analysis of double loop meander line by using iterative process. Microw Opt. Tech. Lett. 26, 396–399 (2000) 12. Sboui, N., Gharsallah, A., Baudrand, H., Gharbi, A.: Global modeling of microwave active circuits by an efficient iterative procedure. IEE Proc-Microw. Antenna Propag. 148(3) (2001)

260

S. Bennour and N. Sboui

13. Sboui, N., Gharsallah, A., Baudrand, H., Gharbi, A.: Global modeling of periodic coplanar waveguide structure for filter applications using an efficient iterative procedure. Microwave and Opt. Technol. Lett 43(2), 157–160 (2004) 14. Sboui, N., Gharsallah, A., Baudrand, H., Gharbi, A.: Design and modeling of RF MEMS switch by reducing the number of interfaces. Microw. and Opt. Technol. Lett 49(5), 1166– 1170 (2007) 15. Baudrand, H., Raveu, N., Sboui, N., Fontgalland, G.: Applications of multiscale waves concept iterative procedure. Inter. Microw. and Opto. Conference. Salvador, BA, Brazil (2007) 16. Sboui, N., Latrach, L., Gharsallah, A., Baudrand, H., Gharbi, A.: A 2D design and modeling of micro strip structures on inhomogeneous substrate. Int. J. RF Microw. Comput.–Aided Eng. 19(3), 346–353 (2009) 17. Ben Romdhan Hajri, J., Ghnimi, S., Sboui, N.: Design of SIW iris-coupled-cavity band-pass filter circuit using wave concept iterative process method. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 209–212, Hammamet (2016) 18. Ansari, R.A., Buddhiraju, K.M.: Noise filtering of remotely sensed images using hybrid wavelet and curvelet transform approach. In: IEEE International Geoscience and Remote Sensing Symposium (IGARSS), pp. 26–31 353 (2015) 19. Barcellona, A., Palmeri, D.: Wavelets image analysis for friction stir processed TiNi functional behavior characterization. Procedia Eng. 109, 8–16 (2015) 20. Geng, L., Bing, Z., Yu-na, Su: Face recognition algorithm using two dimensional locality preserving projection in discrete wavelet domain. Open Autom. Control. Syst. J. 7, 1721– 1728 (2015) 21. Thèse Olivier LE CADET.: Méthodes d’ondelettes pour la segmentation d’images: Applications à l’imagerie médicale et au tatouage d’images (2004) 22. Filali, A., Mokraoui, A., Frikha, T.: High dynamic range image tone mapping approach based on separable stationary wavelet transform decomposition using a coefficients weighted strategy. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 365–369, Hammamet (2016) 23. Gomes Neto, A., Titaouine, M., Baudrand, H., Djahli, F.: WCIP method applied to active frequency selective surfaces. J. Microw.S Optoelectron. 6, 1–16 (2007)

Absorption Enhancement in an Amorphous Silicon Using a Cluster of Plasmonic Hollow Ring Nano-Antennas Abdalem A. Rasheed1(&), Khalil H. Sayidmarie2, and Khalid Khalil Mohammed1 1

Department of Electrical Engineering, College of Engineering, Mosul University, Mosul, Iraq [email protected],[email protected] 2 Department of Communication Engineering, College of Electronic Engineering, Ninevah University, Mosul, Iraq [email protected]

Abstract. Enhancement of absorption and extending its bandwidth is of major interest for solar cells, photodetectors, and variety of applications. This paper presents a nano-structure formed of an array whose elements are in the form of a 3X3 cluster of metal rings having a similar outer diameter but with various inner diameters. Thus, each ring size produces certain resonance frequency and the result of the cluster arrangement is staggered responses that possess larger bandwidth. Simulations using the periodic unit cell approach and the CST microwave studio suite showed that the average absorption power in an amorphous silicon layer has been improved by 3.32 times compared to that without rings. The obtained response covers the frequency range from 230 THz to 360 THz. Keywords: Plasmonic


Nanoantennas
Absorption
Amorphous Si

1 Introduction Surface plasmon resonances in nano-antennas are of interest for solar cells, photo detectors, and other various applications because of the large enhancement of the electromagnetic field, which occurs in the vicinity of the metal surface. The resonance frequency of nano-antenna depends on the size, shape, dielectric environment, and dispersion properties of the used metal [1, 2]. There have been significant efforts aiming to increase the light absorption in photovoltaic (PV) layers, spectroscopic, biomedical, and other optical applications. Robust and versatile light absorption using plasmonic aluminum nanorods [3], and effective coupling of light into an S-shape plasmonic silver nanowire waveguide [4] have been reported recently. The use of plasmonic bowtie nanoantennas has shown absorption enhancement and efficient operation over a wide spectrum [5]. In this work, a new structure of Hollow Ring Nano Antennas (HRNAs) distributed at the surface of a thin layer of amorphous silicon (a-Si) is proposed to enhance the

© Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 261–268, 2020. https://doi.org/10.1007/978-3-030-21009-0_25

262

A. A. Rasheed et al.

absorption and increase the operation bandwidth. The absorption is enhanced at nearinfrared region (800–1500) nm producing a broadband absorption region and then increase the energy conversion in solar cells. Gold, silver, and aluminum plasmons of HRNAs were investigated using the CST microwave studio, which uses the finite integration technique.

2 Plasmonic and Applied Model

Real

m

Imag.

m

Figure 1 presents the real and imaginary parts of the permittivity for silver, gold, copper, chrome and aluminum using the Lorentz-Drude model [6–8]. Silver has the smallest loss in the visible and IR range (400–1500 nm). However, the known degradation problems generally make silver less suitable for plasmonics applications. In contrast, gold has very good chemical stability but presents higher losses below 550 nm [1, 9, 10].

Wavelength nm

Wavelength nm

a

b

Fig. 1. (a) real part e’(x) and (b) imaginary part e”(x) of the permittivity for Ag, Au, Al, Cu and Cr using Lorentz-Drude model [8].

The large losses presented by aluminum around 800 nm, which are associated with the interband transitions at this wavelength, make aluminum not an ideal material for antennas in the visible region [1]. The scattering and absorption cross sections rscat and rabs for a sphere of radius a are given by [11] em  e rabs ¼ K Im½a ¼ 4pK a3 Imð Þ em þ 2e

ð1Þ

K 4 2 8p 4 6 em  e 2 K a j a ¼ j 3 em þ 2e 6p

ð2Þ

rscat ¼

Where a is the polarizability. The em and e are the permittivity of the metal sphere and the surrounding medium, respectively. K = 2p/k and a is the sphere diameter. Equations 1 and 2 show that rabs varies with a3 whereas rscat varies with a6.

Absorption Enhancement in an Amorphous Silicon

263

Consequently, for large particles (in terms of the wavelength), extinction is dominated by scattering, whereas absorption is related to small particles. For PV applications, the ratio of absorption/scattering should be close to unity because the power absorbed by the metal nanoparticles (MNPs) is dissipated as heat.

3 The Hollow Ring Nanoantennas Figure 2 shows the investigated configuration to enhance the light absorption. It consists of a 2-D periodic array of metallic circular rings placed on the surface of a substrate of amorphous silicon (a-Si) having a refractive index of ns = 3.936. The rings are covered by a layer of anti-reflection coating of Indium Tin Oxide (ITO). The CST microwave suit was used in the investigation assuming a unit cell normally illuminated by a plane wave propagating along the z-axis (normal incidence) and is linearly polarized along the x-axis. The periodic boundary condition is assumed in the simulation. The structure was investigated first without the rings and with a back reflector to find the effect of the substrate thickness. The obtained results are shown in Fig. 3a, where the calculated absorption and reflection are at 41% and 59% respectively. Figure 3b shows the effect of the substrate thickness on the frequency of maximum absorption (minimum reflection). Accordingly, the thickness of the (a-Si) layer was chosen as 65 nm to achieve maximum absorption. ITO layer

Background 20nm of Ag

P

2R=43 nm L=130nm

L

P=400nm

L P

P

P

a-Silicon 65nm

Fig. 2. Schematic of unit cell simulation with a 3x3 HRNAs elements

4 The Proposed 3X3 Cluster Rings In order to achieve a response of higher bandwidth in comparison with the responses shown in Fig. 4, a unit cell with 3X3 rings is proposed such that the rings have various dimensions. The inner diameters of the nine rings were set at the values shown in Table 1 and the 3X3 ring cluster was arranged as shown in Fig. 5. In this design, each ring will resonate at a certain frequency depending on its average diameter. The values

264

A. A. Rasheed et al.

Absorption Reflection

0.8

Frequency, THz

Absorption & Reflection

1

0.6 0.4 0.2 0 200

220

240

260

280

300

320

340

320 280 240 200 50

360

Calulated Simulated

380 360

55

60

65

70

75

80

85

d, nm

Frequency, THz

b

a

Fig. 3. (a) Optical absorption and reflection for the a-Si layer of thickness d = 65 nm, without HRNAs (b) Effect of changing the a-Si thickness d on the frequency of maximum absorption (minimum reflection).

of the average diameters (ring sizes) were chosen in an attempt to achieve coverage of the desired band of frequencies (between 230 THz and 360 THz).

Absorption / Reflection

1 0.8 0.6 0.4 0.2 0 200

220

240

260

280

300

320

340

360

Frequency THz 2R=35 2r=0 2R=43 2r=17

2R=43 2r=0

2R=43 2r=9

2R=43 2r=13

2R=43 2r=21

Fig. 4. Absorption (solid lines) and reflection (dashed lines) for various sizes of the rings 2r = 0 means filled ring or solid disk

Figure 6 shows the reflection and absorption responses for the substrate without the rings and those responses with the cluster of rings shown in Fig. 5. It can be seen from Fig. 6a that the absorption in the substrate is low and of narrow bandwidth while that when using the rings (Fig. 6b) is higher and having a much wider bandwidth. The ripples in the response are attributed to the nine individual responses caused by the nine ring having different inner diameters.

Absorption Enhancement in an Amorphous Silicon

265

Table 1. Inner diameters of the nine rings in nm Row/column 1 2 3 1 0 24 17 2 13 0 21 3 9 27 7

Fig. 5. Schematic of the proposed unit cell consisting of nine different HRNAs 1

Absorption Reflection

0.8

1 0.8

0.6

0.6

0.4

0.4

0.2 0 200

240

280

Ag HRNAs

0.2

Without HRNAs 320

0 200

360

Frequency THz

240

280

a 1

0.8

0.8

0.6

0.6

Au HRNAs

0.2 240

320

360

0.4

0.2 0 200

360

b

1

0.4

320

Frequency THz

280

320

360

0 200

Al HRNAs 240

280

Frequency THz

Frequency THz

c

d

Fig. 6. Absorbance and reflectance for rings of three types of metals; (b) silver, (c) gold, and (d) aluminum compared with the case without Rings (a).

To investigate the effect of the type of metal of the used rings, simulations were performed for using the same substrate, ITO and the 3X3 ring cluster while assuming the use of gold and aluminum. Figure 6c and Fig. 6d show the obtained results. The

266

A. A. Rasheed et al.

Lorentz-Drude dispersion relations for the two metals were used to incorporate their properties. Comparing the results, it is seen that for the three metals employed, the inclusion of the HRNAs significantly increases the optical absorption, and the plasmonic phenomenon plays an important role in the obtained results. The best absorption response occurs with a silver HRNAs followed by gold and then aluminum. By return to Fig. 1, which shows the metals dispersion relations with frequency, it is clear that aluminum has inter band transition in the epsilon imaginary part at around 300 THz, where the imaginary part produces negative features of plasmonic, but in contrast, aluminum provides remarkable enhancement in plasmonic behaviour in the high frequency range of UV. The gold is better than aluminum but it has higher loss than silver. Therefore, silver metal has extraordinary features compared to the other two metals especially at this region of the spectral band (IR) and eventual utilization of HRNAs in increasing PV performance. Table 2 shows the absorption and reflection for the three different metals used in the rings. The table also shows the ratio of the absorbed power in the a-Si layer of the HRNAs structure to the absorbed power in the same layer without HRNAs.

5 Power Enhancement in a-Si Layer of the Structure The comparison of absorbed power by the (a-Si) layer in the structure for different materials of HRNAs with the absorbed power without HRNAs is shown in Fig. 7, and the numerical results are listed in Table 2. The results clearly illustrate that the proposed structure of 3X3 rings has significantly improved the absorbance performance where the surface plasmon is generated. The silver metal has better performance, where the optical absorption by the (a-Si) layer increases effectively in the IR range of (230 THz–360 THz). The integrated power under the absorption spectrum shows a gain value equal to 3.32 when silver HRNAs is used in the structure, 1.73 and 0.96 for gold and aluminum respectively. In the case of aluminum rings, although absorption by the structure increases and reflection decreases, the absorbed power by the (a-Si) is less

Absorbed power

0.12 0.1

w/o ring Ag Au Al

0.08 0.06 0.04 0.02 0 200

220

240

260 280 Frequency, THz

300

320

340

360

Fig. 7. Power absorption in the a-Silicon layer for various materials types of HRNAs compared to the case without HRNAs

Absorption Enhancement in an Amorphous Silicon

267

Table 2. Absorption, reflection, and power ratio for various metals with and without rings. Metal type Absorp. A% W.O. rings 41 Ag rings 78.5 Au rings 67 AL rings 56

Reflec. R% 59 21.5 33 44

A/R

absorbed Power with rings absorbed power without rings

0.69 3.65 2.03 1.27

1 3.32 1.73 0.96

than that without rings, and this is attributed to blocking light through the disc, and this increase in absorption represents losses in the structure. Therefore, it has been verified that the silver HRNAs when added to the structure, can enhance the optical absorption in the active layer of the solar cell in the near infrared region. Therefore, the proposed structure will improve the efficiency of the solar cell, plasmonic optical sensors, and optical device applications.

6 Conclusion Absorption enhancement in an (a-Si) layer by using a structure of 3X3 metal rings as a unit cell has been demonstrated. This structure consists of a cluster of 3X3 rings placed on the amorphous silicon layer. The rings have the same outer diameters and various hole sizes, where each ring operates at a certain resonance frequency depending on its hole size so as to cover the desired spectrum range of 230 THz–360 THz. The structure has offered an absorbance enhancement from 41% without rings to 78.5% with silver HRNAs in the 200–360 THz frequency range, and the absorption enhancement in the a-Si layer is 3.32 times that obtainable without the rings. This approach can be applied to the design of various solar cells operating in the visible and infrared frequency range.

References 1. Chu, H.S., Ewe, W.B., Koh, W.S., Li, E.P.: Remarkable influence of the number of nanowires on plasmonic behaviors of the coupled metallic nanowires chain. Appl. Phys. Lett. 92, 103103–103105 (2008) 2. Bohren, C.F., Huffman, R.: Absorption and Scattering of Light by Small Particles. Wiley, New York, (1983) 3. Lecarme, O., Sun, Q., Ueno, K., Misawa, H.: Robust and versatile light absorption at nearinfrared wavelengths by plasmonic aluminum nanorods. ACS Photonics 1, 6538–6546 (2014) 4. Hu, C.-C., Tsai, Tsai, Y.-T., Yang, W., Chau, Y.-F.: Effective coupling of incident light through an air region into an S-shape plasmonic Ag nanowire waveguide with relatively long propagation length. Plasmonics 9, 573–579 (2014) 5. Chau, Y.-F., Chao, C.-T., Rao, J.-Y., Chiang, H.-P., Lim, C.M., Voo, N.-Y.: Tunable optical performances on a periodic array of plasmonic bowtie nanoantennas with hollow cavities. Nanoscale Res. Lett. 11(1), 411–419 (2016)

268

A. A. Rasheed et al.

6. Johnson, P.B. Christy, R.W.: Optical constants of the noble metals. Phys. Rev. B. 6, 4370– 4379 (1972) 7. Ordal, M.A., Bell, R.J., Jr, R.W.A., Long, L.L., Querry, M.R.: Optical properties of fourteen metals in the infrared and far infrared: Al, Co, Cu, Au, Fe, Pb, Mo, Ni, Pd, Pt, Ag, Ti, V, and W. Appl. Opt. 24, 4493–4499 (1985) 8. Vandenbosch, G.A.E., Ma, Z.: Upper bounds for the solar energy harvesting efficiency of nano-antennas. Nano Energy 1(3), 494–502 (2012) 9. Seok, T.J., Jamshidi, A., Kim, M., Dhuey, S., Lakhani, A., Choo, H., Schuck, P.J., Cabrini, S., Schwartzberg, A.M., Bokor, J., Yablonovitch, E., Wu, M.C.: Radiation engineering of optical antennas for maximum field enhancement. Nano Letters 11, 2606–2610 (2011) 10. Oates, T.W.H., Keller, A., Facsko, S., Mucklich, A.: Aligned silver nanoparticles on rippled silicon templates exhibiting anisotropic plasmon absorption. Plasmonics 2, 47–50 (2007) 11. Maier, S.A.: Plasmonics-Fundamentals and Applications, p. 70. Springer (2007)

Telecommunications and Networks

Design of a Content-Based Communication Model Using Caching Technique for VANETs Mohamed Anis Mastouri(&) and Salem Hasnaoui Communication Systems – Sys’Com Laboratory, National School of Engineers of Tunis – ENIT, Tunis, Tunisia [email protected], [email protected]

Abstract. Researchers in VANET has focused essentially on the design of communication protocols in the context of closely spaced vehicles. However the network of VANET can be disconnected. In this case there is a possibility that the communication fails. We have to avoid the discontinuity during data exchange in this kind of network. We propose in this paper a solution based on the caching technique with the store carry and forward algorithm using a communication model based on the content which gives decoupling in space and time to the entities that communicate. This solution is very suitable for such environments. It is designed in order to bypass discontinuity and high mobility. Keywords: VANET


Publish-subscribe
MANET
Opportunistic
Routing

1 Motivation: VANET

VS

MANET

VANET is a set of connected vehicles that move along the roads when the movement paths are controlled. It exhibits a bipolar behavior: VANET can be fully connected or weakly connected [1]. In this case VANET will be represented as a Disconnected MANET (DMANET). This network is characterized by a high mobility of its nodes. Each node can have a different speed which generates a non-predictable topology which tends each time to a set of connected areas. Thus, the vehicles move at a different speed, leading to unpredictable changes in the topology of the network. The result forms a large disconnected network of vehicles. The application of VANETs can be concretized in several forms such as security; traffic conditions; comfort… [2]. It is intended for a broad range of applications used by the drivers. Drivers can get information such as notification of a traffic jam, alert of a dangerous situation, information on traffic lights, notification in case of urgent braking, notification of an accident, obstacle or state of the road, speed limit of the road, traffic conditions or current weather conditions. So VANET is represented as a Disconnected MANET using communication mechanisms which have to be asynchronous and characterized by a strong decoupling. Publish-subscribe is most suitable communication paradigm for building applications under such conditions. The advantage of this model is the decoupling in space and time between the different entities. © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 271–280, 2020. https://doi.org/10.1007/978-3-030-21009-0_26

272

M. A. Mastouri and S. Hasnaoui

So the vehicles will play the role of the event producer/consumer and an event carrier in the same time in order to avoid discontinuity. So each entity will be equipped with a caching.

2 Applications in VANET VANET is used to broadcast various and important information (Fig. 1) such as:

Fig. 1. VANET converging on DMANET

(1) Information about Prevention and road safety: used to prevent road users from seriously injured. Drivers receive information about the road conditions [3]. (2) Information used to optimize traffic: Vehicles collect and share data in order to greatly improve car traffic. (3) Information used to provide comfort to the passengers: vehicles exchange data about comfort such as whether information’s, informative entertainment, internet access … (4) Information about parking: vehicles can also support other applications such as parking. Other services running on the internet can also be applied and shared here. These applications are more suitable for atypical and asynchronous model. The communicating entities are characterized by low coupling interactions. So it is necessary to use a content-based communication model (Fig. 2).

3 Using Publish-Subscribe According to DDS Data exchange in VANETs can help drivers to be informed about traffic conditions in order to give good conditions for driving. This exchange is ensured with publishsubscribe model according to Data Distribution Service - DDS [4].

Design of a Content-Based Communication Model

273

Fig. 2. Examples of applications in VANET

In fact, unlike the client/server model, asynchronous communication systems, based on sending messages, have become more and more interesting in the context of distributed applications [5]. It is agreed that asynchronous communication models are better suited than synchronous models (client/server type) for a network characterized by a strong decoupling between entities. The decoupling results from several factors of a spatial or temporal nature: the geographic distance of the communicating entities, the possibility of temporary disconnection of a node due to a breakdown or an interruption of the communication. The asynchronous communication is the best equipped to deal with these features. For many years the rise of this kind of communications has not been remarkable because of the lack of standardization (as opposed to what has happened for clientserver systems with CORBA) until the arrival of the new specification (DDS - Data Distribution Service) of this famous publish-subscribe paradigm. So for all of these reasons publish-subscribe model according to DDS is very suitable for disconnected MANET notably VANET. DDS enables the development of sensors based services. More specifically, it carries out the communication and distribution of data between information receivers (sensors) and information processors (displays or database). It therefore works in a realtime context to disseminate the requested information, in the requested place in a definite moment. It thus facilitates the various communications between all the actors who intervene during the execution of the application. The system model based on DDS consists of a group of connected nodes via a realtime-transport-layer. Each entity is equipped with an OS with a middle-ware and a publish-subscribe procedure according to data-distribution-service specification [6].

274

M. A. Mastouri and S. Hasnaoui

This model according to dds is based on the interaction between data writer and data reader. The data writer exists in the dds producer and the data reader in the consumer. This is depicted in Fig. 3.

Fig. 3. Topic and data-object in a middleware

The data objects of the distributed systems go through all the participants. Each participant is considered to have a local data object cache. The message on the topic is considered a data object update which is controlled by system. Local modifications of the data-objects are propagated by the middleware; the latter can also update data from different data-objects and manage their deliveries. There are two communication concepts: the publication/subscription communication model and the use of quality of services. These two concepts characterize the behavior of the system and also represent the major qualities of the DDS middleware. We therefore chose to use the publish-subscribe model according to the DDS specification because it allows real-time exchanges according to certain qualities of service that it offers and it is also Object Oriented. Indeed, this model of communication makes it possible to tolerate the intermittent connectivity, as well as the latency in the transmissions thanks to its temporal decoupling and the asynchronous which characterize it.

Design of a Content-Based Communication Model

275

4 Architecture of the Solution Based on Cache Technique In our solution and to make reliable the exchange in this discontinuous network strongly fragmented with moving elements, it is proposed that each node is provided with a cache. This storage capacity is limited and can change from one node to another depending on several parameters. When a node receives information from a neighbor, it begins by checking whether it is interested in it or not. If so, it makes it available to the appropriate local enforcement entity. In any case, it stores it in its cache then it plays the role of mobile carrier and then participates in the spread of this information between connected blocks scattered and met occasionally. Thus, when an application entity of a given terminal publishes information, it puts it first in its local cache. From this one, the dedicated module for opportunistic routing [7] according to the model (store, carry and forward) will be interested in propagating this information from one node to another [8] (Fig. 4).

Fig. 4. Layered architecture of the solution

– Application layer: in this layer we can find a set of services related to the road conditions running in the context of publish subscribe (for example roadside prevention, warning of a traffic jam, a rockslide or an accident). These services exploit the content-based communication model via the interaction between the DDS producer and the DDS consumer. – Management layer: this layer is interested in managing the different application services. It manages the list of interests in the configured entity. The administration layer can have the possibility of acting on the node behavior and it switches the node to a altruistic and proactive behavior. – Caching: Any message just published is stored in caching. Connectivity lost due to frequent nodes mobility can also be recuperated using caching after that it can forward messages. thanks to this cache the node delivers the message to another after an occasional meeting [9]. – List of interests: is the list of topics that the driver subscribes on it. This list is used to be matched with the received information communicated by another entity. In

276

M. A. Mastouri and S. Hasnaoui

fact the information carried by the carrier will be compared with this list to filter the information required. – Opportunistic Dissemination based on publish-subscribe (opp-pub-sub): it consists of using an opportunistic routing protocol adapted to the context of publish subscribe model. This is ensured by a program called opp-pub-sub using epidemic routing with store-carry-and-forward. it allows to save the information (store), carry (carry), route (forward) and distribute it later to new neighbors. – Intra-Bloc Communication: This module ensures the communication between the nodes inside the same block. An example of routing protocol that can be used in this layer is OLSR - Optimized Link State Routing Protocol, which is a proactive routing protocol operating in a mobile environment, distributed without any central entity controlling it. It is used in dense and not very mobile networks which explain its interest inside the block. It is an optimization and adaptation of the principle of link state routing for ad hoc networks. So this model is intended to provide communication between information transmitters and subscribers in a highly fragmented mobile network in small blocks. For this, this proposed model will use the “store carry and forward” protocol to route messages and the DDS middleware to publish them to interested nodes. So it’s a combination of the opportunistic routing protocol with the DDS API. Multi-hop routing occurs in order to reacha remote node in the same block. In addition to its role of storage, routing and dissemination of information, each node must be able to filter this information by distinguishing between its different types and must also subscribe to the information that interests it.

5 Caching Technique The message transited between nodes has to be hosted in a cache when the node meets another that is not interested to it. So we need a persistent cache in the proposed model capable of storing messages exchanged between the publishers and the subscribers. This cache will be managed according to the configuration given to the node. It can be selfish or altruistic behavior. The selfish mode means that the node stores only the requested message that it subscribed on it. The message is stored after a comparison procedure between the list of interests and the received message. Otherwise it is rejected and it isn’t stored in the cache since the mode is selfish. The altruistic mode means that the node can store the information that is not requested by the subscriber. In fact the message is copied in the storage space in each meeting between the current node and the carrier of the message. After that, the node will be waiting for other occasional meeting with another subscriber that it requested the same information. This cache is managed by a program responsible of acting to the message according to the behavior mode of the node and also flushing the space when the storage of the cache reaches the maximum.

Design of a Content-Based Communication Model

277

The owner of the node have also the possibility to manage the cache by searching, adding, removing the different messages stored. The message format is composed of the following fields: – Code: is an integer and unique. Its uniqueness is useful for facilitating cache management later. – Service: The type of service to which the message belongs. – Info: contains information that identify the message (The registration number of the vehicle-publisher, location of a road signaling panel) – The-expiration-date: It’s the life span of the information that reflects the duration of the importance of the information. After that the message becomes useless and must be deleted. – The-data: it represents the content of the information on which the vehicle subscribed to use it. An example of the message exchanged between publishers and subscribes is below: 20

Road-info`

112TUN145

12/05/2021

500 m: radar-detector-Tunis-Beja

Finally the cache is the core of the proposed solution that will circumvent the problems of the high mobility of the network and the frequent disconnections of links.

6 Simulation Context In order to evaluate the performance of the proposed model based on the caching technique, and especially its efficiency in VANET, we programmed a simulator in java based on its storage capacity for messages in transit. In fact the different components that have to be evaluated in each entity are: pub-sub module, the adjacency table manager and the cache manager. In the simulation context the network is initially connected. As a result of the high mobility of its nodes we obtain two separated islets (blocks), so the adjacency table of the mobile node will be changed. In this case the routing technique based on the storecarry-and-forward algorithm will connect the mobile node with the initial network and thanks to the publish-subscribe process the requested message will arrive the subscribers nodes. So our simulator focused on the disconnection of the link between the mobile node and the initial connected block and the recovery of the link cause by the opportunistic routing using publish-subscribe model. The disconnection and the recovery of the link will be simulated as shown in the Fig. 5: After that the reliability of the simulated network will be evaluated according to the capacity of the storage space in number of stored events. The disconnection of the link represents the remoteness of the vehicle and the recovery of the link represents the reconciliation of the link.

278

M. A. Mastouri and S. Hasnaoui

Fig. 5. Disconnection of the link

7 Theoretical Evaluation of the Cache The goal of this part is to evaluate theoretically the utility of the cache in a context of VANET using a content-based communication. For this reason we measure the number of events according to the size of the cache. We assume the following parameters: kpub ksub kLfail tp and ts L ksub Recov and kRecov

Publication-rate subscriber’s access rate of published events. Link-Failure-rate Time-delay of subscribing and publishing. (t = tp + ts) subscriber Recovery-rate and link Recovery-rate

P = probability of i notifications arrived between disconnection and recovery of the link is: p¼

kpub kpub þ ksub recov

!i

ksub recov kpub þ ksub recov

ð1Þ

NL is the maximum cache siz. When the events number exceeds N, the event will be considered lost. In the subscriber, The lost events number is: !i kpub ksub recov EðNÞ ¼ ði  NL Þ sub k þ k k þ ksub pub pub recov recov iNL þ 1 !NL kpub ksub recov ) EðNÞ ¼ sub kpub þ krecov ksub recov 1 X

ð2Þ

The average time per subscriber in which the system is down: T ¼ ts EðNÞ ¼ ts

ksub recov kpub þ ksub recov

!NL

kpub ksub recov

ð3Þ

Design of a Content-Based Communication Model

279

The average number of the lost events according to the cache size is shown in the curve below (Fig. 6):

Fig. 6. The average number of lost events according to the size of the cache (ƛpub = 0.2)

8 Conclusion VANET is represented as a Disconnected MANET using communication mechanisms which have to be asynchronous and characterized by a strong decoupling. For this reason we demanded the use of a communication based on the publish-subscribe model. Since the high mobility of this kind of network, we have overcome the problem of temporary and frequent disconnections of links by using a content-based opportunistic routing and the integration of a cache manager that has been evaluated later.

References 1. Singh, S., Agrawal, S.: VANET Routing protocols: issues and challenges. IEEE (2014) 2. Chouhan, P.: Comparative study MANET and VANET. Int. J. Eng. Comput Sci. 5(04), 16079 (2016) 3. Tulika, Garg, D., Madhav Gore, M.: A publish/subscribe communication infrastructure for VANET applications. IEEE (2011) 4. Yang, J., Sandström, K., Nolte, T., Behnam, M.: Data distribution service for industrial automation. IEEE (2014) 5. Pardo-Castellote, G., Farabaugh, B., Warren, R.: An introduction to DDS and data-centric communications. RTI 6. Vargas, L., Bacon, J., Moody, K.: Integrating databases with publish/subscribe. IEEE (2005) 7. Pelusi, L., Passarella, A., Conti, M.: Opportunistic networking: data forwarding in disconnected mobile ad hoc networks. IEEE Commun. Mag. 44(11), 134–141 (2006)

280

M. A. Mastouri and S. Hasnaoui

8. Kolios, P., Papadaki, K.: Energy-efficient relaying via store-carry and forward within the cell. IEEE (2014) 9. Patel, T., Kamboj, P.: Opportunistic routing in wireless sensor networks: a review. IEEE (2015) 10. Mammas, M., Ghadi, F., Mammas, H.: A software infrastructure for multiservice access based on contactless smartcard and S-OrBAC model. IEEE, Hammamet, Tunisia (2016) 11. Mokhlissi, R., Loutfi, D., El Marraki, M.: A theoretical study of the complexity of complex networks. IEEE, Hammamet, Tunisia (2016) 12. Samir, J., Adnen, C., Sami, B.S., Balas, V.E.: An efficient design of fuel cell electric vehicle with ultra-battery separated by an energy management system. IEEE, Hammamet (2016) 13. Ettaghouzti, T., Hassen, N., Besbes, K.: High performance low voltage low power voltage mode analog multiplier circuit. In: IEEE (2016)

Monitoring of Greenhouse Based on Internet of Things and Wireless Sensor Network Achouak Touhami1(&), Khelifa Benahmed2, and Fateh Bounaama1 1

2

Department of Electrical Engineering, Faculty of Technology, Tahri Mohammed University of Bechar, Bechar, Algeria [email protected] Department of Mathematics and Computer Science, Faculty of Exact Sciences, Tahri Mohammed University of Bechar, Bechar, Algeria

Abstract. Today and especially after the oil crisis, agriculture has become an important sector in the economy of several countries and in particular in the Mediterranean countries. In recent decades, the technology has experienced a great development in greenhouses for good quality production. The greenhouses are enclosures that generate from the local outdoor conditions, a microclimate more favorable to the growth of plants. With the application of new technologies such as: wireless sensor networks (WSN), actuators, controllers and the Internet of Things (IoT), in greenhouses, they come easy to manage where these technologies facilitate the tasks of farmers. The objective of this work is to propose a design based on these new technologies. Keywords: Greenhouses


WSN
Actuators
Controllers
IoT

1 Introduction Nowadays, in the markets, vegetables, fruits, flowers and plants are out of their growing season and out of their physiological cycle. They can only manifest themselves if favorable and particular conditions are met artificially. A greenhouse is an enclosure in which plants are grown. The goal of using greenhouses is to give farmers a precise result on the activities carried out on plants under different environmental conditions. Today, with the rapid development of greenhouse control and monitoring technology, information is obtained accurately and reliably. Over the past decades, wireless sensor networks (WSNs) have received considerable interest [14]. The advances in sensor technology are associated with wireless communication technologies (WSN) [17]. They are easy to install and operate. WSNs for the most part comprise of ease, low power, multi-practical sensor hubs that are little in size and convey over short separations [1]. They gather discovery data from a few points and send it remotely to the base station. At that point, she stores it. While some system hubs may likewise perform information handling and send preprocessed data to the base-station. WSN it can be used for many applications such as precision agriculture, environmental control and health care [14]. So, they can be used in the © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 281–289, 2020. https://doi.org/10.1007/978-3-030-21009-0_27

282

A. Touhami et al.

automation system architecture in modern greenhouses. The advantages of installation of WSN over the wired systems, they are fast, cheap and easy. The reconciliation of new advances in greenhouse the executives and checking framework is one of the approaches to enhance plant development forms. In such manner, with the utilization of new innovations, we will propose in this paper a plan for greenhouse monitoring dependent on the Internet of Things and new correspondence advances. This structure enables the agriculturist to monitor and control the greenhouse utilizing advanced cells or tablet by means of the Internet anyplace and whenever. The rest of paper is organized as follow: in Sect. 2 we will present generalities about greenhouse, IoT and WSN, the Sect. 3 presents some works have been done in monitoring of greenhouse using IoT and WSN, our proposed approach will be presented in Sect. 4 and finally, a conclusion and proposed perspectives of this work will be presented in Sect. 5.

2 Generalities 2.1

Greenhouse

Definition Anuradha Gaikwad et al. [2] define de term greenhouse as: “Greenhouse is an environment which is created by human to grow their crops. It needs to monitor the parameters like temperature, humidity, sunlight”. The fundamental objective of greenhouses is agricultural production outside the natural season of cultivation. Greenhouse modelling Data presentation (input-output) The greenhouse is an agricultural building that contains the inputs. Because of these inputs, outputs are obtained. The input-output vectors are present in Fig. 1.

Fig. 1. The input-output vectors of greenhouse

With: Te: External temperature (oC). Ti: Internal temperature (oC). Rg: Radiation (W/m2). Qs: Floor heating (W/m2).

Monitoring of Greenhouse Based on Internet of Things

283

Qa: Heating area (W/m2). S: Window (m2). V: Wind speed (m/s). Pe: External pressure (mbar). Pi: Internal pressure (mbar). 2.2

Internet of Things

Definition The expression “Internet of Things” was first utilized by the Massachusetts Institute of Technology in the year 1999 [3]. There are a few meanings of the Internet of Things [4, 5]. Definition of ITU-T: “The IoT is a worldwide foundation for the data society, which gives propelled administrations by interconnecting objects (physical and virtual) through data and correspondence advances. By abusing the abilities of distinguishing proof, information catch, preparing and correspondence, IoT takes full favorable position of articles to give administrations to a wide range of utilizations, while guaranteeing consistence with security and secrecy necessities”. Definition of IERC: “A dynamic worldwide system framework with self-arranging capacities dependent on standard and interoperable correspondence conventions where physical and virtual ‘‘things’’ have characters, physical characteristics, and virtual identities and utilize keen interfaces, and are flawlessly incorporated into the data organize”. But loT is smart Internet extension with low-cost intelligent interconnections collecting data and distribute intelligent services and applications [15]. The application domains of IoT There are a lot of applications. We list some application scenarios as follows: Smart Home: life at home is improved, by making it increasingly advantageous and simpler to screen and work home apparatuses and frameworks (e.g., microwave, stove, forced air system, warming frameworks, and so on.) remotely. Industrial Automation: With a negligible human contribution, automated gadgets are mechanized to get done with assembling undertakings. The machines’ activities, functionalities, and profitability rates are consequently controlled and checked. What’s more, time and amount of creation are made strides. Smart Healthcare: Performance of social insurance applications is enhanced, by implanting sensors and actuators in patients and their prescription for checking and following patients. By social occasion and breaking down patients’ body information with sensors and further conveying investigated information to a handling focus, the clinical consideration could screen physiological statuses of patients continuously and make reasonable activities when essential. Smart Grid: The vitality utilization of houses and structures could be improved. For instance, the meters of structures could be associated with the system of vitality suppliers. At that point the vitality suppliers could upgrade their administrations, by gathering, breaking down, controlling, checking, and overseeing vitality utilization. In the interim, the potential disappointments could be decreased. Smart City: Personal satisfaction in the city is improved, by making it increasingly helpful and less demanding for the inhabitants to get data of intrigue. As indicated by

284

A. Touhami et al.

individuals’ needs, different interconnected frameworks astutely offer the attractive administrations (e.g., transportation, utilities, wellbeing, and so on.) to individuals. 2.3

Wireless Sensor Network

Definition Sensor is a measuring device that converts physical grandeur into comprehensive signals for the observer. Wireless Sensor Network (WSN) includes more than a few segments called ‘nodes’ (Fig. 2) [6]. The nodes are smart devices. Their role is the data collection. The basic functions of a sensor network are: Sensing, Communication and Computation [6]. The nodes that gather the data are called source node while the node that assembles the data from all source node is known as the sink node and at some point the door node [6]. The integration of the actuators in the WSN is called Wireless Sensor and Actuator Network (WSAN). The goal of the actuators in the WSN is that they increase the capability of WSN from monitoring to the control [6].

Fig. 2. Wireless Sensor Network (WSN).

Most Used Standard and Technologies 1- ZigBee IEEE 802.15.4: ZigBee is a communication protocol developed by the ZigBee Alliance worldwide and used all over the world for control and monitoring. It aims to achieve the following [7]: – – – – – –

Low expense. Ultra-low power utilization. Use of unlicensed radio groups. Cheap and simple establishment. Flexible and extendable systems. Integrated knowledge for system set-up and message directing [7].

2- WIFI IEEE 802.11: WiFi or Wireless Fidelity is a wireless local area network (WLAN) developed by the American Standards Organization IEEE [8]. This standard is generally considered to be the wireless version of 802.3 (Ethernet). Today, Wi-Fi is the most widely used wireless local area network technology, due to its simplicity and its low implementation cost.

Monitoring of Greenhouse Based on Internet of Things

285

3- Bluetooth IEEE 802.15.1: Bluetooth is a wireless personal area network (WPAN), which allows instantaneous connections between several electronic devices such as cellular telephones, Personal Digital Assistants (PDA) and computers accessible to the users [9].

3 Related Work Many research and projects have been done in order to improve and monitor the climatic conditions under greenhouse using IoT technology and WSN. The authors in [10] have displayed a monitoring and control framework for greenhouse through IoT. The framework will monitor the different natural conditions, such as humidity, soil moisture, temperature, presence of fire, etc. The GSM module is utilized to make an impression on the enlisted number, if any condition crosses certain breaking points. The model was tried under different blends of contributions to their research facility and the exploratory outcomes were found obviously. The authors in [2] have utilized a remote control framework with a resulting programming for the plan and execution of smart greenhouse technology. This control framework handles the fundamental factors, such as sunlight, temperature, humidity, utilizing ongoing clock set and miniaturized scale controller and complete remote transmission of data to remote programming. They have utilized for usage of smart greenhouse monitoring framework the ZigBee technology. This framework is comprised of front-end data acquisition, data processing, data transmission, and data reception. The processed information is send to the middle node through a remote system. Middle of the road node gets all information and sends the information to the PC through sequential port, in the meantime staff may see investigation and capacity of the information. PC gives a continuous information to greenhouse fans another temperature control hardware and accomplish programmed temperature control. [11] have built up a computerized scheduler framework by considerating with all opti-mal plant development necessities for each period of the plant to guarantee that all subjects (mango) will develop superbly. Fundamental equipment segment inside undertaking is Memsic, Zigbee and advanced mobile phone for showcase while MP Lab and LabView are utilized for programming components. With the utilization of this framework work and upkeep cost will be less expensive and the way toward observing and gathering information or data is all the simpler and proficient. Prof. Shirsat, D.O. et al. [12] have depicted the structure of a greenhouse monitoring and controlling framework dependent on IOT utilizing Arduino. The framework will enable him to take appropriate choices by giving the status of the sensors to the rancher with exact data through the IOT web server. Hence this framework encourages rancher to control greenhouse from remote areas.

286

A. Touhami et al.

4 Our Proposed Approach 4.1

The System Design

Our approach proposed in this paper is composed of: wireless sensor network (DHT11 Humidity and Temperature Sensor), the 6LoWPAN smart gateway that connects the Zigbee network with the internet. The sensor set in greenhouse, measure consistently the humidity values, the temperature values of the inside of the greenhouse, then send these values through a ZigBee network to a smart gateway, this data are then sent via a mobile data communication network to a web service that utilizes intelligent software application to automatically analyze the information and act as indicated by the acquired outcomes, by specific enactment of actuators as required. The routing protocol utilized in this proposed design is the RPL protocol. The outputs results are displayed to the client on a smart mobile phone or tablet web application utilizing CoAP or HTTPs interfaces. The objectives of our system are: – The system is easy to deploy, to use, and facilitates management and monitoring of greenhouses. – The system is modular and flexible, making it easy to maintain – The system design is robust and reliable. The inputs of our design are: – The external and internal temperature. – The external and internal humidity. The outputs of our design are: – The optimum internal temperature. – The optimum internal humidity. Sensors The DHT11 humidity and temperature sensor DHT11 sensor is utilized to quantify temperature and humidity values [16], the output of this sensor is advanced and it has quick reaction time, great exactness, and high resolution. Information type of DHT11 is 8bit indispensable RH information + 8bit decimal RH information + 8bit basic T information + 8bit decimal T information + 8bit check aggregate. In the event that the information transmission is substantial, the registration ought to be the last 8bit of “8bit fundamental RH information + 8bit decimal RH information + 8bit vital T information + 8bit decimal T information” [13]. Actuators They are the actuators in window, fan, sprinkler and heater. The role of the actuators is that the equipment is fonctionned automatically (Fig. 3).

Monitoring of Greenhouse Based on Internet of Things

287

Fig. 3. The system design

4.2

Flowchart Representation

Figures 4 and 5 shows how monitoring the temperature and humidity of greenhouse. Firstly, you need to know the type of plant grown in the greenhouse ton know the threshold of temperature and humidity. Then, we check the current temperature (current humidity) with the temperature threshold (the humidity threshold). The test of temperature and humidity is as follow: The test of temperature: If (T > Ttrs) then Fan = ON; Heater = OFF; Else Fan = OFF; Heater = ON; End. The test of humidity: If (H > Htrs) then Window = ON; Sprinkler = OFF; Else Window = OFF; Sprinkler = ON; End.

288

A. Touhami et al.

With: T: current temperature. Ttrs: temperature threshold of plant.

Fig. 4. Flowchart for monitoring temperature

With: H: current humidity. Htrs: humidity threshold of plant.

Fig. 5. Flowchart for monitoring humidity

5 Conclusion In this paper we proposed a system for monitoring a greenhouse. This proposed system is based on IoT technologies and WSN. This system can facilitate the work of farmers who can manage and control greenhouses remotely because of the technologies and materials used.

Monitoring of Greenhouse Based on Internet of Things

289

As a perspective, we plan implemented our solution in simulators dedicated to this application and then applied it in the agricultural lands of south of Algeria. And we hope to treat the problem of energy consumption in sensors by the use of renewable energies

References 1. Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., Cayirci, E.: Wireless sensor networks: a survey. Comput. Netw. 38, 393–422 (2002) 2. Gaikwad, A., Ghatge, A., Kumar, H., Mudliar, K.: Monitoring of smart greenhouse. Int. Res. J. Eng. Technol. 3(November), 573–575 (2016) 3. Uckelmann, D., Harrison, M., Michahelles, F.: Architecting the internet of things, pp. 1–24. Berlin and Heidelberg, Germany, Springer-Verlag (2011) 4. Atzori, L., Iera, A., Morabito, G.: The internet of things: a survey. Comput. Netw. 54(15), 2787–2805 (2010) 5. Shaikh, F.K., Zeadally, S., Exposito, E.: Enabling technologies for green internet of things. IEEE Syst. J., to be published 6. Aqeel-ur-Rehman, Abbasi, A.Z., Islam, N., Shaikh, Z.A.: A review of wireless sensors and networks’ applications in agriculture. Comput. Stand. Interfaces, Elsevier, 36, 263–270 (2011) 7. Patel, N., Kathiriya, H., Bavarva, A.: Wireless sensor network using Zigbee. Int. J. Res. Eng. Technol. 2, 1038–1042 (2013) 8. Varma, V.K: Wireless Fidelity—WiFi. IEEE Emerging Technology portal (2012). https:// www.ieee.org/about/technologies/emerging/wifi.pdf 9. Puy, I.: Bluetooth (2008). http://webuser.hs-furtwangen.de/*heindl/ebte-08ss-bluetoothIngo-Puy-Crespo.pdf 10. Koshy, R., Yaseen, M.D., Fayis, K., Nisil, S., Harish, N.J., Ajay, M.: Greenhouse Monitoring and Control based on IOT using WSN. ITSI Trans. Electr. Electron. Eng. (ITSITEEE), 4(3) (2016) 11. Halim, A.A.A., Hassan, N.M., Zakaria, A., Kamarudin, L.M., Bakar, A.H.A.: Internet of things technology for greenhouse monitoring and management system based on wireless sensor network. ARPN J. Eng. Appl. Sci. 11(22) (2016) 12. Shirsath, D.O., Kamble, P., Mane, R., Kolap, A., More, R.S.: IOT based smart greenhouse automation using Arduino. Int. J. Innov. Res. Comput. Sci. Technol. 5(March), 234–238 (2017) 13. Arif, K.I., Abbas, H.F.: Design and implementation a smart greenhouse. Int. J. Comput. Sci. Mob. Comput. 4(August), 335–347 (2015) 14. EL Ghazi, A., Ahiod, B.: Random waypoint impact on bio-inspired routing protocols in WSN. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT) (2016) 15. Braham, R., Douma, F., Nahali, A.: Medical body area networks: mobility and channel modeling. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT) (2016) 16. Kasmi, M., Bahloul, F., Tkitek, H.: Smart home based on internet of things and cloud computing. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT) (2016) 17. Pasias, V., Karras, D.A., Papademetriou, R.C.: On novel efficient wireless access network design heuristic algorithms for QoS multiservice networks. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT) (2016)

A Fuzzy Queue Scheduling Controller to Enhance QoS for Terminal Communication Jamila Bhar(&) EµE Laboratory, Faculty of Science of Monastir, University of Monastir, Monastir, Tunisia [email protected]

Abstract. Communications of smart devices become widely utilized. When accessing to the network, the node communication should satisfy some quality of service (QoS) requirements. The queue scheduling can extremely affect the QoS support. In order to improve the QoS in network transmission, an innovative fuzzy queue scheduling controller (FQSC) is proposed in this work. This FQSC model is based on fuzzy logic theories and queue scheduling technologies. FQSC is proposed to reduce the transmission delay and packet loss. It adopts an improved generic fuzzy principle to make the buffer length at a stable level by varying a packets number of a queue transmission session and automatically adjusting priority factors of a queue member. Simulation results demonstrate that our approach minimizes considerably the queuing time of data packets in buffer and improves significantly QoS parameters. Results prove also that the proposal improves the network adaptability and stability compared with classic scheduling techniques. Keywords: Fuzzy logic


Scheduling
Smart management

1 Introduction The vision of future networking is that all connecting devices will be smart. Consequently, communication system needs to connect different kinds and a high amount of devices to make smart home, smart organization or smart city [1]. Different sensors and systems need to be linked together. Therefore, exchanged data have to be gathered in queuing buffer through routers. Smart network need great demands on traffic control methods to maintain efficient management of huge number of nodes and massive amounts of data. Particularly, scheduling data is an interesting field for various researchers since there are many open issues including resource sharing, traffic management and QoS efficiency. Scheduling tasks are applied for terminal communications to control and manage network resources based on class of traffic condition and priority of application parameters. A survey of the existing scheduling techniques best suitable for an Internet of Things (IoT) based system is introduced in paper [2]. Another review is presented in [3] where authors classify scheduling proposals regarding main focus and tools. In literature, a massive number of algorithms are designed to improve QoS parameters by enhancing scheduling tasks. Authors in [4] propose a scheduling scheme based on © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 290–301, 2020. https://doi.org/10.1007/978-3-030-21009-0_28

A Fuzzy Queue Scheduling Controller to Enhance QoS

291

Round Robin (RR) Algorithm called MORRA (Modified Round Robin Opportunistic Algorithm). They show that their solution is a well-balanced scheme that outperforms both RR and MaxSNR schedulers. Paper [5] proposes a scheduling system for medical control in a VSN environment. Authors make a virtual system using data classification based on Quality of Service. A huge number of researches demonstrated that scheduling techniques perform better with fuzzy theory. Authors in [6] demonstrate that a fuzzy logic based packet scheduling algorithm presents an advantageous solution to overcome some problems such as complex mathematical calculations and overhead rise. In this context, paper [7] proposes a multicriteria and QoS-based algorithm using the fuzzy logic to make the decision on the appropriate gateway. N. Torjemen and all [8] propose an offload schema utilizing a fuzzy logic system as key metric for offload decision. The work in [9] exploits a Fuzzy based Bayesian Network (FBN) system to resolve the problem of uncertainly QoS caused by the variable performance of Web Service (WS) in the dynamic cloud environment. A fuzzy scheduling proposal is presented by P.E. Mendez-Monroy and all [10]. The idea concerns a scheduler scheme for a networked control system that is physically distributed with a shared communication system. Paper [11] studies a new fuzzy rule associated to scheduling method in order to be used for medical applications. The solutions should guarantee the rigorous QoS demands for body sensor networks optimization. Simulation results presented in reviewed works cited above show good QoS improvement. They prove that Fuzzy logic based technique is one of the most promising approaches that can be coupled with scheduling system. So, achieving greater performance of intelligent systems management can be accomplished by hybrid solution scheme which uses intelligent control to improve conventional control. The presented work proposes a useful and reliable model to efficiently schedule data in a smart network. For this purpose, a robust fuzzy queue scheduling approach is designed, developed and tested. The objectives of the proposed model are to solve problems derived by scheduling techniques and to obtain a generic model that can be utilized for any application domain. The new solution of FQSC combines queuing and fuzzy theories. In experimental phase, VHDL programming language is used. This choice focuses on giving an IP model suitable for any application with various conditions. To illustrate the usefulness of the presented model, we organize this paper as follows: Sect. 2 illustrates the challenges and technological ideas for smart network. Section 3 reviews related works discussing the problem aspects. Details of FQSC model specifications are described in Sect. 4. Section 5 provides the final problem architecture and explores its buffer management strategies implementing the Fuzzy controller. It also describes simulation environment, illustrates simulation results and discusses evaluated parameters. Conclusions and future work are expressed in Sect. 6.

292

J. Bhar

2 Challenges and Technologies Intentions for Smart Networks The smart network exposes multipurpose challenges like scalability, energy saving, intelligent control, communication efficiency, generic model, and standards. To grow to be smart, existing technologies must enhance their functionalities and offered services. Wireless Sensors Networks (WSNs) are considered top potential new technologies that significantly alter the world and the way we live and work. WSNs present most suitable tools for controlling and monitoring smart network. The architecture of WSN is organized of a wide number of low-power, low-cost, small, electronically programmable, and multifunctional sensors. Sensors communicate wirelessly over short distances. They usually operate with restricted energy source and limited memory capacity. The lifetime on each node depends on the efficient use of the energy. Energy consumption can significantly affected by data transmission scenarios. An accurate way of data transfer is then required. In fact, efficient resource utilization has a great effect on improving network performances. Mac protocols are responsible of analyzing network traffic. Well designed algorithms in a MAC layer improve immensely the performance of traffic condition. A considerable amount of research in the area of smart property of application pointed on scheduling techniques. Scheduling disciplines enable the share of a common resource among multiple traffic categories. Data transmission constraints are solved based on the number of priority, class or a weight given to each data flow. These criteria depend on a variety of services co-existing in the considered network [12]. Generally, packets data wait in memory and then it is transmitted. Time of memorization is necessary to serving other buffers. Waiting in buffer can introduce performances degradation and in some cases a loose of data placed at the end of the queue. Furthermore, this problem degrades data rate, augments end-toend delay, introduces jitter, and can cause drop of data in case of queue saturation. In consequence, improving the usage of existing bandwidth and satisfying the level of performances required by application necessitate the resolution of delay problems coming from different traffic nodes. Delay problems are principally due to propagation, processing, and queuing time. Particularly, the queuing holdup has a great reflects on delay problem. So, different scheduling disciplines must attempt to attain equilibrium considering complex constraints. Fuzzy theory shows and tries to surmount the restrictions of wide used scheduling methods and optimizes principally buffer occupancy and decreases the waiting data time. Fuzzy Evaluation Methodology becomes an important technique utilized by researchers for various applications especially those needed to be integrated in a smart environment in order to improve QoS parameters. Fuzzy technologies are integrated to get better data scheduling by defining new way to permit efficient bandwidth distribution, to serve queues that contain packets with dynamic priority level and to minimize data queuing time. The following section presents a brief overview of the existing literature on smart networks and the integration of fuzzy logic on these fields especially in scheduling purpose.

A Fuzzy Queue Scheduling Controller to Enhance QoS

293

3 Related Work It is complicated to make an intelligent network for managing terminals data coming from versatile sources efficiently. A complexity rises when managing a large amount of data and for multi services application. Hence, a sophisticated mechanism of data scheduling is required. Especially, Fuzzy theories play a significant role in intelligent queuing technique. In literature review a comprehensive survey of the Fuzzy field is provided by Lotfi Zadeh [13]. It provides a basic review that is relevant to the essential concepts of the fuzzy theory. Multiple researches works study choices of fuzzy system parameters. Sufficient knowledge about using rules and membership functions is required and can have an effect on the performance of Fuzzy results. Two inputs generally deployed to the Fuzzy Logic Controller: the system transfer function (Input: I), and the error (DI). Author in [14] has modified the control rules of two inputs FLC to three inputs FLC. Simulation results show that the three input properties excel two input FLC properties. Most of deployed Fuzzy controllers consider a triangular membership function. Authors in [15] have showed that the triangular and trapezoidal MFs give analogous result values. For particular behavior, the triangular MF proves superior performance in reaction. The choice of the membership functions number considered to every input can vary from a Fuzzy model to another. In [16–18] authors use 3, 5, 7 or a combination of them. The choice is relative to the system characteristics and preliminaries experimental tests. This makes a design of fuzzy relative to the case study domain. Many case studies in the literature have take advantage of fuzzy theories for modeling their system in spite of classical mathematical methods. These studies are related to various domains. Authors in [19] intend to extract fuzzy opinions allied to sentiment and behavior. They demonstrate that considering Fuzzy Support Vector Machine in classification process outperform basic Support Vector Machine on accuracy and precision. Article presented in [20] gives attention to research works that take advantage of multi-agent system, prediction method, artificial neural network and fuzzy logic. The conclusion proved the success of the combination of different methods to improve system control and implementation. Specially, the number of applications in communication networks based on fuzzy logic is remarkably rising. The rise of fuzzy theory use is motivated by the improved results obtained when resolving communication networks problems by replacing or coupling traditional methods with fuzzy scheme. Examples of the fuzzy applications include power control in cellular systems, congestion control in IP networks, routing and Quality of Service management in wireless sensor networks [21–24]. Author in [25] demonstrates the usefulness of the Fuzzy logic algorithm to monitor the link quality by using a dynamic topology of the network. Several systems controls related to Fuzzy Logic have been successfully deployed in others network subjects with complexes and dynamics problems. Works related to queuing complexity, give in general a high priority for nodes considering the application constraints. In consequence, these nodes have the shortest queues as they are firstly served. Or, this strategy can degrading QoS of application classed with low priority. Diverse interesting ideas of fuzzy queues have been discussed by different

294

J. Bhar

researchers. Ramesh and al [26] applied a Fuzzy method to a batch arrival queuing scheme. Levels of measures of fuzzy models help making optimal decisions for evaluating system performance. Proposed ranking solution in this paper provides practical information for system manager and practitioners. Authors in [27] use fuzzy queuing models employing triangular fuzzy shapes. They demonstrate that queues applying Fuzzy logic provide better results than the discrete queues usually employed. Also, joining scheduling and fuzzy logic together enlarge their practicability. Paper [28] presents an algorithm to improve system performance and resource utilization in congested networks. The simulation results show the superiority of TCP using Fuzzy principles over a normal TCP. Authors in [29] proposed a traffic management for controlling the congestion. They suggest that routers are associated with smart controllers to manage buffers and transfer packets for wired/wireless networks. The diversity of domain exploiting the Fuzzy logic technique has lead to a multiple configurations of the fuzzy system. Our focus in this work is to design a Fuzzy logic controller independently of the domain of application. We consider in its design a dynamic configuration of membership functions number to overcome the variety of level of complexity from one application to another. However, we test our fuzzy proposal on queue scheduling problem. In fact, our approach proposes an intelligent queue selector that schedule data from buffers in efficient way. It uses fuzzy logic theory and WRR scheduling technique in order to provide competent management capabilities for optimizing traffic conditions and attempting highest QoS level.

4 FQSC Approach Overview 4.1

A Fuzzy Scheduling Proposition

Smart systems usually use sensors and communication system. Then, they need to integrate access via a router. Routers require being intelligent, autonomic and rapid to manage data in a fair way. A router usually includes a queues block to save data until insuring its transmission toward final destination. The queues block architecture is composed of a memory subsystem that includes packet buffer memory and a memory management unit called buffer queue manager. The packet buffer memory stores data in the system, while the queue manager is a special purpose mechanism that executes queue management instructions, which enable fast data transfer and queue supervision. Queue buffer can be served considering various scheduling techniques. We define as packet buffer the memory space required to store one packet in the buffer. The queue manager handles packets as units when performing queue operations. Packet buffer scheduling depends on service quality of transferred data. A periodic monitoring of traffic condition and queue state adapts traffic flow to network parameters and provides quality of service (QoS). A fuzzy control solution offers bandwidth optimization and good organization. The greatest characteristic of a fuzzy control system is that it is based on logical and mathematical method that take on continuous values between 0 and 1 rather than classical or digital logic. The use of fuzzy logic makes it possible to create control algorithm with simple and common expressions of If-then type, or linguistic

A Fuzzy Queue Scheduling Controller to Enhance QoS

295

rules. Figure 1 shows the proposed system model that consists of source nodes, FQSC and scheduler. The proposed algorithm is running on the router. This last includes buffers and management block. Various traffic models of services are considered to create different priority services to be sent by source nodes and managed by the router node. The queue scheduling procedure should satisfy the various qualities of service (QoS) requirements. In the flowing section, we explain details of fuzzy controller component.

Fig. 1. Data controller structure

4.2

FQSC Details

Fuzzy logic controller is basically composed by three phases. The fuzzification phase applied to inputs. Inputs data are collected from real world. They are, after that, converted to fuzzy arithmetic to obtain a crisp value. The obtained input is switched to a linguistic term based on its value and then the membership degree of each linguistic variable is determined. A Fuzzy logic Membership functions (MFs) present the significance of the input and output terms used in fuzzy set theory. Accordingly, the forms of MFs are significant for certain problems since they affect on a fuzzy inference system. Precisely, Rules are used where fuzzy inputs are compared and based on the membership functions of each input. The inference phase makes a decision about what should be the input to the system when evaluating which rules are relevant at the current time. The basic fuzzy set operations needed for evaluation decision are min and max rules method. In defuzzification phase, results of fuzzy set operations are aggregated and the advantageous crisp output is computed. The final fuzzy outputs must be converted to a single control output. To develop a mathematical model, we present the assumptions and notations, rules, and steps. We have set vocabularies from (Very Low, Low, Middling Low, Middle, Middling High, High, Very High) to describe the values of variable. A vocabularies number Vn can consider three, five or seven descriptions for input variable depending on a level of fluctuation accorded to the application. The set of rules forms a base. Rules number is equal to a first input vocabulary number multiplied by the second

296

J. Bhar

vocabulary number Vn(In1)*Vn(In2). Every intersection of rules must deliver a partial conclusion. After that, unions provide the overall conclusion. As well, weights obtained by applying the min-max method represent the area of the desired output variable. The maximum law is applied thereafter to define the last weights based on the rule base and according to MFs of the output. The resulting weight values P(i) provide the final output factor value “Zm”. This factor is used to regulate a signal track output. Details of different steps integrated in the proposed FQSC model are depicted in Fig. 2. Giving input value, l(Input) is obtained from l_calcul component. This component is characterized by generic parameters permitting to resolve the equation l(input) = a*Input + b. where a and b are calculated based on limits sub-slots characteristics. Sub-slots boundaries are automatically founded knowing considered input MFs number and the input position on MFs sub-slots. The final output parameter value Zm is calculated with centroid method of defuzzification phase where the Zi arguments indicate the values of the linguistic variables for the output.

Fig. 2. FSM of FQSC model

The queue system in this paper is a fuzzy update model located in every node. The FQSC is responsible of analyzing the QoS requirements, the dynamics of data input to buffers in router, delays and jitter of incoming packets. Taking on account a variety of parameters makes programmed algorithms generic and suitable to different case study.

5 Implementation and Results As it has been already mentioned, the control based fuzzy logic method needs some mathematical and logic concepts. So, a number of functions are defined to develop the Fuzzy algorithm. In order to make a fuzzy control flexible for a state of application parameters, these functions are created with generic parameters. This characteristic permits an elastic command for complicated parameter variation of an application.

A Fuzzy Queue Scheduling Controller to Enhance QoS

297

Developed FQSC uses a set of algorithms running together. The normalization set is also placed behind fuzzy execution phase. Input variables manipulated with Fuzzy controller must perform the three principal steps and needs to execute a predefined mathematical function such as division. Finally, Fuzzy steps module is interactively related with the system command to exchange input and output variables. Developed FQSC uses a set of algorithms running together. The fuzzy logic block components are depicted by a diagram shown in Fig. 3.

Equa on Resolu on

Division

(a,b)

μ(IN)

Q=a/b

Normaliza on & Slot number es ma on

(IN1, IN2, Slot_Nbr)

(IN, Slot_Nbr, Slot_Ord)

Fuzzy Steps Z

IN1 Output

System

Track signal calcula on: Output=Z*IN+IN

Fig. 3. Fuzzy controller steps diagram

To evaluate the developed fuzzy control algorithm, we consider diverse scenarios using different traffic conditions (periodic and sporadic data arrivals). The desired packet number to be delivered from a FIFO is moderated according to the obtained fuzzy output factor. Figures presented in this section show simulations results of the developed FQSC. They demonstrate that the developed mechanism guarantees an automatic, generic and efficient control strategy that minimizes packet drop by tracking every queue state. Figures 4 and 5 show, respectively, the simulation results of the buffer occupancy and the packets drop. Figure 5 shows that a FQSC decrease a percentage of packets drop by 1%. The simulation exhibited a good difference between scheduling using FQSC and scheduling without FQSC. As we can see in the Figs. 4 and 5, when the traffic is very high, there are a little decrease of buffer occupancy with the use of the proposed fuzzy algorithm, however, it was significantly optimized the packets drop compared to WRR. From the same figures, it can also be seen that the evaluated parameters in a low and medium traffic load was well optimized. These results imply that buffers are better served by using a fuzzy controller and then data nodes participate in the traffic considering efficient balance. Figure 6 shows the simulation results of the buffered data delay. It can be observed that the delay of data memorization was considerably minimized using fuzzy logic

J. Bhar 100 80

buffer

60 40

WRR

20 Fuzzy_WRR 1 123 245 367 489 611 733 855 977 1099 1221 1343 1465 1587 1709 1831 1953 2075 2197 2319 2441

0

clock cycles

packets drop

Fig. 4. Buffer occupancy analysis

6

WRR_drop

5

Fuzzy_wrr_drop

4 3 2 1 0 526 654 892 1034 1154 1323 1394 1550 1654 1754 1906 2010 2178 2318 2412 2542

0 clock cycles

Fig. 5. Packet drop analysis

150 Buffering delay (clock cycles)

298

WRR FQSC

100 50 0 1

2

3

4

FIFO N°

Fig. 6. Data waiting delay in FIFOs

A Fuzzy Queue Scheduling Controller to Enhance QoS

299

rules. Results show that the FQSC strategy minimizes the average waiting time in buffer compared with the WRR scheduling technique while considering the same traffic scenarios. Simulation results show the efficiency for the FQSC compared to classical scheduler. We then can conclude that this algorithm not only ensures a good balance in buffer occupancy, but also high QoS improvement as well as.

6 Conclusions The main goal of the FQSC system is reducing loss, low queuing delay and queue balance. To achieve sophisticated queue control functions we have presented in this paper a useful model to efficiently schedule data in a smart network. The FQSC provides important improvement in QoS parameters. The advantage of the explored approach is that the instruction set must be compact and general enough to provide all the required functionality. The proposed work proves a remarkable optimization in network resources management and QoS parameters performances and maintaining best approach for prioritized users. The aim to develop a FQSC prototype model is reached. Our future work, will consider multiple programming environments and various queuing techniques. We will focus on giving conclusions about the choice of suitable programming language and appropriate scheduling methods.

References 1. Kasmi, M., Bahloul, F., Tkitek, H.: Smart home based on Internet of Things and cloud computing. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, Tunisia, 18–20 Dec 2016 2. Muhammad, Z., Saxena, N.: Survey on scheduling mechanisms for wireless sensors in IoT scenarios. In: Proceedings of 102nd IASTEM International Conference, Seoul, South Korea, 18–19 Jan 2018 3. Asadi, A., Mancuso, V.: A survey on opportunistic scheduling in wireless communications. IEEE Surv. Tutor. Commun. 15(4), 1671–1688 (2013) 4. Hamouda, H., Kabaou, M.O., Bouhlel, M.S.: An efficient subcarrier scheduling algorithm for downlink OFDMA-based wireless broadband networks. In: ICWITS 2017: 19th International Conference on Wireless Information Technology and Systems, Lisbon, Portugal, 16–17 Apr 2017 5. Islam, M.M., Huh, E.-N.: A novel data classification and scheduling scheme in the virtualization networks. Int. J. Distrib. Sens. Netw. 25 July 2013. ISSN: 1550-1477 6. Jain, V., Agarwal, S., Goswami, K.: Priority based Fuzzy Decision Packet Scheduling Algorithm for QOS in Wireless Sensor Netork. Int. J. Comput. Appl. (0975 – 8887) 97(3, July) (2014) 7. Zhioua, G., Tabbane, N., Labiod, H., Tabbane, S.: A fuzzy multi-metric QoS-balancing gateway selection algorithm in a clustered VANET to LTE advanced hybrid cellular network. IEEE Trans. Veh. Technol. 64(2), 804 (2015)

300

J. Bhar

8. Torjemen, N., Zhioua, G.e.m., Tabbane, N.: QoE model based on fuzzy logic system for offload decision in HetNets environment. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, Tunisia, 18–20 Dec 2016 9. Chandrasekaran, S., Srinivasan, V.B., Parthiban, L.: Fuzzy based QoS prediction using bayesian network in cloud computing environment. Int. J. Eng. Technol. 7(1.5), 170–175 (2018) 10. Mendez-Monroy, P.E., Sanchez Dominguez, I., Bassam, A., May Tzuc, O.: Controlscheduling codesign for NCS based fuzzy systems. Int. J. Comput. Commun. Control. 13(2), 251–267 (2018). ISSN 1841-9836 11. Otal, B., Alonso, L., Verikoukis, C.: Novel QoS scheduling and energy-saving MAC protocol for body sensor networks optimization. In: BodyNets’08 Proceeding of the ICST 3rd International Conference on Body Area Networks, Temp, Aresona, 13–17 Mar 2008 12. Ridha, O., Jamila, B., Kholdoun, T.: A new scheduling protocol design based on deficit weighted round robin for QoS support in IP networks. J. Circuits, Syst., Comput. 22(3), 21 p. (2013) 13. Zadeh, L.A.: Fuzzy logic—a personal perspective. Fuzzy Sets Syst. 281, 4–20 (2015). ScienceDirect. www.sciencedirect.com 14. Fuyin, D., Weifeng, D.: Design of a three-input fuzzy logic controller and the method of its rules reduction. In: Proceedings of the 2009 International Symposium on Information Processing (ISIP’09), pp. 51–53, Huangshan, P. R. China, 21–23 Aug 2009 15. Gayathri Monicka, J., Sekhar, N.O.G.: Performance evaluation of membership functions on fuzzy logic controlled AC voltage controller for speed control of induction motor drive. Int. J. Comput. Appl. (0975 – 8887) 13(5, January) (2011) 16. Baghli, F.Z., El Bakkali, L., Lakhal, Y.: Multi-input multi-output fuzzy logic controller for complex system: application on two-links manipulator. In: 8th International Conference Interdisciplinary in Engineering, INTER-ENG 2014, Tirgu Mures, Romania, 9–10 Oct 2014 17. Sailan, K., Kuhnert, K.D., Karelia, H.: Modeling, design and implement of steering fuzzy PID control system for DORIS robot. Int. J. Comput. Commun. Eng. 3(1, January) (2014) 18. Omar, A.S., Waweru, M., Rimiru, R.: A Literature survey: fuzzy logic and qualitative performance evaluation of supply chain management. Int. J. Eng. Sci. (IJES) 4(5), 56–63 (2015) 19. Toujani, R., Akaichi, J.: Fuzzy sentiment classification in social network Facebook’ statuses mining. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, Tunisia, 18–20 Dec 2016 20. Reaz, M.B.Ib.: Artificial intelligence techniques for advanced smart home implementation ACTA technical corviniensis. Bulletin of Engineering, ©copyright FACULTY of ENGINEERING HUNEDOARA, ROMANIA (2013) 21. Dzitac, I., Filip, F.G., Manolescu, M.J.: Fuzzy logic is not fuzzy: world-renowned computer scientist Lotfi A. Zadeh. Int. J. Comput. Commun. Control. 12(6), 748–789 (2017). ISSN 1841-9836 22. El Alami, H., Najid, A.: Energy-efficient fuzzy logic cluster head selection in wireless sensor networks. In: Information Technology for Organizations Development (IT4OD), International Conference on Date of Conference: 30 March–1 April 2016. IEEE Xplore (2016). Electronic ISBN: 978-1-4673-7689-1 23. Wang, J., Niu, J., Wang, K., Liu, W.: An energy efficient fuzzy cluster head selection algorithm for WSNs. International Workshop on Advanced Image Technology, IWAIT (2018), 978-1-5386-2615-3 ©2018IEEE

A Fuzzy Queue Scheduling Controller to Enhance QoS

301

24. Quyuan, W., Songtao, G., Jianji, H., Yuanyuan, Y.: Spectral partitioning and fuzzy C-means based clustering algorithm for big data wireless sensor networks. EURASIP J. Wirel. Commun. Netw. 2018, 54 (2018). https://doi.org/10.1186/s13638-018-1067-8 25. Mercilin, R., Raja, K., Indumathi, P.: Fuzzy based faulty link isolation technique in dynamic wireless sensor networks. WSEAS Trans. Comput. 14 (2015). E-ISSN: 2224-2872 26. Ramesh, R., Kumara Ghuru, S.: Cost measures of fuzzy batch arrival queuing model by ranking function method. Int. J. Sci. Res. 4(2277–8179), 234–238 (2015) 27. Sujatha, N., Murthy Akella, V.S.N., Deekshitulu, G.V.S.R.: Analysis of multiple server fuzzy queueing model using a – CUTS. Int. J. Mech. Eng. Technol. (IJMET) 8(10), 35–41 (2017) 28. Gupta, R., Sharma, O.P.: Analysis of QoS for DSR protocol in mobile ad-hoc network using fuzzy scheduler. Int. J. Adv. Res. Electr., Electron. Instrum. Eng. 3(4, April) (2014) 29. Shajahan, B.: A fuzzy based congestion control in distributed wireless network. Int. J. Emerg. Technol. Comput. Sci. Electron. (IJETCSE) 13(2, March) (2015). ISSN: 09761353

The Performance of RFID Tags in Close Proximity to Human Body K. Khoder1(&), K. Kaja1,2, A. Choumane3, S. Boksmati2, and H. Amoud2 1

Department of Innovation and Technology, Azm Institute, 1300 Tripoli, Lebanon [email protected] 2 Faculty of Sciences, Lebanese University, 1300 Tripoli, Lebanon 3 Faculty of Public Health, Lebanese University, Zahle, Lebanon

Abstract. In this paper, the performance of an RFID tag in presence of human body is studied. A commercial RFID tag is designed and simulated using CST Microwave Studio electromagnetic solver. Simulation results show good reflection coefficient and gain values for the RFID tag in free space. A human body model is then introduced and the performance of the tag as a function of the separation distance from body (Z) is investigated. It is found that the reflection coefficient presents considerable alterations at different separation distances. Results show that it is possible to predict a threshold distance Zth above which the RFID tag becomes detectable by the reader antenna. Keywords: CST


RFID tag antenna
Passive tag
UHF
Human model

1 Introduction to Radio Frequency Identification The Radio Frequency Identification (RFID) technology provides the capabilities of wireless identification and tracking of targets. Its use spans over a large spectrum of applications including shopping security, inventory management and tracking of assets to name a few [1–6]. However, the performance of an RFID tag could be significantly worsened in the presence of metallic/liquid objects that degrade the radiation efficiency and gain of the tag antenna [7, 8]. The main components of an RFID system consist of a tag, a reader and a reader antenna. The RFID tag itself could be either passive or active and is mainly formed by a small antenna and a silicon chip. The design and shape of the RFID tag vary largely depending on the frequency range in which the tag will be used, in addition to various other factors. The operation of an RFID identification system starts at the level of the reader. The latter sends a command to the reader antenna to emit an electromagnetic wave. The RFID tag within the radiation range of the antenna collects this wave. A small antenna on the tag itself converts the energy of the received electromagnetic wave into power to trigger the chip. In return, the chip resends the electromagnetic signal back to the reader antenna that allows the reader to identify the tag. Figure 1 shows the different elements of an RFID setup including the reader, the reader antenna and the RFID tag. © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 302–311, 2020. https://doi.org/10.1007/978-3-030-21009-0_29

The Performance of RFID Tags in Close Proximity to Human Body

303

Fig. 1. Elements of an RFID setup.

The frequency bands’ allocations for the operation of RFID is country dependent. In Europe RFID UHF bands are: 866–869 MHz, 902–928 MHz in North and South America, and 950–956 MHz in Japan and some Asian countries [9].

2 The RFID Tag Design and Simulation There are three main types of RFID tags: passive, semi-passive and active. The passive tags are battery-less. They convert the RF waves received by the tag’s antenna to power the chip and backscatter the signal to the reader antenna. In most of the cases, passive tags are used because of their low-cost and zero maintenance need. Passive tags are only detectable when they are interrogated by a reader antenna. If the reader antenna does not send a signal looking for the tag, passive tags cannot interact with the antenna autonomously. In contrast, semi-passive tags make use of a battery source to power the chip, while data transmission is ensured by the power conversion of the RF waves sent by the reader antenna. The active tags, however, use battery source for both the chip powering and data transmission. Therefore, active tags can send signals on their own that could be detected by any RF antenna even if the antenna does not send a probing signal. Active tags usually have higher range and possibly higher gains, but they require maintenance and have commonly higher costs compared to passive tags. Many types of antennas could be used in the design of an RFID tag [10], the most used one is the dipole antenna where the chip is placed in the excitation point of the antenna. The structure of an RFID tag consists of an antenna connected to a chip (application specific integrated circuit). The impedance of the chip, Zc, is purely capacitive and it can be modeled either by a parallel or series equivalent circuit (Fig. 2) with Rp, Cp, Rs and Cs defined by [11]: Rp ¼ Cp ¼

Im ðZc Þ2 þ Re ðZc Þ2 Re ðZc Þ Im ðZc Þ

2 p f ðIm ðZc Þ2 þ Re ðZc Þ2 Þ Rs ¼ Re ðZc Þ

ð1Þ ð2Þ ð3Þ

304

K. Khoder et al.

Fig. 2. The electrical model equivalent to the impedance of the chip, (a) Series circuit and (b) Parallel circuit.

Cs ¼

1 2 p f  Im ðZc Þ

ð4Þ

The RFID tag (Higgs-4) used in this work is a commercially available tag from Alien Technology [12]. The manufacturer datasheet provides an equivalent electric circuit model for this chip as a parallel RC circuit that has the following values: Rp = 1500 Ω and Cp = 0.85 pF. The impedance value and the manufacturer’s model parameters of this chip are shown in Table 1. Table 1. Higgs-3 input impedance and impedance models. Higgs-4 Impedance Zc Rs, Cs Rp, Cp

866 MHz 30.5 – j211 X 30.5 X, 0.867 pF 1500 X, 0.85 pF

Our work primarily focuses on the simulation of the Higgs-4 commercial RFID tag’s response using the CST Microwave Studio electromagnetic solver. For this end, the design of the RFID tag has been implemented in the CST software using the dimensions and parameters provided by the manufacturer. The Higgs-4 tag design has a central operation frequency of 866 MHz in free space. The results of the simulated tag in CST have to be accurately in accordance with the specifications of the commercial tag. Therefore, an improvement of the simulation of the RFID tag in CST is essential to ensure the correct simulation results of the tag’s responses. In the following, we have used the series model of the tag’s chip using the Eqs. (1– 4) in the implementation of the Higgs-4 simulated model in CST. A simple connection of the tag antenna to the chip would result in a mismatch since the chip does not have a 50 X input impedance, contrary to the antenna. Therefore, the design of the antenna is adjusted to incorporate a matching circuit element. The power reflection coefficient defined by (5) has to be considered: C ¼

Zc  Za Zc þ Za

where, Za is the input impedance of the antenna.

ð5Þ

The Performance of RFID Tags in Close Proximity to Human Body

305

The power transmission coefficient between the tag antenna and the chip is [11]: s ¼ 1  jC j2 ¼

4 Ra Rc jZc þ Za j2

; 0\s\1

ð6Þ

where Ra and Rc are the real parts of Za and Zc respectively. Therefore, to improve the power transfer to the chip, the antenna input impedance Za has to be the complex conjugate of the chip impedance Zc. Simply changing the dimensions of a dipole antenna does not lead to an impedance of Za = 30.5 + j211.8 Ω. This is overcome by adding an inductive element in series, which compensates for chip’s capacitance. Several techniques are used to add the inductive element: T-match, inductively coupled loop or nested slot. The most used method is the T-match where the input impedance of the antenna can be changed by introducing a centered short-circuit stub. The antenna source is connected to a second small dipole placed at a close distance from the first and larger dipole (Fig. 3).

Fig. 3. RFID tag geometry.

The simulated impedances of the RFID tag antenna and chip are shown in Fig. 4. The obtained impedance (30.5 + j215 X) at the center frequency (866 MHz) indicates that the matching between the RFID chip and the antenna is achieved. It is important to emphasize that the matching results shown in Fig. 4 and 5 represent the correctness of our simulated model in CST of the Higgs-4 commercial tag. We have not introduced any modifications or changes into the dimensions or parameters of the commercial tag as provided by the manufacturer.

Fig. 4. Impedance of the tag’s antenna and chip (simulation). (a) Real, (b) Imaginary part.

306

K. Khoder et al.

The reflection coefficient of the tag antenna versus frequency is illustrated in Fig. 5. As it can be seen, the reflection coefficient is about −50 dB at the desired frequency 866 MHz.

Fig. 5. Simulated reflection coefficient versus frequency.

Figure 6 shows the simulated 3D radiation pattern of the tag. A maximum gain, at the central frequency of 866 MHz, is found about 1.8 dB.

Fig. 6. 3D radiation pattern for the designed structure at resonant frequency, 866 MHz.

3 Effect of the Body on the Tag Performance RFID tag antenna is particularly sensitive to the environment [13], and the environment of a good location can bring a huge impact on the tag’s antenna. Practical experiments and available literature [14] show that the presence of an RFID tag at the close proximity to the human body results in an alteration of its performance. The tag antenna impedance matching, radiation efficiency, and directivity change accordingly

The Performance of RFID Tags in Close Proximity to Human Body

307

[14]. In general, the closer the tag gets to the human body the lower are the possibilities to detect it by the reader [15]. This creates a major difficulty and hinders the functionality of RFID tags on or near human bodies. In this research, we aim at studying the effects of the human body on the performance of an RFID tag in close proximity and predict a defined set of experimental conditions for the best possible functionality of the RFID tag at the closest possible distance to a human body. The human body is modeled in CST by a stratified threelayer dielectric system representing the skin and fat, muscles and bones as shown in Fig. 7 [16].

Fig. 7. Human body model.

The values of dielectric constants and conductivities of each layer used in the model are given in Table 2 [16]. These values were extracted from CST materials library Bio for tissue, bone skin and muscle. Table 2. Physical parameters of the dielectric layers used for modeling the human body in the CST simulations. Layer Skin + fat Muscle Bone

er 14.5 55.1 20.8

r½S=m 0.25 0.93 0.33

Human body is modeled by a rectangular geometry, which highly reduces the calculation time without practically affecting the results. The CST model of the human body geometry and the RFID tag design is shown in Fig. 8.

Fig. 8. The CST model of the rectangular human body geometry and the RFID tag design.

308

K. Khoder et al.

In order to study the operation of the tag, the S11 parameters are simulated within the frequency range (800–1000 MHz) that encompasses the bandwidth of the reader antenna used in our setup (global UHF band 860–960 MHz). To investigate the proximity effect of the human body on the tag’s performance, S11 parameters were simulated by placing the tag at different separation distances Z from the human body model, as shown in Fig. 8. The curves in Fig. 9 (a) show the S11 parameters at small distances Z ranging from 0 to 5 mm; larger separation distances (Z  20 mm) are shown in Fig. 9 (b). The S11 (dB) response in free space, i.e. in the absence of the human body model, is plotted in blue and labeled as free on both graphs in Fig. 8. It serves as a reference curve to which the RFID tag’s response as a function of the separation distance from the human body model will be compared to. The vertical dashed lines on Fig. 8 indicate the ideal bandwidth limits of the reader antenna between 860 MHz and 960 MHz.

Fig. 9. Reflection coefficient simulation of tag antenna for near body and in free space (a) and for far-body and in free space (b).

In the presence of a human body model, the S11 responses of the RFID tag at different separation distances from the body present considerable alterations compared to the one calculated at the free space position. At small distances between the RFID tag and the human body (0 mm  Z  5 mm), the S11 curves shown in Fig. 9 (a) present singular minimum peaks around 940 MHz closer to the upper limit of the reader antenna’s bandwidth. For larger separation distances (Z  20 mm), the S11 curves in Fig. 9 (b) present two minimum peaks, similar to the free space curve. In fact, the human body affects antennas in two ways: (i) dielectric loading increases the electrical length of the antenna and (ii) the absorption of EM waves by the human body lowers the radiation efficiency. Dielectric loading detunes the operating frequency and lowers the impedance while the absorption decreases the gain. 3.1

Frequency Shifts: Detuning of the Tag Antenna

To better illustrate the variation of the RFID tag’s response depending on the tag –body separation, the shift in frequency is determined between the S11 minimum peak position

The Performance of RFID Tags in Close Proximity to Human Body

309

for each separation distance (Z) and the S11 minimum peak for the free space tag’s position (866 MHz) used a reference. The frequency shift (Δf) is shown in the upper plot of Fig. 10.

Fig. 10. (Upper part): Df versus separation distance Z, (Lower part): minimum reflection coefficient for different tag – body separation distances.

For 0.5 mm  Z  10 mm, Δf is positive and varies between 50 MHz and 110 MHz, i.e. the S11 curves shift to the right with respect to the reference curve at the free space position. Up to Z = 4 mm, the S11 peaks are within the bandwidth of the reader antenna close to its upper limit of 940 MHz, as indicated by the grayed rectangular area in the upper part of Fig. 10. For larger distances up to Z = 10 mm, the frequency shifts remain positive and show that the S11 peaks are very close (within 10– 20 MHz) to the upper limit of the ideal reader antenna’s bandwidth. For Z  20 mm, Δf is negative and is only about 10 MHz. This shows that the decrease in the separation distance between the tag and the human body below 20 mm results in the detuning of the operating frequency of the tag above the free space position value. 3.2

Variation of the Antenna Impedance

The shift in the resonance frequency of the antenna is directly related to the variation of its impedance value at different separation distances. The impedance of the tag antenna in free space has been calculated and optimized for the best matching with the impedance of the chip at 866 MHz. Figure 11 shows the shifts in the reactance (imaginary part) and resistance (real part) of the antenna impedance with respect to the free space position value at different Z distances.

310

K. Khoder et al.

Fig. 11. Shift in reactance (a), and resistance (b) of the antenna impedance with respect to the free space position value at different Z distances.

The simulated results of the impedance shifts correlate with the frequency shifts of the tag antenna. For separation distance smaller than 20 mm, the reactance decreases sharply indicating a capacitive coupling with the human body at close proximity. In fact, a decrease in the imaginary part of the tag antenna compared to its value optimized at the free space position points towards the reduction of the inductance effects initially introduced in the matching procedure. This results in a mismatching between the tag antenna and the chip. Hence the degraded operation of the RFID tags in the close proximity to the human body. For larger separation distances, the curves in Fig. 10 show a practically zero impedance shift. The tag regains its matching values and free space resonant behavior. Threshold functional distance of separation: The horizontal red line drawn on the lower part of Fig. 10 corresponds to S11 = −7 dB. It sets the limit for the acceptable values of the S11 peaks calculated at different tag – body separation distances. It thus becomes possible to determine from Fig. 9 a threshold distance Zth below which the S11 values lie in a region where the RFID tag would not be detectable by the reader antenna. From the simulation data, Zth is found at about 3 mm tag – body separation distance. Primary experimental tests, incorporating the RFID tags close to an actual human body on the level of the upper arm show a very good agreement with the calculation results predicted by our simulations. Air spacing with the human body of about 3 mm seems to provide an acceptable experimental measurement where the RFID tags are detected by the reader antenna within a range of roughly 3–4 meters.

4 Conclusion In this study, the effect of the human body on the RFID tag has been studied. A commercial RFID tag from Alien technology has been successfully designed and simulated using CST. The antenna-chip Matching conditions are studied and respected resulting in a simulated reflection coefficient about −52 dB, and a simulated gain of 1.88 dB at the resonant frequency 866 MHz. The human body is modeled in CST and

The Performance of RFID Tags in Close Proximity to Human Body

311

the results shows considerable alterations of the frequency in function of a separation distance Z between the RFID tag and the Modeled human body. Simulation results shows that a spacing of 3 mm with the human body provide an acceptable experimental measurement where the RFID tags are detected by the reader antenna within an acceptable range.

References 1. Finkenzeller, K.: RFID Handbook: Fundamentals and Applications in Contactless Smart Cards and Identification, 3rd edn. Wiley, New York (2010) 2. Domdouzis, K., Kumar, B., Anumba, C.: Radio-Frequency Identification (RFID) applications: a brief introduction, 2006 Elsevier Ltd. Adv. Eng. Inform. 21, 350–355 (2007) 3. Hui Tan, H.: Application research of RFID in supply chain logistics management. In: IEEE International Conference on Service Operations and Logistics, and Informatics, vol. 2, pp. 2456–2459 (2008) 4. Rao, K.V.S., Nikitin, P.V., Lam, S.F.: Antenna design for UHFRFID tags: a review and a practical application. IEEE Trans. Antennas Propag. 53(12), 3870–3876 (2005) 5. Ahsan, K., Shah, H., Kingston, P.: RFID applications: an introductory and exploratory study. IJCSI Int. J. Comput. Sci. Issues 7 (1, 3) (2010) 6. Jebali, N., Beldi, S., Gharsallah, A.: RFID antennas implanted for pervasive healthcare applications. In: 7th Conference on Science of Electronics, Technologies of Information and Telecommunication (SETIT) (2016) 7. Son, H.W., Choi, G.Y., Pyo, C.S.: Design of wideband RFID tag antenna for metallic surfaces. Electron. Lett. 42(5), 263–265 (2006) 8. Keskilammi, M., Sydanheimo, L., Kivikoski, M.: Radio frequency technology for automated manufacturing and logistics control. Part 1: Passive RFID systems and the effects of antenna parameters on operational distance. Int. J. Adv. Manuf. Technol. 21(10–11), 769–774 (2003) 9. Rao, K.V.S., Nikitin, P.V., Lam, S.F.: Antenna design for UHF RFID tags: a review and a practical application. IEEE Trans. Antennas Propag. 53(12), 3870–3876 (2005) 10. Fairley, M.: RFID Smart Labels – A ‘How to’ Guide to Manufacturing and Performance for the Label Converter, 2nd edn. Labels and Labeling (2007) 11. Ghiotto, A., Vuong, T.P., Wu, K.: Novel design strategy for passive UHF tags. In: 14th International Symposium on Antenna Technology and Applied Electromagnetics (ANTEM) (2010) 12. https://www.alientechnology.com/products/ic/higgs-3/ 13. Raumonen, P., Sydanheimo, L., Ukkonen, L., Keskilammi, M., Kivikoski, M.: Folded dipole antenna near metal plate. In: IEEE Antennas and Propagation Society International Symposium, vol. 1, pp. 848–851 (2003) 14. Oyeka, D.O., Batchelor, J.C., Ziai, A.M.: Effect of skin dielectric properties on the read range of epidermal ultra-high frequency radio-frequency identificatuion tags. Healthc. Technol. Lett. 4(2), 78–91 (2017) 15. Kellomaki, T.: On-body performance of a wearable single-layer RFID tag. IEEE Antennas Propag. Lett. 11, 73–76 (2012) 16. Gemio, J., Parron, J., Soler, J.: Human body effects on implantable antennas for ISM bands applications: models comparison and propagation losses study. Prog. Electromagn. Res. 110, 437–452 (2010)

Impact of Phone and Hand Position on SAR Distribution Using Liquid-Based PIFA Antenna Dina Serhal1(&), Najat Nasser2, Mohamed Rammal3, and Patrick Vaudon2 1

3

ECE Department, Rafik Hariri University, Meshref, Lebanon [email protected] 2 XLIM Laboratory, Limoges University, Limoges, France University Institute of Technology, Lebanese University, Saida, Lebanon

Abstract. This paper presents a new design of PIFA antenna used for mobile phones with a very low specific absorption rate (SAR). Our proposed antenna structure is based on reshaping the ground plane by adding metallic walls filled with an absorbing liquid in order to limit the distribution of the surface current existing on the antenna. This will lead to reducing the electromagnetic radiation toward the human head and therefore the SAR, to attain less than 0.7 W/kg averaged over 10 g in presence of the human head. The effect of the tilting positions of the phone with respect to the human head and the rotation angle of the user’s hand on the SAR is also investigated. Simulation results show that the SAR can be reduced by more than 85% when the mobile phone is tilted upwards by 20° with respect to the horizon. Moreover, the presence of the hand reduces the SAR by 35%. However, the tilt position of the human hand slightly affects the SAR distribution on the human head. Keywords: SAR
Surface current Electromagnetic radiation




PIFA antenna




Absorbing liquid




1 Introduction The use of mobile phones has grown exponentially in the last decade, and now it becomes an essential part of life. Currently, the number of mobile phone users in 2017 is estimated to be about 7.8 billion globally, and expected to reach up to 8.9 billion in 2023 [1]. This in turn, will lead to an unprecedented increase in traffic on the mobile networks, which requires the deployment of additional radio towers. As a result, people will be exposed to massive amounts of electromagnetic radiation. Many concerns have been focused on the interaction between mobile phone and the human head, as the latter is usually in the reactive near field of mobile phone [2]. Mobile phone antenna is estimated to transmit a power between 1 and 2 Watts in the frequency range of the GSM 900 and 1800 bands. An important part of the electromagnetic wave radiation emitted by the mobile antenna is transmitted, reflected, or absorbed by the human head tissue as human body behaves as lossy dielectric material [3]. © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 312–320, 2020. https://doi.org/10.1007/978-3-030-21009-0_30

Impact of Phone and Hand Position on SAR

313

Human exposure to electromagnetic waves can be measured using the Specific Absorption Rate (SAR) metric. This parameter is defined as:   r 2 r J 2 J2 E ¼ SAR ¼ ¼ 2q 2q r 2qr

ð1Þ

Where E represents the electric field (V/m), r (S/m) and q (Kg/m3) are the electric conductivity the mass density of tissue, respectively, and J is the conduction current density (A/m2). The FCC limit for public exposure from cellular telephones is an SAR level of 1.6 W/Kg that is in fact for 6 min per day operation. The limit safety for this SAR is about 3 to 4 min; therefore the utilization of mobile phone must not exceed 18 to 24 min per day [4]. Many researches were conducted to reduce the SAR value. Most of these doings were based on the antenna structure’s modification. Authors in [5] designed a novel triangular metamaterial consisting of two concentric triangular rings of conducting material, positioned between the human head model and PIFA antenna, a reduction of SAR about 50.852% for 1 g tissue mass was observed. Moreover, SAR was reduced by 12.261% at 900 MHz, and about 12% at 1800 MHz for 10 g, by adding RF shield of thickness 1 mm to the dual band slotted PIFA antenna with human head, as presented in [6]. Authors in [7] suggested the use of aluminum sheet, which performs as a conductive armoring material, where SAR averaged over 10 g of tissue mass was reduced up to 25.5%. In [8], the electromagnetic radiation inside the human head model was reduced because of the insertion of an EBG structure which is able to reduce the surface waves and limit the radiation from the ground plane. SAR averaging over 10 g was thus reduced by 38% at 1.8 GHz. In [9], helical antenna at 900, 1800 and 1900 MHz operational frequencies in the nearness of human head was simulated for several values of the respective dielectric properties of human head; authors showed that the conductivity is related by direct variation to SAR, while the permittivity and the density of human head are related by inverse variation to SAR. The antenna designs with low SAR proposed in previous works have relatively complex structures. In this paper, we are proposing a simple structure of PIFA antenna using extended metallic walls filled with absorbing liquid to further limit the propagation of EM radiations toward human head tissues and therefore to further reduce the SAR. On the other hand, the use of liquid material in reducing the SAR was not investigated in literature. Authors in [10] included water in their design in the purpose of designing a tunable or reconfigurable broadband antenna. Liquid can be an excellent candidate for reducing the SAR in antenna design because of its high permittivity, conformability, and reconfigurability. This paper is organized as follows: In Sect. 2, the structure of the PIFA antenna considered in our simulations along with the surface current distribution is presented. In Sect. 3, a new model of antenna consisting of modifying the ground plane in order to limit the EM wave propagation toward the human head tissue is presented. SAR reduction is achieved by extending the ground plane of the original antenna with a U-

314

D. Serhal et al.

shaped wall filled with an attenuating liquid material. Sections 4 and 5 present a study of the best tilt positions of the mobile phone and the human hand that lead to the lowest SAR, respectively. The numerical simulation of 10 g SAR values and other antenna performance parameters were evaluated using the finite-difference time-domain (FDTD) method of CST MWS (Computer Simulation Technology Microwave Studio). The power emitted by the mobile phone is set to be 1 W.

2 Antenna Structure and Design 2.1

Initial PIFA Antenna

Our simulations are based on a simple Planar Inverted-F Antenna (PIFA) operating at 1.8 GHz (see Fig. 1). This type of antenna is widely used in mobile applications due to its attractive properties, like small volume, ease of integration and manufacturing, to name a few [11].

Upper Plate

Ground Plane ShorƟng Plate Feeding Probe

(a)

(b)

Fig. 1. (a) Standard PIFA antenna structure. (b) PIFA antenna near human head phantom.

2.2

Proposed Antenna Structure

Most of antenna structures that were proposed in the literature to reduce the harmful effect of electromagnetic wave propagation along the human head are either complex, or affecting the antenna radiation pattern. Our proposed structure is based on reducing the SAR without affecting the antenna performance. This is done by simply reshaping the ground plane of PIFA antenna shown in Fig. 1 so that surface currents that are responsible of increasing the SAR in the human head are reduced dramatically. The method consists of adding U-shape edge-groove structure at each corner under the ground plane, as depicted in Fig. 2 (a). The dimensions of the walls are optimized in order to get the best mobile antenna performance in terms of impedance matching and radiation pattern. Good results were obtained with a width of k/15, a height of k/11, and a thickness of 0.5 mm, where k is the wavelength at 1.8 GHz.

Impact of Phone and Hand Position on SAR IniƟal PIFA antenna

315

Ground Plane Ground Plane

H= /11 W= /15

(b) W

Absorbing liquid material

H Extended walls filled with an absorbing liquid material (a)

(c)

Fig. 2. (a) Perspective view of the proposed PIFA antenna structure. (b) Side view and (c) perspective view of the extended walls filled with absorbing liquid.

The current density along the human head, which has a direct impact on the SAR, can be further reduced by filling the U- edges with an absorbing liquid material (er = 74, r = 11S/m) at each corner of the ground plane (see Fig. 2 (b) and (c)) [12].

3 Surface Current Limitation Using Liquid and SAR Reduction 3.1

Surface Current Distribution

Figures 3 and 4 show a comparison of the distribution of the current along the simulated head model with and without the liquid-filled walls. Simulation Results indicate that the surface current distribution is further limited in case of our proposed antenna. This means that the propagation of the electromagnetic wave into the head will be limited and therefore the SAR will be reduced, as will be discussed thereafter.

Fig. 3. Surface current distribution on the standard PIFA antenna.

3.2

SAR Reduction

The presence of the absorbing liquid material at the edges of the antenna is responsible for limiting the propagation of the electromagnetic waves into the human head. Therefore, the current density on the antenna will be reduced, which leads to reducing

316

D. Serhal et al.

Fig. 4. Surface current distribution on our proposed PIFA antenna.

the SAR along the human head. Figure 5(a) and (b) show the SAR averaged over 10 g mass tissue using a standard PIFA and our proposed antenna model, respectively. Simulation results show an SAR reduction from 3.4 W/Kg to 0.7 W/Kg, which corresponds to a reduction by 80%.

Fig. 5. Distribution of the SAR on the head phantom (a) using a standard PIFA, (b) using our proposed model.

4 Effect of Phone Tilt Position on SAR In this section, we will investigate the effect of mobile phone tilt position on the SAR. Tilt angles of 10 ; 20 ; and 30 are considered with respect to a reference angle as described in Fig. 6. Figure 7 presents a comparison of the 10 g SAR results obtained for the different tilt positions of the liquid-filled edges PIFA antenna. Results show that the SAR value is reduced to less than 0.01 W/Kg when the mobile antenna is at a tilt position of 20 from the reference angle presented in Fig. 6.

Impact of Phone and Hand Position on SAR

317

Direcon of Rotaon

Reference Angle (0 deg)

Fig. 6. Mobile phone tilt positions setup.

0.8

0.7

Max SAR (W/Kg)

0.7 0.6 0.5 0.4 0.3 0.2

0.1

0.1

< 0.01

< 0.01

20

30

0 0

10

Liquid PIFA Tilt PosiƟon (deg)

Fig. 7. Effect of mobile phone tilt positions on SAR.

5 Antenna Performance and SAR Modification in Presence of Human Hand Figure 8(a) shows the simulation setup indicating the relative position of the head phantom, the hand phantom, and the mobile handset described in Sect. 2. Different tilt positions of the hand will be considered to evaluate the SAR: 10 ; 20 ; 30 and 40 with respect to a reference position as described in Fig. 8(b). 5.1

Antenna Performance in Presence of Hand

The performance in terms of farfield radiation of the liquid-filled PIFA antenna in presence of the human hand is presented in Fig. 9. The hand is considered at the reference angle (see Fig. 8(b)). As can be noticed, the hand is reducing the directivity by 0.7 dB, and the main lobe is tilted downwards by 110 .

318

D. Serhal et al. Head

DirecƟon of RotaƟon of the Hand

Hand Reference Angle (0 deg)

Liquid PIFA Antenna

(a)

(b)

Fig. 8. Simulation setup in presence of the human hand. (a) Perspective view. (b) Tilt positions.

Fig. 9. 3D Farfield radiation of the liquid-filled PIFA antenna, (a) in absence of the human hand and (b) in presence of the human hand. (c) Polar plot of the radiation pattern of the antenna with and without the hand.

5.2

Effect of Hand Tilt Position on SAR

Figure 10 and Table 1 present a comparison of the 10 g SAR results obtained for the different tilt positions of the human hand phantom. Simulation results show that the SAR is reduced at least by 34% in the presence of human hand. This can be explained by the fact that the hand is absorbing a part of the power of the radiated electromagnetic fields. It can be noticed also that the tilt position of the hand is slightly affecting the SAR, in opposition with the effect of mobile phone tilt positions (Sect. 4). 5.3

Best Setup and SAR Distribution

It can be concluded from the previous results that the setup that gives the best result in terms of SAR distribution occurs when both the mobile and the hand are tilted by with respect to the reference angle considered in our simulations. As shown in Fig. 11, an

Impact of Phone and Hand Position on SAR 0.8

0.7

0.7

0.7

0.7

319

0.7

Max SAR (W/Kg)

0.7 0.6 0.46

0.5

0.46

0.45

0.435

0.43

0.4 0.3 0.2 0.1 0 0

10

20

30

40

Hand Tilt PosiƟon (deg) Max SAR with Hand

Max SAR without Hand

Fig. 10. Max SAR values for different hand tilt positions. Table 1. Comparison results in terms of SAR with and without hand. Hand rotation angle (deg) 0 10 20 30 40

Max SAR with hand (W/Kg) 0.46 0.45 0.44 0.435 0.43

Max SAR without hand (W/Kg) 0.7 0.7 0.7 0.7 0.7

Reduction % 34% 36% 38% 39% 39%

SAR < 0.01 W/Kg is obtained with a reduction by 99% with respect to the initial setup: reference angle of mobile phone is zero, in absence with the human hand.

20 deg.

(a)

(b)

(c)

Fig. 11. (a) Initial SAR distribution in absence of hand, (b) the final setup, and (c) the final SAR distribution in presence of hand tilted by 20°.

320

D. Serhal et al.

6 Conclusion In this paper, a new simple model of PIFA antenna with very low SAR is presented. The antenna structure consists of extending the ground plane downwards with metallic walls filled with an absorbing liquid material. Moreover, the effect of tilting positions of the antenna were considered. An important decrease of the SAR is observed when the antenna is tilted by 20 with respect to the human head horizontal axis. The effect of the human hand on the SAR distribution and antenna performance is also investigated. Different tilt positions of the hand were considered. Simulation results show that the SAR is reduced by at least 34% in presence of the hand, with a slight variation of the SAR with respect to the hand tilt positions. Finally, the results in terms of SAR of the best scenario is presented.

References 1. Ericsson Homepage. https://www.ericsson.com/en/mobility-report. Last accessed 30 July 2018 2. Samsuri, N.A.: The Effect of jewellery and the human hand on SAR and antenna performance. In: Loughborough University Institutional Repository (2009) 3. Tsiaras, A.: SAR evaluation in multi-antenna mobile handsets. Master Thesis, Department of Electrical and Information Technology Faculty of Engineering, LTH, Lund University, Sweden (2014) 4. Kumar, G.: Report on Cell Tower Radiation. LAP LAMBERT Academic Publishing, Bombay (2016) 5. Faruque, M.R.I., Islam, M.T.: Novel triangular metamaterial design for electromagnetic absorption reduction in human head. Prog. Electromagn. Res. 141, 463–478. Electromagnetics Academy (2013) 6. Anita Jones Mary, T., Ravichandran, C.S.: SAR reduction in slotted PIFA for mobile handsets using RF shield. ARPN J. Eng. Appl. Sci. 7(11), 1501–1505 (2012) 7. Hanafi, N.H.M., Islam, M.T., Misran, N., Faruque, M.R.I.: Numerical analysis of aluminium sheet for SAR reduction. In: Proceeding of the 2011 IEEE International Conference on Space Science and Communication (IconSpace), Penang, Malaysia (2011) 8. Sultan, K.S., Abdullah, H.H., Abdallah, E.A., Hashish, E.A.: Low SAR, compact and multiband antenna for mobile and wireless communication. In: The 2nd Middle East Conference on Antennas and Propagation, IEEE, Egypt (2013) 9. Husni, N.A., Faruque, M.R.I., Islam, M.T., Misran, N.: Effects of substrate material and dielectric properties on electromagnetic energy absorption over GSM bands. In: 2012 International Conference on Statistics in Science, Business and Engineering (ICSSBE), Langkawi, pp. 1–4 (2012) 10. Xing, L., Huang, Y., Shen, Y., Alja’afreh, S., Xu, Q., Alrawashdeh, R.: Further investigation on water antennas. IET Microwaves Antennas Propag. 9(8), 735–741 (2015) 11. Hirasawa, K., Haneishi, M.: Analysis, Design, and Measurement of Small and Low-Profile Antennas, 1st edn. Artech House Publishers, USA (1991) 12. Nasser, N., Serhal, D., Barake, R., et al.: A novel low SAR water-based mobile handset antenna. Analog Integr. Circuits Signal Process. 96(2), 353–361 (2018)

Study and Design of Time Modulated Antenna Array with Low Sides Bands Levels Alaa Saleh(&), Mohamad El-Khatib, and Mohamad Chakaroun Faculty of Technology Saida, Lebanese University, Beirut, Lebanon [email protected], [email protected], [email protected]

Abstract. The four-dimensional (4D) antenna arrays are achieved by introducing the time as a fourth dimension into traditional antenna arrays. In this paper a novel approach is proposed for designing time modulated array. The methodology of 8 elements (4d) patch antenna array design is presented. The switch on/off time calculation for different elements is shown and a numerical example is provided. Keywords: Time modulated antenna


4d antenna
Switching time sequence

1 Introduction The progress in the telecommunications systems has made the requirements for antenna array parameters very tight. These requirements include the antenna bandwidth as well as antenna radiation pattern and the side lobes levels. Consequently, the need for accurate array and feeding network design as well as manufacturing precision is also increasing. These factors make the antenna cost higher. Therefore it’s necessary to modify the traditional design and implementation process by introducing the time as a new degree of freedom for the design of antenna arrays. The concept of Four-Dimensional (4D) antennas was initially proposed in 1959 [1]. So 4-D antennas consists in using an additional degree of freedom that is time in antenna array design. This technique has become known since 1960 as a candidate for reducing side lobe levels (SLL) in antenna arrays. 4-D antenna arrays take benefits of time weighting in order to control precisely the weighting of different array elements. This is the different from the traditional static weighting used in conventional phased arrays. In order to achieve the time weightings high-speed switches are used. These switches are connected to different array elements. Thus the wanted amplitude/phase of a given element is obtained by turning the switch connected to this element ON and OFF. This technique suppress the need for attenuators and phase shifters used for feeding of traditional phased array elements. One of the drawbacks of time domain switching is the spreading of the spectrum of the modulated signal. New copies of the modulated signals appears around the center frequency. These copies are called the side bands. The side bands may reduce the gain of the antenna and may cause unwanted signal distortion. © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 321–326, 2020. https://doi.org/10.1007/978-3-030-21009-0_31

322

A. Saleh et al.

The suppression of sidebands has recently become possible. [2, 3] show how the application of the differential evolution (DE) reduces the sidebands levels. The particle swarm optimization technique (PSO) is discussed also in [4] for the same purpose. Over the last decade, the concept of 4-D antenna arrays has once again been attracting the attention of antenna designers. New techniques proved the possibility of sideband level (SBL) reduction, such as binary phase center motion (BPCM) and variable aperture sizes (VAS) [5, 6]. These techniques take advantage of the different design parameters offered by different time schemes. The time modulation and the possibility of controlling the radiation through a large number of switching sequences make the 4-D antenna array a good candidate for radars and modern communication systems. Recent studies discuss the potential applications of the side bands. So side bands are no more considered as a shortcoming for the operation of 4-D antennas. They are even considered as an advantage. The Angle of arrival estimation and the beamforming [7, 8] are among the possible applications. For these applications the side band level is enhanced by switching between 0°/180° signals instead ON/OFF state.

2 Antenna Array Design Methodology The proposed method for designing 4d antennas consists in the following steps. First of all the antenna array elements number is fixed (8 elements) each element is a patch antenna operating at 5 GHz. The patch dimensions at the operating frequency are (width = 13.5 mm, Length = 18 mm) and the dielectric substrate is the Fr4 (er = 4.3). After designing the patch antenna, an active element simulation is performed in order to extract the active radiation pattern. The electromagnetic simulations are performed using CST Microwave Studio Software. The next step aim to find the correct weighting (amplitude and phase) for the different elements in order to meet requirements given later. These values are obtained through an optimization procedure in order to reduce side lobes levels. In the last step the on/off time of switches is calculated and sequences that should be applied to different elements are shown. 2.1

Active Element Radiation Pattern Applied to 8 Elements Patch Array

The active element radiation pattern consists in exciting on element of the array while the others are terminated with a matched load (50 Ohms). Figure 1 shows a single patch radiation pattern compared to the active patch radiation pattern. 2.2

Patch Antenna Array Results

In this section an array of 8 patch elements is shown, the radiation pattern is obtained by multiplying the active element pattern by the array factor.

Study and Design of Time Modulated Antenna Array

Active radiation pattern

323

Single Element Radiation pattern

Fig. 1. Comparison between single and active patch antenna radiation pattern

As the size of the array increases, the radiation pattern is governed by the array factor so the element radiation pattern has less influence and the active element approximation of the element radiation pattern is more straightforward. A full electromagnetic simulation in which all elements are excited with the correspondent amplitude and phase is performed. Figure 2 shows that the array radiation pattern based on active element method match the radiation pattern obtained through the full electromagnetic simulation.

Fig. 2. Elements theoretical radiation pattern vs array pattern based on active element method.

3 4d Antennas Theory The main idea behind the 4-d antennas is the control of the amplitude and phase for different by controlling the ON and OFF time for the RF signal. The array factor for a conventional antenna is formulated by: EðhÞ ¼

N X i¼0

ai:EiðhÞ : exp jði:b:d:SinðhÞ þ /i Þ

ð1Þ

324

A. Saleh et al.

Where ai, is the amplitude of the element number i and is the phase /i is the phase applied the element number i. The amplitude and phase for a given element are determined by the Fourier transform of the time sequence applied to the specified element. The periodic sequence is modeled by a time rectangle of width ton and starting time t0 as shown in Fig. 3.

Tp

Switch voltage command

ton

τ

Time

Fig. 3. Switch voltage command for array elements

The final signal for a given element of the array could be written as: V iðtÞ ¼ xðtÞ:sðtÞ ¼ xðtÞ:

n¼ þ1 X

rectton ðt  n  Tp Þ  dðt  sÞ ¼

n¼1

xðtÞ:rectton 

n¼ þ1 X

ð2Þ

dðt  n  Tp Þ  dðt  sÞ

n¼1

x (t) is a sine wave of frequency f0 which represents the carrier of the modulated signal. The time period of the sequence is Tp = 1/fp where fp is the repetition frequency of the sequence. fp is much lower than the operating frequency f0. The frequency spectrum of the periodic sequence is given by: Rðf Þ ¼ ton

m¼ þ1 X Sinðp:f :ton Þ : expðj:2:p:f sÞ : dðf  m:fp Þ p:f :ton m¼1

ð3Þ

When the periodic sequence is multiplied by the carrier, the resulting spectrum represented in Fig. 4 is given by: Rðf Þ ¼

m¼ þ1 X ton Sinðp:ðf  f0 Þ:ton Þ : : expðj:2:p:ðf f 0Þ:sÞ : dðf  m:fp  f0 Þ p:ðf  f0 Þton 2Tp m¼1

ð4Þ

For m = 0, the amplitude of the spectrum at f = f0 is: ton2 this expression shows that if all elements have the same ON time and starting time, the amplitude and phase of the

Study and Design of Time Modulated Antenna Array

325

Fig. 4. Frequency spectrum of periodic time sequence multiplied by a modulated signal

fundamental frequency produced by different elements would be the same m, therefore the main lobe direction will be at theta = 0 (broad side array). For f = f0 + m.fp the spectrum amplitude is: ton Sinðp:m:fp :ton Þ : : expðj:2:p:m:fp:sÞ p:m:fp :ton 2:Tp

ð5Þ

For the first side band frequency f = f0 + fp, the array factor has the following expression. s ton Sinðp:fp :ton Þ ton Sinðp: Tp Þ ¼ : expðj:2:p:fp:sÞ ¼ : expðj:2:p:TpÞ 2Tp p:fp :ton 2Tp p: tTonp

ton

Rðf ¼fpÞ

ð6Þ

By modifying the starting time of different elements with respect to each other’s, a phase increment could be obtained for side band frequencies, while the fundamental frequency radiation remain unchanged, this will steers the side band radiation away from the main lobe radiation. t startðiÞ ¼ i  s The new phase at the first side band frequency for the element (i) is given by: fp ton Sinðp:fp :ton Þ ton Sinðp:fp :ton Þ : expðj:2:p:fp :i:sÞ ¼ : expðj:2:p:k: :iÞ 2 2 p:fp :ton p:fp :ton f0

ð7Þ

In order to steer the radiation of the side band frequencies The array factor should be maximized in the wanted direction, that means bd:SinðhmaxÞ  2:p:fp:s ¼ 0 b:d:Sinðhmax Þ Sinðhmax Þ Sinðhmax Þ :Tp s¼ ¼ ¼ 2:p:fp 2:fp 2 Example for a hmax ðFirst side band frequencyÞ ¼ 100

The time delay that should be applied is s ¼¼ Sinðh2max Þ :Tp ¼ 0:086 Tp

ð8Þ

326

A. Saleh et al.

Normalized Time (t/Tp)

The time sequences that should be applied to the 8 elements array are shown in Fig. 5. 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

1

2

3

4

5

6

7

8

Element number

Fig. 5. Time sequence for different array elements

So for i = 0 (first element of the array) no phase shift is applied. For i = 1! 2:p:k: ffp0, if the time shift = 5.T0 (k = 5) and fp = f0/20, a phase shift of

2:p:5 20

¼ p=2 could be achieved at the first side band frequency.

References 1. Shanks, H.E., Bickmore, R.W.: Four-dimensional electromagnetic radiators. Can. J. Phys. 37 (3), 263–275 (1959) 2. Poli, L., Rocca, P., Manica, L., Massa, A.: Pattern synthesis in time modulated linear arrays through pulse shifting. IET Microwaves Antennas Propag. 4(9), 1157–1164 (2010) 3. Yang, S., Gan, Y.B., Tan, P.K.: Linear antenna arrays with bidirectional phase center motion. IEEE Trans. Antennas Propag. 53(5), 1829–1835 (2005) 4. Li, G., Yang, S., Nie, Z.: Direction of arrival estimation in time modulated linear arrays with unidirectional phase center motion. IEEE Trans. Antennas Propag. 58(4), 1105–1111 (2010) 5. Yang, S., Gan, Y.B., Qing, A.: Sideband suppression in time modulated linear arrays by the differential evolution algorithm. IEEE Antennas Wireless Propag. Lett. 1(1), 173–175 (2002) 6. Yang, S., Gan, Y.B., Tan, P.K.: Linear antenna arrays with bidirectional phase center motion. IEEE Trans. Antennas Propag. 53(5), 1829–1835 (2005) 7. Reyna, A., Balderas, L.I., Panduro, M.A.: Time-modulated antenna arrays for circularly polarized shaped beam patterns. IEEE Antennas Wirel. Propag. Lett. 16, 1537–1540 (2017) 8. He, C., et al.: Direction finding by time-modulated linear array. IEEE Trans. Antennas Propag. 66(7), 3642–3652 (2018)

A Tunable Microwave Bandpass Filter Mohamed Al Khatib, Alaa Saleh, Mohamad Chakaroun, and Mohamed Shehade(&) University Institute of Technology, Saïda, Lebanon [email protected],[email protected], [email protected], [email protected] Abstract. A convenient prototype of a reconfigurable microwave filter is presented in this paper. The transmission on defined frequencies are achieved by tunable combined lines. The measured performance of a narrow-band reconfigurable band-pass filter is realized using diodes, capacitors, and a voltage sweep of DC feed. Coupling networks in this combline filter enable tuning to be reached with less degradation possible in passband performance. Low passband insertion loss and performance are presented, to show how close they are to theoretical expectation, when the design is adjusted from 0.8 GHz to 1.2 GHz on ADS and Momentum EM Simulator. The frequency range is chosen around 1 GHz in the simulation to predict the filter response on low frequencies used in the next 5G technologies. Keywords: Combline


Filter
Microwave
Tunable
Band-pass

1 Introduction Tunable microwave filters applications had expanded to include most modern ESM receiving systems that have need of narrow size, highly selective and reconfigurable filters with low insertion loss [1]. Based on absolute principles, microwave band pass filters rely upon the electromagnetic coupling between resonators to provide the required impedance inverting circuit elements, while coupling is frequency dependent and can be only operated over narrow bandwidths [2]. This has convinced the progress and release of a wide range of filter prototypes [3] with a transmit/receive bands equivalent to the ones for mobile terminals [4] and base stations of mobile communication systems [5]. As an example, wideband coverage including reducing the entire hardware dimensions is obtained by using tunable filter banks. Multiple wireless function would be directed by a reconfigurable passband filter using common hardware to control and eliminate out-of-band interference, also decrease the global system complexity and potentially improving its response and functionality. Reconfigurable bandpass filters class consist of filters with variable bandwidth, frequency center, skirt selectivity, and group delay equalization [6]. The paper purpose is to introduce the design of reconfigurable filters that reshape the response between three frequency centers 0.8 GHz, 1 GHz and 1.2 GHz by sweeping the voltage of two © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 327–336, 2020. https://doi.org/10.1007/978-3-030-21009-0_32

328

M. A. Khatib et al.

DC generators from a positive state to a negative one. This functionality is achieved by designing an architecture of diodes and capacitors in an approach to cut off and join the three different combined lines (comblines) to cover the multiple required frequencies.

2 Theory of Tunable Combline Filters The combline filter consist of a compatible multi-wire transmission line with coupling forced to be only between adjacent lines. Each line is short circuited to the ground from the same end while the opposite ends are connected to lumped capacitors (Fig. 1). Combline filters are compact with a strong stop band designed to have an exceptionally high cutoff rate on the high side of the bandwidth. Generally such filters can be made without dielectric materials so that dielectric losses can be eliminated.

Fig. 1. The combline filter circuit. [7]

3 Filter Implementation The purpose here is to highlight some difficulties we encounter to implement a reconfigurable/tunable filter. Our main job was to eliminate the self from the Chebyshev prototype filter circuit (Fig. 2), transform it into a circuit composed only from capacitors and inverters (Fig. 3) and replacing those inverters by the combline filters come in a later step so that the needed filter architecture is achieved.

Fig. 2. Chebyshev low-pass filter prototype circuit.

Those are the values of a 3rd order Chebyshev low-pass filter with 1 ohm impedance: C11 = 0.85158F, L21 = 1.10316H, C31 = 0.85158 F

A Tunable Microwave Bandpass Filter

329

Fig. 3. Steps to follow to obtain bandpass filter without inductance components. [8]

The diagram below (Fig. 3) represent the steps followed to transform the Chebyshev low-pass filter to a bandpass filter starting with eliminating the inductance L21 from the 3rd order Chebyshev filter circuit and replacing it with a capacitor and two inverters because L = Ck2. In the next step, the two inverters with value equal to 56,9084Ω will be replaced by the architecture shown below (Fig. 4) composed of a chaine of stubs. The final circuit is obtained by dealing with two parallel stubs identical to two parallel resistances in order to obtain one resistance with a value: 1/R total = 1/R1 + 1/R2. Those steps are well presented in (Fig. 5) below.

Fig. 4. The architecture of stubs with the same values of the inverter replaced.

330

M. A. Khatib et al.

Fig. 5. Steps that show how parallel stubs are considered in the transformed circuit.

All steps showed above and the final circuit below is based on a case study of “Parallel Coupled-line Combline Filter” done by Michael Steer [8]. After going through the design the values of the components are shown below as well as the circuit (Fig. 6): • • • • • •

Cb = 1.22170 pF C1 = C3 = 1.64276 pF Z0t1 = Z0t3 = 67.7683 Ω Z0t12 = Z0t23 = 443.232Ω Ct2 = 2.70759 pF Z0t2 = 80Ω

Fig. 6. Final circuit before transforming it into combline filter. [8]

A Tunable Microwave Bandpass Filter

331

Figure 7 shows the filter stubs based circuit performance (Fig. 6) simulated on ADS offering an acceptable transmission loss equal to −0.005 dB and a reflection coefficient lower than −20 dB.

Fig. 7. Performance of the above circuit simulated on ADS.

Replacing an architecture of five stubs with a combline filter (Fig. 8) is an important step to make the circuit feasible, benefit from combline advantages and improve the S-parameter’s response. In a final step to reach the first combline filter, we have to tune the values of lumped capacitors, combined lines widths and lengths to bring a better performance than the stubs based filter circuit at the 1 GHz frequency. The performances represented below show how close results are between bandpass filter based on stubs and the one based on comblines coded by colors. Red graphs represent the transmission loss S(2, 1) and the reflection coefficient S(2, 2) of the stub based circuit (Fig. 6), while blue graphs represent the transmission loss S(6, 5) and the reflection coefficient S(6, 6) of the comblines base (Fig. 8). These graphs shows an acceptable transmission loss above −0.5 dB with S(2, 1) = −0.005 dB and S(6, 5) = −0.005 dB and an acceptable reflection coefficient below −20 dB for both circuits with S(2, 2) = −29.335 dB and S(6, 6) = −33.269 dB (Fig. 9).

332

M. A. Khatib et al.

Fig. 8. Combline filter equivalent to stubs circuits. [8]

Fig. 9. Results comparison between (Fig. 6) and (Fig. 8) simulated circuits on ADS.

4 Triple Band-Pass Filter Architecture The triple band-pass filter is a reconfigurable filter based on the use of three connected MACLIN3s that have the same following characteristics:

A Tunable Microwave Bandpass Filter

333

W1 = 0.84 mm W2 = 1.82 mm, W3 = 0.8 mm, S1 = 780 um and S2 = 780 um But are differentiated by their lengths values to control the covered frequencies: L1 = 10.715 mm, L2 = 3.6 mm and L3 = 6.115 mm To ensure coverage of the frequency range between 0.8 and 1.2 GHz, the combline lengths are the only parameters that are different between the MACLINs which allow us to move the bandwidth from a covered frequency to a new desired one. These three microwave lines will be configured through voltage alternation of two voltage generator, the supply line is surrounded from a side by a diode with electrical characteristics of type M1 connected to a ground, and from the other side, a capacitance with a value of C = 1000 pF connecting this architecture to the MACLIN3s (Figure 10).

Fig. 10. Filter architecture discussed above.

4.1

Configuration of the Filter Circuit

When a voltage generator gives a negative input voltage VDC = −3 V, the diode is blocked and the capacitance is considered as an open circuit so the micro strips remain connected and active. On the other hand, when the voltage input is positive VDC = 1 V, the diode is functional and the micro-strips are connected to the ground, as a result one or two lines will be deactivated. The response of the reconfigurable microstrip filter (Fig. 10) is shown as a final result (Fig. 11) below: 4.2

Results Interpretation

As you can see in the (Fig. 11) the first graph (with a square symbol □, S(2, 1) = −0.017 dB) is centered on 0.8 GHz and related to the first state where

334

M. A. Khatib et al.

Fig. 11. Filter performance according to each configuration of VDC1 & VDC2 voltage.

VDC1 = −3 V and VDC2 = −3 V, as a result the three MACLIN will be connected to reach the maximum length in this circuit with L0.8GHz = L1 + L2 + L3 = 20.43 mm. The second graph (with a round symbol ○, S (2, 1) = −0.141 dB) shows us that the covered frequency is 1 GHz, this state can be reached when VDC1 = −3 V and VDC2 = 1 V so just two MACLIN will be connected and as a result L1GHz = L1 + L2 = 14.35 mm. The third and fourth state presented by the third graph (with two triangle symbol D, S (2, 1) = −0.170 dB) and covering the 1.2 GHz frequency show two added graph with the same filter performance and that is caused when VDC1 = 1 V, no matter what the value of VDC2 is (VDC2 = 1 V or VDC2 = −3 V) only the first MACLIN3 will be activated so the minimum length in this circuit is L1.2GHz = L1 = 10.715 mm. 4.3

Advantages of the Triple Band-Pass Filter

As we can see from the interpretation of the results in the previous part, using such a filter will provide many advantages. The size of the micro strip filter that will replace three different filters to cover three different frequencies is reduced to one filter with a minimum length of almost 2 cm. Economizing the materials and components use,

A Tunable Microwave Bandpass Filter

335

where we can use different part of the architecture two or three times in order to change the filter performance, as well as the elimination of dielectric loss because of the ability of implementing a combline filter without dielectric materials. The figure below (Fig. 12) will provide more details about the circuit schematic completed on ADS showing the variables, values, types of components and presenting the architecture of voltage supply of the triple bandpass filter.

Fig. 12. Filter circuit of the result shown above simulated on ADS.

5 Conclusion The new architecture of the three comblines provide coverage for three different frequencies 0.8 GHz, 1 GHz and 1.2 GHz with less than 0.5 dB of transmission loss. The reconfiguration of the filter transmission varies from side to side to meet the advanced filtering requirements for multifunctional systems. The limitations and complexity of using such filter will be greater with the increase of number of comblines, and also the

336

M. A. Khatib et al.

distortion of the response and the performance will accrued and the transmission loss will increase. Operators have already begun to move to the next generation of microwave antennas that offer greater spectrum utilization, resilience against interference at a reduced costs. This imposes a growing demand on the existing cellular communication infrastructure and especially the microwave filters in order to improve the network performance and save the frequency spectrum.

References 1. Hofman, C.B., Baron, A.R.: Wideband ESM receiving systems- Part 1. Microwave J. 23(9) (1980) 2. Al-Ahmad, M., Matz, R., Russer, P.: 0.8 GHz to 2.4 GHz Tunable Ceramic Microwave Bandpass Filters, 2007 IEEE/MTT-S International Microwave Symposium (2007) 3. Cameron, R.J., Harish, A.R., Radcliffe, C.J.: Synthesis of advanced microwave filters without diagonal cross-couplings. In: IEEE MTT-S Int. Microwave Symp. Dig. 1437–1440 (2002) 4. Field, P.L., Hunter, L.C., Gardiner, J.G.: Asymmetric bandpass filter using a novel microstrip circuit. IEEE Microw. Guided wave lett. 2(6), 247–249 (1992) 5. Hershtig, R., Levy, R., Zaki, K.: Synthesis and design of cascaded trisection (CT) dielectric resonator filters. In: Proc. 27th Eur. Microwave Conf. 784–791 (1997) 6. Wael, M.: Fathelbab, Member IEEEE, Steer, Michael B., Fellow, IEEE A reconfigurable Bandpass Filter for RF/Microwave Multifunctional Systems. IEEE (2005) 7. Hunter, I.C., Rhodes, J.D.: Electronically tunable microwave bandpass filters. IEEE Microw. Theory Tech. MTT-30(9), 1354–1360 (1982) 8. Steer, M.: Microwave and RF design – second edition, 16.1–16.4 (2014)

Power Saving Approach in LTE Using Switching ON/OFF eNodeB and Power UP/DOWN of Neighbors Narjes Lassoued1,2(&) and Noureddine Boujnah3(&) 1

INNOV’COM Research Lab, Higher School of Communication of Tunis SUPCOM, University of Carthage, Carthage, Tunisia [email protected] 2 National Engineering School of Gabes, ENIG, Gabes, Tunisia 3 Faculty of Science of Gabes, FSG, Gabes, Tunisia [email protected]

Abstract. Evolution of wireless technologies during the last decade and the increase of wireless user’s number as well as the traffic amount have led to a significant increase of energy consumption and thus more emission of CO2. Nowadays, reducing energy consumption became a must for both environmental awareness and for the reduction of the operational expenditure of network operators (OPEX). In this paper, we propound a method that consists in lowering down energy consumption in fourth generation cellular network. The main idea is to switch ON/OFF LTE-Advanced Base Station (BS). In parallel, an UP/DOWN power algorithm is implemented to fill coverage holes in the networks and also to maintain an acceptable quality for mobile users. Simulation results proves that our proposed algorithm can reduce significantly the energy consumption with a guarantee of Quality of Service (QoS). Keywords: Switch ON/OFF
Energy consumption
Received power
LTE
Low traffic

1 Introduction Long Term Evolution (LTE) is the successor of 3G technology designed to offer new services and higher data rate. One of the most essential factors for improving performance of LTE is its architecture and integration of new radio and transport functionalities [1]. ENodeB (eNB) is the fourth generation base station (BS), responsible for the most EUTRAN radio operations. eNB requires power supplies, its energy consumption depends mainly on traffic that it supports. According to measurements and statistics published in [2], global mobile data traffic is increasing with an exponential rate between 2015 and 2020. As indicate in Fig. 1, mobile data traffic will grow by 53% from 2015 to 2020; it will reach approximately 30.6 Exabyte per month by 2020. The exponential growth of different fields of Information and Communication Technology (ICT) leads to an environmental and economic problems. On the environmental side, ICT research estimates that the use and the disposal of manufacture of © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 337–349, 2020. https://doi.org/10.1007/978-3-030-21009-0_33

338

N. Lassoued and N. Boujnah

Fig. 1. A projection of mobile data traffic growth: 2015–2020.

ICT equipment accounts for 2% of global CO2 emissions [3]. Furthermore, on the economic side, ICT is responsible for 2% - 10% of annual worldwide energy consumption [4]. Due to OPEX/CAPEX networks indicators, and the considerable CO2 emission caused by ICT sector; lowering down energy consumption in wireless networks becomes necessary. Studying energy consumption in wireless network, it is clear that the base station (BS) consume the biggest part of the energy that can attain 57% among all network [5]. Accordingly, energy saving for mobile network BS has attracted researchers as well as operators particularly switching ON/OFF BSs. Owing to the alternative attitude between day and night of traffic demand in mobile network, switching ON/OFF BSs policies are considered as one of the most active topics among the existing techniques of energy saving. In [6], According to the traffic variation, the authors switch ON/OFF dynamically the BSs by respecting certain conditions of blocking probability to ensure QoS. In [7, 8], the author is interested in UMTS mobile networks, especially during low traffic periods. In this work, the author suggested various Switching ON/OFF algorithms: in [7], BSs are switched ON/OFF randomly to minimize energy consumption in a UMTS cellular network. An improvement of this work is given in [8] where authors proposed a Switch ON/OFF algorithm using a uniform and a hierarchical scenario where the network is dynamically planned. Algorithms of Switch ON/OFF are also applied in LTE and LTE-A networks via many researches. Notably, in [9], Alexandra et al. try to come up with an optimum combination of switching ON/OFF eNBs in order to guarantee the maximum of energy saving. The same authors, in [10] proposed a Switch ON/OFF scheme based on an ascending ranking of the average distance between UEs and their associated eNBs. The algorithm continued when there is no degradation on QoS and stopped in the opposite case. The problem of energy misuse was also studied, and the dynamic BSs Switch ON/OFF algorithm is proposed in [11]. In [12], Narjes et al. implemented a Switch ON/OFF scheme in an LTE-A based HetNet Network. In this work, the switch ON/OFF procedure is performed based on the SINR values of UEs. The authors try to maintain QoS and coverage using a combination between femtocells and CoMP techniques.

Power Saving Approach in LTE

339

In the above-mentioned researches, the main issue is the power aware management in wireless access network. The essential purpose is to reduce power consumption by minimizing the number of active eNodeBs when possible. In this context our work of this paper is situated, our contribution in this work is to perform energy consumption reduction using switch ON/OFF of some eNBs and UP/DOWN power of neighbors without impact on network coverage. Contrary to other works using a uniform distribution, in this work, we use a non-uniform distribution of user equipment’s (UEs). For this reason, the decision to switch ON/OFF an eNB is based on the number of its attached users and the received power from its neighbor’s eNBs to compensate coverage loss. This combination can improve the reduction of energy consumption without scarifying the QoS. Therefore, the non-uniform distribution of UEs makes the representation of mobile network more realistic. The rest of the paper is organized as follows. In Sect. 2, we present the network topology. Section 3 contains a description of the power model. We describe the proposed switching ON/OFF algorithm in Sect. 4. Results of simulation and evaluation of the proposed algorithm are the subject of Sect. 5. The paper concludes with Sect. 6.

2 Network Architecture For the network architecture, we propose an LTE dense urban network, in which, neighboring cells overlaps. The considered topology is formed by N = 7 eNodeBs. As it is shown in the Fig. 2, each eNodeB may serve a single macrocell. These macro cells have the same coverage radius R.

Fig. 2. Network architecture.

We consider a randomly non-uniform traffic distribution of users betwixt cells, for which the number of UEs allocated to one cell varies between 0 and 100 UEs. The use of the non-uniform assignation of UEs makes our representation of traffic more realistic. Figure 3 represents the number of UEs in the entire network from 07:00 pm to 05:00 am.

340

N. Lassoued and N. Boujnah

Fig. 3. Variation of users Vs time.

3 Power Calculation Model We will present, in this section, the power calculation model used to calculate power saving during applying Switch ON/OFF algorithm At an time t, the total power Ptot ðtÞ in the hole network is defined by: Ptot ðtÞ ¼

N X

Pactiv ðiÞ þ Ptrans ðiÞSi ðtÞ þ DP

ð1Þ

i

Pactiv represents the eNB’s activation power, Ptrans represents is the power needed for transmission, Si ðtÞ represents the state of eNB i at time t, DP is the amount of the added power to neighbors eNBs after applying the switch on/off scheme, then N represents the full number of eNodeBs. The activation power Pactiv is calculated by applying the following formula: Pactiv ¼ Pa þ PS þ Pc

ð2Þ

Where; Pa represents the power needed for amplification include feeder, PS represents the power needed for alimentation and Pc represents the power needed for cooling. The power needed for transmission Ptrans is defined by: Ptrans ¼

Nu X j

PTX ðjÞ

ð3Þ

Power Saving Approach in LTE

341

Where; Nu represents the number of users allocated to eNB i and PTX ðjÞ represents the power forwarded by UE j given by: PTX ðjÞ ¼ NRB ðjÞ:PRB

ð4Þ

NRB ðjÞ represents the number of resource blocks (RBs) allocated to UE j, then PRB represents the power forwarded per RB defined by: PRB ¼

Ptrans tot NRB

ð5Þ

tot Where; Ptrans represents the total power needed for transmission and NRB represents the full number of RBs allocated to a single eNodeB. Si ðtÞ represents the state of eNB i:

 Si ðtÞ ¼

1 if ON 0 if OFF

ð6Þ

The received power PRX at UE’s side is computed refereeing to the path loss model described in [13] as: PRX ¼ PTX :G:Da

ð7Þ

Where; PTX is the eNB transmission power given by Eq. 4, G represents gain of the antenna multiplied by the gain of the channel, D represents the distance in km between the base station and UE and a represents the exponent of propagation chosen depending on propagation medium’s type.

4 Proposed Algorithm In this section, we present our proposed Switch ON/OFF algorithm in order to maximize the energy saving. When the number of UEs is small, the eNB is underutilized, resulting in a very significant waste of energy. For this reason, our objective is to Switch OFF the eNB when they are underutilized especially during low traffic periods. We based our algorithm on two assumptions: • First, an eNB i should be switched OFF when the number of its attached active user Ni is lower than a given threshold T. In the opposite case when Ni is greater than T, a wake up procedure is performed. • The second assumption is to calculate the received power Pði;jÞ for user j from the neighbors of the switched off eNB i, then we assign each attached user j to the best neighbor eNB according to the power ranking.

342

N. Lassoued and N. Boujnah

Several algorithms use the Switch ON/OFF procedure during low traffic periods but our contribution here is to use a power compensation procedure in order to guarantee QoS and maintain coverage. First, Switch ON/OFF is executed provided that the cell coverage mustn’t be affected and the QoS must be guaranteed. When we switch OFF the eNB, the coverage will be reduced and thus, some users will be out of cell’s range. Consequently, we must recompense the coverage area in order to provide network coverage to users after a switching OFF procedure. In this case, to achieve coverage we will increase the transmission power of neighbor eNBs that will serve the attached user of the switched OFF eNB. At first, we should compute the number of users Ni assigned to each eNB i then makes an ascending classification of this set of number. We denote Si ðtÞ the eNB i state at time t. The decision to allocate each UE to the correct cell is performed depending on Table 1.

Table 1. eNodeB switch ON/OFF state. Si ðt  1Þ ¼ 1 Si ðt  1Þ ¼ 0 0 Ni ðtÞ  T 0 Ni ðtÞ [ T 1 1

According to the value of Ni , our algorithm have two behaviors: deactivate or reactivate the eNB i. Table 1 shows the evolution of eNB state as function of previous state and number of active users at time t. We compare the selected Ni ðtÞ to a given threshold T and eNB is switched ON or OFF based on Si ðt  1Þ value. 4.1

Switch Off Procedure

An eNB i will be deactivated when the number of its users Ni is lower than T. The following diagram explains more the first behavior of our algorithm (Fig. 4): 4.2

Wake up Procedure

The wake up of an eNB is performed when at an instant t the state Si ðt  1Þ of an eNB i is OFF and the number of its attached UEs Ni ðtÞ is greater than the given threshold T; in this case the base station must be reactivated as explicated in Fig. 5. 4.3

Power Compensation Procedure

Switching ON/OFF eNB is followed by power UP/DOWN of one or more neighbors, the aim is to fill coverage holes then users QoS will not be affected. The following diagram gives details about the evolution state of an eNodeB in the context of our proposed method (Fig. 6).

Power Saving Approach in LTE

Fig. 4. Switch OFF procedure.

Fig. 5. Wake up procedure.

Fig. 6. State diagram of eNodeB.

343

344

N. Lassoued and N. Boujnah

Let Fi ¼ f1; ::; ng be the set of eNB i neighbors switched ON and Nu ðiÞ ¼ f1; ::; jg the set of users assigned to the eNB i. We compute for each user j assigned to eNB i, ðnÞ the received power Pði;jÞ from Fi . Each user j is assigned to the eNB neighbor’s n0 that have the maximum received power: ðnÞ

n0 ¼ arg max Pði;jÞ

ð8Þ

n2Fi

Power compensation process is started when the eNB is deactivated and one or many UEs has no best cell, that is to say that the received power is lower than power sensitivity Pmin : ðnÞ

Pði;nÞ ¼ min ðmax Pði;jÞ Þ\Pmin j

ð9Þ

n

If the clause in (9) is incorrect, we do not add any power. The rate of power that will be added to the total power depends principally on the distance between deactivated eNB and its neighbors: DPðiÞ ¼ Pmin  Pði;nÞ

ð10Þ

Finally, we compute the total power Ptot : Ptot ¼

Fi N X X ½Pactiv ðiÞ þ Ptrans ðiÞ:Si ðtÞ þ ½Pmin  Pði;nÞ ðtÞ

ð11Þ

n

i

To assess performance of the proposed approach, we compute the SINR value for UEs of the switched off eNB especially for users located in the cell extremity. The SINR is calculated respecting the following equation: SINRðjÞ ¼ PN

i6¼j

PRXðjÞ PRX ðiÞ þ KTP W

ð12Þ

P PRX ðjÞ represents the received power given in Eq. 7, Ni6¼j PRX ðiÞ represents the average power of interference signals. KTP W represents the background noise, yet W represents the bandwidth used by UE, K represents the constant of Boltzmann and TP represents the temperature.

5 Performance Evaluation 5.1

Simulation Scenario

To assess performances of our proposed scheme, we use an LTE topology using MATLAB as implementation software. As shown in Fig. 7, we propose an urban

Power Saving Approach in LTE

345

scenario where; the implemented topology is composed by 7 macro cells. The eNBsUEs distance is computed through their positions in the map, thus the distance D among eNodeBs is equal to 500 m.

Fig. 7. Network architecture.

For the allocation of UEs between cells, we suppose a variable distribution of UEs where users are generated randomly for each cell. The maximum number of UEs served by an eNB is fixed to 100 UEs. For the simulation, we consider the eNB’s maximum transmit power for the downlink transmission given in table bellow (Table 2): Table 2. Recommended transmit power [14]. Transmission power [dBm] Bandwidth [MHZ] 43 1.25, 5 46/49 10, 20

The parameters of simulation used in the simulation scenario are defined in Table 3. 5.2

Simulation Results

We will present in this section results of simulation and performance evaluation of the proposed scheme in term of power saving and Quality of Service. Beginning by Fig. 8, representing the behavior of energy consumption in the network over time for different Switch ON/OFF algorithms. In this figure, we compare our proposed algorithm by several cases: the first case represented by the plot in blue

346

N. Lassoued and N. Boujnah Table 3. Simulation parameters. Parameter Value Distribution of user Non-uniform Frequency [MHZ] 2140 Bandwidth [MHZ] 20 Distance among eNodeBs[m] 500 Transmitted power for macrocell [dBm] 35–46 RBs number in a macrocell 100

that describe the energy consumption without the use of any energy saving algorithm. The second curve in green represents the case of random Switch ON/OFF [6] where we chose to turn off 1/3 of the base stations randomly. Finally, the red plot represents the energy consumption after the use of our suggested algorithm.

Fig. 8. Power compensation Vs time.

Like that demonstrated in the figure above, our suggested approach can realize the maximum amount of power saving better than the two other algorithms: No switch on/off and the random algorithm. The efficiency of our algorithm is demonstrated especially in the night zone where the traffic is low. The percentage of power saving is the result of Fig. 9. As shown in this figure, the proposed method achieves the maximum amount of power saving than the random scheme. The percentage of power saving achieved by our proposed scheme can attain 30%.

Power Saving Approach in LTE

347

Fig. 9. Power saving (%) Vs time.

To evaluate efficiency and performances of our algorithm in term of coverage and QoS, we compute the SINR value before and after the switch ON/OFF process for users located in the cell extremity. Looking at the SINR values after the switch on/off procedure, we notice that the quality of service remains constant and has not much affected and this thanks to the use of power adjustment technique (Fig. 10).

Fig. 10. SINR values comparison.

348

N. Lassoued and N. Boujnah

The proposed approach minimize the use of power in LTE access network, moreover, it reduce coverage holes by increasing the power of switched OFF eNBs neighbors.

6 Conclusion The number of telecommunications elements and devices is growing exponentially. In addition, needs for services and new mobile applications lead to a significant growth in energy usage. In this paper, we focused on reduction of energy consumption in LTE. We proposed an algorithm that can reduce the energy consumption with switching OFF some eNBs under some constraints. An eNB is switched OFF when the number of its attached users does not exceed certain threshold. The attached UEs of the switched OFF eNB are served by the best neighbor according to their received power. Simulations results show that the proposed approach realize the maximum amount of power saving compared to other approaches and it can save more than 30% of power consumption especially during the night period. In our future work, we will focus on methods that can realize more energy efficiency and we will compare it to more other existing approaches.

References 1. http://www.3gpp.org/technologies/keywords-acronyms/97-lte-advanced, last visit on February 15 (2018) 2. Author, F.: Cisco visual networking index: global mobile data traffic forecast update, 2015– 2020, white paper 3. Global Action Plan, An inefficient truth. http://www.globalactionactionplan.org.uk, Global Action Plan Report, Dec. (2017) 4. http://green-broadband.blogspot.com/ 5. Han, C., et al.: Green radio: radio techniques to enable energy efficient wireless networks. IEEE Commun. Mag. 49(6), 46–54, June (2011) 6. Gong, J., Zhou, S., Niu, Z., Yang, P.: Traffic-aware base station sleeping in dense cellular networks. In: 18th International Workshop on Quality of Service (IWQoS), pp. 1–2, June (2010) 7. Chiaraviglio, L., Ciullo, D., Meo, M., Marsan, M.A.: Energy-aware UMTS access networks, W-GREEN, Lapland (2008) 8. Chiaraviglio, L., Ciullo, D., Meo, M., Marsan, M.A.: Energy efficient management of UMTS access networks. ITC 21 – 21st International Teletraffic Congress, Paris, France, pp. 1–8, September (2009) 9. Bousia, A., Kartsakli, E., Alonso, L., Verikoukis, C.: Energy efficient base station maximization switch off scheme for LTE-advanced. In: Proceedings of IEEE International Workshop on Computer-Aided Modelling Analysis and Design of Communication Links and Networks (CAMAD), pp. 256–260. Barcelona, Spain, September (2012) 10. Bousia, A., et al.: Green distance- aware base station sleeping algorithm in LTE advanced. In: International Conference on Communications (ICC 2012), pp. 1347–1351, June (2012)

Power Saving Approach in LTE

349

11. Bousia, A., Kartsaklli, E., Alonso, L., Verikoukis, C.: Dynamic energy efficient distanceaware base station switch on/off scheme for LTE advanced. In: Proceedings of IEEE Global Communications Conference (GLOBECOM), pp. 1532–1537. Anaheim, California, USA, December (2012) 12. Lassoued, N., Boujnah, N., Bouallegue, R.: Reducing Power Consumption in HetNet Network Using Power Adjustment and Coordinated Multipoint Technique. 32nd International Conference on Advanced Information Networking and Applications Workshops, AINA 2018 workshops, pp. 539–544. Krakow, Poland, May (2018) 13. Salous, S.: Radio Propagation Measurement and Channel Modeling. Wiley (2013) 14. E-UTRA; further advancements for E-UTRA physical layer aspects, 3GPP, Tech. Rep. TS 36.814 (2010) 15. Samir, J., Adnen, C., Sami, B.S., Balas, V.E.: An efficient design of Fuel Cell Electric Vehicle with Ultra-Battery separated by an energy management system. 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), December (2016) 16. Roukhami, M., Lahbib, Y., Mami, A.: A new efficient energy implementation of K-RLE algorithm for WSN. 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), December (2016)

Formal-Based Modeling and Analysis of a Network Communication Protocol for IoT: MQTT Protocol Jamila Hcine(B) and Imene Ben Hafaiedh Higher Institute of Computer Science (ISI), University of Tunis El Manar (UTM), Tunis, Tunisia [email protected], [email protected]

Abstract. The MQTT protocol is a widespread standard in the IoT world as it is widely deployed in different IoT applications. In the context of the IoT, a certain level of assurance that a given device does not incorporate vulnerabilities based on its protocol compliance is very fundamental to ensure. Such requirements could be reached through formal verification. In this work, we propose a formal model for the description of the MQTT protocol. In particular, we define a generic model expressive enough to model the different Quality of Service levels of the MQTT protocol. Our model is based on the formalism of timed automata. The formal verification of different properties as well as analysis experiments of the proposed model have been performed automatically using Modelchecking.

Keywords: IoT

1

· MQTT protocol · Formal verification · QoS levels

Introduction

Internet of Things (IoT) technology defines a set of objects which interact and communicate with each others in order to exchange information for decision making. IoT does not consider these objects as simple traditional objects but as smart objects by exploiting their underlying technologies. However, these objects are heterogeneous and they will use different communication protocols [1,2] depending on the requirements of the application they will perform. A reliable, efficient and real-time communication technology is an important issue for the development of such IoT applications. Indeed, in the context of IoT, all applications cannot rely on a single communication protocol [3,4]. Consequently, various protocols has been designed to meet different needs of IoT systems [5,6]. The MQTT protocol is one of the most widely accepted and emerging messaging protocols for IoT systems. It is characterized by its efficiency and its flexibility as it does not specify a particular data format. Using formal methods, to evaluate the performance of protocols [7], is considered as an interesting solution which ensures the complying with standards and checking the correctness of the model to the c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 350–360, 2020. https://doi.org/10.1007/978-3-030-21009-0_34

Formal Modeling of MQTT Protocol

351

set of required properties. In this work, we address the formal modeling and verification of the MQTT protocol. In particular, we propose a rigorous formal model allowing the description of different quality of service levels. This model is based on formal semantics of the framework UPPAAL [8]. We also propose a verification of our model using a real-time model checker for different possible network sizes. Using the UPPAAL Tool-set several properties of the MQTT protocol have been verified automatically and a set of performance analysis results have been conducted. The rest of the paper is organized as follows. Our formal model of the MQTT protocol and its different quality of service levels is presented in Sect. 2. The different analysis and verification results are given in Sect. 3. Section 4 discusses related works. Section 5 concludes with the conclusion and possible perspectives.

2 2.1

Formal Model of the MQTT Protocol The Overall Architecture of the Model

The MQTT protocol [9] is based on a Publish/Subscribe mechanism for message exchange. The message exchange is organized in topics and fields. The Quality of Service (QoS) level is an agreement between the sender of a message and the receiver of a message that defines the guarantee of delivery for a specific message. There are 3 QoS levels in MQTT: - 0 (At most once delivery): This service level guarantees a best-effort delivery. There is no guarantee of delivery. - 1 (At least once delivery): The sender stores the message until it gets a PUBACK packet from the receiver that acknowledges receipt of the message. - 2 (Exactly once delivery): This is the highest level of service, in which there is a sequence of four messages between the sender and the receiver, a kind of handshake to confirm that the main message has been sent and that the acknowledgement has been received. The detailed semantics of those messages will be detailed later within the description of our model. Our model is formally defined as a network of timed automata. The overall architecture of our model is partitioned into three parts namely: Broker, Publisher and Subscriber (see Fig. 1). - Broker : is the main part of our model as its is the main controller which receives the incoming messages from the publishers and relays them to subscribers. It consists of five sub-components. A central sub-component called BrokerQos0 ensures the quality of service level 0. For the QoS1, two sub-components ensure the needed communications. A first sub-component, called Side-PublisherQos1 , handles all messages exchanged between the broker and the publisher in the context of QoS1. A second sub-component, called Side-SubscriberQos1 , handles all messages exchanged between the broker and the subscriber in the context of QoS1. Similarly, two sub-components called Side-PublisherQos2 and SideSubscriberQos2 ensure the messages exchange in the case of QoS2. Defining, different components to handle each of the quality of service levels allows our

352

J. Hcine and I. Ben Hafaiedh

Publishers Publisher 1

Broker Side−Publisher

Side−Subscriber Subscribe(A,QoS1)

Publisher 2

Publisher 3

Subscribers

Broker PublishP(A,QoS0)

PublishB(A,QoS0)

Subscriber 2 Subscriber 3

...

... Publisher N

Subscriber 1

Side−Publisher

Side−Subscriber

Subscriber M

PublishP(B,QoS2)

Fig. 1. The overall architecture of the proposed MQTT model

model to be easily extendable to define more or less quality of service levels by adding the corresponding sub-components. - Publisher : We define N publishers {Pi }i∈[1..N ] with different QoS levels. - Subscriber : We define M subscribers {Si }i∈[1..M ] , subscribers are set to communicate with the broker over the required QoS. 2.2

Subscriber and Publisher Components

The behavior of the Publisher is described as follows (see Fig. 2 (a)):

Fig. 2. (a) The Publisher component, (b) the subscriber component

• Initially, the Publisher is in state startPublisher and depending on the value of the variable qos, the publisher decides which transition will be fired. If qos = 0, the publisher fires up a loop edge labeled publishP [pid]! . In this case, the topic has been sent to the broker then deleted. If the defined guard [qos > 0] is satisfied, the internal edge can be taken leading back to the same state.

Formal Modeling of MQTT Protocol

353

We define a function store() modeling the fact of checking the storage of the topic in a temporal variable. By firing up the transition labeled publishP [pid]!, the publisher goes from startPublisher state to published state. x is clock variable initialized whenever the topic is published. • In published state, an invariant Inv=(x ≤ 60s) is defined. This invariant guarantees that the publisher cannot stay in this state more than the predefined period. Whenever the clock reaches this period the publisher fires up the edge resend!. If x reaches 60s, the publisher fires up the edge resend! and turn on the flag rs. While in published state, If qos = 1(respectively qos = 2 ), the publisher may receive an acknowledgement from Side-PublisherQos1 (respectively SidePublisherQos2 ) to confirm that the broker has successfully received the topic. • In prel state, the transition labeled by the synchronization publishP Rel! allowing to synchronize with the Side-PublisherQos1 , but if there are a connection problem and the clock exceeds 60s the publisher fires up resendpubrel! edge and reset the clock. • In pcomp state, when a Publish-complete acknowledgement from the SidePublisherQos2 x is received, then the Publisher deletes the stored topic and return the flag stor to false. If such message is not received until the predefined period is elapsed, then the Publisher goes back prel state. A subscriber subscribes for a given topic. The broker will then transmit that topic messages whenever the new data is available. The behavior of the Subscriber is described as follows (see Fig. 2 (b)): • Initially, the subscriber has the choice between subscribe (respectively unsubscribe) to a topic by firing the edge labeled subscribe[sid]! (respectively unsubscribe[sid]!). Depending on qosS variable, the subscriber can fire publishB? loop edge, if [qosS = 0], it will consume the topic. Else the second edge with the same label is taken up which allows the sending of the topic to the Side-Subscriber components according on its quality of service. • In published state and after receiving that topic from the concerned SideSubscriber depending on the quality of service of the topic, if [qosS = 1], firing the edge labeled pubackS1! to the Side-SubscriberQos1 and the process is competed successfully. Else if [qosS = 2], the function store() is performed and an acknowledgement is then sent to the Side-SubscriberQos2 . • In acknowledged state, the subscriber is waiting for the publication release from the Side-SubscriberQos2 on an external loop edge labeled pubSRel? then sending Publication-Complete message by firing a pubSComp! edge. 2.3

Broker Components

The broker is the central component of the proposed architecture. It takes the role of a server that manages the communication between publishers and subscribers. It consists of a set of components depending on the QoS levels namely:BrokerQos0 , Side-SubscriberQos1 , Side-SubscriberQos2 , SidePublisherQos1 , Side-PublisherQos2 . BrokerQos0 is responsible for ensuring the QoS level of 0 and its behavior is described as follows (see Fig. 3).

354

J. Hcine and I. Ben Hafaiedh

- In startBroker state, the interaction is done with both of publishers and subscribers and it is waiting for topics from publishers by the execution of publishP [e]? edge involving the function publishp() that will store the message if [qos > 0] or will handles it directly to the subscriber using temporary variable. - In subscribed and unsubscribed state, the request of subscription or unsubscription is confirmed by sending an acknowledgement to confirm the request of subscriber. - Once a topic is sent by a publisher, then, 3 scenarios are possible. If [qos = 0], the broker sends directly the topic to the subscribers whose subscribed in using publishB! transition and makes it go back automatically to initial state.

Fig. 3. The BrokerQos0 component

If [qos > 0], the broker waits the sides to complete the process than it removes the message stored. If the message is not acknowledged within 60s, then the message is re-transmitted by the publisher and the process restarted. After that the components Side-Subscriber complete the sending of the topic to the concerned subscribers. Side-PublisherQos1 and Side-SubscriberQos1 are responsible for ensuring the QoS level of 1: - Side-PublisherQos1 sends an acknowledgement to the subscriber after receiving the topic by the broker by firing up an external loop edge labeled publishP Ack1!. Then it publishes the message to the subscriber on the external edge loop publishB! (see Fig. 4 (b)). - Side-SubscriberQos1 keeps waiting for an acknowledgement from the subscriber. Once it is received, Side-PublisherQos1 decrements the number of subscribers that got the topic (see Fig. 4 (c)). The QoS 2 is ensured by: Side-PublisherQos2 and Side-SubscriberQos2 . The behavior of Side-PublisherQos2 is described as follows (see Fig. 4 (a)): - Once the broker receives a message from the publisher, the Side-PublisherQos2 fires up an acknowledgement to the publisher.

Formal Modeling of MQTT Protocol

355

- In pubrel2 state, the Side-PublisherQos2 is waiting for a publish release from the publisher (publishP Rel?). If there are a problem in the connection or timeout, the publisher re-sends the pubrel. - In pubcomp2 state, the Side-PublisherQos2 sends the topic to the subscribers. If there is no subscriber for this topic, it fires up a publishP Comp! and completes the publish process in the side of the publisher. Whenever a problem occurs in sending the publish complete acknowledgement the process restarted from the release state and the pubrel is sent again. The behavior of Side-SubscriberQos2 component is as follows: - Initially, the Side-SubscriberQos2 is waiting for an acknowledgement from the subscriber for the published topic. - After receiving the acknowledgement, it responds with publish-release to the subscriber by firing up the transition pubackSRel!. - In state subComp, the Side-Subscriber is waiting for publish complete.

Fig. 4. Side-PublisherQos2 component (a) and Side-PublisherQos1 component (b) SideSubscriberQos1 component (c) and Side-SubscriberQos2 component (d)

Note that our model allows to define some kind of distributed broker. Indeed, defining different components ensuring each QoS level allows to define independence between these levels, which means that whenever there is an error in one of these components the rest of QoS levels will not be affected. Another important issue, is that we can easily extend our model to define new quality of service levels, by adding the corresponding component.

3

Formal Verification and Analysis

This section is devoted to the formal verification and performance evaluation of the different QoS levels of the MQTT protocol. 3.1

Formal Verification

The purpose of this verification is to check whether the functionalities of the MQTT can be accomplished by the model proposed here. Therefore we perform

356

J. Hcine and I. Ben Hafaiedh

the formal analysis of our model to prove a set of formally defined properties like deadlock-freedom and invariance. Using UPPAAL Model-checker, we have proven a set of formally defined properties (Deadlock, Invariance, Safety and Liveness properties) of the specified MQTT protocol. Deadlock-Freedom: is an important property to be checked when it comes to any protocol in general and dynamic ones in particular. Indeed, in dynamic systems where behaviors and configurations may change at run time deadlock situations have to be avoided. Such situations cannot be proven using simulations. Figure 5 (a) summarizes the verification time taken to check deadlock-freedom of the 3 QoS levels when increasing the number of components. Note that from a certain number of components, deadlock-freedom becomes undecidable because of the state space explosion problem. Invariance: Invariance is one of the most important issues when studying qualities of service 1 and 2 in MQTT protocol. Indeed, for both QoS levels, we have defined an invariant which guarantees that a publisher cannot wait for an acknowledgment infinitely. This invariant is formally described using TemporalLogic as follows : ∀w : pidt , A[] (P ublisher(w).idle ∧ rs = true) =⇒ x ≤ 60 Figure 5 (b) summarizes the verification time taken for checking invariance for our model. Store Property: Store is an important property to be checked when it comes to MQTT protocol. In particular, for QoS of levels 1 and 2. Indeed, for both

Qos0

Qos1

Qos2

Qos1

Qos2

300

Time of verification

Time of verification

250

120 100 71.219

67.25

80 60

42.343

36.75

40 20

3.39

5.891

20 11.625 12.109

24.938

19.969

175.703

200 150

162.344

102.516 59.922

100 50

7.797

90.5

38.968 13.281 25.719

0

0 10

(a)

264.843

126.75

140

15

20

20

25

Number of components

Qos1

40

60

80

100

Number of components

(b) Qos2 448.765

450

Time of verification

400 350 300

243.922

250 200

77.062

100 50

143.703

123.843

150 7.938

25.985

39.453

0 20

(c)

50

70

100

Number of components

Fig. 5. Verification time of (a) Deadlock-Freedom (b) Invariance (c) Store property

Formal Modeling of MQTT Protocol

357

levels, before delivering the topic to broker,it is mandatory that the publisher store it. This property is formally described (Temporal Logic) as follows : ∀w : pidt A[] P ublisher(w).published =⇒ P ublisher(w).stor == true Figure 5 (c) shows the verification time taken by UPPAAL for the store verification. CTL Properties: Based on the UPPAAL model-checker different properties could be verified. These properties have to be written as Temporal logic formulas, then checked automatically. Table 1 presents the verification results of different CTL properties defined as follows (for a configuration of 13 components): • Property 1 : Each topic is received once by the broker: A[] Broker.published =⇒ Broker.nbrec = 1 • Property 2 : Each topic is received once by the subscriber: ∀w : sidtA[]Subscriber(w).published ∧ j < SU ∧ Subscriber(w).subscrib[subtop[j]].sub = 1 =⇒ Subscriber(w).nbrec = 1 • Property 3 : A subscriber get topics only if it was subscribed in that topics: ∀w : sidtA[] Subscriber(w).published ∧ j < SU =⇒ Subscriber(w).subscrib[subtop[j]].sub = 1

Table 1. Verification-time (second) of variety of CTL Properties Properties

3.2

QoS Level Result

Verification Time (s)

Property 1

1 2

Not Satisfied Satisfied

0.016 229.562

Property 2

1 2

Not Satisfied Satisfied

0.11 256.875

Property 3

1 2

Satisfied Satisfied

165.812 257.703

Performance Analysis

We have formally proven the correctness of the model with respect to a set of formally defined properties. Further to qualify the performance of the three types of quality of services refers to MQTT protocol explained above, a list of simulation tests have been conducted using UPPAAL simulator.

358

J. Hcine and I. Ben Hafaiedh

100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00%

10

20

50

70

100

100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00%

Nb MsgSend

9.13%

18.50%

17.75%

63%

89.38%

Nb MsgSend

Nb MsgRec

4.63%

9.63%

22%

19.63%

45.63%

Nb MsgLost

4.33%

8.83%

20.75%

30.88%

43.75%

(b)

(a) 100.00% 90.00% 80.00% 70.00% 60.00% 50.00% 40.00% 30.00% 20.00% 10.00% 0.00%

10

Nb MsgSend 16.66%

(c)

10

20

50

70

100

5.50%

9.50%

23.75%

28.75%

37.25%

Nb MsgRec

9.25%

18%

87%

54.50%

64.75%

Nb MsgLost

13.25%

24%

63.50%

80%

76%

20

50

70

100

23.33%

63.33%

85.55%

85.55%

Nb MsgRec

15.55%

22.22%

62.22%

84.44%

85.55%

Nb MsgLost

18.88%

31.11%

64.44%

88.88%

73.33%

Fig. 6. Simulation of (a) QoS0 (b) QoS1 (c) QoS2

Indeed, For each quality of service protocol different metrics could be measured in different possible configurations such the percentage of the messages sent by the publishers, messages received by the subscribers and lost messages. Such aspects are very interesting in the context of IoT. In Fig. 6 (a), we can observe that the percentage of sent messages increases considerably with the time and still more higher than received messages cause of the resent principle. In Fig. 6 (b), As we know that in the QoS1 , the topic received more than once that’s why we can see that the number of received messages is greater than that the already sent. In Fig. 6 (c), The last protocol represents the highest quality of service, indicated by the QoS 2. The point here is to avoid duplicate messages as the broker can recognise them itself imply that in the simulation as represented in the figure that the number of sent messages almost equal to the received.

4

Related Work

In this section, we provide a discussion about the different existing research in the field of IoT, in particular, those related to communication protocols. In this context, the Publish/Subscribe paradigm is increasingly an important communication model, especially in the area of networks and IoT where messages can be communicated with more efficiency and less power consumption. In [10], the authors have presented a model for analyzing the publish/subscribe paradigm

Formal Modeling of MQTT Protocol

359

from a formal point of view. Their model addressed several aspects such as system extensibility, however, it was too general, so no particular performance analysis corresponding to a particular protocol could be performed. This is not applicable when one wants to compare different existing protocols. In terms of formal verification of communication protocols in IoT, there is little work in the literature mainly because of the very recent arrival of protocols. Indeed, the existing approaches for the validation of such protocols are basically based on test and simulation sequences [11]. In the context of formal verification, [12] has proposed a formal specification for the MQTT protocol based on the process of algebra. However, their analysis results are completely static and do not take into account the dynamic aspects of the system. The authors in [13], have proposed an environment using Event-B to model IoT protocols. Then they have used this platform to model some of the widely used IoT protocols namely MQTT, MQTT-SN, and COAP. They have formally verified properties related to connection establishment, duplication of messages etc. Their verification is not expressive enough to define CTL properties, it handles only properties based on invariant satisfaction. Recently, a similar approach based on the formal specification of the MQTT protocol is presented in [14]. Their work is based on the MSC UPPAAL and deals with the study of the probability of occurrence of events. In our work, we specify completely different kinds of properties like those related to QoSs and duplication of messages. To give a more precise idea about our approach with respect to the existing ones, specifically in [14] and [12], we have performed a comparative study based on the obtained results at different levels. At the level of deadlock verification, our approach results have reached up to a system with 25 components, while for existing approaches this aspect is whether undecidable or not studied. For Invariance verification, our model performs better than the two existing works as it checks Invariance for a system with 100 components. Furthermore, we propose a model that it is easily extensible to model new quality of services.

5

Conclusion

In the current research, we have proposed a high-level and a formal model for a well-known IoT communication protocol namely MQTT. The proposed model provides a way to describe the protocol in an abstract manner which allows to easily analyze and verify different properties without having to implement it in a concrete system. In particular, we have focused on the different QoS levels of the protocol. As future work, we are working in a formal model of COAP protocol. So that a comparative analysis between both protocols could be performed.

360

J. Hcine and I. Ben Hafaiedh

References 1. Jaikar, S.P., Iyer, D.K.R.: A survey of messaging protocols for IOT systems. Int. J. Adv. Manag. Technol. Eng. Sci. 510–514 (2018) 2. Benkerrou, H., Heddad, S., Omar, M.: Credit and honesty-based trust assessment for hierarchical collaborative IOT systems. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 295–299, December 2016 3. Naik, N., Jenkins, P.: Web protocols and challenges of Web latency in the Web of Things. In: Eighth International Conference on Ubiquitous and Future Networks (ICUFN), pp. 845–850 (2016) 4. Kasmi, M., Bahloul, F., Tkitek, H.: Smart home based on Internet of Things and cloud computing. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 82–86, December 2016 5. Tan, L., Wang, N.: Future Internet: the Internet of Things. In: 3rd International Conference on Advanced Computer Theory and Engineering (ICACTE), vol. 5, V5–376–V5–380 (2010) 6. Mokhlissi, R., Lotfi, D., Marraki, M.E.: A theoretical study of the complexity of complex networks. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 24–28, December 2016 7. Braham, R., Douma, F., Nahali, A.: Medical body area networks: mobility and channel modeling. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 1–6, December 2016 8. Wedyan, F., Freihat, R., Wedyan, S., Bani-Salameh, H., Yousef, H.: Domain analysis of formal model checking tools. In: International Conference on Engineering and Technology (ICET), pp. 1–4 (2017) 9. Amaran, M.H., Noh, N.A.M., Rohmad, M.S., Hashim, H.: A comparison of lightweight communication protocols in robotic applications. IEEE Int. Symp. Robot. Intell. Sens. 76, 400–405 (2015) 10. D´ıaz, G., Cambronero, M.E., Maci´ a, H., Valero, V.: Model-checking verification of publish-subscribe architectures in web service contexts. In: Proceedings of the 30th Annual ACM Symposium on Applied Computing, SAC 2015, pp. 1688–1695 (2015) 11. Aziz, B.: Towards a mutation analysis of IoT protocols. Inf. Softw. Technol. 100, 183–184 (2018) 12. Aziz, B.: A formal model and analysis of the MQ telemetry transport protocol. In: International Conference on Availability, Reliability and Security, pp. 59–68 (2014) 13. Diwan, M., D’Souza, M.: A framework for modeling and verifying IoT communication protocols. In: International Symposium on Dependable Software Engineering: Theories, Tools, and Applications (2017) 14. Houimli, M., Kahloul, L., Benaoun, S.: Formal specification, verification and evaluation of the MQTT protocol in the Internet of Things. In: International Conference on Mathematics and Information Technology (ICMIT), pp. 214–221 (2017)

Model Based Validation of Real Time QoS for NCDCLA Protocol in Wireless Sensor Networks Amra Sghaier1(B) and Aref Meddeb2 1

2

NOCCS, ENISO, University of Sousse, Sousse, Tunisia sghaier [email protected] National Engineering School of Sousse University of Sousse, Sousse, Tunisia [email protected]

Abstract. The important requirement in the context of network providing systems is the priori determination of temporal behavior. Such as is an important requirement in the context of network providing real time QoS guarantee has to be considered in packet delivery ratio and end-to-end delay. With this concern, NCDCLA, Network Coding based duty cycle learning algorithm provides a powerful method for improving performance of wireless sensor network. We specify in this paper a formal model for NCDCLA protocol based UPPAAL tools hence in order to satisfy QoS requirements in WSNs. Indeed, experiment’s results show that NCDCLA does demonstrate a significant improvement in terms of both end-to-end delay and packet delivery ratio. Keywords: NCDCLA UPPAAL

1

· Real time QoS · Verification formal ·

Introduction

The development of wireless joins the micro- electromechanical technology to give birth of wireless sensor networks. WSN consists of autonomous sensors nodes distributed in an environment. Each node is composed of a capture unit, a processing unit, a communication unit and a power supply. Although, its similarity to others wireless networks, WSN are distinguishable by a set of specific constraints. These constraints of WSNs nodes are manifested in the scarcity of energy resources, limited memory capacity and low computing power [1]. These constraints contribute on increasing of the complexity of these systems. Moreover, all nodes will eventually be deployed on a large scale and their number will be larger of the current internet of things. Indeed, equipment currently connected to the internet is devices that require from users their maintaining to use them properly. Wireless sensor networks, on the other hand, are stand alone devices by definition. They don’t require human intervention after deployment. In addition, in most applications, these nodes are deployed on a very large scale to collect as much data as possible. c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 361–372, 2020. https://doi.org/10.1007/978-3-030-21009-0_35

362

A. Sghaier and A. Meddeb

The fields of application for the wireless sensor networks are virtually unlimited such as medicine, military, precision agriculture, home automation, industrial automation and other IOT applications [2,3]. These applications can be grouped into different classes according to their objectives, the traffic characteristics and data delivery requirements. They can be also classified to the types of WSNs nodes used as well, or the topology used industrial applications are characterized by a set of essential needs that can be summarized in: energy savings, quality of service, auto-configuration, mobility and security. Regarding this aspects, Quality of service (QoS) is defined as the most important concepts in networks modern [4]. It is increasingly required because of the integration of several services with different needs which require a high QoS in terms of packet delivery ratio, energy efficiency, and end-to-end delay. As part of the networks of wireless sensors, the problem becomes even more complex because of the constraints intrinsic to WSN nodes. The limitation of resources (memory, processor, and energy), the unreliable wireless communication, high density and distributed nature of nodes represent the main issues for the development of any communication protocol. Obviously, because of these constraints, the guarantee of QoS in a network of sensors wireless poses non-trivial research problems. All mechanisms and protocols must be adapted and simplified so that they are compatible with the constraints of the resources of a WSN. We specify in this paper a formal model for NCDCLA protocol [5] based UPPAAL model checker tool [6] in order to verify the configuration of our paradigm which guarantees QoS like packet delivery ratio and end-to-end delay. Both simulation and model checker aims to analyze the worst case behavior. We have used OPNET Modeler 14.5 in [5] as simulator to determine the QoS requirements of NCDCLA protocol. As a reminder this paper is organized of 6 sections: Sect. 2 introduces related work. Section 3 presents the NCDCLA protocol modeling. Section 4 presents the protocol verification. Section 5 presents a comparison between UPPAAL and discrete event simulation. Finally, the conclusions are offered in Sect. 6.

2

Related Works

Critical applications call for a huge requirements and correct behavior. For this way, the protocols must be formally verified. Author [7] used UPPAAL model checker as verification tools to validate and tune F-MAC protocol. Verification shows that up to 6 nodes, F-MAC protocol is a reliable real time protocol. In [8], the author presents Dual Mode Adaptative MAC a real time protocol for process application in WSN. The protocol is verified using UPPAAL tools. In [9], the goal of the verification was to explore the behavior of a new Real Time MAC protocol with realistic assumptions on sensor network. The formal verification was done by using model checking methodology and UPPAAL tools. In [10] a model checking design drive framework was proposed in the way to designing the QoS based routing protocol. This model includes the light weight

NCDCLA Protocol Modeling

363

design process, the timed automata model and the alternative QoS verification properties. As result, the model checking driven design framework is a strainghtformed and modular method. Author [11] used UPPAAL tools in order to validate the MAC protocol of CSMA/CA on terms of message sending/ receiving and energy harvesting WSNs. As a result, the formal verification proves some performance trends. Moreover, the application of systems and methods for high QoS in Biometrical Sensors in medical domain can be harmful for the lives of patients. The authors [12] verified the QoS parameters of the network such as connectivity and packet delivery ratio and end-to-end delay against different temporal configuration settings for BSN using the chipcon CC2420 when the ultimate goal is to verify temporal configuration against different topologies in order to improve QoS properties. Similarly, in [13] authors presented the timed automata models for BSN using chipcon CC2420 transceiver so as to verify temporal configuration that satisfy QoS requirements. Authors [14] used Statistical Model Checking (SMC) as verification tools to verify and analyze the qualitative and quantitative properties of CFMA protocol (Collision Free Mobility Adaptive). In [15], Authors used UPPAAL-STRATEGO for automatic synthesis. The goal of this model is to find a strategy for verifying dynamic power management. This methodology presents a powerful method for designers in their construction strategies. Grichi et al. [16] introduced a novel methodology named RWiN using unified modeling language to verify reconfigurable wireless sensors networks. They used UPPAAL to analyze the originality of RWiN. Too, authors [17] presented a stochastic real time modeling formalism based on timed automata for verifying real time systems. This model aims to express stochastic timing constraints and to analyze several case studies. As well, authors [18] presented an approach based on hybrid automata in order to deriving QoS for SDF-modelled streaming applications. This approach significantly improves the scability. Ayoub et al. [19] introduced a new approach named 2AFM-QoS. This approach aims to automate the simulation of QoS mechanisms in digital networks.

3 3.1

NCDCLA Protocol Modeling Overview of UPPAAL

UPPAAL [6,20] is a model checking tool of real time systems developed jointly by BRICS and UPPSALA University. UPPAAL includes the following parts: – The description language allows describing system behavior with data type such as arrays, bounded integers . . . – The simulator is a validation tool. – Then, the model checker explores the state space of system for checking invariant and reachability properties.

364

A. Sghaier and A. Meddeb

A UPPAAL model is defined by a composition of a number of basic components called process or automata. Then, an automaton is defined as a composition of states (locations) and transitions (edges). A location could be: – Initial locations: each template begins with an initial location and it is marked by a double circle. – Urgent locations: when a process in an urgent location, time cannot pass. So, no delay in urgent location. – Committed locations: like urgent location, no delay in committed locations. Furthermore, when a process in a committed location, the next transition must involve automata in committed location. Each edge could be: – Selection: Selection binds non-deterministic identifier to a value in a given range. – Guard: An edge is activated if and only if the guard is evaluated to true. – Synchronization: A synchronization label can be found in the form c! or c?. Synchronization label must be without side effect and must be evaluated on a channel. Then, synchronization label could only refer an integer, constants and channels. – Updates: An update label could refer clocks, integer variables and constants. An update label can also call functions. UPPAAL is characterized by the efficiency and ease of usage. UPPAAL is an extension of the timed automata. A timed automaton (TA) is an extended finite state machine with clock variables. Data variables such as integer and array are the one used as a further extension on timed automata in UPPAAL, this set is used to ease the modeling tasks. Formally, a timed automaton is defined as follow [3]:  , T, Inv) T A = (Q, q0 , H, – – – – –

Q: presents a finite set of states; q0 :presents the initial state; H:  presents the finite set of clocks; :presents a finite set of actions; T :presents the set of transitions between locations;  T ⊆ Q × ζ(H) × ×2H × Q

– Inv:presents the mapping associating with each location q an invariant Inv(q). The semantic interpretation of a timed automaton is a TLTS for all states that are defined by (l, v) where l ∈ Q is a state of the automaton et v ∈ V (H), V () is the set of clocks of H.

NCDCLA Protocol Modeling

365

The parallel composition of several timed automatons allows to represent an N T A (Network Timed Automata) where this composition can be synchronized on common actions for all T A. Formally an N T A is defined by:  , T, Inv) T Ai = (Qi , q0 , H, N T A offers a better model than T A because it allows to model time, competition and communications. This is done by modifying the definition of the T A by adding: – V : is the data variable set; – T ⊆ Q × ζ(H) × ζ(V ) × ×2H × up(2v ) × Q where ζ(V ) represents the constraints on the clock and up(2v ) presents the associated updates. Allowing defining templates which have the same control structure is up to modeling language using different parameters, this is considered as the perfect feature for modeling sensor nodes. 3.2

NCDCLA Protocol

The NCDCLA protocol combines two operational mechanisms: Network coding in application layer and duty cycle learning algorithm (DCLA) in the MAC layer. The basic functioning of the protocol is based on the adaptation of duty cycle in order to select the optimal duty cycle so as to improve the QoS requirements in one hand and network coding, which combines data with anXOR operator to satisfy the requirements of end-to-end delay, energy consumption, packet delivery ratio in other hand [5]. The network topology of the NCDCLA protocol consists of 10 Micaz nodes and a PAN coordinator and an analyzer (sink) as shown in Fig. 1.

Fig. 1. NCDCLA protocol layer

366

A. Sghaier and A. Meddeb

Sensor nodes are required to deliver critical data to a sink node going through the PAN coordinator. At the beginning, the PAN coordinator must determinate the optimal duty cycle without any knowledge using Q Learning technique based on reinforcement learning. In active time, sensor nodes embed in the MAC header their queue occupation and delay values. According with nodes traffic load as well as its incoming traffic characteristics, the PAN coordinator must deduce the optimal policy Π ∗ (s). Mathematically, the optimal policy is defined as: Π ∗ (s) = argmax(Q∗ (s, a))

(1)

In order to reach the optimal policy Π ∗ (s), the Q learning algorithm is based on a two-dimensional table indexed by the state-action pair Q(s, a). The idea is to maximize exploration while randomly selecting actions over a large number of iterations, which makes it possible to perform several cycles of searching for rewards R(s, a), from the initial state to a goal state and to reinforce at each step the quality of the action that leads to rewards. The algorithm stops when all possible states are visited and the exploration rate is reduced to a probability  which allows the determination the optimal Q function Q∗ , which presents an indication for how good it is for an agent to pick action a while being in state s. Rewards function allows the PAN coordinator to learn the optimal behavior. Then, the PAN coordinator makes decisions on the optimal duty cycle, the beacon order BO and the superframe order SO are selected in order to find a compromise between queuing delay and overhead. R(s, a) is defined as follows : R(s, a) = rt + γmaxa ∈A(s ) Q(st+1 , at+1 )

(2)

All received packet from end devices to the PAN coordinator in the ready queue will be encoded using the mechanism of network coding, which aims to reduce the queue delay and improve all of throughput, energy efficiency and packet delivery ratio. The PAN coordinator forwards the encoded packet to the sink. Then the sink decodes the encoded packet and recovers data. 3.3

Formal Model

We use timed automata model to verify the properties of the NCDCLA protocol. In usual scenarios, sensor nodes and the PAN coordinator are modeled with UPPAAL. At the beginning, the parameters used in our modeling will be explained as follows: – The PAN coordinator seeks the optimal duty cycle with Q Learning() function. – If the period of duty cycle is active, the PAN coordinator sendP acket() and rcvP acket().

NCDCLA Protocol Modeling

367

– The PAN coordinator periodically sends probe information channel! And is synchronization with channel? – SendP [id] and RcvP [id] means respectively the number of encoded packet forwarded to the sink and the number of packet received from end devices. – Id refers the number of node sensor. – All packets received will be encoded with Xored() function. – Cnx[P AN id][id] indicates the connection matrix between PAN coordinator and end devices. – Clk refers the packet delay. – backof f D means the backoff duration. – queuing D means the queue delay. Figure 2 describes the timed automaton of PAN coordinator. The timed automaton begins with an initial location marked with a double circle.

Fig. 2. Timed automaton model of PAN coordinator

The PAN coordinator starts with adapting the duty cycle using the Q learning() function, knowing that the necessary parameters are initialized such as the energy level currentstate = e and the number of messages m = 0.

368

A. Sghaier and A. Meddeb

If the reward number is less than 4 max rewards 0 and clk 0 and the number of packets in the queue is other than zero queuingP acket >= 1 and there is synchronization channel!. The PAN coordinator uses the Xored() function to code the packets. The PAN coordinator switches to the wait location if queue packet = backof f D, the connection is not completed Cnx[P AN id][id] = 0 and the number of messages is 0. If the connection is made Cnx[P AN id][id] > 0 ,the energy is greater than 0 and does not exist a collision state collision?, the sensor node changes to the rcv location and the number of messages is increased RcvP [id]++. If there is a collision case clk >= backof f D, the sensor node changes to the Idle location. When sensor node wishes to transfer data to the PAN coordinator, the synchronization is performed by channel! and the sensor node changes to the send location as long as clk 0

(6)

We verify NCDCLA about the properties like end-to-end delay, packet delivery ratio, no deadlock and network connectivity. The model checker can manage networks from 5 to 10 nodes. The verification results are summarized in Table 1 which shows a good performance of the protocol.

5

Comparison of Simulation Results

The main purpose of verification tools is to analyze real time systems in order to obtain satisfied analysis results. To implement our results, we have made a comparison of the obtained results between OPNET MODELER 14.5 [5] and UPPAAL model checker.

370

A. Sghaier and A. Meddeb Table 1. Verification results

Property

Network size Results

end-to-end delay

6 6 8 6

Satisfied Unsatisfied Satisfied Satisfied

CPU time (sec) Memory (MB) 94.74 3.37 96.65 83.93

31.87 50.42 43.95 30.72

Packet delivery ratio 5 6 8 6

Satisfied Satisfied Satisfied Unsatisfied

616.54 232.28 657.31 353.2

66.08 220.03 375.47 663.87

No deadlock

8 8 8 6

Satisfied Satisfied Unsatisfied Satisfied

496.22 104.10 347.06 688.1

84.80 103.54 523.78 307.2

No deadlock

6 6 8 6

Unsatisfied Satisfied Satisfied Satisfied

68.24 93.17 87.56 89.73

90.05 15.52 33.76 31.48

Table 2. Comparison analysis Property

Sensors

Packet delivery ratio Node Node Node Node Node Node Node Node Node Node

1 2 3 4 5 6 7 8 9 10

UPPAAL (%) OPNET 14.5 (%) 99.1 91.5 93.9 89.7 97.4 98.7 95.3 92.9 88.6 97.1

95.3 92.1 87.4 90.7 95.8 98.1 93.6 90.9 90.3 95.7

To ensure comparison, we used the same scenario with the same parameters such as datarate(kbps) = 250, packetsize = 120, numberof nodes = 10 and the same simulation time. According to the simulation results obtained in Table 2, we can confirm that the results coincide with a slight difference due to the simplicity of the proposed model with UPPAAL. As well, we can confirm that UPPAAL present a powerful method to tune and validate the temporal behavior of NCDCLA.

NCDCLA Protocol Modeling

6

371

Conclusion

In this article, a behavioral and timeliness validation of NCDCLA protocol has been presented. The NCDCLA protocol is designed for critical applications which involve many verifications techniques such as testing, simulation and formal verification. Furthermore, we have used the UPPAAL model checking tool for modeling and verification of NCDCLA protocol. A formal model has been presented using timed automaton for sensor node and a PAN coordinator in order to tune and validate the QoS requirements such as energy efficiency, end-to-end delay and packet delivery ratio. The results of our study show a significant improvement of QoS requirements with NCDCLA protocol.

References 1. EL Brak, M., Essaaidi, M.: Wireless sensor network in smart grid technology: challenges and opportunities. In: 2012 6th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 578–583. IEEE (2012) 2. Akyildiz, I.F., Su, W., Sankarasubramaniam, Y., et al.: A survey on sensor networks. IEEE Commun. Mag. 40(8), 102–114 (2002) 3. Kasmi, M., Bahloul, F., Tkitek, H.: Smart home based on Internet of Things and cloud computing. In : 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 82–86. IEEE (2016) 4. Pasias, V., Karras, D.A., Papademetriou, R.C.: On novel efficient wireless access network design heuristic algorithms for QoS multiservice networks. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 153–160. IEEE (2016) 5. Sghaier, A.M.R.A., Meddeb, A.R.E.F.: NCDCLA: QoS aware network coding based duty cycle learning algorithm for real time and reliable wireless sensors networks. In: International Conference on Sensor, Systems, Signals and Advanced Technologies, Hammamat-Tunisia (2018) 6. Mouradian, A., Aug-Blum, I.: Formal verification of real-time wireless sensor networks protocols with realistic radio links. In: Proceedings of the 21st International conference on Real-Time Networks and Systems, pp. 213–222. ACM (2013) 7. Somappa, A.A.K., Prinz, A., Kristensen, L.M.: Model-based verification of the DMAMAC protocol for real-time process control. In : VECoS, pp. 81–96 (2015) 8. Watteyne, T., Aug-Blum, I., Ubda, S.: Formal QoS validation approach on a realtime MAC protocol for wireless sensor networks. Thse de doctorat. INRIA (2005) 9. Chen, Z., Peng, Y., Yue, W.: Model-checking driven design of QoS-based routing protocol for wireless sensor networks. J. Sens. 2015, 7 (2015) 10. Chen, Z., Peng, Y., Yue, W.: Modeling and analyzing CSMA/CA protocol for energy-harvesting wireless sensor networks. Int. J. Distrib. Sens. Netw. 11(9), 257157 (2015) 11. Tschirner, S., Xuedong, L., Yi, W.: Model-based validation of QoS properties of biomedical sensor networks. In: Proceedings of the 8th ACM International Conference on Embedded software, pp. 69–78. ACM (2008)

372

A. Sghaier and A. Meddeb

12. Fanourgakis, E., Schupp, S.: Modelling and verification of QoS properties of a biomedical wireless sensor network. Project Work, University of Hamburg-Harbug (2012) 13. Behrmann, G., David, A., Larsen, K.G.: A tutorial on Uppaal. In : Formal methods for the design of real-time systems, pp. 200–236. Springer, Heidelberg (2004) 14. Houimli, M., Kahloul, L.: Modeling and performance evaluation of protocols in mobile wireless sensor networks. In: International Conference on Broadband and Wireless Computing, Communication and Applications, pp. 328–339. Springer, Cham (2017) 15. Dai, S., Hong, M., Guo, B.: Synthesizing power management strategies for wireless sensor networks with UPPAAL-STRATEGO. Int. J. Distrib. Sens. Netw. 13(4), 1550147717700900 (2017) 16. Grichi, H., Mosbahi, O., Khalgui, M., et al.: RWiN: new methodology for the development of reconfigurable WSN. IEEE Trans. Autom. Sci. Eng. 14(1), 109– 125 (2017) 17. Nouri, A., Mediouni, B.L., Bozga, M., et al.: Performance evaluation of stochastic real-time systems with the SBIP Framework. Technical report TR-2017-6, Verimag Research Report (2017). 3.1, 4.1, 4.1, 6, 2017 18. Ahmad, W., Jongerden, M., Stoelinga, M., et al.: Model checking and evaluating QoS of batteries in MPSoC dataflow applications via hybrid automata. In: 2016 16th International Conference on Application of Concurrency to System Design (ACSD), pp. 114–123. IEEE (2016) 19. Bahnasse, A., Badri, A., Talea, M., et al.: Towards a new approach for automating the simulation of QoS mechanisms in a smart digital environment. Proc. Comput. Sci. 134, 227–234 (2018) 20. Wimmer, S., Lammich, P.: Verified model checking of timed automata. In: International Conference on Tools and Algorithms for the Construction and Analysis of Systems, pp. 61–78. Springer, Cham (2018)

SDR-Based Transmitter of Digital Communication System Using USRP and GNU Radio Nabiha Ben Abid1(&) and Chokri Souani1,2 1

Université de Monastir, Faculté des Sciences de Monastir, Laboratoire de Microélectronique et Instrumentation, 5000 Monastir, Tunisie [email protected], [email protected] 2 Université de Sousse, Institut Supérieur des Sciences Appliquées et de Technologie de Sousse, 4003 Sousse, Tunisie

Abstract. Today, an experimental research on wireless communication protocols is a progressively emerging field of study. Actually, users require in the existing technology a data flow that is at a time high rate and low-cost, as well as high efficiency and flexibility in order to fit the necessarily upgradeable system requirements. Software-defined Radio is a more important modern technology in the communication system. SDR is a radio that allows the software control of different modulation and demodulation schemes. It ensures as well the tuning of any frequency band whether large or narrow, a security function in communications and the wave form required by the new norms evolving over a large range of frequencies. In this article, we are presenting the conception of a wireless communication system using a software radio (SDR). SDR, as associated to real-time signal treatment structures, this signal processing protocol uses USRP as hardware and GNU Radio as software. The system was designed for an FM transmitter adjustable from 88 MHz to 108 MHz, with maximum efficiency and high quality audio signal. The audio file is transmitted to the USRP card using wide band frequency modulation (WBFM). Keywords: Software Defined Radio
GNU radio Digital communication
QFH antenna


USRP


1 Introduction The radio is a device that transmits or receives signals in the radio frequency (RF) band of the electromagnetic spectrum to facilitate the exchange of data. In today’s world, radios exist in a huge number of items such as computers, vehicles, cell phones and satellite receiver. Software defined radio technology offers an affective and relatively reasonable solution to the problem of traditional radio devices from higher production costs and minimal flexibility [1]. This article present a SDR-based digital communication transmitter system using GNU Radio software coupled with Universal Peripheral Radio Software (USRP) hardware. In this paper, first we will defined the SDR, in the second part we present an overview of USRP concept and architecture, as well, a brief on the antenna © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 373–381, 2020. https://doi.org/10.1007/978-3-030-21009-0_36

374

N. B. Abid and C. Souani

description to improve the sound quality, after that we will present some applications of transmitter radio FM.

2 Background 2.1

Software Defined Radio (SDR)

A number of definitions can be found to describe Software Defined Radio, also known as SDR. Starting with a clear and basic definition from the SDR Forum, working in collaboration with the Institute of Electrical and Electronic Engineers (IEEE) P1900.1 group, Software Defined Radio is a Radio in which some or all of the physical layer functions are software defined [2]. In other words, the Software Defined radio, SDR has a generic hardware platform on which software runs to provide functions including modulation and demodulation, filtering, frequency selection and many other functions. In addition, Software defined radio is a multifunctional, programmable, and simple to upgrade radio that can support a variety of services [1]. It is an enabling technology, applicable across a wide range of areas within the wireless industry, which gives efficient and relatively low cost solutions to several of the problems inherent in more traditional radio architectures [3]. The software defined radio system includes digital signal processing (DSP) processors and general-purpose processors for implementing radio functions that transmit or receive signals in the RF band of the electromagnetic spectrum. Fig. 1 shows the concept of Software Defined Radio. This figure shows that the ADC process is taking place after the Front End (FE) circuit, how is needed due to the limitation of the speed of current Commercial. The ADC will digitize signal and pass it to the baseband processor for further processes; demodulation, channel coding, source coding and etc. [4].

Fig. 1. Software defined radio block diagram.

An ideal receiver would be attaching an ADC to an antenna. A Digital Signal Processor (DSP) is used for signal processing. The digital signal processor generates a stream of output which is given to digital to analog converter. An ideal transmitter is also attached. The output is connected to the radio antenna [5].

SDR-Based Transmitter of Digital Communication System

2.2

375

Universal Peripheral Radio Software (USRP)

Universal Peripheral Radio Software (USRP) is a hardware interface used, which constructs the interface between a host PC and the RF data by mixing the transmitted and received signals with a software-definable IF. USRP is a unified map that incorporates digital-to-analogue converters (DACs) and analogue-to-digital converters (ADCs), front-end RF, and an FPGA that performs significant pre-processing of the input signal, making it the most usable hardware with the GNU radio. The main advantage of USRP card is fast and economical, for this reason, GNU Radio users choose to work on this card to implement applications in real time [6]. Figure 2 shows a USRP motherboard that is built from different components: USB 2.0 controller, ADC, CNA, PGA, daughter board and FPGA, it is combined with four daughter boards, each block of a typical USRP product consists of two parts: a motherboard with a high-speed signal processing FPGA processor and several daughter cards covering different frequency ranges. The receiver is bit stream data from the antenna to the host computer, while the transmitter is from the host computer to the antenna [7].

Fig. 2. Universal software radio peripheral [7].

2.3

GNU Radio

GNU radio is a free and open source software development tool which buildssignal processing blocks to implement software defined radio systems [4]. The goal of GNU Radio is to write applications to transmit and receive data in digital form. It has filters, channel encoders, modulators, and demodulators, found in radio systems. The GNU Radio applications are written using the Python programming language for high-level

376

N. B. Abid and C. Souani

blocks as well as the creation of graphs, the so-called low-level blocks are implemented in C++ which is used for the critical signal processing path using the floating-point extensions of the processor. GNU Radio is used with external RF hardware; in our case we are using a universal software radio device as RF hardware [8] (Fig. 3).

Python application developement creating flow graphs Signal Processing in module C++ RF Generic Frond End Fig. 3. GNU radio development model.

GRC COMPONION.GRC is a graphical programming tool that allows to build graphs in gnuradio flow [7]. GUIs for GNU Radio applications are built in Python. We recommend wxPython to optimize multi-platform portability [9]. GRC is a better tool to learn the basics of GNU Radio. then the user chooses the blocks of the library when the GRC executes a model, then slides and drops in the window GRC, then connect the blocks between them finally to edit various parameters of block.

3 Applications of Software Defined Radio Software Defined Radio is a Radio in which some or all of the physical layer functions are software defined; it is a technique for turning a computer into a radio. SDR uses computing power; we can listen to and decode a wide variety of programs. In this work, the figure shows the digital communication elements used in this project, we will discuss the implementation of FM transmitter using GNU Radio and USRP (Fig. 4). 3.1

Implementation of FM Transmitter

In this part, we discuss in detail the design of FM transmitter communications system using the broadband frequency modulation technique. In this technique, the highest frequency component of a modulating signal is greater than that of the peak frequency deviation of modulated signal. Figure 5 shows the Functional diagram of the transmitter detailed flowchart of the Fm transmitter. The main goal is to transmit an audio file to the destination, so we will use a WAV file source. In this source we have to choose the input file source where we store our audio file. Then, this input file is sent to the Rational Resampler block passing through the multiplier block, this block converts the frequency by performing interpolation or decimation as needed. Now, the output of the rational Resampler is sent to the WBFM block, here the quadrature rate is set to the multiple of 48 kHz and then sent to the multiplier block with the incoming sampling frequency. Figure 6 shows the detailed flowchart of the Fm transmitter.

SDR-Based Transmitter of Digital Communication System

377

Fig. 4. Block diagram of a software radio Composed of a USRP and GNU Radio. TX

Computer

USRP

GNU Radio Traitement signal

FPGA

Modulation

Interpolation

USB/Gbit

ADC DAC

Démodulation Etc… USRP Motherboard

RF IF Daughterboard

Fig. 5. Functional diagram of the transmitter.

Now we need a good antenna linked to the signals in order to transmit an audio sequence. The making of transmitter with a good sound quality is based on a QFH omnidirectional antenna allowing the perfect detection of FM signals using a USRP. Moreover, QFH is easy to make which means that we do not need to point at it so that it can detect the signals and have excellent performances all over the surrounding space. This antenna is fixed at the USRP which directly connected to the computer with an Ethernet wire, and on the computer a GRC session demodulates the incoming signal and sends the raw base-band audio data to the loudspeakers. The conceptual diagram thereof is presented below (Fig. 7).

378

N. B. Abid and C. Souani

Fig. 6. GRC flowchart for FM transmit.

Fig. 7. Description of FM transmitter system design.

3.2

Building the QFH Antenna

First of all the QFH antenna is relatively easy to build; it can be a homemade antenna. There are a few things need to be highlighted before construction can begin on the QHF. First it’s actually two antennas, a small one inside a bigger one that is of the proper dimensions so that it blocks out surrounding interference and receives signals very well on the 88 MHz frequency band FM. Second, you can build QHF antennas of different sizes that that are delicate to various frequency bands, because the dimensions of the antenna are directly proportional to the frequency you wish to send, but we will be concentrating on the dimensions for 88 MHz. We should calculate the size of our QFH antenna: the central mast and the supports, which will hold the components. to do this we used the following great online calculator made by John Coppens [10]. Use we motioned before the most important variable to select on the calculator is your design frequency; this is because the design frequency chooses what FM band your antenna will be advanced for. In our case, we selected a

SDR-Based Transmitter of Digital Communication System

379

design frequency of around 88 MHz to receive and transmit radio frequency signals, we connect this antenna to USRP to give a good sound quality, it’s represented in this Fig. 8.

Fig. 8. QFH Antenna.

3.3

Spectrum of the Detected Signal

This application has tried to present an efficient and affordable method to create a spectrum analyser based on SDR to detect the spectrum in the FM band. Here our main is to transmit some audio file to the destination so we are going to use a file source (here wav file source) and not the USRP. In the Wav file source we have to choose the input file source where we store our audio file. Figure 9 shows the graph for our Wav file before and after the WBFM block. The next step now and the most important is placing the FM demodulator. The output of the rational Resampler is sent to the WBFM block where the audio rate is predefined. Here quadrature rate is set in the multiples of 48 “Audio rate: 48 kHz, Quadrature rate: 48 kHz * 4 = 192 kHz, Tua: 75u” After the WBFM Transmit block, we show the properties of the FFT Sink to see exactly how the signal is taken into account after modulation. Now, these multiplied signals are sent to the USRP receiver, here we use 88 MHz as the center frequency but it can change using the slider, an antenna that is connected to the Tx/Rx daughter board. Similarly, from the two curves of Fig. 9, we note that the power level of the demodulated signal has decreased. In fact, it goes from −58 to −80 dB. This means that the broadband FM signal has been successfully demodulated In addition, the following figure shows the output of the same previous example, but with a different kind of frequency spectrogram. The use of the Waterfall mode to display all the information provided after demodulation of signal from the FM band (Fig. 10). The diagram above is named the visualization of a waterfall plot model. The spectrogram presents the surrounding strong FM bands and the signal flow from which it drops from the top to the bottom, and the reform comes the designation of the display as waterfall. The strong line in the middle shows the presence or the reception of the 88 MHz FM band signal. The signals surrounding this line show the presence of other low bands called “lateral bands” with some noise.

380

N. B. Abid and C. Souani

Fig. 9. Wav file signal before and after the WBFM transmission block.

Fig. 10. The waterfall plot of the 88 MHz FM station detected using the GRC flow.

SDR-Based Transmitter of Digital Communication System

381

4 Conclusion In this paper, we properly understand the concept of Software Radio and GNU Radio with the help of USRP and Python; we demonstrated the usability of SDR to implement the communication system. The use of reliable USRP and GNU Radio tools has proved to be very useful in emulating the FM transmitter system and understanding the digital processing of communication signals, while reducing costs compared to other options considered. In this article, we explained how easy it is to use SDR to implement the current wireless network development using GNU Radio and USRP. The flexibility of this platform at the software level facilitates various communication standards. Thus, to obtain and build a SDR system, in real time, realized in a simple and profitable way on the USRP box and with a good sound quality, we realized the FM transmission on USRP, based on a simple omnidirectional QHF antenna. This antenna plugged in instead of one of two USRP antenna, which is directly connected to the computer.

References 1. Rouphael, T.J.: RF and Digital Signal Processing for Software-Defined Radio: A MultiStandard Multi-Mode Approach. Newnes (2009) 2. The Wireless Innovation Forum. http://www.wirelessinnovation.org, last accessed (2017) 3. Pucker, L.: SDR Architecture. https://www.wirelessinnovation.org/assets/documents/tutSDR_Architectures.pdf 4. Al Masri, A.: Localisation sur une plateforme radio définie par logiciel. Université du Québec en Abitibi-Témiscamingue, Diss (2012) 5. Machado-Fernández, J.R.: Software defined radio: basic principles and applications. Facultad de Ingeniería 24(38), 79–96 (2015) 6. Barve, S., et al.: Open source software defined radio using GNU radio and USRP’. Int. J. Sci. Technol. Res. 3(5) (2014) 7. Vachhani, K., Rao Arvind, M.: Experimental study on wide band fm receiver using gnuradio and rtl-sdr. Advances in Computing, Communications and Informatics (ICACCI), 2015 International Conference on. IEEE (2015) 8. Muslimin, J., et al.: Sdr-based transceiver of digital communication system using usrp and gnu radio. Computer and Communication Engineering (ICCCE), 2016 International Conference on. IEEE (2016) 9. Blossom, E.: GNU radio: tools for exploring the radio frequency spectrum. Linux Journal 2004(122), 4 (2004) 10. Coppens, J.: http://www.jcoppens.com/ant/qfh/calc.en.php

Performance Evaluation of Nonlinear LMMSE-SVR Equalizer for High-Speed Radio Systems Anis Charrada(B) SERCOM-Labs, EPT, University of Carthage, 2078 Tunis, La Marsa, Tunisia [email protected]

Abstract. In this paper, we examine the channel frequency response equalization under the existence of impulse noise of a high-speed radio wireless system. Thus, we progress, by means of Linear Minimum Mean Squares Error-Support Vector Regression (LMMSE-SVR), an outspread algorithm to estimate complex values of the selective channel from the transmitted pilot symbols and then perform equalization task. This process mixes initially the estimation at reference symbols, performs at data information signals the nonlinear interpolation and finally accomplishes channel equalization. Numerical simulations are established in terms of Mean Squares Error (MSE) as well as Bit Error rate (BER) performances for a Long Term Evolution system using a high level modulation scheme (64-QAM Qadrature Amplitude Modulation) under high speed mobility (350 Km/h) in the existence of impulsive noise.

Keywords: LMMSE-SVR Impulsive noise · 3GGP

1

· Equalization · Very high-speed ·

Introduction

In order to reduce the Doppler effects and ensure the efficiency of the transmitted information, efficient estimation and equalization of the channel are required in highly mobility environments [1]. Alternatively, Orthogonal Frequency Division Multiplexing (OFDM) technique can enhance the spectral efficiency of the mobile system by affording resilience to the data information symbols towards the fading channel effects [2]. A good tool to face the nonlinearities caused by the high mobility environment is the Support Vector Machine (SVM) thanks to its great generalization capabilities. Thus, if the support vector regression learning process is correctly achieved, then the totality of unknown information data symbols may be recognized. In this paper, a Linear Minimum Mean Squares-Support Vector Regression (LMMSE-SVR) method is described using the Gaussian kernel and applied in frequency dimension. Channel estimation is firstly achieved at indices of reference c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 382–389, 2020. https://doi.org/10.1007/978-3-030-21009-0_37

Performance Evaluation of Nonlinear LMMSE-SVR Equalizer

383

signals using Linear MMSE approach and secondly SVR interpolation mechanism is executed with the aim to determine the sub-carriers frequency responses at each OFDM symbol. This paper is structured as following: The model of the system is presented in Sect. 2. Section 3 gives the details of the Linear MMSE-SVR functional. Simulation analysis are provided in Sect. 4. Section 5 concludes this paper.

2

System Model

In time domain, the expression of a received OFDM symbol comprising Nf subcarriers can be stated as the following baseband discrete expression: y(m) =



X d (s)H(s) e

2π jN sm f

s∈Θ / P

+



X p (s)H(s) e

2π jN sm f

+ wg (m) + b(m)

(1)

s∈ΘP

with X d (s) represent the data information symbols delivered at the sth subcarrier and X p (s) stand for the pilot reference signals; ΘP symbolizes the set of NP reference signals subcarriers, and H(s) = DF TNf {h(m)} stands for the sth subcarrier response in frequency domain. Into a fading frequency-selective channel, a 2 ) is introduced. Gaussian noise process (AWGN) N (0, σw On the other hand, in realistic world environments, impulse noise may be found frequently. Generally, the impulse noise is modelled as b(m) = μ(m)κ(m), with

Fig. 1. Variance of impulsive noise.

384

A. Charrada

μ(m) characterizes a random process with RBF (Radial Basis Function) distri2 , whereas κ(m) symbolizes a Bernoulli random bution with a power value of σBG process as follows [3]:  p, κ=1 Pr (κ(m)) = (2) 1 − p, κ = 0 An instance of impulsive noise variance is presented in Fig. 1. After DFT transformation, and after assuming eliminating symbols interferences, we obtain Y (s) = X(s)H(s) + WG (s) + B(s) = X(s)H(s) + E(s), s = 1, · · · , Nf

(3)

where E(s) symbolizes the overall noise in frequency domain.

3

LMMSE-SVR Approach

First, assume that we dispose of a Long Term Evolution frame containing a number of OFDM symbols equal to Ns and each one having a number of sub-carriers equal to Nf . The delivered pilot symbols may be represented as Xp =diag(X(m, nΔP )), n = 0, 1, · · · , NP − 1, with ΔP refers to the pilot symbol distance in frequency direction, m characterizes index in time direction and n stands for indices in frequency domain. It should be noted that the suggested Linear MMSE mechanism contains a twice distinct levels: training and estimation via interpolation. At the first level, the sub-carriers reference signals are determined using the Linear MMSE process by [5]:  −1 † P , P RP + σ 2 (XP XP )−1 H H = RP (4) M M SE

HH

HH

w

LS

P†

P where RHH = E[H P H ] characterizes the auto-correlation matrix of the chanP  stands for reference signals sub-carriers responses obtained by Least nel and H LS Squares technique: the cost function of Least  Squares can be given by :  P P P P P P † minimize (Y − X H )(Y − X H ) [6], and then

 P = XP −1 Y P , H LS

(5)

where Y P = Y (m, nΔP ) represents the received reference signals at the (mΔP )th  P = H(m,  reference signal index for the mth OFDM symbol, whereas H nΔP ) LS characterizes the obtained LS frequency responses. The overall OFDM channel sub-carriers responses can be obtained in the interpolation level by means of    P (m, nΔP ) ,  H(m, z) = g H (6)

Performance Evaluation of Nonlinear LMMSE-SVR Equalizer

385

with z = 0, · · · , Nf −1, and g(·) is the linear minimum mean squares error-support vector regression interpolation function. In the Reproducing Kernel Hilbert Space, g(·) can be expressed as  H(m, nΔP ) = w † ϕ(m, nΔP ) + q + en , n = 0, · · · , NP − 1

(7)

with q characterizes the bias, w are the weights of the problem functional, and en refer to the total noise. Moreover, we used the ε-Huber formulation [7] to enhance the estimation performance. After constructing and minimizing the corresponding primal problem, we get the following optimal weights solution: w=

N P −1

ψn ϕ(i, nΔP ),

(8)

n=0 ∗ ∗ ∗ ∗ ) + j(αI,n − αI,n ) knowing that αR,n , αR,n , αI,n , αI,n where ψn = (αR,n − αR,n represent real and imaginary components of the Lagrange multipliers. The Radial Basis Function kernel matrix can be represented as following:

G(r, t) = < ϕ (m, rΔP ) , ϕ (m, tΔP ) > = K ((m, rΔP ) , (m, tΔP )),

(9)

where K is the Mercer’s RBF [8]. After calculating the Lagrangian and deriving its derivatives to primal variables and annulling derivatives, we obtain the following dual functional: max− −

  1 † ψ (G + γI ) ψ +  ψ † Y P 2 (αR + αR ∗ + αI + αI ∗ ) 1 ε),

(10)

∗ ∗ subject to 0 ≤ αR,n , αR,n , αI,n , αI,n ≤ C [8] Finally, all OFDM data information symbols can be determined by Np −1

ˆ H(m, z) =



ψn K (z, (m, nΔP )) + q,

z = 0, · · · , Nf − 1.

(11)

n=0

4

Simulation Results and Analysis

Consider the following model of multipath time-varying channel impulse response h(τ, t) =

Q−1 

hq (t)δ(t − τq ),

(12)

q=0

with hq (t) symbolizes the attenuation of the q th path (representing the q th channel impulse response), Q denotes the total replicas of the channel, τq stands for

386

A. Charrada

160

Row indices of G

140 120 100 80 60 40 20

20

40

60

80

100

120

140

160

Column indices of G

Fig. 2. The contour plot of the Gram matrix. 2.5 2 1.5

Output values

1 0.5 0 -0.5 -1 -1.5 -2 -2.5 0

20

40

60

80

100

120

Number of rows of G

140

160

180

Fig. 3. The area graph of the Gram matrix.

the delay of q th replica. Furthermore, the used channel simulation parameters are stated in [4], as well as the tuning parameters of the LMMSE-SVR algorithm. We used two objectives criterias (SNR and SIR) expressed as follows [3]:  

E |y(m) − wg (m) − b(m)|2 (13) SN RdB = 10log10 2 σw  SIRdB = 10log10



E |y(m) − wg (m) − b(m)|2 . 2 σBG

(14)

Performance Evaluation of Nonlinear LMMSE-SVR Equalizer

387

The parameters of simulation are stated in [8] and are in agreement with 3GPP standards [9–12]. Our proposed approach estimates one thousand four hundreds OFDM symbols that corresponds to ten LTE frames. First, Fig. 2 displays the contour plot of the kernel matrix isolines. Moreover, the area graph in Fig. 3 displays elements in the kernel matrix as fluctuations presenting at each x step, each row participation to overall curve height. Figures 4 (a, b) show a sample of the values of support vectors (real and imaginary components) from simulating the corresponding scenario. These dual coefficients constitute the solution of the support vector regression problem. Figure 5 gives performance in terms of Bit Error Rate for LS, LMMSE and our technique for p = .1 and SIR = −5 dB. It is clear that the worst results appear for LS technique, whereas the best performance occurs to our proposed complex LMMSE-SVR algorithm. 9 8 7

Re (

i

)

6 5 4 3 2 1 0 0

20

40

60

80

100

120

140

160

180

120

140

160

180

Support Vectors

(a) 12

10

Im ( i)

8

6

4

2

0 0

20

40

60

80

100

Support Vectors

(b) Fig. 4. Support vectors components (a, b) (real, imaginary) dual variables.

388

A. Charrada 100

BER

10-1

10-2

64QAM-LS Estimation 64QAM-LMMSE Estimation 64QAM-LMMSE-SVR Estimation 64QAM-Perfect Estimation

10-3

0

5

10

15 SNR (dB)

20

25

30

Fig. 5. Bit error rate performance (mobility = 350 Kmph, modulation = 64-QAM)

Fig. 6. 3D-mean squares error performance.

Figure 6 displays a 3D mean squares error performance of the wireless radio system transmission scenario under consideration. This 3D curve illustrates the errors distribution versus time and frequency.

5

Conclusion

The aim of this work is to strictly perform estimation of multipath frequencyselective channel in high-speed radio systems (up to 350 Kmph) under nonlinear noise joint with high level modulation scheme (64-QAM) in order to increase

Performance Evaluation of Nonlinear LMMSE-SVR Equalizer

389

the overall transmission throughput of the system. Our approach is designed to enhance estimation of the channel using OFDM reference signals with the intention of enhancing the system spectral efficiency and then increasing QoS in high-mobility situations. According to 3GPP specifications, BER and MSE prove through simulation, the abilities of the proposed method to reduce the nonlinear effects of the impulsive noise as well as the deep fading caused by high-speed rapid nonlinear fluctuations.

References 1. Li, T., Fan, P., Xiong, K., Ben Letaief, K.: QoS-distinguished achievable rate region for high speed railway wireless communications. In: IEEE Wireless Communications and Networking Conference (WCNC), pp. 2044–2049 (2015) 2. Dai, X., Zhang, W., Xu, J., Mitchell, J.E., Yang, Y.: Kalman interpolation filter for channel estimation of LTE downlink in high-mobility environments. EURASIP J. Wirel. Commun. Netw. 1–14 (2012) 3. Charrada, A., Samet, A.: Joint interpolation for LTE downlink channel estimation in very high-mobility environments with support vector machine regression. IET Commun. J. 10(17), 2435–2444 (2016) 4. Charrada, A., Samet, A.: Estimation of highly selective channels for OFDM system by complex least squares support vector machines. Int. J. Electron. Commun. ¨ 66(8), 687–692 (2012) (AEU) 5. Jianning, Y., Kun, L., Xie, Z.: An improved channel estimation method based on jointly preprocessing of time-frequency domain in TD-LTE system. J. Netw. 9(4), 1047–1054 (2014) 6. Charrada, A., Samet, A.: Nonlinear complex M-SVR for LTE MIMO-OFDM channel with impulsive noise. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 10–13 (2016) 7. Charrada, A., Samet, A.: Nonlinear complex LS-SVM for highly selective OFDM channel with impulse noise. In: 6th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 696– 700 (2012) 8. Charrada, A.: SVM based on LMMSE for high-speed coded OFDM channel with normal and extended cyclic prefix. Phys. Commun. J. 29(2018), 288–295 (2018) 9. 3rd Generation Partnership Project, Technical Specification Group Radio Access Network; evolved Universal Terrestrial Radio Access (UTRA): Base Station (BS) radio transmission and reception, TS 36.104, V8.7.0, pp. 22–33 (2009) 10. 3rd Generation Partnership Project, Technical Specification Group Radio Access Network; evolved Universal Terrestrial Radio Access (UTRA): Physical Channels and Modulation layer, TS 36.211, V8.8.0, pp. 50–58 (2009) 11. 3rd Generation Partnership Project, Technical Specification Group Radio Access Network; Physical layer aspects for evolved Universal Terrestrial Radio Access (UTRA), TR 25.814, V7.1.0, pp. 20–29 (2006) 12. 3rd Generation Partnership Project, Technical Specification Group Radio Access Network; evolved Universal Terrestrial Radio Access (UTRA): Physical layer procedures, TS 36.213, V8.8.0, pp. 23–31 (2009)

A Real-Time Flash-Floods Alerting System Based on WSN and IBM Bluemix Cloud Platform Hamadi Lirathni(&), Amira Zrelli, Med Hchemi Jridi(&), and Tahar Ezzedine(&) Communication System Laboratory Sys’Com, National Engineering School of Tunis, University Tunis El Manar, Tunis, Tunisia [email protected],[email protected], {amira.zrelli,tahar.ezzedine}@enit.utm.tn

Abstract. In this paper, our aim is to present a Real-Time Flash Flood Alerting System using wireless sensors Network. In fact, we applied novel wireless technologies to monitor physical environment, among these technologies we deal in this paper with the famous application of wireless technology: “Wirelees Sensor Networks (WSNs)”. Our System aims to predict water level and different weather conditions such as temperature, humidity soil moisture. The collected data can be used to report and to forecast alarms for possible and future disasters. In our proposal system, the collected flooding information will be send at real time to IBM Blue mix cloud platform. This cloud platform helps the regulatory and welfare authorities to take suitable action and allow proper flood alerts notifications in particularly for citizen. In this work, we address the problem of real time floods alerting, so we propose a novel design of floods prediction system based on Wsn and Internet of things (Iot) technologies. Using a regression model we have been able to deduce about a true or a false alarms. Keywords: Wireless sensor network Climate change
Sensors
Iot




Flood




Bluemix




Alerting system




1 Introduction Wireless sensor networks are actually applied to monitor environmental changes such as flooding. Floods and fires are considered the first responsible for damages. Among these damages, we can cite the precious destruction of human lives, the degradation and the devastation of huge number of buildings each years. Bangladesh, Philippines and India are the most countries affected by floods [2]. In these countries, people suffer of disasters produced due to natural catastrophic. Nowadays, a many efforts are putted to develop intelligent systems which are able to reduce risks by an early disaster predictions system. Wireless sensors network and internet of things technologies can be applied in the prediction systems models. Indeed, there are severe limitations of old prediction system which may be solved using novel technologies [1]. In Tunisia, the Tunisian Civil Protection (TCP) has responsibility for controlling the national disaster management system and this organization will provide effective relief machinery for © Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 390–399, 2020. https://doi.org/10.1007/978-3-030-21009-0_38

A Real-Time Flash-Floods Alerting System

391

readjustment following flooding disaster. Floods are considered as a disaster which may cause extensive damage. Many villages are submerged by the flood waters and low-lying areas turn into huge lakes. The marooned people have to be evacuated to higher and safer places. Those who cannot be transported, have to be supplied with food etc. In Tunisia this problem had happened twice this year. It has a significant effect on residents, businesses and commuters in flood areas. The cost of damage caused by flooding can be reduced significantly by the warning time given before a flood event, and this makes flood monitoring and prediction critical to minimizing the cost of flood damage. A wireless sensor network is applied to help data communications and to resolve problems. In our work, gains of a wireless technology “WSNs”, has to be asseted to ‘Weather Monitoring Stations’. Many sensor stations measure and send parameters through a wireless network server. Added to that, the wireless sensor network is easy to maintain, devices are low-cost and uses less energy. Present-day, the wireless sensor network is used in many applications beauce it’s a cheaper technology and moreover it’s an efficient applications. The flexibility of “WSNs” have been applied with various ways according to different values measurement of the environment, it can be used to monitor the amount of carbon dioxide using MQT sensor nodes. Natural disasters are a worldwide aspect and require serious cooperation to address. Last hurricanes, floods, and other events have embellished this along with the differences of the effects of disasters on developed countries oppose to undeveloped countries. In Tunisia, the flood phenomenon is ancient [1]. Throughout history, there are dozens of times that regions have been affected. The most well-known events, most of which are still in the minds of the people, have been those recorded since the beginning of the last century, particularly after the 1950s. Tunisia had been affected by floods damages: 1969 floods (the whole country, (Sabbalet Ben Ammar), 2009 (Redayef), etc. Thus, many episodes had been marked for a long time the hydrological chronicles of the country. Given the scientific uncertainty surrounding the issue of climate change in general, and especially about their impact on rainfall trends, there is nevertheless a certain element: the hydrological changes inherent in an excessive urbanization and the various management actions, sometimes reckless, continue to increase the vulnerability of our cities and our spaces to the risk of flooding. In this paper, firstly we present goals of real flash floods alerting. In the second section we provide a novel architecture of flash-floods alerting system based on WSN and Iot. Indeed, we report our own algorithm model used to predict floods damages at real time.

2 Goals of Real-Time Flash-Floods Alerting System There have been many investigation and formal researches on early warning system in general as well as the specific purpose of flood detection [2]. Several approaches have led to solve the problem at different times. In this time, we have are following this approach. First of all, we just collect information and data including: water level, temperature degree and rainfall prediction. Then, we compare these values to conclude about possible floods. Flash-floods alerting system is able to deduce about real time floods and generate data. The current system is directed into an flash flood warning to develop the original proposal, other research produced more insight into existing

392

H. Lirathni et al.

technology especially its implementation and its dilemma. Indeed, an wireless technology exists to monitor the environment and to include water point (Fig. 1).

Water level sensing network

Analyzing equipment Monitoring System

Server

Data Base

Gateway

Delta of Majerda

Fig. 1. Design of water level monitoring system (Delta of Majerda)

Companies with sensors in-place indicated the pressure transducer sensors, when protected perfectly, can be more durable. These sensors are used primarily where water is continuously above the sensor. Ultrasonic sensors can be seated above an area where water exists or is expected to exist and can calculate the height of water below the sensor. Data loggers are very common items and available from a number of sources [3]. Secondly, the sink of river side can maintain the water and also flow of water. We get the information from the riverside sink. In our case, we suppose that the sink is using water level sensor node which flow of water level. This node can transmit data information through WSN. Therefore, our hydrological system provide a rainfall and weather report, the department give us process real time rain and weather forecasting information. This information can be taken as an input to predict and prevent how much rainfall may be there in special and particular region. Finally, the information will be directly sent from the FTP server applied in our system to the application servers to predict floods at real-time monitoring.

3 Design of Real-Time Flash Floods Alerting System Actually, “IoT” can be presented as an emergent technology which can enable several things and devices to become an intelligent nodes. IoT is considered as one of the more technologies used in the construction of real time flash floods [5]. There are various

A Real-Time Flash-Floods Alerting System

393

limitations of floods warning system in rural and developed countries this paper proposes a distributed network model based on wireless sensors. Patil et al. (2015) introduce dynamic internet protocol based on ARM9 embedded webserver to detect floods [11]. Azam et al. (2017) present a mobile application to detect flood and to alert for dangerous flash flood [12]. Their proposed system provides timely and correct data information, so they can reduce the potentially risks and damages at real times [12]. In their work, Ferris et al. (2016) show that wireless and mobile alert messages for damages and flash floods warnings can reduce accidents by 16% compared with other alerted protocols [13]. The obtained results in [13] confirm that wireless alert messages effectively reduce hazards damages associated to extreme weather especially floods. In this section, we introduce the design of our own flash floods alerting system. The design of our presented system consists of two different nodes which depend from the task they perform: sensing nodes and gateway node. Routing protocols of disasters and damages detection system have been largely studied in our previous work [14, 15]. We can affirm that LEACH and CTP (Collect Tree Protocol) will be the most adapted to our system. The comparison affected in [14] between these protocols affirm that LEACH protocols will be more suitable to route data between nodes. John et al. (2017) studied RPL (Routing protocol for Low-Power and Lossy) based on specific routing metrics. They showed that RPL can be appropriate for IoT applications [16]. Wireless Sensor Network or WSN are considered a special type of Ad hoc networks with a lot of nodes that are micro-sensors capable of collecting and transmitting environmental data in an autonomous way and the fixed communication infrastructure and centralized administration are absent and the nodes play both the role of the host and the routers. It’s one of the new technologies that are shaking up the world and the way we live and work t responds to the emergence in recent decades of supply and increased need of diffuse and automatic observation and control of complex physical and biological phenomena, in different fields. The sensing nodes as their names suggest collect different data of the corresponding environment including but not limited to water level, soil moisture, humidity, air temperature, rainfall, and water flow data. These nodes cover a dense area. Once these nodes collect the data, they channel it to the gateway node. The gateway node send it to the IBM Bluemix cloud platform. Finally, the servers collect data and alert the community with social media, SMS and/or Email (Fig. 2). Typically, wireless networks nodes [5] are applied in one of three types of network topologies: Star Topology, Mesh topology and Cluster topology. In the case of star topology, we can see that each sensor node can connect directly to a sink (the gateway of our network). In the other case, the cluster tree network, each sensor node is connected to other node higher as a tree and then the higher node is connected to the gateway, thus all data are routed from the lowest node on the tree to the gateway. In the case of mesh topology, mesh feature nodes are connected to multiple nodes in the system and can send data through a reliable path which is available (Fig. 3). Sensor nodes can take self-decisions in floods detection systems to accomplish sens-ing tasks, constructing network topology and routing policy [6].

394

H. Lirathni et al.

Information from metecrological Institute Alert Real Time flooding forecasting report

MQTT SMS or social Media Gateway

IBM BlueMix IoT

Motes

Information

Fig. 2. Design of the floods monitoring system

Star

Cluster Tree

Mesh

Gateway Node

Fig. 3. Common WSN networking topologies

3.1

Bluemix

IBM Bluemix is a cloud platform for basic mobile application. It can be considered as a service (PaaS) developed by IBM [3]. It may supported various programming languages and also services as well as integrated DevOps to build, run, deploy and manage applications on the cloud. Bluemix is based on Cloud Foundry open technology and runs on Soft Layer infrastructure. The goal of [3] for the IBM Mobile Cloud Services was to provide robust and scalable services for any mobile applications

A Real-Time Flash-Floods Alerting System

395

coupled with Android, SDKs for iOS, and JavaScript clients which integrate well with the native development environments of each platform. The IBM Bluemix IoT service provides a simple and powerful capability to interconnect different kinds of devices and applications all over the world [9]. 3.2

MQTT

MQTT (MQ Telemetry Transport) has been designed as a novel messaging transportation protocol, this protocol can bring as a topic aimed to publish “subscribe mode” [5]. MQTT protocol is used to provide home control services in smart cities and buildings systems. The design of principles MQTT is to minimize network bandwidth and device resource requirements. MQTT turn out to make the protocol ideal of the emerging Internet of Things world of connected devices, and then for mobile applications the bandwidth and the battery power are considered at a premium. 3.3

Zigbee

ZigBee called also IEEE 802.15.4 [6, 7] is considered as a wireless and wifi standards based on technology presented to address needs of low-cost, low-power wireless sensor and control networks in just about any market. Zigbee shares the 2.4-GHz license-free band with IEEE 802.11b/g [7]. An attacker may jam the band using radio channels which are designed to transmit strong signals as part of test modes. In our design system, we use wireless sensor Waspmote as new generation of wireless sensor mote which has been recently released by Libelium. In the following Fig. 4. XBEE TRANCEIVER 3 is the bloc diagram of the hardware components of a sensor mote. ZigBee Alliance industry community, relatively recent based on the IEEE 802.15.4, aiming to provide a simpler protocol, less expensive and to overcome the problem of too high energy consumption as Bluetooth. An amalgam is often made between the ZigBee standard and the IEEE 802.15.4 standard, which nevertheless apply to different levels. Any standard Zigbee transmitter may be bought for a few dollars [7] in the form of a tiny evaluation board). XBee is a device used to send and receive data wirelessly and can build up a IEEE 802.15.4 network reference standard. XBee functions can be divided by network topology in different ways including the coordinator, the router and the end device. 3.4

Gateway Architecture

As shown Fig. 4, the central element of our gateway is a raspberry pi and is currently routing all network traffic between the ZigBee network and the internet, while also providing DHCP and DNS services API Operation [8].

396

H. Lirathni et al.

Temperature Sensor XBEE Transmitter

Libelium Controller Water Level and Speed Sensor

Alarm

Battery Supply

Fig. 4. Mote bloc diagram

Read Data (L,R,D,Xi, SC)

Compute Linear Regression Model

Deduce the Weighted Mean P and S

Predict corresponding value of L

Lnow >Lpast

Lnow < Lpast

Floods is not expected

Recalibrate time prediction and In-

If Counter > < C1;2 ¼ max C1;3 > > > : C1;4 C1;5

11.2640 19.0951 30.1834 39.9769

10.7546 18.8301 29.8711 39.6426

10.9123 18.9714 30.0490 39.8404

¼ 0; 252 ¼ 0; 831 ¼ 1; 064 ¼ 1; 349 ¼ 1; 191

  Cmax ¼ max C1;1 ; C1;2 ; C1;3 ; C1;4 ; C1;5 ¼ 1; 349

2nd Secondary User: Cmax

8 C2;1 > > > < C2;2 ¼ max C2;3 > > > : C2;4 C2;5

¼ 0; 983 ¼ 0; 205 ¼ 0; 407 ¼ 0; 997 ¼ 0; 494

  Cmax ¼ max C2;1 ; C2;2 ; C2;3 ; C2;4 ; C2;5 ¼ 0; 997

3rd Secondary User: Cmax

8 C3;1 > > > < C3;2 ¼ max C3;3 > > > : C3;4 C3;5

¼ 0; 591 ¼ 0; 123 ¼ 1; 061 ¼ 0; 925 ¼ 0; 885

  Cmax ¼ max C3;1 ; C3;2 ; C3;3 ; C3;4 ; C3;5 ¼ 1; 061 Number of users = 4 The global sum of spectral system capacity: Cglobal ¼ 11; 164 As shown in Fig. 4. Our proposed algorithm (for M > 1) improves the spectral efficiency of the system compared to the simple orthogonal algorithm (for M = 1). Moreover, it is remarkable that the increase in the number M of users in parallel with each access to the spectrum generates an increase in capacity. This is very logical, because this way we increase the probability of finding a user that allows a better use of the spectrum thus, a better optimization of spectral capacity. But, this contribution increases the number of high capacity users. The evaluation of our study shows that the capacity values of our model give a greater capacity than the simple orthogonal approach as shown in Fig. 4. The conclusion of this comparison justifies that the idea of an orthogonal model with competition is more efficient compared to the simple orthogonal model.

428

H. Farhi and A. Messai

Fig. 4. Comparison between the capacity of simple orthogonal model and orthogonal with competition

4 Conclusion Our research, examined the cognitive radio system in which cognitive utilizers profit from the used spectrum. At start, the research allowed quantifying a basic parameter in cognitive radio systems that is known as the global capacity, using a model of competition between secondary users of cognitive radio compared to the devices of the simple orthogonal model. In fact, we have defined the sum of the overall capacity of each system, and then got a description of the maximum realizable spectral capacity. Besides, the great number of users in that plan. Our results of analyzes and simulations highlight and prove the value of our orthogonal approach by competition. The latter has allowed us to increase the spectral capacity of the secondary users as well as the spectral efficiency of the network; these two factors play a leading, essential and major role in the improvement of quality of service.

References 1. Cognitive radio networks: a survey. In: Proceedings of IEEE International Conference on Innovations in Electrical, Electronics, Instrumentation and Media Technology ICIEEIMT 2017. IEEE (2017) 2. Raut, R.D., Kulat, K.D.: SDR design for cognitive radio. IEEE (2011) 3. Wang, Y.: Lockheed Martin TSS. Cognitive Radio for Aeronautical Air-Ground Communications. IEEE (2010)

Spectral Capacity in Cognitive Networks

429

4. Salem, M.B., Ettabaâ, K.S., Bouhlel, M.S.: Anomaly detection in hyperspectral images based spatial spectral classification. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp 166–170 (2016) 5. Glossner, J.: Special session on software defined radio (SDR) and cognitive radio (CR). In: Sandbridge Technologies. IEEE (2010) 6. Aissa, I., Frikha, M., Tabbane, S.: A dynamic power management procedure in cognitive radio. Tunisian High School of Communications. IEEE (2010) 7. Sinanovic, S., Serafimovski, N., Haas, H., Auer, G.: Optimum spectral efficiency of horizontally spectrum sharing 2-link system. Institute for Digital Communications School of Engineering and Electronics, The University of Edinburgh, UK. IEEE (2008) 8. Foukalas, F.: The performance gain of cognitive radio in adaptive modulation scheme. In: Wireless Engineering and Technology. SciRes (2010) 9. Benzarti, M., Abdella Ouf, Z.: Comparative study of frequency synchronization in SISO and MIMO-OFDM systems. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp 197–203. IEEE (2016) 10. Devroye, N., Mitran, P., Tarokh, V.: Achievable rates in cognitive channels. IEEE Trans. IT 52(5), 1813–1827 (2006) 11. Haddad, M., Hayar, A.M., Debbah, M.: Spectral efficiency of cognitive radio systems. In: Mobile Communications Group, Institut Eurecom (2009) 12. Khan, F.A., Ratnarajah, T., Sellathurai, M.: Multiuser diversity analysis in spectrum sharing cognitive radio networks. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 117–122. IEEE (2016) 13. Pasias, V., Karras, D.A., Papademetriou, R.C.: On novel efficient wireless access network design heuristic algorithms for QoS multiservice networks. In: 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), pp. 153–160. IEEE (2016)

Cloud Service for Edge Configuration in Home-Based Healthcare Context Imen Ben Ida1(&) and Abderrazak Jemai1,2 1

SERCOM Laboratory, Department of ETIC, Polytechnic School of Tunisia, Carthage University, Tunis, Tunisia [email protected] 2 INSAT, Carthage University, Tunis, Tunisia

Abstract. The expansion of Internet of Things (IoT) and the evolution in communication technologies have an important promising trend in healthcare domain. By exploring connected medical devices, some healthcare services are moved from the hospital to the home environment and the control of patients is continuous in a comfortable home environment. Various solutions based on Cloud services and IoT technologies are explored to collect and analyze the patient data. However, availability limitation can arise when transferring huge amounts of patient data from IoT devices onto cloud. To support these challenges, a third layer between the cloud and devices which is known as Edge layer is proposed. In this paper, we present the benefits of using edge computing for home-based healthcare and we propose an edge-based architecture that enables the medical staff to change the edge layer configuration using a cloud service. We focus specially on the integration of cloud services and edge services to ensure an effective control and monitoring of the patient data. Keywords: Edge computing


Cloud service
IoT
E-health

1 Introduction In the healthcare field, applying Internet of Things (IoT) technologies which connect small devices to the internet have attracted attention. It has the potential of improving the quality of medical services with automating tasks which were previously achieved by humans. The Advances in smart devices or objects in term of sensing data in different ways is a core part of new patient health status monitoring systems [1]. These medical devices can be connected to the virtually unlimited resources of cloud to ensure the easy automation of the process of collecting and delivering personalized on-demand services. By exploring Cloud and IoT technologies, A rising trend is to move some healthcare services from hospital (Hospital-Centric) to the home environment (HomeCentric). One advantage of this trend is the fact that patients can benefit from different healthcare services at any time in a home environment. Added to that, healthcare costs

© Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 430–439, 2020. https://doi.org/10.1007/978-3-030-21009-0_42

Cloud Service for Edge Configuration in Home-Based Healthcare Context

431

could be highly reduced by remote control and limited hospital resources can be more disponible [2]. In Home-based medical systems, connected devices are used to collect data of the patient health status such as temperature, heartbeat and glucose level in their home. The sensors and electronic devices that collect patient data may be fixed or implanted within the body which known as Body Area Networks [3]. Patient context data are also important to collect such as room temperature and humidity because they affect in many cases on the patient situation. Sending the collected data to the cloud gives the caregivers the possibility of remotely controlling the patient [4]. This service is beneficial specially for the elderly person, it offers the medical staff the ability to regularly monitor his behaviors. Hospital with limited resources will be more disponible for the patients who need an emergency care. However, integrating IoT and Cloud technologies introduces several challenges in home-based healthcare solutions. First, patient control depends on the network reliability and the availability of cloud services. Added to that, the embedded applications usually need to send continuous flux of data where availability should be considered [5]. To compensate these challenges, Edge computing is a promoting solution for Patient control in a distributed way. Several tasks could be moved from cloud to a local Edge in order to ensure real-time responsiveness, data pre-processing and security features. Smart gateway is an example of edge component that is reinforced with sufficient power, intelligence, storage and communication capabilities to manage the interchange of information between the connected objects and the cloud services. In this paper, we explore the edge computing paradigm and we propose a cloud service that enables the remote configuration of the edge layer. The main role of the proposed cloud service is the possibility of changing the local processing of a homebased gateway without code modification. The rest of the paper is organized as follows. Section 2 presents a standard architecture based on edge computing and its benefits in home-based healthcare case. The related works and the motivation are analyzed in Sect. 3. Section 4 focuses on prototype implementation of the control cloud service. Finally, the paper concludes in Sect. 4.

2 Edge Computing for Home-Based Healthcare 2.1

Edge Based Architecture

Edge computing refers to the enabling technologies allowing computation to be performed at the proximity of data sources. “Edge” is an intermediate layer between the data sources such as medical devices and the cloud data centers. It offers several computing and network resources [5]. At the edge layer, the things are considered as data producers. Edge can perform computing offloading, data storage, caching and processing, as well as distribute

432

I. Ben Ida and A. Jemai

request and delivery service from cloud to user. As a result, the edge should be well designed to meet specific requirements such as reliability and data security [6]. The Fig. 1 presents the 3 main layers of a basic architecture that explores Edge computing for home-based healthcare system:

Fig. 1. Edge-based architecture for home-based healthcare system.

a. Medical Sensors Network: The sensors are enabled by the capability of sensing and communicating data about the patient health status. Communication protocols are used then to transmit these data to the gateway. b. Gateway at the Edge layer: A smart gateway is used to support different communication protocols. It acts as a dynamic intermediate layer between the home-based sensor network and the hospital server. The gateway receives data from the connected medical devices to provide higher-level services such as local data storage, filtering and security functionalities. The collected data are transferred then to the cloud. c. Cloud services: Cloud computing environment has the role of a Back-end system. It provides a long term storage of the patients data and several analytic functions. Medical staff can benefit from online graphical user interfaces for patients control, reporting and feedback. 2.2

Edge Services in Healthcare Context

The edge components such as gateways are in a unique position between both the devices and the cloud for home-based healthcare. This promising opportunity is exploited by different means:

Cloud Service for Edge Configuration in Home-Based Healthcare Context

433

a. Pre-processing: The gateway supports extra processing functionalities before sending data to the cloud. For example, bio-signals collected from users’ body usually contain noise frequencies that distort the signal quality. This issue can be addressed by embedding filtering functionalities in the gateway to remove the noises and reconstruct data into useful form [7]. b. Local cache storage: The gateway initially stores the data in a local database before sending it to the cloud. The local storage can be considered as cache storage that contains the application’s recently collected data. Its main role is to offer a low-latency access to the data. For each data request, the gateway checks the local storage before sending a request to the cloud. Added to that, in case of internet interruption, the local storage prevents the loos of the patients collected data [8]. c. Reliability: Medical sensors could have lost connection to the cloud very easily due to battery outage, loose of internet connection, hardware or software misfunction, etc. One of the principal benefits of the Edge layer is to ensure patient data reliability when the internet connection is lost and to inform the responsible which device is not responding. Added to that, the Edge layer ensures the interoperability when using different IoT protocols which improves the communication reliability [5]. d. Security and privacy: Securing the data at the Edge layer is a recommended solution to ensure the privacy for patient data. To reinforce the security of the general solution, local security mechanisms could Provide multitenant access with dynamic, zone-based and contextaware security policies. Added to that, the gateway improves of the low computational capability of IoT devices to encrypt data [9]. e. Devices control: The Gateway could manage various aspects of underlying IoT objects by keeping check on their activities and energy consumption of power [4].

3 Related Work and Motivation A cloud service that relies on a rule-based platform is proposed in [10]. It enables users to manage and monitor rules running on both local gateway and a cloud server. Added to that, it provides a messaging service to ensure the exchange of data between a gateway and the cloud. In [11] the authors focus on the requirements of the network for a remote monitoring system. The requirements are the real-time event update, bandwidth proprieties and data generation. They propose a remote health monitoring architecture based on

434

I. Ben Ida and A. Jemai

IoT and called IReHMo. The prototype implementation reduces the volume of transmitted data and the required bandwidth for a specific use case. Negash et al. [8] present an e-health gateway that assist IoT-enabled medical services. The gateways are placed in geographically distributed network and each one is responsible for managing a set of IoT medical devices. The gateways are supported with data analytics and adaptive configurations in order to monitor the patients independent from their movements. A cloud platform is considered as a back-end system where the data persists and users get access to the system through web or mobile interface. It allows the communication with other systems such as hospital information system. Another smart e-health gateway named UTGATE is presented by Rahmani et al. in [7]. It offers at the Edge layer real-time local storage, data processing and data analysis. The system takes in consideration reliability, mobility and energy related issues. For the system performance evaluation, an IoT-Cloud-based Early Warning Score (EWS) case study is used. In [12], the authors present a smart home solution based on different nodes that are connected to a gateway using a ZigBee network. A cloud platfom is used to record the behavior of the patients in real time and also to control the different equipments of a house in remote way. Despite the efforts of using edge-based architectures for patient control, there are only small contributions that allow a remote edge configuration. The presented efforts focus on general implementations based on the use of a gateway which contains a predefined strategy of local processing and data storage. Added to that, the cloud layer is used only for data storage and data processing services and it is not explored to enable remote configuration depending on the patient situation. Motivated by the need of remote configuration for home-based gateways to ensure customized healthcare solutions, we focus on the 3rd layer of the edge-based architecture. We observed that there is an opportunity to contribute with a cloud service that enables remote control and configuration of a home-based gateway in a personalized and flexible way.

4 The Proposed Solution The proposed solution illustrated in Fig. 2 is based on 3 layers architecture presented in Section 2. • Cloud server: a trusted cloud used by the hospital to offer a configuration service, a cloud storage and a monitoring dashboard. • Home-based Gateway: Smart gateway in the Edge layer which offer several services that depend on the medical staff configurations. • Medical devices: Considered as simple data publishers. The configuration service gives the possibility to take in consideration different situations of patients in order to reduce the unnecessary use of the gateway resources and to offer a remote personalized configuration.

Cloud Service for Edge Configuration in Home-Based Healthcare Context

435

Fig. 2. The proposed architecture based on edge computing

In the following sub-sections, we start with presenting the data model then we describe the main services of the Edge layer configuration. 4.1

Data Model

The communication between the cloud layer and the Edge layer is based on publish/subscribe pattern. Publish-subscribe messaging systems support data-centric communication and have been widely used in IoT systems. With the publish-subscribe pattern, the exchange of messages between clients is ensured using a broker that manages topics and sub-topics. A publisher on a given topic can send messages to other clients acting as subscribers to the topic without the need to know about the existence of the receiving clients [13]. To organize the topics and sub-topics in both local and cloud brokers, we propose a common meta-model presented in Fig. 3. The meta-model is used to define two categories of operations: • Collection: The collection of time-series data sensed by the medical devices. • Configuration: The definition of the gateway parameters by the medical staff. For each category, there are corresponding proprieties. First, the proprieties of the collection are the names of the collected data such as temperature and heart beats. Second, the proprieties of configuration are the status of each concerned data and the time range for both local and cloud storage. For example, to disable the collection of temperature in the gateway1, the cloud broker sends the following message: Configuration/gateway1/temperature/False “False” is the message value that will be received by the client subscriber of the topic: “Configuration/gateway1/temperature”

436

I. Ben Ida and A. Jemai

Fig. 3. Data Meta-model

In the other side, To publish a body temperature value equal to 37° which is collected by the gateway1, the message is: Collection/gateway1/temperatue/37 4.2

Cloud Layer Services

The cloud layer services ensure the flexibility of the configuration and the monitoring of the gateway. a. Cloud Broker: The cloud broker organizes the data exchange between the home-based gateways and the cloud server. It uses the topics defined by the medical staff and it respect the predefined data model. We use for the implementation the Mosquitto server and the client implementations of the MQTT protocol [14]. MQTT is a publish/subscribe protocol with low network overhead which can be implemented also on constrained devices with limited resources. b. Cloud Middleware: The cloud middleware is based on Node Js [15]. It handles the requests of configuration from the cloud configuration service in real-time. It publishes the submitted configurations as messages to the corresponding topics. The private key is used by the cloud middleware to decrypt the received encrypted data. c. Configuration service: The configuration service is a web application offered to the medical staff as an entry point on cloud to manage the following functions: • Local storage: The medical staff can modify the interval of storing data in the local database, this reduces the unnecessary data storage and processor usage in the gateway. • Devices status: The cloud service gives the possibility of adding and removing the medical devices by a simple modification of their status.

Cloud Service for Edge Configuration in Home-Based Healthcare Context

437

• Cloud storage: The storage strategy depends on the medical staff specification, they can remotely modify the interval of sending data to the cloud. The remote modification process using the configuration service is illustrated in Fig. 4.

Fig. 4. Sequence diagram of configuration process.

d. Storage service: To support the storage of the big data generated by different devices continuously, a No-SQL time series database is requested in the cloud server. e. Monitoring dashboard: The stored data are accessible via a dashboard which display the collected data depending on the user access role. 4.3

Edge Layer Services

The gateway ensures a continuous monitoring of physical parameters of the patient’s health. Several functions are embedded in the proposed gateway such as receiving and saving the devices data, sending data to the cloud and ensuring data encryption. a. Local Broker: The local broker uses the predefined data model to manage the topics. It is based on Mosquitto server and uses the MQTT protocol to receive the patient data from the devices and the configuration data from the cloud server. b. Local middleware: The local publish/subscribe middleware handles the requests from the cloud configuration service and the collected data with low use of resources. The main functionalities of the local middleware are:

438

I. Ben Ida and A. Jemai

• Subscription to the cloud configuration topics. • Sending the collected data to the hospital cloud with the predefined configurations. • Subscription to the local broker and saving the received data with the predefined configuration. • Encryption of the received patient data. c. Local Database: The local database supports saving No-SQL time series data which are generated by the home-based devices. The main role of the local storage is to prevent the loos of the patients data in case of internet interruption.

5 Conclusion In this paper, we present the benefits of edge computing in healthcare context and we propose a home-based solution that enable the remote configuration of the edge layer. The solution combines the benefits of cloud services and IoT technologies. The proposed configuration service gives to medical staff the possibility to customize the monitoring of patient data which optimize the edge resources use such as memory and energy. A data model is proposed to ensure the interoperability between the local and cloud processing. In future works, we will focus on the applicability of context-aware and scale-up techniques at the Edge level.

References 1. Ghanavati, S., Abawajy, J.H., Izadi, D., et al.: Cloud-assisted IoT-based health status monitoring framework. Cluster Comput. 20, 1843 (2017) 2. Yang, G., et al.: A health-IoT platform based on the integration of intelligent packaging, unobtrusive bio-sensor, and intelligent medicine box. IEEE Trans. Ind. Inform. 10(4), 2180– 2191 (2014) 3. Braham, R., Douma, F., Nahali, A.: Medical body area networks: mobility and channel modeling. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, pp. 1–6 (2016) 4. Bahar Farahani, I., Firouzi, F., Chang, V., Badaroglu, M., Constant, N., Mankodiya, K.: Towards fog-driven IoT eHealth: promises and challenges of IoT in medicine and healthcare. Futur. Gener. Comput. Syst. 78, Part 2, 659–676 (2018) 5. Shi, W., Cao, J., Zhang, Q., Li, Y., Xu, L.: Edge computing: vision and challenges. IEEE Internet Things J. 3(5), 637–646 (2016) 6. Ren, J., Guo, H., Xu, C., Zhang, Y.: Serving at the edge: a scalable IoT architecture based on transparent computing. IEEE Netw. 31(5), 96–105 (2017) 7. Rahmani, A.M., Gia, T.N., Negash, B., Anzanpour, A., Azimi, I., Jiang, M., Liljeberg, P.: Exploiting smart e-Health gateways at the edge of healthcare Internet-of-Things: a fog computing approach.Futur. Gener. Comput. Syst. 78, Part 2, 641–658 (2018)

Cloud Service for Edge Configuration in Home-Based Healthcare Context

439

8. Negash, B., Gia, T.N., Anzanpour, A., Azimi, I., Jiang, M., Westerlund, T., Rahmani, A.M., Liljeberg, P., Tenhunen, H.: Leveraging Fog Computing for Healthcare IoT, pp. 145–169. Springer International Publishing, Cham (2018) 9. Ben Ida, I., Jemai, A., Loukil, A.: A survey on security of IoT in the context of eHealth and clouds. In: 2016 11th International Design & Test Symposium (IDT), Hammamet, pp. 25–30 (2016) 10. Xu, X., Huang, S., Feagan, L., Chen, Y., Qiu, Y., Wang, Y.: EAaaS: edge analytics as a service. In: 2017 IEEE International Conference on Web Services (ICWS), Honolulu, HI, pp. 349–356 (2017) 11. Khoi, N.M., Saguna, S., Mitra, K., hlund, C.: IReHMo: An efficient IoT-based remote health monitoring system for smart regions. In: 2015 17th International Conference on Ehealth Networking, Application & Services (HealthCom), Boston, MA, pp. 563–568 (2015) 12. Kasmi, M., Bahloul, F., Tkitek, H.: Smart home based on Internet of Things and cloud computing. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), Hammamet, pp. 82–86 (2016) 13. Hakiri, A., Berthou, P., Gokhale, A., Abdellatif, S.: Publish/subscribe-enabled software defined networking for efficient and scalable IoT communications. IEEE Commun. Mag. 53 (9), 48–54 (2015) 14. Light, R.A.: Mosquitto: server and client implementation of the MQTT protocol. J. Open Source Softw. 2(13), 265 (2017) 15. Node Js Homepage. https://nodejs.org/en/. Last accessed 9 Nov 2018

Least Squares Channel Estimation of an OFDM Massive MIMO System for 5G Wireless Communications Abdelhamid Riadi1 , Mohamed Boulouird1,2(B) , and Moha M’Rabet Hassani1 1

Instrumentation, Signals and Physical Systems (I2SP) Group, Faculty of Sciences Semlalia, Cadi Ayyad University, Marrakesh, Morocco [email protected], [email protected], [email protected] 2 National School of Applied Sciences of Marrakesh (ENSA-M), Cadi Ayyad University, Marrakesh, Morocco

Abstract. A Least Squares Channel Estimation (LSCE) method is designated for an Massive MIMO systems combined with Orthogonal Frequency Division Multiplexing (OFDM) system. In this paper, the OFDM technique is founded on pilot tones. The Mean Square Error (MSE) of the LSCE is described in which the estimated channel is computed first. Simulation shows that when we increase the Base Station (BS) antennas, the linear detectors such as MMSE and ZF provide a best performance of Bit Error Rate (BER). Keywords: Massive MIMO · OFDM · Channel estimation Least squares · MMSE detector · ZF detector

1

·

Introduction

Nowadays, in a world of great mobility, the speed and capacity of communication systems are essential elements in order to keep people from all over the world in communication. Synchronization is crucial in all wireless communication systems and especially in OFDM and Multiple Input Multiple Output (MIMO) OFDM, to produce a good data rate as well as quality of service [1]. The complexity of a network is an effective tool to analyze its structure [2] and provide a proper structure that facilitates the real system. Massive MIMO systems will become a promising solution technique for 5th generation (5G) cellular network; increasing the BS antennas and combining with the OFDM, Massive MIMO can support very high throughput and/or performance of the links as well as spectral efficiency [3]. The pilot contamination is a crucial issue that degrades the performance of Massive MIMO system, by accounting of the non orthogonal pilot sequences overlap with each other [4,5]. To follow the multichannel with double selection for a MIMO-OFDM an efficient approach is study, the non-linear intricate Muli-Support-Vector-Machine-Regression technicality [6]. In an OFDM system, like all wireless communications system, in all the time the data is misshapen, due to channel effect. Hence, to know the transmitted data with a low c Springer Nature Switzerland AG 2020  M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 440–450, 2020. https://doi.org/10.1007/978-3-030-21009-0_43

OFDM Massive MIMO System for 5G Wireless Communications

441

error probability, the channel must be estimated at the receiver [7]. Otherwise, the flat Rayleigh fading channel is using by parameter estimation of non data aided vector, generalized by Cramer Rao lower bound expressions [8]. Hence, the LS technique is widely used for channel estimation. The organization of this paper is as follow. Section 2, the system model is illustrated in the Uplink (Up) transmission, in which the received data is contaminated by a White Noise Additive Gaussian (AWGN). In Sect. 3, the LSCE is introduced. The MSE of the LSCE is presented in Sect. 4; in Sect. 5, linear detectors such as MMSE and ZF are discussed. Section 6 presents the simulation results. In the end of this paper, the conclusion is done in the Sect. 7.

2

Massive-MIMO OFDM System

Consider a Massive MIMO system in Up transmission from Nt terminals with single antennas to a single BS with Nr antennas. The considered system is presented in Fig. 1. It’s a Massive-MIMO-OFDM system with Nr and Nt receive and transmit antennas respectively. The length of sub-carriers and the cyclic prefix (CP) are defined by K and ν respectively. The CP is inserted on each transmit antenna to achieve a full OFDM symbol. In this paper, the CP is superior than the utmost multi-path delay [7,9].

Fig. 1. System model.

In the same way, at the reception the CP is removed on each receive antenna, taking for example the qth receive antenna, the received signal vector y q (n) is K × 1 and expressed as follow:

442

A. Riadi et al. q

y (n) =

Nt  r=1

q,r H r Hcir F X (n) + z q (n)

(1)

q,r From the Eq. 1, the circulant matrix Hcir has a first column defined by T q,r T , 01×(K−L) ] , in addition to that L is the length of the channel impulse [h response and hq,r present L × 1 vector. The OFDM vector that is transmitted on each transmit antenna is defined by X r (n) with K × 1 dimension, r and n are index of the number of transmit antenna and time respectively. as shown in the Fig. 3 and z q (n) is additive Gaussian noise at Time Index (TI) n with zero mean and variance of σn2 . Moreover, the unitary DFT matrix the dimension K × K is presented by F; from √ the eigenvalue decomposition of the circulant T q,r = FH diag{ KF[hq,r , 01×(K−L) ]T }F [9]. Finally the FFT matrix becomes Hcir of the received signal y q (n) is given as follow:

q

Y (n) =

Nt 

√ T diag{ KF[hq,r , 01×(K−L) ]T }

r=1

×X r (n) + Ξ q (n)

(2)

where Ξ q (n) = Fz q (n).

3

Massive-MIMO Estimation

Based on the same system presented in Fig. 1, the LSCE scheme is presented. Then, (2) can be written as: Y q (n) =

Nt 

diag{X r (n)}Fhq,r + Ξ q (n)

(3)

r=1

√ From the Eq. 3, F is K × l of F, where l is the 1st column of F. noting r (n) = diag{X r (n)}. Hence, the Eq. 3 becomes: Xdiag Y q (n) =

Nt 

r Xdiag (n)Fhq,r + Ξ q (n)

(4)

r=1

Furthermore, in this work the training of all OFDM symbols is done at maximum value g and the TI is n ∈ {0, · · · , g − 1} , we consider the data model: Y q = Ahq + Ξ q T

T

(5) T

T

where Y q = [Y q (0), · · · , Y q (g − 1)]T , Ξ q = [Ξ q (0), · · · , Ξ q (g − 1)]T , ⎡ ⎤ 1 Nt (0)F · · · Xdiag (0)F Xdiag ⎢ ⎥ .. .. A=⎣ ⎦ . . 1 Nt Xdiag (g − 1)F · · · Xdiag (g − 1)F

(6)

OFDM Massive MIMO System for 5G Wireless Communications T

443

T

and hq = [hq,1 , · · · , hq,N t ]T . The LSCE technique minimize the noise defined in Eq. 5, basing on the cost ˆq function (Eq. 7), to obtain the estimated channel noted by h ˆ q ) = ||Y q − Ah ˆ q ||2 J(h ˆ q )H (Y q − Ah ˆq) = (Y q − Ah H

H

H

(7) H

ˆq − h ˆ q AH Y q + h ˆ q AH Ah ˆq = Y q Y q − Y q Ah ˆ q variable, In the next taking the derivation of the Eq. (7) relative to h ˆq) ∂J(h ˆ q )∗ ) = 0 = 2 ∗ (−(AH Y q )∗ + (AH Ah ˆq ∂h

(8)

ˆ q = AH Y q and the solution of the LSCE, is given by Finally, we have AH Ah the following expression: ˆ q = A+ Y q (9) h where A+ is the pseudo-inverse that equal to (AH A)−1 AH if gK  LNt . Because rank(A) = min(gK, LNt ), the necessary and sufficient condition to have unique LSCE is gK  LNt . This LS method presents a low complexity and a high simplicity, in addition to that also taking the information about the channel and the noise are not necessary [9–11].

4

Mean Square Error of LS Estimator

This section presents, the MSE of LSCE, basing on estimated channel calculated above (Eq. (9)); the MSE ca be expressed as follow: 1 ˆ q − hq ||2 } E{||h LNt 1 E{||A+ Ξ q ||2 } = LNt H H 1 tr{A+ E(Ξ q Ξ q )A+ } = LNt

M SE =

(10)

H

Where E(Ξ q Ξ q ) = σn2 . Then the MSE become: M SE =

σn2 tr{(AH A)−1 } LNt

(11)

Basing on the same proofs as in [7,10]. To obtain the minimum MSE of the LSCE in such a way the power P devoted for training is fixed, we need AH A = PILNt . Hence, the (M SEmin ) is expressed by the following equation: M SEmin =

σn2 P

(12)

444

5

A. Riadi et al.

Linear Detectors

The channel estimate for the uplink is done at the base station by letting all users send Pilot Sequences (PS). The time required for transmission of the uplink pilot is independent of the BS antennas number. To estimate the channel of each users in their cells the BS use these PS. Then, the BS use the estimated channel to detect the uplink data. The performance of linear detectors such as ZF and MMSE is evaluated [12,13]. Linear MIMO detectors are based in general on a multiplication of the received signal by T (Fig. 2): d = Ty (13) In this case, the symbol T define the linear transformation matrix in which presented according to different specification [13]. The Fig. 2 shows the basic principle of MIMO linear detectors.

Fig. 2. Basic principle of MIMO linear detectors.

5.1

Zero Forcing Detector

The ZF is a linear detector used for focusing the interference to zero. Accordingly, the noise level can increase [12–14]. The linear transformation matrix is given by: ˆ+ (14) TZF = H ˆ 1, · · · , h ˆq, · · · , h ˆ Nr ]T satisfies ˆ −1 H ˆ H , the matrix H ˆ + = (H ˆ H H) ˆ = [h where H Nr > Nt and a complete column rank of Nt . 5.2

Minimum Mean Square Error Detector

From the Eq. (13), T can be defined for MMSE detector. Hence, the main goal is to minimize the MSE between the actual transmitted signal and the received signal multiplied by T as shown below T [12–14]. TM M SE = arg min E(||X − TM M SE Y ||22 ) TM M SE

(15)

OFDM Massive MIMO System for 5G Wireless Communications

445

Finally, transformation TM M SE can be shown as: ˆ + 2σ 2 I)−1 H ˆH ˆHH TM M SE = (H n

(16)

where σn2 is the noise power and Y = [Y 1 , · · · , Y q , · · · , Y Nr ]T and X = [X , · · · , X r , · · · , X Nt ]T . 1

6

Simulation Results

6.1

Simulation Setup

In this part, a Massive MIMO system is considered to evaluate the performance of the LSCE and ZF/MMSE detectors in a different ways. This is done under the variants of the following scenarios: – – – – – –

Massive MIMO System (Nt × Nr ): 50 × 100, 50 × 150, 50 × 200 and 50 × 250. Modulation: 64-QAM. OFDM Subcarrier: 512. Channel: Rayleigh Fading. Noise: AWGN with (0,σn2 ) mean and variance respectively. Number of channel taps between each transmit-receive antenna(L − tap): 5.

6.2

Simulation Results

Figure 3 shows the resource mapping of pilot and data OFDM subcarriers for each transmit antenna for all Nt antennas of Massive MIMO system. Figure 4 presents the estimated channel defined by using the LSCE discussed in Sect. 3 compared with the true channel. The OFDM Massive MIMO system based on LSCE has been focused to enhance the estimation performance; comparing True channel with Estimated channel, it is clear that the power of the two channels to little pretends confuses on all the subcarrier OFDM. In addition to that, the MSE of LS channel estimation in Fig. 5 shown the best performance in function of Signal to Noise Ratio (SNR). In the case when the number of antenna at the transceiver equal to 50 (Fig. 6), the BER of the MMSE detector decrease more then ZF detector over the range of SNR. But its performance is not very important. Therefore, let us increase the number of antenna at the receiver from the 50 to 250, Figs. 7 and 8 shows the performance curve of ZF and MMSE detectors comparison of various antennas configuration system in Massive MIMO enumeration in (50 × 100), (50 × 150), (50 × 200), (50 × 250). The BER is decreased more if the number of antennas is increased in the BS for the MMSE detector or ZF detector. Table 1 shows the performance of ZF and MMSE detectors with Nt = 50 antennas of each terminals for the SN R = 6 dB, it is clear that the ZF and MMSE detector have a same performance in term of BER on the range of SNR presented.

446

A. Riadi et al.

Fig. 3. OFDM Subcarrier.

Fig. 4. Power of OFDM subcarrier for true channel and estimated channel. Table 1. A comparative table for MMSE and ZF detectors at SN R = 7 dB for (50 × 100), (50 × 150), (50 × 200)), (50 × 250) Massive MIMO antennas. Nr

ZF detector

MMSE detector −

100 BER = 91.93 × 10 4 BER = 92.19 × 10− 4 150 BER = 61.2 × 10− 5

BER = 61.2 × 10− 5

200 BER = 78.13 × 10− 6 BER = 78.13 × 10− 6 250 BER = 26.04 × 10− 6 BER = 26.04 × 10− 6

OFDM Massive MIMO System for 5G Wireless Communications

Fig. 5. Channel estimation error vs. SNR.

Fig. 6. BER vs. SNR using MMSE and ZF detectors.

447

448

A. Riadi et al.

Fig. 7. BER vs. SNR using MMSE detector.

Fig. 8. BER vs. SNR using ZF detector.

OFDM Massive MIMO System for 5G Wireless Communications

7

449

Conclusion

This paper evaluates the LSCE for an OFDM Massive MIMO systems. In the Up transmission the LS provides a good estimation of Rayleigh channel. In the case of the number of antennas at the transceiver are equal, the MMSE detector performs better then the ZF but its performance is not important. Therefore increasing the number of antennas at the BS favored more independent of the Rayleigh channel coefficient consequently the performance both the ZF and MMSE detectors is best over a small range of SNR.

References 1. Benzarti, M., Abdellaoui, Z.: Comparative study of frequency synchronization in SISO and MIMO-OFDM systems. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), 18–20 December 2016 2. Mokhlissi, R., Lotfi, D., El Marraki, M.: A theoretical study of the complexity of complex networks. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), 18–20 December 2016 3. Rusek, F., Persson, D., Lau, B.K., Larsson, E.G., Marzetta, T.L., Edfors, O., Tufvesson, F.: Scaling Up MIMO: opportunities and challenges with very large arrays. IEEE Signal Process. Mag. 30(1), 40–60 (2013) 4. Boulouird, M., Riadi, A., Hassani, M.M.: Pilot contamination in multi-cell massiveMIMO systems in 5G wireless communications. In: 2017 International Conference on Electrical and Information Technologies (ICEIT), Rabat, Morocco, 15–18 November 2017 5. Belhabib, A., Boulouird, M., Hassani, M.M.: Pilot contamination in a massive MIMO multi-cell environment of 5G communication systems. 4` eme Edition de la Journ´ ee de la Recherche Scientifique - JRS 2018, Marrakesh, Morocco, February 2018 6. Charrada, A.: Nonlinear complex M-SVR for LTE MIMO-OFDM channel with impulsive noise. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), 18–20 December 2016 7. Moradi, A., Bakhshi, H., Najafpoor, V.: Pilot placement for time-varying MIMO OFDM channels with virtual subcarriers. Commun. Netw. 3(1), 31–38 (2011) 8. Salem, M., Bouall`egue, A., Jarboui, S.: Vector parameter estimation over flat Rayleigh fading channel: Cramer-Rao lower bound generalized expressions. In: 2016 7th International Conference on Sciences of Electronics, Technologies of Information and Telecommunications (SETIT), 18–20 December 2016 9. Barhumi, I., Leus, G., Moonen, M.: Optimal training design for MIMO OFDM systems in mobile wireless channels. IEEE Signal Process. Mag. 51(6), 1615–1624 (2003) 10. Tung, T.-L., Yao, K., Hudson, R.E.: Channel estimation and adaptive power allocation for performance arid capacity improvement of multiple-antenna OFDM systems. In: IEEE Third Workshop on Signal Processing Advances in Wireless Communications(SPAWC 2001), Taiwan, China, 20–23 March 2001

450

A. Riadi et al.

11. Youssefi, M.A., El Abbadi, J.: Pilot-symbol patterns for MIMO OFDM systems under time varying channels. In: 2015 International Conference on Electrical and Information Technologies (ICEIT), Marrakech, Morocco, 25–27 March 2015 12. Riadi, A., Boulouird, M., Hassani, M.M.: An overview of massive-MIMO in 5G wireless communications. In: Colloque International TELECOM 2017 & 10`emes JFMMA, EMI - Rabat, Morocco, 10–12 Mai 2017 13. Yang, S., Hanzo, L.: Fifty years of MIMO detection: the road to large-scale MIMOs. IEEE Commun. Surv. Tutorials 17(4), 1941–1988 (2015) 14. Rajeev, P., Prabhat, A., Norsang, L.: Sphere detection technique: an optimum detection scheme for MIMO system. Int. J. Comput. Appl. 100(2), 975–8887 (2014)

Author Index

A Abdelhamid, Chafai, 124, 148 Abdelmalek, Raja, 24 Abid, Nabiha Ben, 373 Amoud, H., 302 Aouaouda, Sabrina, 240 Atia, Salim, 89 B Badis, Afef, 156 Bchir, Yosr, 204 Ben Hafaiedh, Imene, 350 Ben Hmida, Fayçal, 49 Ben Ida, Imen, 430 Ben Mansour, H., 167 Benahmed, Khelifa, 281 Benmerzoug, Amine, 400 Bennour, S., 254 Bensoussan, Alain, 185 Benzarti, Faouzi, 24 Besbes, Kamel, 101 Bhar, Jamila, 290 Boksmati, S., 302 Bouani, Faouzi, 230 Boujmil, Mohamed Habib, 156 Boujnah, Noureddine, 337 Boulouird, Mohamed, 440 Bounaama, Fateh, 281 Bousbia-Salah, Mounir, 14 Bousselmi, Souha, 37, 77 Bouziri, Hend, 412 C Chaarabi, L., 167 Chakaroun, Mohamad, 321, 327

Charrada, Anis, 382 Chehade, Rafic Hage, 194 Cherif, Adnen, 37, 77 Choumane, A., 302 Cresson, P. Y., 177 D Daghari, Marwa, 124, 148 Dufour, Arnaud, 185 E El Ghor, Hussein, 194 El-Khatib, Mohamad, 321 Ettaghzouti, Thouraya, 101 Ezzedine, Tahar, 390 F Farhi, Haider, 423 Feneniche, Wafa, 89 Flissi, Mustapha, 89 G Gaamouri, Sabah, 14 Garradhi, Karima, 101 Gdaim, Soufien, 204 Ghozzi, Rim, 66 González-Sentís, Manuel A., 185 Guermoudi, A. A., 177 H Hachaïchi, Yassine, 111 Hafsi, Sami, 230 Hage, Julia, 194 Hajlaoui, El Amjed, 220 Hamadeh, Nizar, 194

© Springer Nature Switzerland AG 2020 M. S. Bouhlel and S. Rovetta (Eds.): SETIT 2018, SIST 147, pp. 451–452, 2020. https://doi.org/10.1007/978-3-030-21009-0

452 Hamdi, Rachid, 14 Hamrouni, Chafaa, 148 Hamza, Djilali, 204 Hasnaoui, Salem, 271 Hassani, Moha M’Rabet, 440 Hassen, Néjib, 101 Hcine, Jamila, 350 J Jelassi, K., 167 Jemai, Abderrazak, 430 Jridi, Med Hchemi, 390 K Kaja, K., 302 Khatib, Mohamed Al, 327 Khoder, K., 302 L Lahbib, Younes, 111 Lahouar, Ali, 134 Lahouar, Samer, 66 Lasri, T., 177 Lassoued, Narjes, 337 Lirathni, Hamadi, 390 M Mabrouk, Nabila, 49 Mahi, A., 213 Mami, Sonia, 111 Mastouri, Mohamed Anis, 271 Meddeb, Aref, 361 Meguellati, Sabrina, 89 Messai, Abderraouf, 423 Mezaache, Salah Eddine, 89 Mnasri, Zied, 24 Mohammed, Khalid Khalil, 261 Moulahi, Med Hedi, 49 Moussaoui, Lotfi, 240 Mtibaa, Abdellatif, 204

Author Index N Nafkha, Kamel, 124 Nasr, Mouhamed Ben, 77 Nasri, Sonia, 412 Nasser, Najat, 312 O Ouldabbes, A., 177 R Rammal, Mohamed, 312 Rasheed, Abdalem A., 261 Rhouma, Aymen, 230 Riadi, Abdelhamid, 440 Righi, Ines, 240 Rouabah, Khaled, 89 S Sakli, Hedi, 124, 148 Saleh, Alaa, 321, 327 Saoud, Safa, 37, 77 Saoudi, Lalia, 400 Sayidmarie, Khalil H., 261 Sboui, N., 254 Serhal, Dina, 312 Sghaier, Amra, 361 Shehade, Mohamed, 327 Souani, Chokri, 66, 373 T Talbi, Mourad, 3 Touhami, Achouak, 281 Tounsi, Patrick, 185 V Vaudon, Patrick, 312 Z Zrelli, Amira, 390