Microwave and Wireless Synthesizer [2nd ed.] 9781119666004

Table of contents :
Authors
Preface
Notation
1. Loop Fundamentals
2. Almost All About Phase Noise
3. Special Loops
4. Loop Components
5. Digital PLL Synthesizers
6. A High-Performance Hybrid Synthesizer
Appendices
Bibliography

Citation preview

M I C R O WAV E A N D WIRELESS

SYNTHESIZERS Theory and Design SECOND EDITION

Ulrich L. Rohde Synergy Microwave Corp. Paterson NJ, USA Rohde

&Schwarz

Munich Germany University of the Armed Forces, Munich

Federal Republic of Germany

Enrico Rubiola FEMTO-ST Institute, CNRS and UBFC Besangon, France

Observatory THETA, Besangon, France INRiM, Torino, Italy

Jerry C. Whitaker Advanced Television Systems Committee Washington, DC, USA

W i l e y

This second edition rst published 2021 ©2021 John Wiley &Sons, Inc. Edition History John Wiley &Sons (Je, 1997)

All rights reserved. No part of this publication may be reproduced, stored in aretrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as permitted by law. Advice on how to obtain permission to reuse material from this title is available at http://www.wiley.com/go/permissions. The right of Ulrich L. Rohde, Enrico Rubiola, and Jerry C. Whitaker to be identi ed as the authors of this work has been asserted in a c c o r d a n c e w i t h l a w.

Registered Of ce

John Wiley &Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA Editorial Of ce 111 River Street, Hoboken, NJ 07030, USA

For details of our global editorial of ces, customer services, and more information about Wiley products visit us at wwv/.wiley.com, Wiley also publishes its books in avariety of electronic formats and by print-on-demand. Some content that appears in standard print versions of this book may not be available in other formats. Limit of Liahility/Disclaiiner of Warranty While the publisher and authors have used their best efforts in preparing this work, they make no representations or warranties with

respect to the accuracy or completeness of the contents of this work and speci cally disclaim all warranties, including without limitation any implied warranties of merchantability or tness for aparticular purpose. No warranty may be created or extended by sales representatives, written sales materials or promotional statements for this work. The fact that an organization, website, or product is referred to in this work as acitation and/or potential source of further information does not mean that the publisher and authors endorse the information or services the organization, website, or product may provide or recommendations it may make. This work is sold with the understanding that the publisher is not engaged in rendering professional services. The advice and strategies contained herein may not be suitable for your situation. You should consult with aspecialist where appropriate. Further, readers should be aware that websites

listed in this work may have changed or disappeared between when this work was written and when it is read. Neither the publisher nor authors shall be liable for any loss of pro t or any other commercial damages, including but not limited to special, incidental, consequential, or other damages. Library of Congress Cataloging-in-Puhlication Data Names: Rohde, Ulrich L., author. IRubiola, Enrico, 1957- author. I Whitaker, Jemy C., author. Title: Microwave and wireless synthesizers: theory and design /Ulrich L.

Rohde, Synergy Microwave Corp., Paterson NJ, USA, Rohde &Schwarz, Munich, Germany, University of the Armed Forces, Munich Federal Republic of Germany, Enrico Rubiola, FEMTO-ST Institute, CNRS and UBFC,

Besanqon, France, Observatory THETA, Besanqon, France, INRiM,Torino, Italy, Jerry C. Whitaker, Advanced Television Systems Committee, Washington, DC, USA. Description: Second edition. IHoboken, NJ :John Wiley &Sons, Inc., 2021. IIncludes bibliographical references and index.

Identi ers: LCCN 2020020727 (print) ILCCN 2020020728 (ebook) IISBN 9781119666004 (cloth) IISBN 9781119666097 (adobe pdf) IISBN 9781119666110 (epub) Subjects: LCSH: Frequency synthesizers-Design and construction. I Phase-locked loops. IDigital electronics. IMicrowave circuits-Design and construction. IRadio frequency. Classi cation: LCC TK7872.F73 R62 2021 (print) ILCC TK7872.F73 (ebook) IDDC 621.3815/486-dc23

LC record available at https://lccn.loc.gOv/2020020727

LC ebook record available at https://lccn.loc.gOv/2020020728 Cover design by Wiley Cover image: Courtesy of Ulrich L. Rohde Set in 9/1 Ipt, TimesLTStd by SPi Global, Chennai, India

fi

fi

fi

fi

fi

fi

fi

fi

fi

fi

SKY10025794_032421

Microwave and Wireless Synthesizers: Theory and Design, Second Edition. Ulrich L. Rohde, Enrico Rubiola, and Jerry C. Whitaker.

©2021 John Wiley &Sons, Inc. Published 2021 by John Wiley &Sons, Inc.

CONTENTS

Author Biography

xii

Preface

X V I

Important Notations

X X

1Loop Fundamentals

1

Introduction to Linear Loops /1 1-2

Characteristics of aLoop /3

1-3 1-4

Digital Loops /7 Type 1First-Order Loop /10

1-5

Type 1Second-Order Loop /12

1-6

Type 2Second-Order Loop /20 1-6-1

1-7

Transient Behavior of Digital Loops Using Tri-state Phase Detectors /22

Type 2Third-Order Loop /27 1-7-1

Transfer Function of Type 2Third-Order Loop /28

1-7-2

FM Noise Suppression /35

Higher-Order Loops /36 1-9

Fifth-Order Loop Transient Response /36 Digital Loops with Mixers /40

1-10

Acquisition /44 Example 1/48 1-10-1 Pull-in Performance of the Digital Loop /49 1-10-2 Coarse Steering of the VCO as an Acquisition Aid /52 1-10-3 Loop Stability /54 References /62

Suggested Reading /62 2Almost 2-1

all

About

Phase

Noise

65

Introduction to Phase Noise /65 2-1-1

The Clock Signal /65

2-1-2

The Power Spectral Density (PSD) /68

2-1-3

Basics of Noise /71

2-1-4

Phase and Frequency Noise /78 V

Vi

CONTENTS 2-2

2-3

2-4

2-5

The Allan Variance and Other Two-Sample Variances /88 2-2-1

Frequency Counters /89

2-2-2

The Two-Sample Variances AVAR, MVAR, and PVAR /94

2-2-3

Conversion from Spectra to Two-Sample Variances /96

Phase Noise in Components /100 2-3-1

Ampli ers /100

2-3-2

Frequency Dividers /104

2-3-3

Frequency Multipliers /112

2-3-4

Direct Digital Synthesizer (DDS) /117

2-3-5

Phase Detectors /128

2-3-6

Noise Contribution from Power Supplies /132

Phase Noise in Oscillators /133 2-4-1

Modern View of the Leeson Model /134

2-4-2

Circumventing the Resonator’s Thermal Noise /144

2-4-3

Oscillator Hacking /146

The Measurement of Phase Noise /153 2-5-1

Double-Balanced Mixer Instruments /154

2-5-2

The Cross-Spectrum Method /166

2-5-3

Digital Instruments /171

2-5-4

Pitfalls and Limitations of the Cross-Spectrum Measurements /180

2-5-5

The Bridge (Interferometric) Method /187

2-5-6

Artifacts and Oddities Often Found in the Real World /190

References /193

Suggested Readings /197 3Special Loops

201

3-1

Introduction /201

3-2

Direct Digital Synthesis Techniques /201 3-2-1

AFirst Look at Fractional N/202

3-2-

Digital Waveform Synthesizer / 2 0 3

3-2-

Signal Quality /220

3-2-

Future Prospects /235

3-3

Loops with Delay Line as Phase Comparators /236

3-4

Fractional Division VSynthesizers /237 3-4-1

Example Implementation /240

3-4-2

Some Special Past Patents for Fractional Division NSynthesizers /253

References /255

Bibliography /256

Fractional Division NReadings /256 259

4-1

I n t r o d u c t i o n t o O s c i l l a t o r s a n d T h e i r M a t h e m a t i c a l Tr e a t m e n t / 2 5 9

4-2

The Colpitts Oscillator /259

s

fi

2

4

3

4Loop Components

CONTENTS

4-2-1

Linear Approach /260

4-2-2

Design Example for a350 MHz Fixed-Frequency Colpitts Oscillator /269

4-2-3

Va l i d a t i o n C i r c u i t s / 2 8 2

4-2-4

Series Feedback Oscillator /314

4-2-5

2400 MHz MOSFET-Based Push-Pull Oscillator /319

4-2-6

Oscillators for IC Applications /336

4-2-7

Noise in Semiconductors and Circuits /337

4-2-8

Summary /339

4-3 Use of Tuning Diodes /339 4-3-1 Diode Tuned Resonant Circuits /340 4-3-2 Practical Circuits /344 4-4 Use of Diode Switches /345

4-4-1 Diode Switches for Electronic Band Selection /346

4-4-2 Use of Diodes for Frequency Multiplication /347 4-5 Reference Frequency Standards /351 4-5-1 Specifying Oscillators /351 4-5-2 Typical Examples of Crystal Oscillator Speci cations /352 4-

Mixer Applications /354

4-

Phase/Frequency Comparators /357 4-7-

Diode Rings /357

4-7-

Exclusive ORs /358

4-7-3 Sample/Hold Detectors /362 4-7-4 Edge-Triggered JK Master/Slave Flip-Flops /368 4-7-5 Digital Tri-State Comparators /369 4-8

Wideband High-Gain Ampli ers /378 4-8-1

Summation Ampli ers /378

4-8-2

Differential Limiters /382

4-8-3

Isolation Ampli ers /382

4-8-4 Example Implementations /387 4-9

Programmable Dividers /393 4-9-1 Asynchronous Counters /393 4-9-2 Programmable Synchronous Up-/Down-Counters /394 4-9-3

Advanced Implementation Example /405

4-9-4

Swallow Counters/Dual-Modulus Counters /407

4-9-5 Look-Ahead and Delay Compensation /411 4-10 Loop Filters /421 4-10-1 Passive RC Filters /421 4-10-2 Active RC Fi Iters /422 4-10-3 Active Second-Order Low-Pass Filters /423 4-10-4 Passive LC Filters /426

4-10-5 Spur-Suppression Techniques /427 4-11 Microwave Oscillator Design /430 4-11-1 The Compressed Smith Chart /432

fi

fi

fi

fi

1

2

7

6

4 - 11 - 2 S e r i e s o r P a r a l l e l R e s o n a n c e / 4 3 4

V I I

CONTENTS

V I I I

4-11-3 Two-Port Oscillator Design /435 4-12 Microwave Resonators /444 4-12-1 SAW Oscillators /445 4-12-2 Dielectric Resonators /445 4-12-3 YIG Oscillators /448 4 - 1 2 - 4 Va r a c t o r R e s o n a t o r s / 4 5 2 4-12-5 Ceramic Resonators /455 References /461

Suggested Readings /464 5Digital PLL Synthesizers 5-1

471

Multiloop Synthesizers Using Different Techniques /471 5-1-1

Direct Frequency Synthesis /471

5-1-2

Multiple Loops /473

5-2

System Analysis /477

5-3

Low-Noise Microwave Synthesizers /484 5-3-1

Building Blocks /485

5-3-2

Output Loop Response /489

5-3-3

Low Phase Noise References: Frequency Standards /490

5-3-4

Critical Stage /493

5-3-5

Time Domain Analysis /503

5-3-6

Summary /508

5-3-7

Two Commercial Synthesizer Examples /512

5-4

Microprocessor Applications in Synthesizers /518

5-5

Transceiver Applications /523

5-6

About Bits, Symbols, and Waveforms /526 5-6-1

Representation of aModulated RF Carrier /527

5-6-2

Generation of the Modulated Carrier /529

5-6-3

Putting It all Together /533

5-6-4

Combination of Techniques /535

Acknowledgments /537 References /540

Bibliography and Suggested Reading /540 6AHigh-Performance Hybrid Synthesizer 6-1

543

Introduction /543

6-2

Basic Synthesizer Approach /544

6-3

Loop Filter Design /548

6-4

Summary /556 Bibliography /557

AMathematical

Review

5 5 9

CONTENTS

A-1

Functions of aComplex Variable /559

A-2

Complex Planes /561

I X

A-2-1 Functions in the Complex Frequency Plane /565 A-3

Bode Diagram /568

A-4

Laplace Transform /582

A-4-1 The Step Function /583 A-4-2 The Ramp /584 A-4-

Linearity Theorem /584

A-4-

Differentiation and Integration /585

A - 4 - 5 I n i t i a l Va l u e T h e o r e m / 5 8 5 A - 4 - 6 F i n a l Va l u e T h e o r e m / 5 8 5

A-4-7 The Active Integrator /585 A-4-8 A-5

Locking Behavior of the pll /587

Low-Noise Oscillator Design /590 A-5-1 Example Implementation /590

A-6 A-7

Oscillator Amplitude Stabilization /594 Very Low Phase Noise VCO for 800 MHZ /602 References /605

BAGeneral-Purpose Nonlinear Approach to the Computation of Sideband Phase Noise in Free-Running Microwave and RF Oscillators B-1

Introduction /607

B-2

Noise Generation in Oscillators /608

B-3

Bias-Dependent Noise Model /609 B-3-1 Bias-Dependent Model /617 B-3-2 Derivation of the Model /617

B-4

General Concept of Noisy Circuits /619 B-4-1 Noise from Linear Elements /620

B-5

Noise Figure of Mixer Circuits /622

B-6

Oscillator Noise Analysis /624 Limitations of the Frequency-Conversion Approach /625

B-7

B-7-

Assumptions /626

B-7-

Conversion and Modulation Noise /626

B-7-3 Properties of Modulation Noise /626 B-7-4 Noise Analysis of Autonomous Circuits /627 B-7-5 Conversion Noise Analysis Results /627 B-7-6 Modulation Noise Analysis Results /627 B-8 B-9

Summary of the Phase Noise Spectrum of the Oscillator /628 Veri cation Examples for the Calculation of Phase Noise in Oscillators Using Nonlinear Te c h n i q u e s / 6 2 8

B-9-1 Example 1: High-2 Case Microstrip DRO /628 B-9-2 Example 2: 10 MHz Crystal Oscillator /629

1

4

3

fi

2

B-9-3 Example 3: The 1-GHz Ceramic Resonator VCO /630

607

X

CONTENTS

B-9-4 Example 4: Low Phase Noise FET Oscillator /632 B-9-5 Example 5: Millimeter-Wave Applications /636 B-9-6 Example 6: Discriminator Stabilized DRO /639 B-10 Summary /641 References /643

CExample of Wireless Synthesizers Using Commercial ICs

645

DMMIC-Based Synthesizers

665

D-1

Introduction /665

Bibliography /668 EArticles on Design of Dielectric Resonator Oscillator E-l

671

The Design of an Ultra-Low Phase Noise DRO /671 E-1-1 Basic Considerations and Component Selection /671 E-l-

Component Selection /672

E-l-

DRO Topologies /675

E-1 -4 Small Signal Design Approach for the Parallel Feedback Type DRO /677 E - l - 5 S i m u l a t e d Ve r s u s M e a s u r e d R e s u l t s / 6 8 3

E-l-6 Physical Embodiment /685 E-lE-l-8

Acknowledgments /685 Final

Remarks

/688

References /692

Bibliography /692 E-2

ANovel Oscillator Design with Metamaterial-MbBius Coupling to aDielectric Resonator /692 E-2-1

Abstract

/692

E-2-2

Introduction

/693

References /699

FOpto-Electronically Stabilized RF Oscillators F-l

F-2

Introduction /701 F-l-l

Oscillator Basics /701

F-l-2

Resonator Technologies /701

F-l-3

Motivation for OEO /704

F-1-4 Operation Principle of the OEO /704 Experimental Evaluation and Thermal Stability of OEO /705 F-2-1 Experimental Setup /705 F-2-2 Phase Noise Measurements /708

F-2-3 Thermal Sensitivity Analysis of Standard Fibers /709 F-2-4 Temperature Sensitivity Measurements /710 F-2-5 Temperature Sensitivity Improvement with HC-PCF /712

7

3

2

F-2-6 Improve Thermal Stability Versus Phase Noise Degradation /712

701

CONTENTS F-2-7

Xi

Passive Temperature Compensation /713

F-2-8 F-3

Improving Effective 2with Raman Ampli cation /714 Forced Oscillation Techniques of OEO /718 F-3-1 Analysis of Standard Injection-Locked (IL) Oscillators /718 F-3-2 Analysis of Self-Injection Locked (SIL) Oscillators /720 F-3-3 Experimental Veri cation of Self-Injection Locked (SIL) Oscillators /721 F-3-4 Analysis of Standard Phase Locked Loop (PLL) Oscillators /723 F-3-5 Analysis of Self Phase Locked Loop (SPLL) Oscillators /725 F-3-6 Experimental Veri cation of Self-Phase Locked Loop (SPLL) Oscillators /726 F-3-7 Analysis of Self-Injection Locked Phase Locked Loop (SILPLL) Oscillators /728

F-4

SILPLL Based X- and jV-Band Frequency Synthesizers /731 F-4-1 A-Band Frequency Synthesizer /732 F-4-

19" Rack-Mountable A-Band Frequency Synthesizer /737

F-5

Integrated OEO Realization Using Si-Photonics /742

F-6

Compact OEO Using InP Multi-Mode Semiconductor Laser /744

F-7

F-6-1

Structure of Multi-mode InP Laser /744

F-6-2

Multi-mode Laser and Inter-Modal RF Oscillation /745

F-6-3

Self-Forced Frequency Stabilizations /747

Discussions /752

Acknowledgments /753 References /754

GPhase Noise Analysis, then and Today G-1

Introduction /761

G-2

Large-Signal Noise Analysis /762

761

References /769

HANovel Approach to Frequency and Phase Settling Time Measurements on PLL Circuits

771

H-1

Introduction /771

H-2

Settling Time Measurement Overview /771

H-2-1 Theoretical Background of Frequency Settling Time /771 H-2-2 Frequency Settling Measurement in the Past /772 H-3

R&S FSWP Phase Noise Analyzer /774 H-3-1 Phase Noise Analyzer Architecture /774 H-3-2 Typical Test Setup for Settling Time Measurements /776

H-4

Frequency Hopping and Settling Time Measurements in Practice /776 H-4-

Trigger on Wideband Frequency Hopping Signals /776

H-4-2 Frequency and Phase Settling Time Measurement /777 H-5

Conclusion /780

fi

fi

7 8 3

fi

1

2

Index

AUTHOR BIOGRAPHY

ULRICH L. ROHDE

%

i

n

p r ,

in rii b-wtiu4k«vu Fttwtftkv k«xttri'i(

I

I

ft A

i ^

■?^v

■S,^i Member of the Faculty of the Institute for Technical Informatics, INF 3, Universitat der Bunderwehr, Munich, Germany. https://www.unibw,de/home-en. “I like solving problems in my eld of expertise where others have failed. Most people lack the drive or staying power to see things through.”

Prof. Dr.-Ing. habil. Dr. h.c. mult. Ulrich L. Rohde is apartner of Rohde &Schwarz, Munich, Germany; Chairman of Synergy Microwave Corp., Paterson, NJ; President of Communications Consulting Corporation; an honorary member of the Senate of the Armed Forces University Munich; honorary member of the Senate of the Brandenburg University of Technology Cottbus-Senftenberg; and past member of the Board of Directors of Ansoft Corporation, Pittsburgh, PA. Dr. Rohde serves as an honorary professor at IIT Delhi, full professor at Oradea Uni¬ versity, Romania, and visiting professor at Technical University, Munich, Germany. Prior to Honorary Professor of RF and Microwave Technologies at the University of Cottbus, Dr. Rohde was amember of the staff at George Washington University (1982) and as an adjunct professor at the University of Florida, Gainesville, teaching in the Electrical Engineering and Computer Sciences departments, gave numerous lectures worldwide regarding com¬ munications theory and digital frequency synthesizers.

fi

fi

X I I

X I I I

Dr, Rohde is an IEEE life Fellow, published more than 350 scienti c papers in professional journals and con¬

ferences, co-authored of 15 technical books and book chapters, and holds over 4dozen patents. Dr. Rohde has received anumber of awards, including:

Recipient of the 2019 Honorary Fellowship Award from lETE (Institution of Electronics and Telecommuni¬

cations Engineers), which recognizes the outstanding contributions for the applications of microwave theory and techniques.

Recipient of the 2019 IEEE CAS (Circuits and Systems Society) Industrial Pioneer Award. Recipient of the 2017 RCA (Radio Club of America) Lifetime achievement award for “For signi cant achievements and amajor body of work accomplished over alifetime that has advanced the art and science of radio and wireless technology.”

Recipient of the 2017 IEEE UFFC-S W. G. Cady Award for “Pioneering research, development, and com¬ mercialization of signal generating and processing devices for commercial and scienti c applications.” Recipient of the 2017 IEEE AP-S Distinguish achievement award for “Outstanding career achievement in the eld of antennas and propagation.”

Recipient of the 2016 IEEE MTT-S Applications Award for “Signi cant contributions to the development of low-noise oscillators.”

Recipient of the 2015 IFCS 1.1. Rabi Award, IEEE International Frequency Control Symposium &European Frequency and Time Forum.

Recipient of the 2015 IEEE Region 1Award for outstanding scienti c contributions and leadership in the design and implementation of sophisticated RF technologies. Recipient of the 2014 IFCS C. B. Sawyer Award recipient, IEEE International Frequency Control Sympo¬ sium.

Dr. Rohde is amember of the following: Fellow Member of the IEEE; Member of the IEEE Technical

Committee for HF, VHF, and UHF Technology MTT-17; Member of the IEEE Signal Generation and Fre¬ quency Conversion MTT-22; Member of the Board of Trustees Fraunhofer Gesellschaft (EMFT) for Modular

Solid State Technology, Member of the Board of Trustees of the Bavarian Academy of Science, and Hon¬ orary Member of the Academy of Science, all in Munich; ETA KAPPA NU (EPSILON SIGMA CHAPTER)

Honor Society; Executive Association of the Graduate School of Business-Columbia University, New York; The Armed Forces Communications &Electronics Association; Fellow of the Radio Club of America; and

former Chairman of the Electrical and Computer Engineering Advisory Board at New Jersey Institute of Technology.

In 2006, Dr. Rohde was honored as Microwave Legend by Microwave &RF Magazine; the selection was based on global voting. In 2009, Dr. Rohde was selected in the list of Divine Innovators of November 2011, Microwave Journal. Based on Dr. Rohde’s ve-decade of scienti c creativity and pioneer contributions in the eld of microwave and antenna, IEEE has established three awards on his name:

●IEEE Ulrich L. Rohde Innovative Conference Paper Award on Antenna Measurements and Applications ●IEEE Ulrich L. Rohde Innovative Conference Paper Award on Computational Techniques in Electromagnet¬ i c s

●IEEE Ulrich L. Rohde Humanitarian Technical Field Award.

His hobbies including sailing, U.S. Merchant Marine Of cer, Master of Steam or Motor Vessels, photography,

fi

fi

fi

fi

fi

fi

fi

fi

and ham radio (NlUL).

fi

fi

AUTHOR BIOGRAPHY

Xiv AUTHOR BIOGRAPHY Enrico Rubiola

Enrico Rubiola is aprofessor at the Universite de Franche-Comte and aresearcher with the Department of Time and Frequency of the CNRS FEMTO-ST Institute, Besanqon, France, and an associated researcher at INRiM, the Italian institute of primary metrology in Torino. Formerly, he was aprofessor at the Universite Henri Poincare, Nancy, France, and an assistant professor at the Politecnico di Torino. He was also aguest professor at the Universita di Parma, Italy, and aguest scientist at the NASA/CALTECH Jet Propulsion Laboratory. After graduating in Electronic Engineering at the Politecnico di Torino in 1983, Enrico received aPh.D. in Metrology from the Italian Minister of University and Research, Roma (1989), and aSc.D. degree from the Uni¬ versite de Franche-Comte in 1999.

Enrico’s primary interests are high-spectral purity oscillators from low RF to optics, phase noise, amplitude noise, noise in digital systems, general time and frequency metrology, frequency synthesis, spectral analysis, wavelet statistics (Allan variance and similar variances), microwave photonics, precision electronics form dc to microwaves, and precision instrumentation. He designed the frequency synthesis of the FEMTO-ST cryogenic

oscillators, achieving 3x10“'^ stability. Enrico is known for innovative instruments for AM/PM noise measure¬ ment with ultimate sensitivity, -210 dBc/Hz and below, for the theory underlying the “Leeson effect," for theory of modern frequency counters (H, A, and Q), for dedicated signal-processing methods, and for hacking oscil¬ lators from the phase noise plots. In 2018, Enrico received the IEEE W. G. Cady Award “for groundbreaking contributions” in the

eld.

In 2012, Enrico founded the Oscillator IMP project, aplatform lor the measurement of short-term frequency stability and AM/PM noise of oscillators and related components. In 2013, he founded the European Frequency and Time Seminar (http://efts.eu), ano-pro t crash course in Time and Frequency, and he has been chairing and running it since.

Awealth of articles, reports, conference presentations, and lectures for PhD students and young scientists are available on the Enrico’s home page http://rubioIa.org. Jerry C. Whitaker

Jerry C. Whitaker is Vice President for Standards Development at the Advanced Television Systems Committee;

fi

fi

Washington, D.C. Mr, Whitaker supports the work of the various ATSC technology and specialist groups and assists

AUTHOR

BIOGRAPHY

XV

in the development of ATSC Standards and related documents. He currently serves as Secretary of the Technol¬ ogy Group on Next Generation Broadcast Television. He is also closely involved in work relating to educational programs.

Mr. Whitaker joined ATSC in 2000 and has participated in all facets of the organization, from development of standards and recommended practices to representing ATSC at various organizations and venues. Prior to joining the ATSC, Mr. Whitaker headed the publishing company Technical Press, based in Morgan Hill, Calif. Mr. Whitaker is the author and editor of more than 35 books on technical topics, including: The SBE Broadcast Engineering Handbook', The Standard Handbook of Video and Television Engineering, 4th ed.; NAB Engineering Handbook, 9th ed.; DTV Handbook, 3rd ed.; and The Electronics Handbook, 2nd ed. He is the coauthor with Dr. Rohde of Communications Receivers, 2nd, 3rd, and 4th ed,

Mr. Whitaker is aLife Fellow the Society of Broadcast Engineers and aFellow of the Society of Motion Pic¬ ture and Television Engineers. He has served as aBoard member and Vice President of the Society of Broadcast Engineers. He served as Chair of the NAB Broadcast Engineering Conference Committee from 1993 until 2000, and as Chair of the SMPTE Fall Technical Conference Program Committee from 2007 until 2013.

Mr, Whitaker was previously Editor, Editorial Director, and Associate Publisher of Broadcast Engineering magazine and Video Systems magazine. In aprevious life, he was Chief Engineer for radio stations KRED-AM and KPDJ-FM in Eureka, CA. He also

worked in radio and television news in Sacramento, CA, at KCRA-AM and KCRA-TV. His rst experience in broadcast engineering came at KERS-FM, the campus radio station at California State University, Sacramento. Mr. Whitaker twice received the Jesse H. Neal Editorial Achievement Award from the Association of Business

Publishers (ABP). He was also named “Educator of the Year” by the Society of Broadcast Engineers in 2002. His hobbies include building high-end vacuum tube audio ampli ers, restoring vintage radio broadcast hard¬ ware, and attending his children’s numerous school sporting events. Mr. Whitaker lives with his wife and daughters

fi

fi

(and two dogs) in Morgan Hill, CA.

P R E FA C E

Since 1997 when the rst edition of Microwave and Wireless Synthesizers was published, enormous progress has been made both in semiconductors (discrete devices and integrated circuits) and in fundamental technologies.

Speci cally, much better implementations of direct digital synthesis (DDS) architectures have helped to improve performance in anumber of key areas, notably reduced spurious elements and noise. Another critical part of a synthesizer is the voltage-controlled oscillator (VCO). New insights into its functionality have made the out-of-loop bandwidth much better. This also applies in the understanding of crystal and other high-2 based oscillators. These portions of the rst edition have been greatly enhanced and updated in this Second Edition. The radio frequency (RF)/microwave synthesizer text has also been expanded considerably, and the very impor¬ tant element of signal generation has been added, for example, arbitrary waveform generation and vector signal generation with very fast switching and high frequency resolution, and low noise/low spurious response.

TECHNOLOGY

A D VA N C E M E N T S

In order to properly set the stage for the Second Edition, some introductory comments are warranted on the topic of signal generation. Today, the trend is toward full digital generation of modulated RF signals, that is, the traditional analog VCO and phase-locked loop (PEL) are no longer always necessary. Currently, the required functionality and the subsequent digital-to-analog converter (DAC) are incorporated into large-scale application-speci c integrated

circuits (ASICs), The following two links are to datasheets of such devices*: http;//www.analog.com/media/en/ technical-documentation/data-sheets/AD9164.pdf and http://www.ti.com/lit/ds/symlink/dac38j84.pdf. From Analog Devices, the authors would like to thank John Morrissey and Ian Collins for their technical con¬ tribution and permission to use the Analog Devices material. Modern waveform synthesis is based on the approach that the DAC directly generates complex modulated signals, that is, subsequent analog modulation is no longer required. The eld programmable gate array (FPGA) generates the UQ modulation data as adigital data stream. High-speed serial interfaces are used to transmit the HQ data from the FPGA to the DAC, for example, in accordance with the JESD204B standard. To improve data ef ciency, only the baseband data is transmitted, not the sampled RF signal. To achieve the DAC’s nal sampling

rate, which at 2.5 x/,^ typically lies signi cantly above the sampling rate for baseband data, the baseband data is interpolated up to the DAC sampling rate (i.e., initially greatly oversampled). The carrier is generated by a numerically controlled oscillator (NCO). Today’s state-of-the-art phase accumulators have resolutions of 48 bit (or higher). If the NCO’s frequency resolution is insuf cient, the remaining frequency difference can be pre-calculated in baseband to achieve the required nal resolution. Adigital multiplier mixes the up-sampled baseband data with the NCO carrier to yield the digital modulated RF signal. This digital RF data is then fed into the RF DAC and output as an analog signal. The maximum frequency of the RF signal is limited by the Nyquist rate (DAC sampling rate). If the required RF frequency lies above this frequency, the signal can be up-converted with ascalar mixer. Overall, this results in very compact solutions for digitally generating RF signals without impairments. Addi¬ tional advantages are the option of very fast frequency hopping, including phase-locked frequency hopping, that is. *The hyperlinks in this book were functional at the time of publication. Some hyperlinks change over time, and so it may be necessary to search on the document lename to resolve amodi ed link.

fi

fi

fi

fi

fi

fi

fi

fi

fi

fi

fi

fi

X V I

X V I I

phase-accurate hopping back to the previous frequency. Because this method of signal generation is mathematically deterministic, clock synchronization allows very ef cient multiple-in/multiple-out (MIMO) signal generation. Digital synthesis has the disadvantage of inherent spurs (DAC nonlinearities, nite resolution, and power source timing in the DAC). Depending on the RF frequency, these spurs show up between 40 and 80dBc and need to be eliminated through appropriate frequency planning and ltering. Avery good clock source is necessary to achieve good phase noise in high-performance DACs. From asys¬ tem theory perspective, aDAC functions as afrequency divider, delivering very good phase noise values close to the carrier. Unlike traditional synthesizers, there is no “noise shoulder” in the region of the PLL bandwidth. Another advantage of an RF DAC compared with atraditional PLL is that the clock is generated with anarrowband, high-quality oscillator instead of atunable broadband VCO. AMAJOR

REVISION

Preparation of the Second Edition of Microwave and Wireless Synthesizers has included acomprehensive review of all of the original text and the addition of new updated text and illustrations. The new edition has been acollective effort among the authors. Jerry C. Whitaker took charge of the organization of the book, streamlining it for easy reading and attending to the details of publication, and Enrico Rubiola added his insightful knowledge of noise and noise contribution in circuits and system to the new edition. Another useful item we introduced is aclear mathe¬ matical method to treat the oscillators design for best noise performance, both at RF and microwave frequencies. This is of great value since most companies still prefer to design their oscillators or have them custom-optimized. In addition. Prof. Afshin Daryoush agreed to write atheoretical introduction about opto-electric synthesizer based on ber optic cable. Also, some of our joint publications on this topic are included. Rohde &Schwarz Munich provided some of the important material including photographs. They came from the Business Unit. Measurement and Test Equipment, speci cally from Mr. Pauly and Mr. Pointner. Also, our col¬ leagues at Synergy Microwave Corp., NJ, USA, helped with important contributions, speci cally Dr. Ajay Poddar. This book is based on theoretical and practical work done over many years, courses given at George Washington University 1983, very recent developments done at Synergy Microwave Corp., and mathematical treatments found in the Ansys HFSS circuit analysis program (which was supported by avariety of government contracts). We have used mostly software from Ansys Corp. and Keysight. There are other suppliers of similar high-performance software; however, we tend to have less access to those.

For individuals who are getting acquainted with oscillators and synthesizers, we strongly recommend purchas¬ ing the ARRL Handbook for Radio Communications, annually ptiblished by the American Radio Relay League, 225 Main Street, Newington, CT 06111. The chapter on oscillators and synthesizers is acomplete and well written rst-time introduction to this topic. Microwave and Wireless Synthesizers, Second Edition, is divided into six chapters beginning with Chapter 1 on loop fundamentals, which provides detailed insight into settling time and other characteristics of the loop. The clear differentiation between analog and digital loops has proven to be quite useful, and topics such as pull-in performance and acquisition are discussed in detail. This mathematically based presentation remains very much up to date.

Chapter 2outlines noise and spurious responses of the loops. The linear approach of oscillator phase noise is very detailed and walks the reader through all the important steps and contributions, both inside and outside the loop. We also look at the noise contribution of the various parts of the loop, such as frequency dividers, phase detec¬ tors, and even power supplies. Finally, the noise analysis of the entire system and its measurements are covered. We added clari cation and guidance to many parts. Here, modern IC design has helped to substantially improve the PLL.

fi

fi

fi

fi

fi

fl

fi

In Chapter 3we look at special loops. Here, the DDS technique—explained in detail—should prove most inter¬ esting to the reader. The fractional division Msynthesizer technology (FN) competes with DDS. Details regarding amixed approach are also shown in the appendices. Most of these FN loops can be found in cell phones and battery-operated two-way radios. The fractional division Nsynthesis principle is quite complex. This is best seen from the various patent applica¬ tions. Most RF and microwave companies now use this technique. The digital implementation of the accumulator and its compensating network has the greatest in uence on performance. This area is very exciting, but only major

fi

fi

P R E FA C E

P R E FA C E

houses will be able to alford the large-scale custom integrated circuits (LSls) and gate arrays that will solve these problems. The appearance of high performance fractional-/V synthesizer chips for low-power applications further illustrates this trend when cost and power consumption are an issue. However, when highest performance is more critical than cost and power consumption (e.g., test equipment applications), modern DDS system implementations clearly win. If the readers are interested in CMOS PLL Synthesizers, there is an interesting book by Keliu Shu and Edgar Sanchez-Sinencio, published by Springer. ISBN 0-387-23668-6. ©2005. Also we found aPhD disserta¬ tion on the topic, “Low Phase Noise CMOS PLL Frequency Synthesizer—Design and Analysis,” by Xinhua He, which can be found at: https://drum.lib.urnd.edu/bitstream/handle/1903/7337/umi-umd-4746.pdf?sequence=l& isAllowed=y. This book does not address foundry designs speci cally, but the two examples earlier (the book and the dis¬ sertation) require afoundry and provide many useful information about the topic. We are sure there are more and newer publications about Synthesizer CMOS IC-design. The book mentioned previously provides an interesting discussion on Sigma-Delta modulators. Chapter 4provides adetailed overview of loop components. Many practical circuit details are found in this chapter as it addresses low-noise oscillator design, including the use of linear CAD tools. The section on refer¬ ence frequency standards provides thorough insight into the design of crystal oscillators, which are avital part of synthesizers and which must provide both low aging and optimum phase noise. Other important components include mixers, phase/frequency discriminators, wideband high-gain ampli ers, programmable dividers, and loop lters. The microwave oscillator design section in Chapter 4is unique because it is the rst systematic evaluation of all aspects of microwave design techniques—including different resonators, such surface acoustic wave (SAW) oscillators, dielectric resonators, and ceramic resonators—and addresses the use of tuning diodes. We apply non¬ linear time domain analysis to calculate the oscillators, and we employ CAD tools to verify the oscillator noise performance. Chapter 5provides in-depth details about multiloop synthesizers. The section on microwave synthesizers deals with analysis, architectures, and trade-offs. Another unique section is the survey of critical stages and the examina¬ tion of their behavior. Microprocessors are used to optimize parts of the synthesizer architecture. Various techniques are found in synthesizer for military communication equipment. Chapter 6is dedicated to practical synthesizer examples, which combine the techniques outlined in previous chapters. The design of ahigh-performance hybrid synthesizer and the related performance measurement tech¬ niques enable the engineer to follow the various design steps and design rules. The core of this chapter teaches the reader to understand the essentials for success in the rst go-round of adesign. The appendices comprise eight sections. Appendix Aprovides amathematical overview for individuals who want to write their own CAD programs. Also, veri ed designs are provided for very low phase noise very high frequency (VHF) and ultra high frequency (UHF) oscillators. Analog Devices and others provide aset of free CAD tools, such as aSPICE version, to analyze many types of systems, for example: http://www.analog.com/LTSpice. Appendix Bis amathematical treatment of the nonlinear time domain approach to calculating phase noi.se in afree-running oscillator. This may be the rst complete treatment of its kind in abook and is based on recent publications by the renowned expert team of Synergy Microwave Corporation. Appendix Cis areprint of selected application notes provided by Analog Devices. This is useful to get afeel for the state-of-the-art synthesizer chips available and how to use them. Appendix Ddiscusses monolithic microwave integrated circuit (MMlC)-based synthesizers. Appendix Eis about advanced dielectric resonator oscillator (DRO) design. Appendix Fdescribes opto-electronic oscillator/synthesizer. Appendix Gdiscusses noise analysis, then and today. Appendix Hdescribes anovel approach to frequency and phase settling time measurements on PLL circuits. The following link documents some interesting frequency synthesizer approaches: https://www,nature.com/

fi

fi

fi

fi

fi

articles/ncomms3097.

fi

fi

X V i l l

PREFACE

XiX

CONTRIBUTIONS AND RESOURCES

Many renowned experts have contributed to the Second Edition of Microwave and Wireless Synthesizers. It is a concise collection of practical and theoretical information, and we are very grateful to have received input from many engineers interested in speci c applications. Anote of caution: Currently, it appears that the cleanest and best single or dual-loop synthesizers can be built by using custom made fractional-division NASIC chips. While many companies are competing for the PEL chip market, at the time of publication of this book, it appears that the performance of the Analog Devices family member PEL chips is leading the market. As technology changes, adifferent company may be leading tomorrow. We strongly advise athorough evaluation of all available chips (including how long those chips will be in stock) before committing to a nal design. Many come and go, and the lifetimes of these devices vary widely. Since adesigner needs to test awhole range of parameters for VCO/PEEs—including VCO tuning character¬ istics, RF power atness, PEE transients (both spurious and harmonic), and of course phase noise as well as added phase noise and burst phase noise—having the necessary signal analysis tools is essential. The recently introduced

FSWP signal analyzer from R&S® can handle all the needed measurements and more at afrequency range of 1MHz to 26.5 GHz and higher (e.g., 50 GHz). The FSWP, for example, can measure the phase noise of a1.8 GHz VCO/PEE in less than afew seconds covering 100-F frequency offset points. The R&S FSWP phase noise analyzer and VCO tester combines extremely low-noise internal sources and cross-correlation technology, delivering high sensitivity for phase noise measurements. As aresult, it takes just seconds to measure even highly stable sources such as those found in radar applications. Additional options such as pulsed signal measurements, additive phase noise (including pulsed) characterization, and integrated high-end signal and spectrum analysis make the analyzer aunique test instrument. Some additional interesting facts about Synthesizers can be found in U. E. Rohde, Matthias Rudolph, “RF/Microwave Circuit Design for Wireless Applications,” 2nd Ed., John Wiley &Sons. pp. 745-746, ISBN 1118431405, New York, 2013.

The completion of this new edition was an enjoyable project. We are very grateful to the various individuals who supported us throughout the work, speci cally Prof. Afshin Daryoush, Dr. Ajay Poddar, Anisha Apte from Synergy Microwave, and, of course, thanks to our publisher, John Wiley &Sons, Inc., for their continued support of Microwave and Wireless Synthesizers, Second Edition. Ulrich E. Rohde, Marco Island, FE, USA

fi

fi

fi

fl

Enrico Rubiola, Besan^on, France Jerry C. Whitaker, Morgan Hill, CA, USA

I M P O R TA N T N O TAT I O N S

Symbol

Meaning Loop pai'ameters of /ith-oider PLL Digital data values Ampli er gain Open-loop gain Amplitude of transfer function Amplitude response of network Bandwidth of input bandpass lter (Hz)

aj(i =I, 2,, n) A

Ms) A(co)

Bl

Noise bandwidth of PLL (Hz)

B(s)

Closed-loop gain of PLL

E(s)

Error function

F

Noise gure Frequency (Hz) Corner frequency of icker noise

fJi fc f,

Fourier frequency (sideband, offset, modulation, baseband frequency) Carrier frequency Instantaneous frequency Peak frequency deviation (Hz) Instantaneous frequency lluctuation Peak deviation of sinusoidal frequency modulation

m

fo lit) A / A/(r)

^./peak A /r e s

Residual FM

Fis). F(jw)

Transfer function of loop

G(s), G(jw)

Feedforward function (rad/s)

G„is)

Transfer characteristic of adivider

F(s), H{jco)

Feedback transfer function

K

Loop gain (rad/s)

lter

Phase detector gain before lock (V/rad) Phase detector gain factor (V/rad) Multiplier gain VCO gain factor (rad/s V)

Ke K m

K O

Shaper constant

K

dc Gain of PLL or velocity error coef cient (rad/s) An integer or an integer index on asequence 1.4 X10-2^ Ws/K

Fix)

Laplace transform of x

‘^ilj

Single-sideband phase noise to total signal power in a1-Hz bandwidth An integer Modulation waveshape An integer denoting frequency multiplication or division An integer

k

m

ni{t) M n

Loop order

n

n{t)

fi

fi

fi

X

fl

fi

fi

X

Noise voltage (V)

I M P O R TA N T N O TAT I O N S

Symbol

Meaning

nit)

Time average of noise An integer representing frequency division or multiplication Fourier ttansform of pulse waveshape Signal power (W) Power of single sideband Available signal power Quality factor of unloaded resonator Laplace transform complex variable Signal-to-noise ratio Signal-to-noise ratio in loop bandwidth One-sided spectral density of white noise (dB/Hz) Spectral density of frequency uctuations

V, V, Piico) P. P P

ssB ■s a v

2unl s=a+jco SNR

SNRi S o

/, and are the phases of the input and VCO output signals, respectively. Kg is the phase detector gain factor and has the dimensions of volts per radian. It will also be assured that the

VCO can be modeled as alinear device whose output frequency deviates from its free-running frequency by an increment of frequency. (1-2)

where V^ is the voltage at the output of the low-pass lter and A",, is the VCO gain factor, with the dimensions of rad/s per volt. Since frequency is the time derivative of phase, the VCO operation can be described as de,

A(u =

dt

(1-3)

= K X

With these assumptions, the PLL can be represented by the linear model shown in Figure 1-2. F(s) is the transfer function of the low-pass lter. The linear transfer function relating 6^{s) and 0,(j) is B(s) =

e„{s) KgK^F{s)/s e.{s 1 -F KgK^F{s)/s

(1-4)

If no low-pass lter is used, the transfer function is KgK

B(s) =-^

2-2-

K

e, s-i- KgK^ s+K

(1-5)

fi

fi

fi

fi

fi

fi

fi

fi

=

)

which is equivalent to the transfer function of asimple low-pass lter with unity dc gain and bandwidth equal to K. This is really the minimum con guration of aPLL. Since there is no divider in the chain, the output frequency and the reference frequency are the same. The rst PLL built probably used aring modulator as aphase detector.

CHARACTERISTICS OF ALOOP 3

>

K,

Ko

F{s)

s

Figure 1-2 Block diagram of aPLL using alinearized model.

V1

Four-quadrant multiplier V.

^2 >^2

Figure 1-3 Waveform and transient characteristic of alinear phase detector.

The ring modulator or diode bridge is electrically the same as afour-quadrant multiplier and operates from -re to +7C of phase range.

Since the VCO probably has asine-wave output and the reference frequency also has asine wave, it is referred to as asinusoidal phase detector. This does not really mean that the phase detector is sinusoidal; it means that the waves applied to the phase detector are sinusoidal.

Since there are no digital components in this basic loop, it is correctly called an analog phase-locked loop and, as stated earlier, is the minimum form of aPLL. To model it correctly, anumber of assumptions are required.

We have already stated that, for our initial consideration, the loop is locked and that the transfer characteristic of the phase/frequency detector is linear in the area of operation. Afour-quadrant multiplier or diode quad has a sinusoidal output voltage, as shown in Figure 1-3, and is only piecewise linear for 0=0or in the center of operation. This minimum con guration of aPLL has several drawbacks. The absence of a lter does not allow one to choose parameters for optimized performance; the diode ring has only several hundred millivolts output, and an additional loop ampli er will add noise to the system. Therefore, for frequency synthesizer applications, asimple analog loop without any lter is rarely used. Such aloop would be called a rst-order type 1loop, and we will deal with it later.

There are some applications for a1:1 loop, as we will see. The 1:1 loop is used to clean up an existing fre¬ quency, whereby the loop bandwidth is kept narrow enough to allow fast locking, and wide enough to permit fast acquisition. However, the loop bandwidth is narrower than the spurious frequencies present at the reference input. The attenuation of the loop lter of such aloop, which should be asecond-order loop, will clean up the output signal relative to the reference.

1-2

CHARACTERISTICS

OF

ALOOP

We have met the minimum-con guration analog PLL and already learned that there are several limitations to this loop. The rst step in increasing the output voltage delivered from the diode quad and avoiding an operational ampli er is to use adifferent phase detector.

fi

fi

fi

fi

fi

fi

fi

fi

fi

fi

If the diode bridge arrangement is changed and the diodes are overdriven by the reference signal and the input signal from the VCO remains suf ciently small relative to the reference, the phase detector is still in alinear operation mode, and the output signal of the phase detector is no longer asinusoidal curve but rather has alinear

sawtooth form. The range over which the phase detector operates is still from -n to +;r, and the 'd

V', I

''s

-jn-JTMn-

0

'^2 0=down

C JK master/slave FF

G2 s

Case 1 V

/

Up

c >

● o

F F

1

^2

R

Up—u—u—Lr

Q

Down Case 2

L-
=arctan cuTj -arctan cutj -E k

(1-150)

The phase margin is

assuming that (B~T2T3{(o) dco

= 0

d(t>

f

fi

dw 1-E {0)T2 1 -E {cor^p

(1-151) = 0

(1-152)

3 2 L O O P F U N D A M E N TA L S and therefore

(1-153)

CO O

If we set

a=arctan aiTj

(1-154)

p=arctan cotj

(1-155)

(j) =a~p+K

(1-156)

and

imp) =tan[(a -P) +k] tan(a -P) +0 1 - 0

(1-157)

=tan(ff -P)

then t a n a

tan(a -p) =

tan/J

0)^2 -COTj 1

1-F tan atan /?

=imp)

(1-158)

If we set

(1-159)

then

tan

lter

Active

£ M i l H

-

2

lter

Passive

AwtjH —> 00

N

Af

„ H

=K

lter

M

l N

Capture range

T2#0

n

T2=0

Awl

Pull-in range

A(i)p

W-

Pull-in time

2Cco„K„K,i N

n/5 2CCO,Ml

Acop «

Tp

N

-

C

IAcop«

O

I M i M d

Tp '

(coi

A wP O

l.8®„(f-M)

A®py =7T ®„ exp

f > l

Af po =«: ®„ exp

N

-wl

x^(co},

Pull-out range ? < 1

I M m M a

AcOp ~tr Y

N

(-jL= arctan £

arctan

VC^ (

w

Figure 1-27 Magnitude of the open-loop gain of aPLL system.

The straight-line approximation of the magnitude of this open-loop transfer function is plotted in Figure 1-27. The magnitude of the response decreases at the rate of 6dB/octave until the frequency is equal to the -3-dB frequency of the low-pass lter (1/2); for higher frequencies the magnitude decreases at arate of-12dB/octave. Several rules of thumb, developed by Bode for feedback ampli ers, are useful in selecting the loop parameters. The rst has to do with selecting the lter bandwidth cop =1/t. The approximation is: If the open-loop frequency response crosses the 0-dB line with aslope of -6dB/octave, the system is stable; If the slope is -12 dB/octave or greater, the system is unstable. The second-order system under consideration is inherently stable, but the model is an approximation to ahigher-order system. If the open-loop second-order model crosses the 0-dB line at -12 dB/octave. there is little room left for error. Additional phase shift from the VCO or phase detector could cause the loop to go unstable.

To have the open-loop gain cross the 0-dB line at -6 dB/octave, it is necessary that C0p> K^. The larger cop is, the better will be the loop stability. From the ltering viewpoint, the smaller cOp, the smaller the loop bandwidth and the less noise that will reach the VCO. should be as small as possible to minimize the bandwidth. The larger the K^,, the smaller the steady-state error and the faster the loop response. Hence, in PLL design, compromises among noise performance, loop stability, steady-state error, and transient performance

fi

fi

fi

fl

fl

fi

fi

fi

fi

fi

fi

fl

fi

must be made.

5 6 L O O P F U N D A M E N TA L S

Figure 1-28 Open-loop frequency response in the case =K^.

Another rule of thumb that is helpful in PLL design is that the frequency co^ at which the magnitude of the open-loop transfer function is unity, = 1

(1-236)

is approximately the closed-loop 3-dB bandwidth. This relation is exact for the case where F(jco) =constant.

If F{s) is asimple low-pass lter response and co^ >K^„ the open-loop frequency response will be as shown in Figure 1-28. In this case, the loop bandwidth is approximately equal to K^.

If 0, the temperature Tf^ is the probe signal th enables the measurement of the power gain A-, and F, gives the equivalent temperature F^ after talcing away the gain A^. In actual experimental conditions, the output noise power is

P,=A^k{T, +T^)B

P,=Ah{T,. +T^)B where Bis the bandwidth of a lter at the device output. The solution of the system is F„ -YT, Y-

Tf, (high)

L Device

R

(2-25)

1

B

Ph (high)

o

o

o

o

P, (low)

Tf (low) v w R

fi

fi

fi

:

)

fl

Figure 2-9 The Ymethod for the measurement of the equivalent noise temperature of adevice.

INTRODUCTION TO PHASE NOISE 77

where Yis the power ratio de ned as Y = - l Pi

(2-26)

The solution of the system is Pe =

T, -YT, K-

1

with

Ph Th +Pe

Y = - ^ PI

> I

(2-27)

Tl +Te

The main virtue of the Ymethod is that the factor A^/5 cancels in the evaluation of Y. This simpli es the cali¬ bration and results in improved accuracy because Agenerally suffers from atness defect, while Bis the equivalent noise bandwidth, which results from integrating the transfer function. It is worth mentioning that the equivalent noise temperature includes thermal noise in strict sense, the shot noise, and any other noise process. For this reason, people with abackground in optics may nd this concept particularly

misleading. In fact, in optical systems there is no temperature and, in high SNR condition, the electrical noise at the detector output is chie y shot noise.

Flicker Noise

Flicker noise is characterized by the PSD proportional to 1//, or close to 1//, in awide range of frequency. The

digression about whether icker noise is fundamental or not, is more academic than pragmatic, and goes far beyond our scopes. The most interesting fact about icker is its ubiquity [14, 15]. After being discovered in carbon micro¬ phones [16], it is found in geophysical phenomena, climatology, mechanics, optics, classical music, Internet traf c, and in avariety of other domains [14], and of course electronics. Flicker noise originates around dc. Flicker of phase and icker of frequency are parametric noise types, gen¬ erated by anear-dc process, which modulates the phase or the frequency of asignal. Flicker is of paramount importance for us because it turns out to be amajor limitation in the noise of synthesizers, and of oscillators as well.

Spurs and Other Unwanted Signals

fi

fi

fi

fl

fi

fi

fl

fl

fi

fi

fl

fl

fi

fi

The generation of aclean microwave signal, free from spurs, interferences, and other unwanted signals is ablend of engineering, experience, and art. We all are used to the presence of unwanted signals at 60 Hz (50 Hz in Europe) and multiples, from the power grid. Such signals show up as spectral lines in phase noise, and as ahum sound in audio-frequency. They get in microwave circuits in several different ways, like the ripple of supply lines, unequal potential of different ground points, ground loops, and magnetic elds captured by loops and turned into emf. These signals are added in the low-frequency part of phase lock, and transposed to the carrier as parametric noise. Other interferences have similar behavior, like the ripple from switching power supplies, and the high-voltage raster signals from cathode ray tubes. Unshielded ac magnetic elds affect the magnetic permeability of ferrite cores, which modulate the phase of RF signals. Acoustic noise gets in microwave circuits via the sensitivity to acceleration. Most of such noise comes from fans, from the mechanical vibration of transformers, and again from unshielded magnetic elds via the ac attractive force on iron parts. Disturbances from 50 to 60 Hz power grid usually extends up to approximately 1kHz, becoming progressively smaller as the number of harmonic decreases. Odd-order harmonics are generally stronger than even-order harmon¬ ics. Acoustic noise is most present between 1and 2kHz, while switching power supplies and cathode-ray tubes are typically in the 10-100 kHz region. Quartz oscillators and other electro-mechanical oscillators are highly sensitive to acoustic noise. Some are also sensitive to magnetic elds, mainly because of the presence of magnetic materials in packaging and springs. YIG materials are highly sensitive to magnetic elds, but packaged YIG oscillators are generally well shielded. Digital circuits can be an annoying source of spurs and disturbances because of the variety of effects. Radiation occurs at the clock frequency, or at the bus frequency, which occurs from 100 MHz, or less, up to 1GHz. High

peaks of current on supply lines or ground are driven by software in microprocessors and eld programmable gate arrays (FPGAs), which sometimes cause alarge number of transistors to switch with random, pseudo-random, or pseudo-periodical appearance. Spurs may be observed in awide range of frequency from Hz to MHz. Impressively large spurs may be observed in digital phase noise test equipment, if the user removes the post-processing lters that hide the spurs. Digital circuits can also interfere with other parts of asystem in another subtle way, via thermal uctuations. The problem arises from modern very large-scale integration (VLSI) integrated circuits, where high dissipated power per unit of silicon surface is necessary to achieve the computing power. Adissipation of afew watts is usual in FPGAs, direct digital synthesizers (DDSs), etc., in asmall chip, and the surface is proportionally hot. If not appropriately shielded, temperature uctuations show up generally below 10 Hz with asteep spectrum,

of slope Mp or higher. Every circuit is aspecial case, and the literature provides little or no help. To complete the picture, the electromagnetic interference between different part of asystem is one of the earliest known forms of spurious signals, and however sometimes dif cult to model and predict precisely. The electromag¬ netic interference impacts on systems as an additive disturbance, or through intermodulation in junctions. Some classic reference books are available on this topic, by Goedbloed, Ott, Paul, and Perez.

2-1-4 Phase and Frequency Noise

The Quantities Sg(f)

(f), L{ f), and S^{ f)

The PSD of the random phase 9(t), denoted with Sg(f) [rad^/Hz], is the obvious choice to characterize the phase noise in the frequency domain. Its use already appeared as S^(f) in an article [17] presented at an NASA symposium [18] intended to clarify spectral purity and related problems. Similarly, the phase time uctuation can be characterized in the frequency domain in terms ol' S„(/), which

is PSD of x(r). Because it holds that x(/) =9{t) /2nfg, the quantities Sg(f) and S^{f) are fully equivalent, and related by 1

S.(f) =

(2-28)

Pg(f)

The quantity L(f), de ned as 1

(2-29)

Lif) =^Sg(f) (de nition)

is the most widely used measure for phase noise. L{f) is generally given in dBc/Hz using 10 log|Q L(f). Some authors include 10 log|Q in the de nition of L{f). According to (2-29), L(f) and Sg(f) are fully equivalent and

differonlyintheunitofangle.TheunitistheradianforSg{f),and'Jlrad='Jlx180/;r=81.03°forL{f). If we started from the scratch now, we would use Sg{f), and L(/) would not exist. The reason is that Sg(f) is aproper SI quantity, L(f) is not. The problem originates in the early attempts to measure the phase noise with a spectrum analyzer. At that time, the phase noise was measured as Power in 1Hz bandwidth at a

Uf) =

frequency/ off the carrier frequency/, Carrier power

(obsolete de nition)

(2-30)

The true measurement of phase noise became common in the 1970s [19], when the double-balanced mixer (DBM) was available as an off-the-shelf component, suitable to awide range of carrier frequency. Notice that the IEEE Standard 1139 replaces (2-30) with (2-29). This was done since the rst edition published in 1988 [20] and the second [21] and third edition [22] of this Standard, published in 1999 and 2009. respectively, con rm this choice.

fi

fi

fi

fi

fl

fi

fl

fi

fi

,

fi

fi

The obsolete de nition (2-30) is conceptually incorrect and experhnentally incorrect. Let us discuss why. First and foremost, the sideband power originates any combination of amplitude noise and phase noise. The obvious consequence is that (2-30) is aconceptually wrong representation of phase noise. For example, there is a

fi

fl

78 ALMOST ALL ABOUT PHASE NOISE

INTRODUCTION

TO

PHASE

NOISE

79

discrepancy of 3dB in the actual phase uctuation of two signals having the same spectrum, one affected by equal amount of AM and PM noise, and the other having negligible AM noise.

Second, phase noise is measured with aphase detector. Consequently, “the SSB power in 1Hz bandwidth” does not match the operation of the instrument.

Third, phase noise is pure angular modulation, thus the total power is the same at any modulation index. The random nature of noise does not change this fundamental property. The de nition (2-30) can be used only for small modulation, where most of the power is in the carrier, and the power associated to the sidebands is comparatively small. Slow phenomena, like frequency random walk and drift, yields to large phase swing, exceeding 1rad- in 1Hz bandwidth. In the presence of such phenomena, (2-30) gives nonsensical results. By contrast, the correct de nition (2-29) is perfectly suitable to large phase swing. Finally, the obsolete de nition (2-30) suffers from several pathologies. What happens with an odd signal affected by strong AM noise, and small PM noise? What happens if aspur occurs only in the upper (or lower) sideband, in the PM noise measurement range? In both cases the sideband-to-carrier ratio gives anonsensical picture of the phase uctuations.

It is acommon belief that L(/), or equivalently Sg{f), is avalid measure only for small angles. In reality, there is no reason for such limitation, and L(f) is valid even if 6(t) accumulates alarge number of cycles. In other words, there is no reason to restrict L(J) to values below 0dBc/Hz. In optics, measuring lasers one may encounter values of -1-40 or 60 dBc/Hz, which are theoretically and experimentally correct. Of course, the phase detector has to work correctly in this regime. Heuristic Derivation of L(f) and Sg(f) in the Simple Case of Additive Noise

It is instructive to derive the quantities L(f) and Sg{f) for the simple case of white noise having PSD equal to N [W/Hz] added to asinusoidal signal of power Pg [W]. Before proceeding, we have to make clear that the case described does not match the de nition of L(f) and Sg(f), but approximates it for small N. The catch is that a true random PM keeps the total power constant, while the added noise does not. With this caveat, our heuristic derivation gives useful results. Let us start with L(/), with the help of Figure 2-10. In the standard notation for microwave circuits, asinusoidal

signal v(?) =Vg cos(cOgt -I- 9) can be represented as the complex vector

V=\/,/«

(2-31)

The power of such signal is Pg =IV'gl^/2P, thus the vector length is Vq =y/2RPg. Similarly, anarrow noise slot of bandwidth Bcentered at/g can be represented as avector V

=

(2-32)

of random amplitude V^(t) rotating at the frequencyand average absolute value V„ =y/lRNB.Adding carrier and sideband under the approximation of small noise-to-signal ratio, we get 9{t) =

^/2RNB

sin(2;r/0 =^jNB/PgSinilnft)

(2-33)

The RMS value of 6{t) for P=1Hz is \jNB/2Pg. Accordingly, it holds that 1

N

2

P0

(2-34)

The aforementioned formula can be rewritten in terms of equivalent noise temperature or of the noise factor F a s

fl

fi

fi

fl

fi

^0

1 F k T

or Uf) =^ 2

fi

1k{T^ +T,)

P0

(2-35)

80 ALMOST ALL ABOUT PHASE NOISE

Po

Q W 0-

B

B

N I

LSB

USB

/ f .

I

/,

m

I

/o-/

/o

m

fo

+f

m

-

ep =^/mp^

'

USB,

I%

^rms =VA^

V2^ 'i

1 I

Carrier

\

/

/

H

a p

\ \

I )

\

Figure 2-10 Heuristic derivation of L(f) in the case of addi ve white noise. The resistance Ris not shown. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

Asimilar procedure can be used to derive with the help of Figure 2-11. While the carrier is the same as

earlier, now we have two symmetric narrow sideband slots of bandwidth Bcentered a and at/g +/„, V——

4-

Vp+

These sidebands have random amplitude Vlsb(0 an

’tfm'

(2-36)

The power associated to each sideband is NB, equally

split into AM and PM noise. Thus, the absolute value of the vectors that contribute to PM noise is V,U S B = sJRNB. Combining carrier and sidebands under the approximation of small noise-to-signal ratio, we get

6(t) =

2y/RNB

S'n(2;r/„,f)

V ,L S B

“

(2-37)

and

2NB

6{t) =

.

(2-38)

s i n

P0

The RMS value of 0(t) for B=1Hz is ^JNB/Pf^. Thus N

(2-39)

So(f) =f Using the equivalent noise temperature or the noise factor F, the aforementioned formula becomes

d

fi

fl

V

FkT o

(2-40)

r

' 0

t

kT,

Se(f) =y

INTRODUCTION TO PHASE NOISE 81

A

i i Q w D -

Po

B

B

N LSB

USB

I

▶

/

/

m

I

m

I

I

i

/o-/

I

io

m

/

- - H /o +/

%/m i

\ / 2 ^0

Carrier

9p=2%/NB/2P

= N / P0

«pV^

h-

Up=2%/NB/2Pq

- fm

Carrier

%/m

I

S „a = N / P.0

/m

Figure 2-11 Heuristic derivation of S(,(Oand oiS^(f) in the case of additive white noise. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

The same development can be used to derive the amplitude noise, in this case, the vector representing the lower side band (LSB) has opposite sign with respect to phase noise; hence the sum of the two sideband vectors is parallel to the carrier. The result is N

(2-41)

S.(f) =~ M )

Using Figure 2-10 instead, we notice that the old de nition of L{f) based on the sideband-to-carrier ratio gives both amplitude uctuations and phase uctuations, of equal amount. After our digression on the reason why the de nition of L(J) has be changed to L{f) this unpleasant fact does not come as asurprise. Should the

reader have to face both PM and AM noise, we recommend the use of Sg(f) and or Lg(f) and L^if) with obvious meaning. The notation M(f) is sometimes encountered as the AM-noise counterpart of L(J). The smallest amount of white noise for asource characterized by aresistance at atemperature Tis kT

fi

fl

=

fl

fi

' 0

(2-42)

’

82 ALMOST ALL ABOUT PHASE NOISE Table 2-4 Scaling rules for anoise-free synthesizer delivering fo

u t

={n/d)f.„

Spectrai domain

Quantity

Time domain

Phase

^out(0 =^^in(0

Frequency

(A/;„)(0 =(^)(A/;jw

Phase time

Xou.(') =

5x..ut(/) =5xi„(/)

Fractional frequency

Vou.(') =

5y„„,(/) =Syi„(/)

Scaling the carrier frequency with an ideal noise-free synthesizer, which delivers an output frequency/^ =(n/d)f-, the quantities associated to phase noise scale according to the rules listed in Table 2-4. We want to draw the attention of the reader to the simple fact that the synthesizer scales up or down the input phase noise and the input frequency noise in the same way. By contrast, the amplitude at the output of the ideal synthesizer is determined by the output stage, rather being sensitive to the input amplitude. Example 2Noise Factor. Asystem has anoise factor of 1.8 dB and receives at the input asinusoid of power P(, =100 pW (-10 dBm). The phase noise L(f) is A(/) =A

= 1IO' **/'" X1.386 X10““ X290

2Po “2

10

- 4

= 3 x 1 0

- 1 7

(-165.2 dBc/Hz)

If the reader can think in dB, the aforementioned formula becomes

=-3 +1,8 -174 -(-10) =-165.2dBc/Hz

Additive and Parametric Noise

Experience shows that in all oscillators the sideband noise increases greatly as we observe very close to the carrier. Slowly tuning the oscillator to adifferent frequency, we are faced to the evidence that the noise sidebands are attached to the carrier and follows the carrier frequency. This behavior is incompatible with the noise model we

have used in Section 2-1-4 to derive Sg(f). How could the additive noise have asharp peak centered exactly at the carrier frequency, “know” when the oscillator is re-tuned or drifts, and track the carrier by shifting the peak to the new frequency? No way. Our derivation related to Figures 2-10 and 2-11 is correct, but it does not explain this behavior. The answer is that there are two types of phase noise, and of amplitude noise as well, called additive noise and parametric noise. They already appeared in the seminal article [17], at that time called additive and multiplicative noise. The basic mechanisms are represented in Figure 2-12. Understanding the difference between these types of noise is of paramount importance to master phase noise. The additive noise is exactly what we

have explained in Section 2-1-4 when we derived Sg{f) by adding carrier and sideband vectors. White, or nearly white, noise is present in awide radiofrequency and microwave spectrum, and it adds to the carrier. By contrast, the parametric noise originates from anear-dc noise, which modulates the carrier phase, frequency or amplitude, or any combination of. The noise spectrum of the near-dc process is transposed to the sidebands around the carrier with the appropriate rules and symmetry. In this way, it is perfectly sound that anoise pattern with spurs in the microwave spectrum is centered at the carrier, and it appears unchanged around the new frequency after tuning or drift. Notice that the power conservation inherent in the angular modulation of any index or phase swing applies only to parametric phase or frequency noise. By contrast, adding awhite noise process results in higher total power. Two-port components show asimilar behavior, with the difference that they have no frequency drift because the output frequency is rigidly determined by the input frequency. The digression about additive and parametric noise and their difference deserves further attention, because the term added noise is sometimes encountered in the specs of components and of test equipment. This term denotes the

INTRODUCTION TO PHASE NOISE

Additive noise

83

Parametric noise

a P S D

PSD Internal

Internal

Output

noise

Output

noise

E 3

CO

fo

fo

f

f

Figure 2-12 Additive and parametric noise. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

phase noise, and sometimes also the amplitude noise, that the component “adds” to the incoming carrier. The choice of the term “added noise” is unfortunate because it is too easily mistaken for “additive noise,” while it refers to both the additive noise and the parametric noise that the component “adds” to the incoming carrier.

The Polynomial Law

Amodel that is found useful to describe the phase noise of oscillators and components is the polynomial law, often also called power law [2, 23] 0

5«(/) =ZbJ'

(2-43)

/j\. where Tand 5/are the RMS values. The equality holds for

Gaussian distributions. The theory underneath is found in many textbooks on the Fourier integral, among which we prefer Papoulis, 1962. Actual instruments work with the acquisition time T^, which is a nite and well identi ed quantity. The acquisition time is associated to the window (taper) function used in the FFT analysis, the most

popular of which is the Hanning window. In practice, the resolution is governed by i5/>C,, with C, is in most cases of at least 2-3. The lowest frequency of the FFT is/, =1/T. However, the rst points may not be plotted, or discarded, because of the poor resolution 5//and because of artifacts related to the window function. The consequence is that the minimum plotted frequency is ruled b > C, with Cof the order of 5-10. For example, the acquisition of asingle spectrum down to =10 mHz gives —600 s(C =6). If we decide to average on 12 spectra in order to get acomfortable con dence level, the measurement takes 2h. By contrast, the AVAR and the other wavelet variances, described later in this chapter, are way more ef cient at estimating the long-term behavior of oscillators in ashort measurement time. Interestingly, all these variances are easily calculated from atime series of phase data, the same used to calculate the PM noise PSD. The RF Spectrum of the Oscillator Signal

fi

fi

fi

fi

fi

fi

y

fi

fi

fi

fl

fi

k

fl

m

The oscillator signal, observed with an RF spectrum analyzer, looks like arather narrow bell-shaped pattern, wob¬ bling, wandering, and drifting. How does it relate to the PM noise PSD, which is seemingly unbounded at low Fourier frequencies?

8

ALMOST ALL ABOUT PHASE NOISE

However naive the question may seem, the problem underneath is surprisingly complex. All dif culties start

from the fact that the variance does not exist for icker (1/f) and for steeper processes (l//^ l/f\ etc.), and from the fact that the variance of the truncated Sg(f). high-passed at afrequency/^, diverges as/^ 0. Thus, the use of ahigh-pass lter does not help. The problem of the line shape is addressed in [24-26]. These references rely on arather dif cult the mathemat¬ ical framework and provide aseparate solution for each PM noise process. Reference [26] is particularly useful for optics, where the line broadening phenomena are ampli ed by beating two laser beams down to RR However, the language used in these references may be unusual or dif cult to microwave engineer. That said, we try to grab the main physical facts. First of all, the angular modulation does not affect the total electrical power, and the same happens with phase

noise. Consequently, the generalized PM power, given in rad^, is allowed to be quite large, and even to diverge in the long run, without violating the energy conservation principle. The pattern seen on the display of the spectrum analyzer reveals the average electrical power. The averaging time is set by 1/RBW or by 1/VBW, which is longer (RBW is the resolution bandwidth, and VBW is the video lter bandwidth). Second, the electrical power is spread in sidebands, governed by the in nite series of Bessel functions i„(/5), where is the modulation index. The sidebands are separate entities at small /?, but they collapse in asingle line at large p. Something happens in between. Third, the spectrum analyzer has alimited RBW. In the traditional scanning spectrum analyzer, the transfer function associated to the RBW is aLorentzian line shape, determined by the intermediate frequitncy (IF) lter. The displayed shape results from the convolution of the input spectrum and the IF frequency response. The behavior of modem spectrum analyzers based on the FFT of awide-band IF signal is rather similar, however more dif cult to understand in rigorous mathematical terms. For the sake of clarity, let us say that the RBW is associated to a Lorentzian

l t e r.

Now we apply the previous concepts to the oscillator signal. When the oscillator delivers apure and stable signal, the RF spectrum is aclean and sharp line, narrower than the instmment RBW. The displayed spectrum is aLorentzian pattern determined by IF lter. Such pattern wanders and drifts slowly, following the oscillator frequency. The spectrum of white FM noise is aLorentzian. Thus, if the dominant oscillator noise is white FM, wider than the RBW, the analyzer displays aLorentzian determined by the oscillator FM noise. When

icker FM or FM random walk is dominant, the numerous random sidebands tend to

cluster in Gaussian shape. However, if the width of the Gaussian is still comparable to the Lorentzian RBW, there results aVoigt distribution [27]. Until now, we have implicitly assumed that the frequency reference inside the spectrum analyzer is stable and free from noise. Actually, the displayed pattern results from the PM noise of both the oscillator umler test and the analyzer’s internal oscillator, and the two contributions are indistinguishable.

2 - 2 T H E A L L A N VA R I A N C E A N D O T H E R T W O - S A M P L E VA R I A N C E S

We study the frequency uctuations of areal oscillator around the nominal value, as afunction of the measurement time. Such

uctuations contain tme random terms, plus other phenomena like drift and environmental effects.

fi

fi

fi

fl

fi

fl

fi

fi

fi

fi

fi

fi

fl

fl

fl

fl

fi

fi

fl

8

However, the environmental effects are of systematic origin; they can only be described statisticalW for the part of the environment that escapes from quantitative understanding. This goes under the term in uence quantities in the formal language of metrology. Before tackling the analysis of frequency uctuations, we have to study the basic operation of ths frequency counters, sometimes cdWtAfrequency-to-digital converters. Other options are possible, chie y time analyzers and phase meters. These instalments are highly specialized and seldom found in general laboratories. Modern frequency counters are rather complex, and their statistical properties may be dif cult to understand. Reference [28] reviews the high resolution counting techniques, and References [29, 30] provide insight in the statistical processing techniques.

THE ALLAN VARIANCE AND OTHER TWO-SAMPLE VARIANCES

8 9

2-2-1 Frequency Counters The nFrequency Counter

The ncounter is an instrument that measures frequency or period by counting anumber of events occurring during the gate time t. The classical frequency counter, shown in Figure 2-15, is the simplest example. The instrument counts the integer number of cycles of the input signal in the gate time tgenerated by the reference clock. The classical frequency counter is seldom used because of the poor resolution at low input frequency. If t=1s, the number is equal to the frequency expressed in Hz, thus the measurement of the 50 Hz frequency of the European power grid suffers from aquantization of 2%. For this reason, the classical frequency counter is generally replaced with classical reciprocal counter. The roles of the clock signal and of the input signal are interchanged, and the instrument measures the average period by counting the clock cycles in asuitable multiple of the input period that approximates r. With aclock frequency of 10 MHz, the quantization is 10“'' in 1smeasurement time, regardless of the input frequency. More sophisticated instruments can measure afraction of aclock cycle by interpolating between edges of the clock signal. Combining reciprocal counting and clock interpolation boosts the resolution up to 10 digits and more, but the noise mechanism of the Hcounter remains. In this type of instruments, the noise is determined by the uctuations found at the start and at the stop event that de ne the gate time t, and the uctuations occurring between start and stop do not contribute to the result. In practice, the time uctuations originate from the quantization noise, from the interpolator, and from the noise of the input trigger. With sophisticated interpolators,

the remaining noise from the input trigger is the dominant noise source. The contribution of the frequency reference is not accounted here, and must be considered separately.

Since the input frequency uctuates, it is useful to replace/q with/g +(A/q)(0. The average frequency is given by I

/o +(A/o)(r)df

0

(2-56)

Introducing the fractional frequency uctuation y(r) =(A/g)(r)//g, we can replace the average (2-56) with the weighted average C

O

y=J/y(t)w^{t)dt o

(2-57)

where

1/ w

n

“

0

0 ’;'stheaverage.Theproblemisthat(2-72)isusefulonlyfor

certain types of uctuations, and of course it works perfectly with the white PM noise we were really concerned about. In such cases (2-72) converges to the “ideal” value called mathematical expectation, and the con dence improves progressively as Aincreases.

The problem is that real oscillators suffer from frequency icker, random walk, and drift, and other processes. These processes make (2-72) depend on both Aand t, and the mathematical expectation does not exist. That something goes wrong with Aand is clearly seen with linear drift (constant aging). Try yourself to calculate the deviation ain the case of a100 MHz oscillator drifting by -1-1 mHz/s, measured with sgate time of Is. After taking A=10, 10^, IO-\ etc., we agree that the experimental variance is not agood choice in this case. The problem with -r is alittle bit subtler. Let us take the distance from two points on asheet of sandpaper as asimple example. Ahuman-sized ruler may indicate adistance quite close to 6in., which we perceive as “correct” at our scale, and that’s it. By contrast, atiny insect experiences amuch longer path of consisting of high obstacles it can surmount only thanks to its incredible agility. What happens? The answer is that the dis¬

tance depends on the size of the ruler used to measure it. Not surprisingly, something similar happens with the oscillator uctuations y(t) and the measurement time t. Still on the example of the 100 MHz oscillator drift¬

ing at H-l mHz/s, try yourself to calculate Cy as using (2-72) after switching the gate time to 0.1, 1. and IOs. The difference between results should convince the reader that (2-72) is to be replaced with amore appropriate tool. Here the family of two-sample (Allan and Allan-like) variances gets on the stage. The key points are a s follows:

●The measurement time ris made explicit

fi

fi

fi

fi

fl

H

fl

fl

fi

●The variance is calculated using asimple and perfectly de ned pattern.

THE ALLAN VARIANCE AND OTHER TWO-SAMPLE VARIANCES

95

The Allan Variance (AVAR)

Let us rst introduce the average fractional frequency y^fr), de ned as k r

VkM=[

y(0 dt

(2-73)

This is the fractional frequency ywe are familiar with, measured with aIT counter where the gate time has duration

Tstarting at r=(/:- 1)t and ending at t=kr. So, y, (t) is averaged between 0and r, yjC-r) is averaged between Tand 2t. etc. It is important that the measures are exactly contiguous, with no dead time in between, otherwise the nal result will differ from the AVAR. This requirement is not areal limitation because the speed of nowadays digital hardware is generally suf cient to avoid such dead time. Understanding what happens in the presence of a dead time is aspecial topic for experts. Having de ned t, we solve the dependence on Nby setting A=2in (2-72). Notice that A=2is the smallest value that gives avalid variance. Rewriting (2-72) for A=2

W=[V2W -h]^+ [Vi(i^) and expanding using

(2-74)

y, )/2, we nd the two-sample variance

(Ty(T) =^[y2(r) -yj(T)]^

(2-75)

This variance is also called the Allan variance [2] or AVAR. The obvious notation Allan deviation (ADEV) is

often used for Uyfr). Equation (2-75) should be intended in the sense of the mathematical expectation, that is, the ideal average on in nite values of

Digital 3:TTL

\

■ D

* 0

z

5 T

CD

W - . .

-120

-"

NDigital 3: ECL 125 MHz

c r

D. Digital 3: TTL

-130

-r10 2MHz

Note: Frequency given is output

-180 10

102

1Q2

X

-140

10''

10''

Offset frequency from carrier (Hz)

10®

10

102

1

10®

£ 10''

I

10®

10®

Offset frequency from carrier (Hz)

Figure 2-27 Phase noise of some dividers as all data refer to the output frequency (A), and are scaled up to 10 GHz (B). Reprinted with permission from [47].

In accordance with Figure 2-27, the practical noise limit is of the order of-170 dBc/Hzforthe old TTL dividers and -155 dBc/Hz for the ECL dividers. Yet, these values are not easy to achieve in practice. Small-scale integration complementary MOS (CMOS) dividers, up to an input frequency of 10 MHz, are similar in phase noise to the TTL devices. However, the close-in noise or noise between 1and 10 Hz offthe carrier is higher than that of TTL devices. TTL devices require higher shielding and better power supply decoupling to prevent external crosstalk between the various stages, which otherwise results in unwanted spurious and sidebands. Programmable logic devices offer new design perspectives, but they are only usable up to 1GHz. Commercial dividers up to 12-26 GHz input frequency are available from Analog Devices, Hittite (now with Analog Devices), Keysight Technologies, Microsemi, ON S e m i c o n d u c t o r. P a s t e r n a c k , a n d o t h e r m a n u f a c t u r e r s .

Most of these dividers have another unpleasant effect, in the form of internal crosstalk. Crosstalk is de ned as the amount of input signal appearing at the output. In high-performance synthesizers, it is necessary to use a

fi

fi

Asynchronous

2

fi

fi

fi

105

PHASE NOISE IN COMPONENTS

low-pass lter after the reference or the programmable divider and apulse shaper to translate the resulting sine wave back into asquare wave for appropriate suppression of the crosstalk.

Phase Noise Scaling

In anoise-free divider, the device divides the input phase by Nfor the same reason it divides by Nthe input frequency and 0„(O =

fo =T, N

0,(0

(2-84)

N

Naively, we may be inclined to extend this result to Sg^(J)

/N^. Reality is more complex because aliasing

strikes on some types of signals, but not on others. Some relevant cases will be analyzed in the following text.

Let us start from white noise. In digital circuits, phase noise is sampled at the frequency 2/g for the simple reason that the phase uctuations exist only during the rising and falling edges of the signal, while the pulse level has no effect on the output. Thus, the bandwidth of phase noise is equal to/p, which is half the sampling frequency. Some circuits switch only on one active edge, either rising or falling. In such case the sampling process takes place at/,,

and the noise bandwidth is equal to/p/2. We take/p as the bandwidth, leaving to the reader the extension to/,/2. The easiest way to understand aliasing is to derive the spectrum from the time uctuation x(r) and its mean square value (x-). In principle, anoise-free divider transfers the time uctuation from the input to the output. By virtue of the Parseval identity, it holds that

(x^)=5,(/)5

(2-85)

where the bandwidth Bis equal to/ at the input, and to/„ =fJN at the output. The consequence is that (with aliasing)

(2-86)

Thus, the bandwidth reduction by ]/Nh compensated by increasing the noise PSD by the same fai;tor N. The time

uctuation can be converted into phase uctuation using d(t)

This applies to the input frequency /and

to the output frequency/„. There results

=7N

(with aliasing)

(2-87)

By contrast, aliasing produces no detectable effect on icker noise because the aliases are attenuated by afactor

1/2/, 1/3/, etc. Thus, the output phase noise is described by o(/)

/ (/) (no aliasing)

(2-88)

and 1

5 „(/) =^Sg;(f) (no aliasing)

(2-89)

Afurther phenomenon occurs when an analog (sinusoidal) is sent to the input, instead of aclean digital signal from the same logic family. The phase noise PSD in the input stage increases because of nonlinearity and other phenomena.

Figure 2-28 shows the interplay of noise levels in adigital divider. The input PM noise increases by afew dB in the rst stage, and then it is scaled down by afactor of \/N or 1//V“. depending on the noise type. As an unpleasant consequence, a-E10 divider may scale down the input white PM noise by amere 5-6 dB instead of the 20 dB of the

fl

=

fl

=

fl

fl

fl

fi

=

fi

7

1//Wrtile.

fl

fl

106 ALMOST ALL ABOUT PHASE NOISE

107

S«(/)“ Sin

Input

1/A/

1/A/2

White noise

Flicker and higher slopes, and pedestal

Figure 2-28 Phase noise scaling in adigital frequency divider.

The case of asignal with apedestal (Figure 2-29) is so common in PLLs and frequency synthesizers, that it deserves special attention. For this purpose, we can divide the input phase uctuation 6(t) into pedestal and oor, denoted with the subscriptsp and/. Additionally, pedestal and oor have different origins, thus they are statistically independent. The pedestal is rather narrow as compared with the carrier, but it contains most of the mean square

uctuation(9^).Inmostpracticalcases,Aisnotlargeenoughtoshrinkthebandwidthofthepedestal{N1is allowed. Of course, the conversion ef ciency is not auseful parameter to describe PLLs and injection locked oscillators. Designing aPLL multiplier, harmonic sampling is generally used for upper microwave frequencies, where there are no convenient dividers. However, nowadays commercial microwave prescalers work up to at least 26 GHz. Attempts to use tunnel diodes for this purpose, or parametric effects in tunnel diodes, show up in the literature from time to time but they have failed to show reliable performance, due to complexity and component tolerances in temperature. Aharmonic sampler is typically abalanced modulator that uses hot carrier diodes, driven from apulse or comb generator with extremely high harmonic contents. Atypical application for such acircuit is in spectrum analyzers, where the input frequency and the YIG oscillator can be locked together. Asimilar application is where aharmonic comb is generated from a1-MHz reference, and locking can occur every 1MHz up to several GHz. These circuits require apre-tuning mechanism to make sure that the desired harmonic is selected, preventing false locking. This type of multiplication is used in systems where the frequency of the VCO is changed frequently and low spurious contents and high SNR are required. For higher frequency ranges, IMPact ionization Avalanche transit-time (IMPATT) diodes or other exotic devices can generate the necessary frequencies, and some of these multipliers are also built as injection locked oscillators. An injection locked oscillator can be considered as afrequency multiplier within acertain pulling range, where the oscillator locks up with the reference frequency. These are highly nonlinear phenomena, described in the literature

ALMOST ALL ABOUT PHASE NOISE

from time to time. Explanations and mathematical models are built primarily around experimental data, not always reliable for extrapolation or new design. Low-frequency injection locking is avery convenient way of combining extremely high stability in certain types of crystal oscillators, which are being used as areference for extremely low-noise crystal oscillators operating at the same frequency. For single-frequency applications, we nd synthesizer loops using high-frequency crystal oscillators at discrete frequencies between 70 and 150 MHz, locked to afrequency standard with anarrow-band loop, so that the output PM noise depends only on the VHF oscillator, rather than on the input frequency. These loops have bandwidth of afew Hz or less, and therefore compensate only for temperature effects and aging. The step recovery diode (SRD) generates acomb of sharp pulses at the transition between the on and the off region, and in turn acomb spectrum extending to high harmonics. Such diodes may be useful to generate frequencies up to tens of GHz and generally require apower of -I-20 dBm or higher. Aproblem with the SRD is the relatively high phase noise, inherent in abrupt change of capacitance used to generate the sharp pulse. The typical design of ahigh-order multiplier may include alow-noise pre-multiplier, followed by an SRD. Example 8Frequency Mlultiplication. The frequency multiplier from astable 5MHz oscillator to 9.18 GHz

for aCs frequency standard' can be implemented with aXl8 low-noise multiplier, followed by axl02 SRD multiplier. If the oor of the quartz oscillator is -152 dBc/Hz, anoise-free multiplier would rise such oor to -127 dBc/Hz after multiplication xl8 (-1-25 dB), and to -87 dBc/Hz after further multiplication by 102 (-1-40 dB). The -127 dBc/Hz at output of the x18 multiplier sets relaxed specs for the PM noise of the X102 SRD multiplier at approximately -130 dBc/Hz (3 dB less than the PM noise at its input). Introducing the contribution of alow-noise rst stage, the result may differ by -f 1... -I- 2dB. This reinforces the conclusion that alow-noise pre-multiplier is useful to relax the noise specs of ahigh-order second stage ■ For xed-frequency applications, tuned frequency multipliers with transistors work well up to the order of a GHz. Saturation should be avoided because the charge stored in the forward-biased CB junction slows down the transistor and introduces phase noise. Class-C push-push con gurations can be used to improve the ef ciency by reinforcing the even harmonics, and canceling the odd harmonics. Likewise, the push-pull con guration reinforces the odd harmonics and cancels the even harmonics. The differential pair is an appealing option of aswitch because it generates asquare wave with very sharp edges, making ef cient use of the transistors bandwidth and keeping the transistors in the linear region. Baugh [58] suggested the use of an inductor to turn the edges into sharp pulses for ef cient low-noise multiplication, but this idea is seldom seen in practical applications. Varactors [57] are agood choice for high power ef ciency and low noise because they exploit smooth nonlinear

capacitance. The ef ciency of an ideal reactive nonlinear reactance, that is, loss-free varactor, is t] -- 1, regardless of the order of multiplication. In practice, ri is limited by the small series resistance of varactors. Aproblem with the varactor multiplier is that it works at xed frequency or in arather narrow bandwidth. The design can be rather complex because it requires resonant circuits at the input and at the output, and idlers at all intermediate harmonics

if N> 2. Complexity may limit Ato aconvenient maximum of 3-5. However, ef ciency can be so high (up to -2 ... -1dB) that the varactor multiplier is an option for the output stage in small radio transmitters, instead of an active multiplier. The NLTL is another application of varactors suitable to high-order frequency multiplication [59], It exhibits low phase noise, and relatively wide input bandwidth (one octave). Such device is aladder LC network with series inductors, and the parallel capacitors replaced with varactors. The mechanism derives from studies on soliton

waves. The input signal is turned into apulse comb, and progressively shrunk and sharpened as the signal propagates through the line. Commercial devices are available, delivering multiplication up to xlOO at amaximum output frequency up to tens of GHz. The typical input power is of -t-20 dBm. and in some cases up to -1-27 dBm. The loss

is of 10-40 dB depending on the order of multiplication, on power, and on other parameters. These components are often dif cult to use, and the phase noise is highly dependent on the input power [60]. Figure 2-37 shows the

'The clock atomic transition is, by de nition, 9.192631770GHz for the unperturbed atom. An offset of the order of 1-2 Hz applies, due to the magnetic C- eld needed to align the magnetic momentum of all atoms. Therefore, ahigh-resolution

fl

fi

fi

fi

fi

fi

.

fi

fi

fi

fi

fi

fi

fl

fi

fi

frequency of f»12.6 MHz must be added to the 9.18 GHz carrier. More modern schemes start from 9.2 GHz, and subtract a high-resolution 7.4 MHz frequency.

fi

fi

11 4

x 2

100 MHz

250 MHz

input

D>

200 MHz

11 5

fo N LT L

K>

—

1 6 - 1 7

9.4 GHz

dBm

9.4 GHz

output

- 6 0

Take away 3dB for one chain -70

-80 N

I C M

■ a

■9 0

(b) m T 3

-100

□ B

5s CO

-110

-120 @9.4 GHz -130

n

1 0

102

103

10'*

105

/(Hz)

Figure 2-37 Phase noise of apair of muitiplication chains from 100 MHz to 9.4GHz [61]. The noise is shown for two different values of the input power between -H6 and +17 dBm. The phase noise spectrum is ©2009 IEEE, reprinted with permission from [61].

phase noise of apair of multiplication chains from 100 MHz to 9.4 GHz implemented with nonlinear delay lines. This chain is apart of aminiature atomic clock. The multiplication x94, inherently, increases the phase noise

by 39.5 dB (20 log,o(94) =39.5). So, a oor of approximately -120dBc/Hz at 9.4GHz output is equivalent to -160dBc/Hz at the 100 MHz input. For frequency synthesis, the critical region is beyond about 100 Hz. A icker PM of -80 dBc/Hz at 1Hz is extremely low for a10 GHz signal and exceeds practical needs. For comparison, the PM noise of the best 10 GHz cryogenic oscillators is of the order of -90 dBc/Hz, which gives afrequency stability of 10“‘5 (ADEV at 1s).

Schottky recti ers are an appealing option for low-order multiplication (usually 2-3, but up to 7) because low-noise packaged components are available, requiring only an external lter. Figure 2-38 shows some examples of such multiplier. Theoretical ef ciency is limited to (2-96)

ntx \/N~

but practical ef ciency is lower. However, the low ef ciency is partially compensated by the xA^ increase in PM noise, inherent in the frequency multiplication. In the end, these multipliers may be agood option as the rst stages of high-order multiplication. An example will be provided.

The Schottky diode multiplier deserves attention for special applications, where the thermal and long-term stability of the phase is of paramount importance (timekeeping for space application and for radio navigation). The multiplier can be implemented using only x2 stages, frequency converters, low-pass lters, and notch lters.

The point is that bandpass lters are to be avoided because they suffer from aphase drift determined by the drift of the internal components multiplied by the quality factor Q. Oppositely, low-pass lters and notch lters are immune from this phenomenon if the main signal falls in aregion where the frequency response is at. Accordingly, a

multiplier x4 from 100 MHz can be implemented with two x2 stages followed by alow-pass at 250 and 500 MHz.

fi

fl

fi

fi

fl

fi

fi

fi

fi

fl

fi

fi

fi

fi

These lters are stable because the cutoff frequency of each is far from the carrier frequency. Amultiplier x5 starts from the same x4, followed by aDBM which adds 100 MHz. Since the mixer delivers 400 ±100 MHz as the main

fi

fi

PHASE NOISE IN COMPONENTS

ALMOST ALL ABOUT PHASE NOISE

(A)

/; Input

o K3

/o =2/,

a

Output

(B)

DC block R F

Input

/o =2/,-

IF

Output

LO

Figure 2-38 Two examples of the Schottky diode multiplier: (A) half bridge, and (B) double balanced mixer. With the scheme (B), frequency doubling (W =2) is preferred.

products, anotch removes the unwanted 300 MHz, and a550 MHz low-pass cleans the output. The notch lter may suffer from thermal effects at 300 MHz, yielding minor changes in the spur rejection, but the phase of the 500 MHz signal is stable.

Example 9Frequency Multiplication. We analyze the frequency doubling of alow-noise 100 MHz ovencontrolled crystal oscillator (OCXO) using aMini Circuits LK3000-I- frequency doubler. The OCXO has aPM

noise oor of -175 dBc/Hz and output power P=-F 14 dBm. The doubler has aloss of 10.5 dB. It is followed by a250 MHz low-pass lter (0.5 dB loss at 200 MHz) and an ampli er (noise factor F=2dB). Let’s evaluate the phase noise.

First, the thermal noise at the oscillator output is kTtP =-174-14 =-l88dB rad^/Hz, thus -191 dBc/Hz.

This is 16 dB lower than the oscillator oor, thus guessing 2dB noise factor for the multiplier will not change the result. Second, the power at the ampli er input is 4-3 dBm (4-14- 10.5-0.5 =3). Hence, the PM noise of the ampli er is FkT/P =4-2-174-3

—176 dB rad^/Hz, thus —179 dBc/Hz. Third, the oscillator noise scaled up

to 200 MHz is -169 dBc/Hz (-175 4-6 =-169), which is lOdB higher than the ampli er noise. Iht conclusion is

that the noise is set by the Sg„(f) =N^Sgi(f) rule, and that the overall noise of the multiplier is negligible. Should we v/mi further multiplication, we start from -169 dBc/Hz instead of -175 dBc/Hz. Therefore, the second stage will be more tolerant to the noise of the components, and it will be easier to keep with the xAA law ■

For extremely high frequencies, hundreds of GHz or some THz, it is necessary to start from aclean microwave oscillator, optionally locked to astable HF/VHF reference. Conversely, the direct inultiplication of aHF oscillator

will probably fail because of an intrinsic property of frequency multiplication. PM is ruled by the Bessel J{f) functions, where /? is the modulation index. Thus, y,, for the carrier, 7, for the rst-order sidebands, 7, for the second-order sidebands, etc. The xkf scaling rule is an approximation that holds for small angle modulation, where only carrier and rst-order sidebands are considered. Since the total RF power is independent of the modulation index, energy conservation requires that the sideband power comes at expenses of the carrier. When the modulation

index approaches 2.4, nulls, and the carrier sinks abruptly in the noise pedestal. Unlike the simple case of sinusoidal modulation, where the carrier re-appears at higher modulation index following the oscillating behavior of Jq(P), the carrier is lost because of the statistical nature of the many spectral component, which constitutes the random PM. This phenomenon, called carrier collapse, challenged the early attempts to design high frequency

.

fi

fi

fi

fi

fi

fl

fi

fi

fl

synthesizers starting from the stable 5or 10 MHz OCXOs [62, 63].

fi

fi

11 6

PHASE NOISE IN COMPONENTS

11 7

Two technologies are in competition for THz frequency multipliers, the heterostructure barrier varactor, and the Schottky diode [64]. The former is suited for the generation of odd harmonics due to internal symmetry. The latter

is the simplest devices, and indeed probably the best for high ef ciency. For these reasons it will probably be the preferred option for future applications. Anyway, the THz region is beyond our scope, and the reader should refer to the literature.

2-3-4 Direct Digital Synthesizer (DDS)

The DDS is such an important block in modern frequency synthesis that deserves special attention. This section describes the general principles and the phase noise of commercial components. Most of the material is based on our earlier article [65]. The reader may learn more about the DDS from several references listed in the Suggested Readings at the end of this chapter.

Theory of Operation

The principle of operation follows immediately from the block diagram shown on Figure 2-39. The register is a/n-bit D-type ip- op called phase accumulator. The accumulator content at the discrete time kis the integer

number x^, which takes avalue from 0to 2"' -1. At each clock cycle, a: is incremented by Nmodulo 2'". This means that, when xreaches or exceeds 2"', the over ow is ignored and counting goes on from the reminder. In formula

m

-(●*n^, so that the total noise is chie y limited by cr^. With 2-4 bits in excess, the quantization noise is 12-24dB smaller than the analog noise. This choice also results in signi cantiy reduced distortion and spurs because these artifacts originate from the non-random nature of the quantization. The equivalent number of bit (ENOB) results from an attempt of simpli cation, describing total noise a~ = +Ca with asingle parameter. So, the engineer uses the formula R/2F S R

1 (7“

=

12 22ENOB

(2-104)

and that’s it. The catch is that the ENOB hides the difference between analog noise and quantization noise, and their statistical proprieties. The ENOB is suitable to signals with suf ciently good random characteristics, like audio communication. Conversely, synthesizers deliver highly coherent signals. In this case, the ENOB fails to describe the quality of the output because the quantization noise yields artifacts, harmonics, and spurs, while the analog noise does not.

Truncation Spurs

fi

fi

fi

fi

fl

e

y

fi

fi

>

The output DAC has a nite number nof bits, which in turn determines the number pof address bits that gives distinct values at the LUT output. Higher number of bits results in duplicated codes at the LUT output, and in no improvement. The value /? =n-t- 2or p=n-I- 3is often found. The full m-bit word of the accumulator represents the exact instantaneous phase, given b + Nmod 2'". However, the voltage delivered to the output is deter¬ mined only by the higher pbits. In other words, the accumulator de nes 2'” possible phases, or states (all the dots on the circle of Figure 2-40), but only 2'’ distinct phases (the thick dots on the circle) can be delivered to the output. The resolution Ax =2"'“'' results in around off phase error distributed from 0to Ai9, where Ai9 =2kI2"'~>’ rad. Such error is of pseudo-random nature because it results from the fully deterministic operation of the accumulator (the DAC analog noise is not counted here). Pseudo-randomness, as opposite as true randomness, originates spurs. This can be understood by analyzing carefully the Accumulator and the Frequency Control Word (Figure 2-41).

120

ALMOST ALL ABOUT PHASE NOISE

(A) m w

Output phase

Phase error

A H

AL

m CO

Xl

m-p Im-p- 1

m - 1

pbits

0

b=m-p bits mbits

(B) CO CO

FWL

F W H

NH

Nl m-p\

m - 1

CD CO

0

m-p-\ b=m-pbits

pbits mbits

Equivalent frequency word (EFW)

Ne b - 1

0

b=m-p bits Figure 2-41 (A) Accumulator and (B) frequency word.

For easier interpretation, it is useful to divide the accumulator into two sections, accumulator high (AH) and accumulator low (AL), thus .r =x^+Xj^. AH has the size of pbits and generates the output phase. AL has the size of m-p bits and generates the truncation spurs. Similarly, we divide the frequency word (FW) in two parts, FWH (the higher pbits), and FWL (the lower m-pbits. Notice that the most signi cant bit (MSB) of FVf must be zero, otherwis exceeds/^|;/2 and the output freqtiency results from aliasing. An important case is FWH containing an odd number, and FWL containing all zeros F W H

F W L

Oxxxxxxl

0000000000000000

Accordingly, the value x,. is incremented in steps equal to Ax starting from zero. Thus, at each clock cycle, X). jumps to the next thick dots on the circle and AL contains all zeros. There is no phase truncation, and no truncation spurs.

The case of FWH containing at least one “1,” and FWL containing all zeros is an obvious extension. Now x^. walks through the thick dots in steps multiple of Ax, the step being determined by the position of the rightmost “1,” and there is no tnincation.

F W H

F W L

Oxxxxxxx

1000000000000000

fi

e

Another important case is FWL containing all zeros, but the MSB is equal to one

PHASE

NOISE

IN

COMPONENTS

121

The accumulator content xincrements in steps exactly equal to A.v/2, hence the truncation error is asquare wave of peak-to-peak amplitude Aa/2. This is the condition of maximum spurs. The maximum spur-to-carrier ratio is

2-2", that is, -6.02 pdBc/Hz. The accumulator contents is periodic. It starts from zero and it rst returns to zero after anumber of clock cycles

called Grand Repetition Period^ (GRP). The maximum GRP is equal to 2'” clock periods, thus 2"7/^^ seconds, and occurs when is an odd number (the FW has the LSB equal one). The GRP can be rather long. For example, with

/^,i( =1GHz and m=48 bits, the GRP is 2"^** =2.81 X10'“* clock periods, thus 2.81 X10^ s(3.25 days). If the FW contains rtrailing zeros F W H

F W L

Oxxxxxxx

0000000001000000

m - 1

0

the GRP is given by GRP =2"'-'

(2-105)

This is rather obvious because xis incremented in steps odd multiples of 2'", thus the lowest rbits of the accumulator will always be zero. We can see this as asmaller DDS, where the accumulator has m-rbits. The simplest way to understand the truncation spurs is to interpret AL as an accumulator, which behaves in the same way as the full accumulator (Figure 2-41). In fact, the content increments in steps and over ows, exactly as Xdoes. The trivial difference, that the over ow of AL goes to AH while the over ow of AH is discarded, is not relevant here.

Since AL has anumber b=m-p oibits, its operation is described by the equation

Hmi =^L.k +mod 2''

(2-106)

where the increment

Ne =

for/7^ 2

(2-107)

6 - 1

is the content of the equivalent frequency word (EFW). results from the following reasoning. If the MSB of the

FWL is zero (A'^ /^. With agood design, the white noise of the buffer should not degrade the oscillator noise. We expect this because the Friis formula applies to white noise, and the power level is generally higher at the input of the buffer than at the input of the sustaining ampli er. By contrast, cascading several ampli ers, the b_, coef cients add up in away that is independent, or almost

independent of the carrier power. When/^ /l= 1.25 Hz

fl

fi

fl

fi

fi

fl

fi

fi

fl

Figure 2-64 Phase noise spectrum of the Rakon HSO 14 OCXO measured with aMicrosemi 5120A test set. The spectrum is ©2019 Rakon France SAS, courtesy of Patrice Canzian and Vincent Candelier. Graphical adaptation and comments are ours.

MEASUREMENT

OF

PHASE

NOISE

153

Using the formulas of Table 2-6, the modi ed Allan deviation is 4.9 XlO

- I 6

5 . 2 X l O-

modc7y(T) =

I S

-F4.9X lO

- 1 4

Unlike in the previous examples, white and icker PM (the 1/t-\/7 and l/r terms) provide useful information because the corners where white PM crosses

icker PM and where

icker PM crosses

icker FM, are of 8.9 and

106 ms, respectively, which is still in the range of practical measurements. Random walk, temperature uctuations, and frequency drift are not visible on Figure 2-64. Such phenomena will inevitably show up, however, only for T>lO s, which is the reciprocal of the lowest frequency (O.l Hz) available on the phase noise plot. Given the applications this oscillator is intended for, rather specialized in the long-term performances, the Allan

deviation is preferred to the modi ed Allan deviation. Setting/^ =5Hz, we nd - L S

9.6 X10

ffy(T) =

-F5.9X 10

- 1 4

T

with acorner at r=162 ms.

It is interesting to compare mod o-y(r) to ffy(r). As we have seen, mod (7^(t) provides separate values for white

PM and icker PM. with no need of alow-pass lter. The low-pass, however, is implied in the sampling interval Tq. By contrast, ct^{t) provides asingle value for both, proportional to l/r, with astrong effect of the low-pass lter

on the contribution of white PM noise. Additionally, mod fTy(r) always gives values lower than those of

2-5 THE MEASUREMENT OF PHASE NOISE

We have already seen that the measurement of SSB noise referred to the carrier power has been abandoned long time ago, replaced with the direct measurement of the phase uctuations versus an appropriate reference. Some general-purpose spectrum analyzers include the dedicated hardware that enables the measurement of the phase noise associated to an input signal. However, these instruments are limited by the stability and by the noise of their internal oscillator and synthesizer, and they are usable only for the measurement of some rather noisy oscillators. Instead, dedicated instruments are the right choice. Three basic ingredients are needed for the measurement of phase noise: ●Phase reference ●Phase detector

●Signal processing unit based on FFT and averaging.

fi

fl

fl

fi

fl

fl

fi

fl

fl

fi

fi

fi

fl

The phase reference is an oscillator or asynthesizer, which provides asuitably pure signal. The phase detec¬ tor converts the phase difference, input versus reference, into avoltage or other signal. The DBM—or diode ring—saturated at both inputs is in most cases the preferred phase detector because of its low background noise. Digital detectors, like the XOR gate and the PFD, are not suitable to general test equipment, mainly because of their background noise. All these detectors require that the phase reference is at the same frequency of the input signal. The DBM is not the only option for the phase detector. Other types of instruments are found, based on direct digitization of the input signal, and on software de ned radio (SDR) techniques. These digital techniques are more exible, overcome some of the problems of the saturated mixer, and enable to compare the phase of two signals that are not at the same frequency. However, the noise of the ADCs is the major problem of such instalments. Most modern instruments make use of two separate and equal channels that measure simultaneously the input signal. The background noise is rejected thanks to an appropriate correlation-and-averaging algorithm which relies on the hypothesis that the two channels are statistically independent. The use of coaelation relaxes the noise speci cations for the reference oscillator and for the phase detector, at the cost of longer measurement time. The measurement of noise below the background noise of asingle channel is possible. As aconsequence, correlation

fi

fl

THE

154 ALMOST ALL ABOUT PHASE NOISE

and averaging allow the use of asynthesizer as the reference in each channel, which is generally noisier than a dedicated low-noise oscillator. Without synthesizers, aspeci c low-noise reference oscillator is necessary for each frequency of interest. In this section, we will learn about this type of equipment, principles, background noise and other limitations, and some tricks useful to extend the range of application.

2-5-1 Double-Balanced Mixer Instruments

The basic measurement scheme, shown on Figure 2-65, is straightforward. Unlike the regular use of the mixer, the

local oscillator (LO) and RF signals are synchronous (/lo =/rf =/o) quadrature, and large enough to saturate the input. In this condition, the difference/^q -/j^p degenerates to adc signal sensitive to the phase 9 V=K„e

(2-164)

The sum/pQ -H/jjp falls at 2/„, which is ltered out. The value of Kg can be up to 0.3-0.7 V/rad in favorable conditions (Figure 2-66). Alow-noise ampli er (LNA) is needed at the mixer output to raise the small signal to alevel suitable to the FFT analyzer. For an introduction the DBM, the reader can refer to the author’s earlier work [80], to an old but good white paper from Watkins Johnson [81], and to the classic Maas book [82].

COs(aiQf-I-6) RF

0 N

Cut 2o),0 IF

o

>

e

o

,

03

LNA

c

03

LO

sin((Uof) Figure 2-65 Basic phase noise measurement. Reprinted from [1], CC BY Rubioia, and adapted to our notation.

+1.0

„ + 0 . 5 > 03 D )

o >

0

13 Q . a

O - 0 . 5

- 1 . 0

0°

4 5 “

90“

135°

360“

P h a s e d i ff e r e n c e

’

fi

fi

fi

Figure 2-66 Phase-to-voitage conversion of adoubie-balanced mixer saturated at 15-20 dBm power at each input, plotted for different vaiues of the ioad resistance. Reproduced with permission from [80].

THE

MEASUREMENT

LO in

OF

PHASE

NOISE

155

R F i n

©

■ a c

§

! 0

IF out

Figure 2-67 Double-balanced mixer. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

The mixer is implemented with adiode ring and baluns, similar to the circuit shown on Figure 2-67. How¬ ever, the actual implementation may be more complex. Baiuns are present at both RF and LO input to match the

unbalanced-mode input to the diode ring, which is balanced. Two types of balun are often found. At microwaves frequencies, multi-section microstrip lines are preferred, providing atypical bandwidth is of 1-3 octaves. Wider bandwidth comes at the cost of larger physical size and higher loss. The HF-UHF implementation is based on iron powder transformers that exhibit abandwidth of up to three decades. Awider bandwidth, up to four decades, is achieved with asmart transformer, where twisted pairs are wound on an iron powder core shaped in toroidal form or in binocular form. Primary and secondary windings are coupled magnetically lower frequencies, while capacitive or electromagnetic-line coupling takes over at high frequencies. The boundary between RF-type and microwave-type implementation is in the region of 3GHz, depending on design and manufacturing choices. The Schottky diodes are preferred because of low threshold and fast switching time. High-level mixers, up

to 200 mW (-f23 dBm) or more input power, are convenient because of the higher value of Kg. and in turn the lower background noise. These mixers differ from Figure 2-67 in the use of 2-3 diodes in series in each arm of the ring. In each arm, an appropriate network distributes power and reverse bias equally between the diodes. By

contrast, the double-DBM (sometimes called triple balanced mixer) cannot be used as aphase detector because the IF output cannot be dc coupled. Special mixers intended as phase detectors achieve higher gain by increasing the IF impedance to 500 Q. typical.

Proper switching operation requires that the IF output current can circulate at both dc and 2/^. The problem arises in wideband mixers, where the circulation of the 2yj, current has to be ensured by the load at the IF output, the reason being that the upper IF frequency falls in the LO/RF frequency range. In such cases, the low-pass lter must have resistive or capacitive input impedance, not inductive impedance. It has been reported that aseries resonator at the IF output, tuned at 2/q, is useful in that it maximizes the IF current at 2/g, and increases Kg. Of course, this trick is reserved to special cases, where the experimentalist is interested in asingle value of/(,, or at most in asmall set of frequencies, and has full access to the system inside. The DBM is an appealing choice for aphase detector because of the low background noise, the wide range of operating frequency, and the overall simplicity of the system. Most general-purpose DBMs are suitable as phase detectors. In the absence of speci c information, one can assume that best power is 3dB above the nominal LO power, and that the same power should be used for the LO input and for the RF input. One can also assume that phase-detector bandwidth is of the nominal bandwidth.

fi

fl

fi

fi

The narrow power range, typical of the DBM, can be annoying. The problem is that the input power must be suf cient to saturate the mixer, but smaller than the absolute maximum rating level, with asafe margin. Unfortu¬ nately, the gap between nominal and maximum power is not comfortable and leaves approximately ±5 dB around anominal power of 10-15 dBm. At lower power, the background noise increases. Further decreasing the power. Kg drops abruptly and the mixer is no longer usable. The mixer inputs, strongly saturated, have highly nonlinear behavior and the input impedance changes with frequency. Strong odd harmonics of the carrier frequency are re ected back, combining in rather unpredictable

156 ALMOST ALL ABOUT PHASE NOISE

way depending on cable length. For this reason, it is agood practice to introduce a3-dB attenuator as close as possible to the mixer inputs.

The Measurement of Oscillators

The basic scheme for the measurement of the PM noise of oscillators is shown on Figure 2-68. Taking the error voltage Vas the output of the mixer, the PLL is used as ahigh-pass lter. So, beyond acutoff frequencythe error signal is asymptotically equal to V B e l o w t h e c u t o ff , t h e e r r o r s i g n a l i s s m a l l , b u t t h e phase noise can still be calculated using the equation of the PLL. It is useful to bring the reference oscillator as close as possible to the device under test (DUT) frequency by adjusting the dc offset, so that the detector and the

control work close to 0V. In this condition Kg is the highest, and the measurement starts with the instmment in the middle of the dynamic range. In production and in industrial applications, it is generally possible to rely on areference oscillator whose PM

noise can be neglected, being L^pp(/)with asuf cient margin. In rare cases, we may have to test special low-noise oscillators, where no lower-noise reference is available. As a rst approximation, we can measure two equal oscillators, so that the noise of each is half (-3 dB) of the result displayed by the test set. Dropping the hypothesis that the two oscillators are equal, the reliable measurement of a single oscillator with the scheme of Figure 2-68 is acomplex and time-consuming task because we need to compare all the possible pairs in aset of at least three similar oscillators and to solve for the noise of each. However, the cross-spectrum method provides asimple and practical solution, discussed later in this chapter. Notice that in Figure 2-68 we have kept the reference oscillator outside the test set. This is often necessary in ageneral-purpose instrument because alow-noise reference is needed, at the same frequency of the oscillator under test. Neither awideband VCO nor asynthesizer would feature the low noise needed to meastire high purity oscillators. Introducing asynthesizer for exible operation requires the dual-channel scheme, which we will study l a t e r.

The PLL error function E(s) =9^{s)/9j{s) is given by E(s) =]-B{s) =

(2-165)

1+G(s)H{s)

Including the gain of the LNA in Kg, the closed-loop error voltage KgEis) is described by the transfer function T{s) =

V{s) 9is)

=Kg[\ -B{s)]

(2-166)

General-purpose PM noise test set RF

IF

V = K na O

-LN^

LO

0 ) N CO c CO

REF(/v Control V,V C O

^DC

o

DC REF

fi

fi

fl

=

fi

Figure 2-68 Phase noise measurement of an oscillator using the error signal of aPLL.

THE

MEASUREMENT

OF

PHASE

1 5 7

NOISE

For the simplest loop, where G(s)H(s) =KgKJs, the function T{s) is a rst order (single pole) high pass lter KgS

T{s) =

(2-167)

S+KgK„ or equivalently

\T{f)?=Kl

f

(2-168)

where is the cutoff frequency 1

(2-169)

/hp -271^^^°

In commercial test sets,/fjp is generally chosen by an internal algorithm, and only advanced users can take control on it. However, the implications of/j,p deserve attention. Naively, one may be inclined to set/pjp at avalue lower than the lowest analysis frequency. For example, being interested in L(/) from 10 Hz to 100 kHz, we would

choose/[jp =1... 2Hz, so that \T(J)\^ =Kj (constant) in the full span. However, atighter loop is abetter choice, with/jjp set approximately at the corner between the 1//^ noise, or the 1//^ noise, and the white region (Figure 2-69). The instrument measures S^(f), that is, the PSD of v, and calculates L(J) as Lif) =

1

2|T(/)|2

Of course, this relies on the accurate measurement of IT(/)F in actual conditions, which can be accomplished by modulating the VCO signal. A rst advantage of this approach is the reduced burden for the FFT’s dynamic range.

This is quite obvious from Figure 2-69. The oscillator Sg(f) has awide dynamic range (plot A) because of the 1//^

and 1//'* behavior. By contrast, Vrequires acomparatively smaller dynamic range because its spectrum (plot B)

S(f)

1//'^

(A) Oscillator S,,(/), calculated

(B) Sp (/), measured

1// White

IP

(C) High-pass IV/0F log-log scale

/

Ap

Figure 2-69 Tight PLL for the phase noise measurement of oscillators. An arbitrary constant is added to the plots for better

fi

fi

fi

readability of the plot.

158

ALMOST

ALL

ABOUT

PHASE

NOISE

contains at most Hf- components at low frequency. Asecond and more subtle advantage is that the tighter lock overrides some uncontrolled effects of electromagnetic interferences and in turn provides more reliable results. Electromagnetic interference is sometimes asource of erratic or wrong results, dif cult to identify and x. RF/microwave leakage is to some extent inevitable, due to connectors, coaxial cables, power lines, grounding, insuf cient shielding, etc. The problem arises from the fact that the reference oscillator and the oscillator under test are at the same frequency. The power leaking from one oscillator builds up as asigni cant energy in the second oscillator, after integration over the relaxation time of the internal resonator. The resonator’s relaxation time may be unexpectedly long, up to hundreds of milliseconds in the case of high stability 5-10 MHz OCXOs. Of course, reciprocity makes the stray coupling bidirectional. Leakage may injection-lock the two oscillators to one another or corrupt the transfer function T{s) if coupling is insuf cient for locking. Interestingly, injection locking is aphase sensitive phenomenon. With the same amount of power leakage, the oscillator may lock or not, depending on the electrical length of the path. Should aphase noise spectmm be suspected of being corrupted by leakage, the following tests are recommended: Opening the loop when the two oscillators are set as close as possible to the same frequency, they phase lock t o o n e a n o t h e r.

In open loop condition as earlier, the beat note is not sinusoidal. Instead, it slows down or almost stops when certain phase relationships are met. In this case, it is likely that the two oscillators try periodically to lock to one another when the phase relationships are favorable, but coupling is insuf cient and injection locking fails.

In the 1//^ or 1//'* region of Sg(f), the slope tends to decrease or get at toward low frequencies instead of getting steeper.

Changing the length of critical cables affects the low-frequency region Sg(f). The critical cables are those connecting the oscillator under test to the mixer, or the reference oscillator to the mixer. The shape of Sg(f) changes after introducing acommon mode lter—a ferrite ring or clamp—along acable, RF output, VCO input, or power supply. If any of the previous symptoms show up, the experimentalist should be aware that the phase noise measurement is unreliable. Investigate on the transfer function T(s) is recommended.

Background Noise, Spurs, and Other Experimental Issues

Recalling the scheme of Figure 2-68, we identify the following contributions to the instrument background noise: ●Mixer

●LNA between mixer and FFT analyzer ●Reference oscillator

●The dc reference, providing the tuning voltage ●Pollution from AM noise (CF section “The Effect of AM Noise”)

Let us start with the mixer and the LNA. They must be analyzed together. The typical background noise is shown on Figure 2-70 and discussed in the following text. Lowest noise operation requires high driving power, at least

●+●10 dBm. The mixer adds little white PM noise because its noise factor is of some dB [83], but it has alow gain Kg. The consequence is that the white noise oor is set by the LNA after the mixer. The reader can refer to [84] for the design of LNAs speci cally intended for the lowest background noise in this type of applications. Unfortunately,

we do not have analytical expressions for Kg and for the mixer noise factor. The icker noise is an experimental

fi

fi

fi

fi

fl

fi

fl

fi

fi

fi

fl

fi

fi

fi

fl

parameter, for both the mixer and the ampli er. We see on Figure 2-70 that the total noise is signi cantly higher than the noise of agood ampli er divided by the mixer gain. The following examples show typical values of white and icker PM noise, and their origin.

THE

MEASUREMENT

OF

PHASE

NOISE

1 5 9

N

X - 1 0 0 C\J ■ a

CO CQ ■ O

-120- 1// Microwave

CO -140

J// HF-UHF

- 1 6 0 ! >1

H -

fc \

-180 1

1Q

1 0

10

10“

106 /(Hz)

1 0 ®

Figure 2-70 Typical background noise of amixer, inciuding the low-noise ampli er that follows. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

Example 19 Mixer and LNA white noise. Let us calculate the white PM noise background assuming that the

mixerhasKg=500mV/raddrivenat40mW(-1-16dBm),andthatthewhitenoiseoftheLNAis=1.25nV/\/Hz. including the 50 £1 input load.

We rst convert e„ into S,,(/) =e® =1.56 X10“'* VVHz, that is, -178 dB VVHz. Going backward to the input, the background noise is - 1 8

rad^Hz

Sg{f) =SAf)/Kj =6.25x]0

(-172 dB radVHz)

For comparison, the thermal noise at the mixer input is

Sg{f) =kT/P =4x\0-^'/4x\0~^ =10“'VadVHz (-190 dB rad^/Hz) In this example, it takes anoise factor of 18 dB for the mixer noise to match the noise of the ampli er. Choosing

different components and parameters, the result does not change signi cantly. For example, the highest Kg found in acommercial instrument is of 1V/rad, which requires -1-20 dBm input power.

Notice that the value of of this example is quite optimistic because it includes the thermal noise \/4kTR of the input resistor, 0.9 nV/-\/Hz with /? =50 Qat room temperature

■

Example 20 Mixer and LNA icker noise. We use the mixer of the previous example. Kg =500 mV/rad, and the LNA designed for PM noise applications [84], which exhibits 1.6 nV icker (-176 dB V®). Referring this value to the input, we

nd

Seif) =S,{f)/Kl =(2.5 X10“'*//) /0.5“ =10“'7/rad® (-170 dB rad^) This is 30 dB lower than the overall noise shown on Figure 2-70, which refers to the HF-UHF mixers. Such margin may be reduced by 10 dB with better mixers, if any, and with anot-as-good ampli er. Anyway, the result yields safely to the conclusion that the background noise is chie y originated in the mixer ■

At low frequencies, the dominant phase noise in oscillators is 1//^, 1//®, and higher slope types, while mixer and low-noise dc ampli er have only white and 1//noise. In this region, the phase noise of the reference oscillator

fi

fi

.

fi

fl

.

fi

fl

fl

fi

*

fi

®

fi

2

is generally the most severe limitation to the measurement.

It is often necessary to provide adc voltage at the VCO input to bring the oscillator at the nominal frequency, as in Figure 2-68. The voltage noise of this source turns into FM noise at the oscillator output. Thus, white and

icker noise show up as white and icker FM noise, whose slope is \lf~ and 1//^ on the phase noise spectrum. In principle, the contribution of the dc source should be made smaller than the oscillator noise. This is not always possible, chie y in the case of oscillators having low phase noise and high voltage-to-frequency gain. By contrast, the control provides only the small correction needed to keep the oscillator locked during the meiisurement. For this reason, in agood design the uctuations coming from the control fall below other noise contributions. Having seen unexpected experimental mistakes, we strongly recommend at least aquick check on the noise sent to the VCO input. The FFT analyzer is preceded by an LNA. Thus, an appropriate choice of the ampli er and of its gain makes the noise of the analyzer negligible. That said, the noise of the analyzer deserves more attention in earlier FFT analyzers, where the low-frequency decades were obtained by reducing the sampling frequency of the converter.

The problem comes from the quantization noise, whose variance is a- =F^j,j,/12. The Parseval identity states that where 5,,(/) is the white noise oor, and//2 is the bandwidth of the quantization noise, equal to half the sampling frequency/^.. Thus, the quantization noise is 5,,(/) =Flsb/^-^s analyzer, this is seen as a staircase-shaped noise oor, increasing steadily toward the low-frequency decades, where the sampling frequency is progressively lower. The problem is solved in modern analyzers. The input ADC runs always at full speed, and the lower sampling rate is obtained by data decimation after digital low-pass ltering.

Asymmetric Driving for Low-Power Signais

The mixer is unsuitable to low power signals because Kg decreases. This impacts strongly on the white noise oor, and icker PM noise tends to increase at low power. Below athreshold power, Kg drops suddenly, and the mixer is no longer usable. However, asymmetric power driving is possible, with the LO input saturated, and the RF input in the linear regime, say, at apower 10 dB lower than the LO nominal power, or even less. This may be convenient for the measurement of oscillators and of two-port components, when only the reference signal has apower suf cient to saturate the mixer.

In the asymmetric power driving, the mixer works as asynchronous detector. This mode is broadly similar to the regular “superheterodyne receiver,” differing in that LO and RF frequency is the same, thu degenerates to dc. The LO signal is

'^Lo(0=V^satSin(®oO

(2-170)

where the peak voltage results from saturation. Using the approximation cos 6^1and sin

0 for small 0,

the RF signal Vr[,(0 =Vq cos(®„?+ 6>) becomes

='^0 cos {mot) -eVo sin [mot)

(2-171)

Dropping the 2®,, term, the detected signal at the IF port is V = V ,0A 6

(2-172)

where Ais the mixer loss written as a“gain.” For example, aloss of 6dB translates into A=0.5 because

10-6/20 _Q5Prom the de nition of the phase-to-voltage gain Kg =VW, we nd K ^9

Example 21

= F 0. A

(2-173)

We use amixer that has aloss of 8dB (A =0.4) when the LO port is saturated at -1-18 dBm

(Vja, =2.5 Vacross 50Q load). Sending a-12dBm signal {Vq =80 mV across SOD load) to the RF port, in

fl

s

0

fi

'

fi

fi

.

fl

fl

fl

fi

fl

fl

fi

■

quadrature with the LO signal, the phase-to-voltage gain is Kg =32 mV/rad

fl

fl

160 ALMOST ALL ABOUT PHASE NOISE

MEASUREMENT

OF

PHASE

NOISE

161

Heterodyne Measurement of Oscillators

The heterodyne method (Figure 2-71) is agood option to extend the range of aphase noise test set to higher frequencies by exploiting alow-frequency beat at= |/.withand alsoIn atypical case, we compare two microwave oscillators by bringing the beat down to the HF region. Of course, asuitable reference oscillator must be available, and some auxiliary pieces of hardware. In open-loop conditions, the phase uctuation of the beat note is

e, =e,.~e^

(2-174)

So.b(f) =S,i(f) +Sg,.(J)

(2-175)

thus

Interestingly, the scheme of Figure 2-71 takes bene t from aleverage effect, which relaxes the frequencystability speci cation for the VCO and for the synthesizer by afactor of fjfj. This leverage effect is adirect consequence of the fact that the beat mechanism stretches the time associated to aunit of phase (radian) by the

factor/j.//^. This is particularly useful in the l//% I//^ and steeper regions of L(/). For the purpose of extending the frequency range, the heterodyne scheme is preferred to afrequency divider because of the lower background noise.

(A) Speci c hardware

OUT

Generai-purpose PM noise test set (D N

DC

R R

IF

— L N / V

o

o

CO c CD

-LN/5

LO

REF

fr Synthesizer VCO

Control

(B) DUT

Speci c hardware

Generai-purpose PM noise test set

DC IF

Aux REF

LN^

CD N CO c

CO

t

LO

Synthesizer

VCO Control

fl

fi

fi

fi

Figure 2-71 Heterodyne (beat) method for the phase noise measurement of oscillators with asingle VCO (A), and with an auxiliary VCO (B). Reprinted from [1], CC BY Rubiola, and adapted to our notation. ER Slideshows, public domain material.

fi

fi

THE

162

ALMOST

ALL

ABOUT

PHASE

NOISE

In the scheme of Figure 2-71 A, the reference oscillator also drives the synthesizer. This may be imprac¬ tical because commercial synthesizers accept only some round values of the reference frequency, typically 5-10-100 MHz. An alternate scheme is possible, shown in Figure 2-7IB. In this case, the main reference oscillator

is free running, with no control, and the auxiliary reference is phase-locked to the beat note= \ ~f,\. The hardware is clearly simpler than on Figure 2-71 A, and the bene t of the leverage effect is the same. Besides the microwave practice, Figure 2-7IB solves some dif cult problems of PM noise measurements, beyond the scope of this book. For example, the metal-semiconductor (Schottky) diode can be used to down convert from the THz region to HF or VHF. Similarly, the fast PIN InGaAs photodetector is routinely used in metrology labs to beat 1550-nm telecom lasers down to microwaves.

The Measurement of Ampli ers and Other Two-Port Components

In the case of two-port components, we opt for the differential measurement scheme shown in Figure 2-72. Once the quadrature condition is set, the mixer delivers avoltage proportional to the instantaneous phase uctuation of the DUT. The oscillator PM noise is common mode, thus it is rejected. In spite of this, practical measurements are way more dif cult than Figure 2-72 lets us believe. The PLL scheme (Figure 2-68) is simple to use, to the extent that the quadrature condition is set automatically and precisely by the feedback. By contrast, in Figure 2-72 the quadrature condition relies on an adjustable phase shifter manually set by the operator. The reference arm is the preferred location for this phase shifter because it is independent of the DUT. Atuning range of 180°, with a comfortable margin, is suf cient because any of the two quadrature points at ±90° can be used, with equivalent results. Different types of phase shifters can be used, depending on frequency. Mechanical phase shifters (U-shaped line stretchers) are appealing for their low noise, ne tuning capability, and wide frequency range. One of us (ER) has used extensively the phase shifters manufactured by ARRA for research applications. Aproblem with the line stretchers is the small delay range, related to the physical size. For reference, arange of 1ns is equivalent to 30 cm change in the electrical length, thus of approximately 25 cm physical excursion. Of course, the range can be extended with aset of electrical cables of known length, but the operation is tedious and time consuming. A90° directional coupler terminated to varactors at two ports is an excellent phase shifter, provided the noise of such varactors is low enough compared with the DUT. The frequency range is limited by the 90° coupler. For lower noise, the varactors can be replaced with variable capacitors, but in this case the adjustment is dif cult and time consuming. Aclassic solution suitable to HF-VHF (Figure 2-73) is found in an article by Phillips [85]. In summary, experience suggests that it is almost impossible to combine the suitable range of phase with awide range of frequency. The reason is that electronics gives wideband control on delay, or narrowband control on phase. The fact that delay and phase are related does not really help to get the >180” excursion we need. The measurement of two-port components challenges the background noise of the instrument. This happens because these components often exhibit very low noise and because the noise processes are of the same type of those of the instrument, that is, white and icker PM. For example, the icker PM noise of an RF ampli er may be of the same order of that of aDBM. The background noise is discussed in section “Background Noise, Spurs, and Other Experimental Issues”. The measurement of ampli ers is always tricky because it is necessary to match both input povt'er and output power to the instrument. Aproblem is that the mixer has anarrow power range. Another problem is that white

3 d B

cpI

D U T

RF

o N

I F

7

A

LNA

c

os

LO

90° adjust

fi

fi

fl

fi

fl

fi

fi

fi

fl

fi

fi

fi

fi

Figure 2-72 Phase noise measurement of atwo-port component. Reprinted from [1], CC BY Rubiola.

THE MEASUREMENT OF PHASE NOISE 163

In

I Out

a

O o

Figure 2-73 Example of variable phase shifter.

3 d B OUT

cpI

R F

0 ) N

IF

o

o

CO c

CO

DUT

LO

7 90° adjust

Figure 2-74 Phase noise measurement of frequency dividers and multipliers, and other devices whose output frequency is not

equal to the input frequency. Reprinted from [1], CC BY Rubiola.

phase noise increases at low input power. In practice, it is often necessary to introduce appropriate attenuators at both input and output of the ampli er, whose attenuation must be determined for each case. Finally, frequency multipliers, dividers, and synthesizers are aspecial case because they deliver an output fre¬ quency that is not equal to the input frequency. The scheme of Figure 2-74 solves the problem by using two equal DUTs, so that the mixer receives the same frequency at the two inputs. Of course, this method gives the total noise of the two DUTs, with no means to divide the noise contribution of each. We rely on the assumption that the phase noise of the two DUTs is the same, and we take away 3dB for the phase noise of one. This method may also be useful in other cases, for example, in LNAs, where the enhanced sensitivity due to the presence of two DUTs helps to get out of the background noise.

The Discriminator Method

Figure 2-75 shows amethod to measure the PM noise of an oscillator using adelay line as the frequency reference, so that the signal at the mixer output is Kg[6(t) -9(t -t)]. The measurement is possible because the delay line de-correlates the phase noise, under some conditions. This method is useful for fast phenomena, not for random walk and drift. The dynamic range is limited by the amount of delay that can be introduced without excessive attenuation and by the background noise of the mixer and of the following circuits. Lance et al. [86] used coaxial cables, enhancing the sensitivity with the cross-spectrum method discussed later in Chapter 7. For long delay, up to 10-20 ps, the optical ber proved to be an ef cient solution [87] because of the extremely low attenuation, 0.2dB/km, or 0.04dB/ps delay. The delay at microwave frequency is obtained by modulating and detecting the intensity of alaser beam. Improved sensitivity is achieved with the cross-spectrum method, using two statistically independent instruments, which measure simultaneously the same oscillator [88, 89].

The response of the system is de ned as

fi

fi

fi

fi

|T(/)l' =

V{f) ' Q(f)

(2-176)

16

ALMOST ALL ABOUT PHASE NOISE

(A)

3 d B

cpI

delay x

RF CD

OUT

IF

●

o

t N

o

LL CD c

(0

LO

90° adjust (B) e(s) exp(-sr) V(s) K

0

+

Figure 2-75 (A) Phase noise measurement of an oscillator using adelay line as the reference. (B) The phase-noise equivalent circuit. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

where V(f) and 0(/) are the Fourier transform of the output voltage, and of the oscillator randcrm phase. This transfer function is easy to derive analytically using the phase step method that we have seen in section “The Resonator and Its Impulse Response” with the response of the resonator. We use the Laplace transforms V{s) and 0(j), where j=o- -l- jxa is the complex variable. By inspection on Figure 2-75, the line delays the phase perturbation and the impulse of phase by the same amount t. The output voltage is (2-177)

F(.v) =/f,[l -e-'^]0(T) With simple manipulations, we nd

\V(jt =K,[\ =Kj[ | 0 ( / ) =AK]%mH2Kft)\Q(f)\^ and

f

-

nally

\T(S)? =AKl&m\27xfx)

(2-178)

The aforementioned equation is exploited to calculate 1

2-AK^gSm'{2nfx)

S^f)

(2-179)

from the PSD S^,(f) of the voltage measured by the FFT analyzer. Notice that (2-179) has singularities at/= m/2t, integer n, where L(J) cannot be calculated. At/0, the transfer

function is approximated with \T(f)\~ =IbK^Tu^r-f^, and the instrument has apoor sensitivity due to background noise. For n>\, L(f) shows large and sharp peaks due to the background noise. In practice, the system is usable u p t o / « 0 . 8 / 2 T.

The delay line can be replaced with areference resonator, as shown on Figure 2-76. In the rst scheme from the top, the resonator is used as the reference for the measurement of an oscillator. Of course, it is necessary that

fi

fi

fi

\

4

the resonator is more stable than the oscillator.

THE

(A)

MEASUREMENT

OF

PHASE

165

NOISE

3 d B

u

cpI

RF

0 ) N

REF

DUT

IF

O

CT3

O

c

03

LO

7 90° adjust (B) 3 d B

cpI

u

R F

DUT

REF

0 N

IF

CO

oo

c

CO

LO

7 DUT 90° adjust (C)

0(s)

1/t S+1/T

V(s)

Figure 2-76 Phase noise measurement of an oscillator using aresonator as the reference (A). The roles can be inverted (B), using the oscillator as the reference for the measurement of aresonator. In this case, it is convenient to use two equal DUTs. (C) We

see the phase-noise equivalent circuit of the rst scheme. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

In the second scheme the roles are interchanged, and the resonator is the device under test. In this case, it is convenient to use two equal DUTs having the same resonant frequency and the same Qbecause in this case the noise of the reference oscillator is rejected. The third scheme, on the bottom of Figure 2-76, is the phase-noise equivalent circuit used to measure the input oscillator versus the reference resonator. We use this scheme to derive the frequency transfer function

\T{fy? assuming that the resonator natural frequency/„ is equal to the carrier frequency/q. We follow the same methods just used for the delay line, but the resonator’s phase response is B{s) “The Resonator and Its Impulse Response”. There follows that

T{s) =Kg \-

1/t i-F 1/t

5-FI/t

+ 1/t), as seen in section

(2-180)

thus

f

(2-181)

I

=

fi



7 +

166

ALMOST ALL ABOUT PHASE NOISE

Using the quality factor Qand the natural frequency /„ =/j, of the resonator, the transfer function is better rewritten as

f r- +

(2-182)

fn 42^

Finally, the oscillator phase noise is 1

Ssif) =

\T(f)\

(2-183)

s,(f)

r-

0

At/-> 0, the response of the instrument is poor, and dominated by the background noise. However, the measurement does not suffer from the in nite series of singularities as in the case of the delay line. The resonator has its minimum

loss at/„ =/„. This enables to pump correctly the mixer at the two inputs.

2-5-2 The Cross-Spectrum Method

The scheme of the dual-channel measurement, shown on Figure 2-77, consists of two equal branches that measure the same oscillator using the PLL method. The main point is that the noise of the reference oscillators, of the mixers, and of the LNAs can be rejected using correlation and averaging. This is possible because the devices are physically separate, thus we can assume that their noise processes are statistically independent. By contrast, the DUT is common to the two branches, thus it is fully correlated, and captured by the statistical process. The cross-PSD relates to the Fourier transform of the correlation function. Thus, averaging on mmeasures of

Sg(f),thesingle-channelbackgroundnoiseisrejectedbyafactorofapproximately1/^/m.Itisthereforepossible to measure aphase noise Sg(f) lower than the background noise of asingle branch. Figure 2-78 shows what happens during the measurement process. The DUT noise (C) is lower than the back¬ ground noise in single-channel mode (A), thus the measurement is possible only after rejecting the background. The instrument displays the cross PSD averaged no macquisitions. The cross PSD starts from the single-channel back¬

ground (A) and is progressively reduced proportionally to 1/-y/m. With small m, the single-channel background is not suf ciently rejected, and the instrument displays the plot (B). When mis large enough, the single-channel background is well rejected (D), and the instrument displays the DUT noise (C).

Control VCO

REF

Synthesizer

IF

V L O ,

■ 5

0

LNA

X=Kg(e -ip)

RF

t u

.

0) Q) c N

e

DUT /\y

C

cpI 3dB

>

>

V § f

RF R E F

Synthesizer

W

LO,

y=Kg{0-ip)

IF O

O

Q

LN,

VCO Control

fi

fi

fl

Figure 2-77 Dual-channel phase-noise measurement system. Reprinted from [1], CC BY Rubiola, and adapted to our notation.

THE MEASUREMENT OF PHASE NOISE

167

Soif)

_A) Single channel Dackground* noise; T

S,

(B) Averaging limit

"lihsuf cientm)

i

S^(C) PUT noise ♦

+

(D) Averaging limit

log-log scale

/

Figure 2-78 Rejection of the background noise in the dual-channel phase noise measurement.

In this version of the PLL method, we have introduced asynthesizer between the reference oscillator and the

phase detector. The obvious bene t is that the system is exible and suitable to awide range of frequencies, with¬ out need of aseparate reference oscillator for each frequency of interest. The higher noise of the synthesizer, as compared with an oscillator, can be tolerated, thanks to the noise rejection of the dual-channel scheme. Figure 2-79 shows asimpler version of the dual-channel system. It differs from the previous version in that there is only one reference oscillator, driving both the synthesizers. The trick is that each synthesizer has an internal frequency refer¬ ence, locked to the main reference. The appropriate cutoff frequency inside such synthesizers may be of the order of 0.1-1 Hz, depending on the interplay between the stability and phase noise spectra of the internal and external ref¬ erences. Thus, for/beyond the cutoff, the two synthesizers are statistically independent, and their noise is rejected. Below the cutoff, the entire measurement relies on the stability and on the spectral purity of the main reference. The Rejection of the Background Noise

After the heuristic reasoning, we explain the mathematics underneath the rejection of the single-channel noise. Amore detailed treatise of the cross-spectrum is available in [90].

Control

P L L

FC|

LO

I F

^ — L N ,

Synthesizer R F O U T

Q) 0) C N

S

cpI 3dB

P S CO 3

R F

LO

P L L

Q

IF

—LN,

Synthesizer FC R E F

Control

fl

fi

fi

Figure 2-79 Alternate dual-channel phase-noise measurement system. Reprinted from [1], CC BY Rubiola.

168 ALMOST ALL ABOUT PHASE NOISE

With reference to Figure 2-77, the two signals at the input of the FFT analyzer are x{t) =K, Wit) -cpit)]

(2-184)

y(r) =Kg Wit) -

(2-185)

where the gain is the same for the two channels and includes the trivial gain of the LNAs. The random phases cpit) and ipit) account for the noise of the references, the mixers, and the LNAs. Accordingly, 6it) is the random phase of the oscillator under test, with no additional terms. It is sound to assume that 0(0, (pit), and if/it) are statistically

independent because they come from separate hardware. Only 0(0 appears in both a(0 and y(0. As usual, we denote the Fourier transform with the uppercase letters

Xia) =Kg [©(») -®(®)]

(2-186)

Yico) =Kg[eico)-'i>iw)]

(2-187)

The cross PSD is given by

s,M)=I

(2-188)

where Tis the measurement time, the superscript stands for complex conjugate, and the factor “2” xes the scale factor from the two-sided Fourier transform into to the one-sided PSD. Using alighter notation where the frequency is implied, the aforementioned formula is expanded as

5,,^

- 0 O * - TO * - t - TO * )

(2-189)

The dual-channel FFT analyzer measures the cross-spectrum (Sy^if)),,, averaged on mdata records of .r(0 and yit) acquired simultaneously

-(0O*>„,-(T0*),„ +(TO*>

m

(2-190)

Arather intuitive theorem states that if two random variables are statistically independent in the time domain,

their Fourier transforms are also statistically independent. Thus, we expect that (0O*),„ ^0, (T0*)„, ^0, and (TO*)^ -> 0for large m. Consequently (2-191)

The process takes atime mT, not counting the computing time. It is useful to write the instrument readout as the estimation, denoted with the “hat” accent

Sgif) =^

abs

aliasing

Amplitude

c o s

s i n

(optional)

N C O -F

A

Frequency

Phase

control

word (in) Frequency control

Clock

word (ref)

V

N C O c o s

s i n

● -

atan —x(///r)

V

Ref o

c

-

A D C

Anti

abs

aliasing

Amplitude

(optional)

fi

fi

fi

fi

fi

fi

fi

fi

fi

Figure 2-82 Basic scheme of the direct-digitization phase detector.

1 7 2

ALMOST ALL ABOUT PHASE NOISE

limited by the lter roll off. However, more stringent limitations may apply, due to the architecture of the instrument, and to the processing speed. The CORDIC algorithm [92, 93] is most often used to calculate the phase. Interestingly, the digital technology enables the calculation of phase, and also of amplitude, with so high accuracy that it exceeds the general metro¬ logical performance of the instrument. The phase of the reference signal is scaled according to the 'equency ratio yi//]., so that it can be compared with the input phase. An alternate and elegant solution consists of converting the phase of both input and reference to phase-time. The two-branch con guration is necessary to bring the external reference (5-10-100 MHz, or arbitrary frequency) in the machine because the clock frequency takes xed values determined by design considerations. The con guration of Figure 2-82 has three relevant features, advantageous versus the DBM scheme ●It operates at arbitrary frequencies, with no need for the input and the external reference to be at the same frequency. ●The oscillator under test and the reference are free running, with no need of phase or frequency lock. ●Measuring atwo-port device, there is no need for aline stretcher or for avariable phase shifter to set the quadrature condition. These features enable the measurement of frequency dividers, multipliers, etc. in astraightforward way, without need of comparing two equal DUTs. The main problem of the scheme shown is the background noise, generally limited by the noise of the ADCs. For reference, the noise of aselected 12-14bit ADC at 100MHz sampling frequency, operated at full range is - 1 5

5e(/) =io-‘7/+io

radVHz

(2-199)

that is, -110 dB rad^ icker, and -150 dB rad’/Hz white oor. The icker PM noise is atechnical parameter of the ADC. as we have seen with ampli ers. The white noise results from the quantization noise and from the clock jitter (2-200)

The quantization noise can be calculated as follows. At full range input =V,H S R ,the carrier power on a1-D resistance is

P

V,?F, S R

=

(2-201)

With nbits (ENOB). the quantization noise power is F.2L S B (T“

FSR

12

(2-202)

12 x2^''

uniformly distributed from 0to the bandwidth B=fJ2. Thus FSR

B

(2-203)

6X2^"/,

The phase noise is given by Sg(f) =NIP, thus 4

(2-204)

fi

fi

fi

fl

fl

fi

fl

fi

fi

3X2^"/,

173

For example, a14-bit ADC with ENOB =12 bits, and sampling at 32 MS/s, has aquantization noise s

4 e.q

-

= 2 . 5 x 1 0

- 1 5

3x2^^x(32x lO®)

that is, -146 dB radVHz, or -149 dBc/Hz,

The quantity 5^ is the clock-distribution PM noise, which is atechnical parameter of the ADC, It hits on PM noise only, not on AM noise. For this reason, there is an asymmetry between AM and PM noise oor, and can be measured as B ^0,ck -SaS

Values of 0-3 dB and more are observed, depending on the operating conditions and on frequency. In fact, 5^ is

of the time type, while Sg ^is of the phase type. Spurs and artifacts are another problem of digital systems. A rst type of spurs results from sampling and digital synthesis. The sampling process produces (2-205)

4u,=Al/,-Af/,

(2-206)

Additionally, the NCO produces spurs at multiples of the grand repetition rate (GRR) /sS p u r

/ct

(2-207)

2^

where/j.,. sthe NCO clock frequency, and nis the number of bits of the NCO. However, the equivalent value of to be used here can be smaller than the actual number of bits in the NCO register, depending on the frequency control word. Torosyan suggests that the equivalent ti is the number of bits of the frequency control word from the MSB to the rightmost “1,” which of course depends on the output frequency. For example, a24-bit NCO has n=bits when the control word is 01110101 110101 111 1000001, and «● =18 bits when the control word is

01110101 11010111 11000000. Details are found in [94] and [95]. Distortion produces spurs at high frequencies, and aliasing brings them down to baseband. The digression we have seen with the DDS applies. At the state of the knowledge, the spurs cannot be eliminated, so they are generally removed from the displayed data in order to give the best representation of the DUT noise. Because of the high noise of the ADCs, across-spectrum con guration is necessary to reduce the background noise of the instalment. Two equal blocks like Figure 2-82 are used instead of the DBM and measure simultane¬

ously the quantity 6=6^- 0,.. As aresult of design choices, and probably also of marketing choices, commercial instruments often use asingle input for the external reference. Thus, the noise of the external reference cannot be rejected. Asmall number of digital instruments are commercially available, listed in Table 2-11. Some of them will be brie y discussed in the following pages.

The Microsemi Family of Phase Noise and Allan Deviation Tester

Microsemi (formerly Symmetricom) manufactures three instruments for the measurement of phase noise and Allan deviation. All these instruments are based on ascheme broadly similar to Figure 2-82. The background noise is rejected thanks to the cross-spectrum method. Input and reference are symmetrical, so they can be chosen indepen¬ dently in the range shown. Unfortunately, there is only one input for the external reference, thus it is impossible to reject the noise of the reference as we did in Figure 2-77. Figure 2-83 shows the block diagram of the 5120A [96], and Table 2-12 shows the background noise. This

instrument, probably the rst implemented with fully digital architecture, has no capability to work beyond the

fl

fi

fi

fi

rst Nyquist zone, thus the input frequency is limited to 30 MHz.

fl

fi

fi

THE MEASUREMENT OF PHASE NOISE

174

ALMOST ALL ABOUT PHASE NOISE

Ta b l e 2 - 11 D i g i t a l p h a s e - n o i s e a n a l y z e r s

Type and brand

Input frequency

5125 A

1-400 MHz

Analysis frequency

Note

100 pHz-lMHz

Performs L(f) and ADEV

Microseini

Discontinued, June 2018

5120A

1-30 MHz

100 pHz-1 MHz

Performs L(f) and ADEV

0.5-30 MHz

1Hz-100 kHz

Performs L(/), AM noise, and ADEV Requires ahost PC for all measurements Supersedes the Miles Design 5330A

1-200

1mHz-1

Performs L(/), AM noise, and ADEV Requires ahost PC for all measurements

Microsemi

3120A Microsemi

PhaseStation 53100A

MHz

MHz

Jackson Labs

1MHz-8/26.5/50 GHz

FSWP8

10 mHz-300 MHz

Performs L(f) and AM noise

FSWP26

Additional functions (VCO test, baseband,

FSWP50

and microwave spectrum analyzer, pulsed signals, etc.), some optional

Rohde Schwarz

Input

Split

^ADC \_

Frequency convert and

9 :i n

phase detect -F

I

9 i :n - 0 ^r e f

Synthesizer

/ADC N-

DFT

Frequency convert and

phase detect

9 ref

Cross

spectrum

I

Synthesizer

0 C

H^ TO kHz

dBc/Hz at 1Hz

100 kHz

1 M H z

Elapsed

Instrument

-138.2

58 min 49 s 33 min 5s

PhaseStation 53100A PhaseSta.tion 53100A

-148.2

8 h

PhaseStE.tion 53100A

100.0 MHz

11 d B m

-121.1

10.0 MHz

12 dBm

5.0 MHz

12 dBm

Figure 2-85 Background phase noise of the PhaseStation 531OOA. Courtesy of Jackson Labs Technologies Inc. and Miles Design LLC, used with permission. Graphical editing and comments are ours.

unit of time in the segments at low frequency. Accordingly, the icker region is seen as aslope of -15 dB/decade, as expected. Because the spectrum is quite irregular in the icker region, the frequency segments cannot be identi ed, and alonger measurement time may result in further reduction of the background noise. At 5MHz the white noise

appears at and regular. This makes us think that mis large enough for the average to hit the ultimate limit set by the hardware. The same is less clear for the other plots, obtained with shorter measurement time.

The Rohde &Schwarz FSWP Family of Phase Noise Analyzers

The FSWP is arecent and highly innovative, sophisticated, and complex family of microwave phase noise analyzers working from 1MHz up to 8/26.5/50 GHz, using the cross-spectrum technique. These instruments provide avariety of features, the AM and PM measurement of oscillators, the AM and PM measurement of two-port components via an internal synthesizer, the analysis of pulsed signals, the measurement of noise factor, and the test of VCOs, to mention the most important. Additionally, they can be used as aregular microwave spectrum analyzer, and also as adual-channel FFT analyzer up to 10 MHz. This section is based on the product documentation available online, on the article [101], and on personal experience.

The scheme (Figure 2-86), however derived from the general principles stated earlier in this Chapter, looks rather different. The input signal is split into two channels, down-converted to an appropriate IF using two sepa¬ rate references, separate synthesizers, and separate mixers. The IF signal is digitized, at 100 MS/s on 16 bits, and processed by asophisticated FPGA module (Figure 2-87).

fi

fl

fl

fi

fl

fl

fl

fl

The LO signals are derived from two different reference oscillators, one of which is phase-locked to the other with abandwidth of less than 0.1 Hz. Consequently, the PM noise in the two channels de-correlates progressively starting from 0.1 Hz, and the full bene t of the cross-spectrum method is achieved one decade beyond, thus at

!

Ref 1

1 7 9

II

/ a d c

Synth I/Q mixer

Q1

C^ADC

Input

< o

Atten —Split

CL

12

14 would not improve system performance mea.surably since the amplitude quantization noise would always predominate. So, L=K+2k& reasonable design guideline.

Finally, the designer has achoice of fciock- ^ue to the Nyquist sampling theorem and practical LPF lter design considerations, the maximum useful output frequency is 0.4 The choice will often depend on the DAC speed and the rate at which one can economically generate the required sampled digital data. Other

Considerations

The design of the PAC poses some interesting design challenges. Abrute force method uses aROM (or RAM) with

2^ addresses and an output width of A" bits. With L=14, a16A'x 12 bit lookup (K =12) table would be needed. An alternative to straight lookup is astructure that uses piecewise interpolation and quadrant logic to form the sin Xoutput. One design uses 32 segments per quadrant to form the output. It uses only 640 bits of coef cient ROM (see Ref. [12]).

Piecewise methods are frequently based on the partitioning of the PAC phase term. One method splits the input phase into upper and lower pieces. Let us call these terms the bottom ( ) and top (T). In this case. PAC input phase =B+{2')T where the top bits are shifted by 7bits from the LSB position. So

sin(phase) =sin(B) cos(2"') -i- sin(2''7’) cos( )

(3-17)

cos(B) =1and sin(S) =B

(3-18)

sin (phase) =Bcos{2^T) +sin (2'^T')

(3-19)

For small B.

So

fi

fi

fi

fi

fi

fi

Let us look at an example with K=12 bits. If the top term, 7’, is only 6bits wide then the ROM storage needed is 128 words. This is so since you must store 654 points of sin (2-'7) and 64 points of cos(2-'7). Also, Bis 6bits, so the required multiplication is 6x12. For 7=B-6 bits, the error using this approximation is less than 0.015%,

D I R E C T D I G I TA L S Y N T H E S I S T E C H N I Q U E S 2 1 5

which is better than the required 12-bit resolution. Implementing this architecture yields good results when the synthesizer must be compact and composed of only afew very large scale integration (VLSI)-type chips. Incidentally, the piecewise approximation method of sine generation surprisingly is spectrally clean. For only two segments from 0° to 90°, the total harmonic distortion is only 2.3% (see Refs. [13, 14]). Another scheme to increase the sample rate is through parallelism. Using multiple channels of computation/PAC yields very favorable results [15].

The phase accumulator structure is exible and lends itself to modi cation for implementing AM, FM, and PM. For an example of acommercial product, see the Hewlett-Packard 8791 synthesizer.

Modulation with the Phase Accumulator Synthesizer

Fortunately, this structure is amenable to AM. PM, and FM simultaneously and in real time. Let us investigate a stnicture with all three modulations. The expression that describes the ltered output is Y(n) =Am(0 sin[SF,.(;) -E Pm(0] where Am(t)

desired AM

Pm(0

desired PM

z F,.(f)

phase accumulator operator (identical to discrete integrator) instantaneous frequency

(3-20)

F,. is composed of acarrier term and any desired modulation. So ^,(0 =/"earner+ Fm(0

(3-21)

The AM may be added with areal-time hardware multiplier after the PAC. The PM is easily implemented

by adding the Pm(f) term to the phase accumulator register output. Likewise, the F,(r) term is formed with the inclusion of an FM adder block. The output of this adder is the input to the phase accumulator. So, with minor architectural adjustments, real-time modulation is available. The hardware must be designed so that anew point on Y{n) is computed each clock cycle. From where will the Am(0, Pm(r), and Fm(r) data come? In general, the data come from two sources. One is areal-time user-suppled input. Here, the user must be able to supply high-speed digital data to the synthesizer. Another source of modulation data is the RAM. In this case the AM, PM, and FM data are stored in different

dedicated RAMs. The RAMs are addressed and the data are combined to produce Y(n). Adding modulation to the phase accumulator structure provides avery exible synthesizer. For signals that may be described by their AM, PM, or FM components, this method provides acompletely deterministic approach to signal synthesis.

RAM-Based Synthesis The third architecture to be discussed is RAM-based synthesis. Figure 3-9 shows ablock diagram of such asyn¬ thesizer. The major blocks are: ●Fast static RAM for waveform storage ●Memory address sequencer ●DAC/LPF subassembly (as in Figure 3-3a) ●Waveform development station

fi

fl

fi

fl

●Waveform development station

216 SPECIAL LOOPS

Clocks

Address

sequence —/■ generator

Markers, clocks,

Fast static

DAC

RAM

K

system

LPF

amplitude,

Signal out

attenuation

Microprocessor system

triggers

HPIB bus Waveform

development

16-bit

station

Fast port

●Waveform lookup from RAM

●Frequency resolution =(Fg|Q^|^)/(array size) ●Bandwidth =dc to 0.4

clock

●Amplitude resolution (norm) =1/2^^ Figure 3-9 Block diagram of aRAM-based synthesizer.

In essence, the method uses asampled image of the desired nal waveform. This image is stored in the waveform RAM. The sequencer scans the desired waveform samples and these samples, in turn, are sent to the DAC for conversion to the analog domain. The theory of sampled data referred to and discussed in the section on the Phase Accumulator Method is the basis for understanding this method as well. Before the components are discussed, let us consider an analogy to the RAM-based synthesizer, namely, the compact disc (CD) player. The CD player has arotating platter that contains the music as sampled data. In the RAM-based synthesizer, a memory contains the sampled data to be played back. In aCD player, the data are sampled at a44.1-kHz rate. In amemory-based synthesizer, the rate is selected by the user and is usually much higher. CD players feature two 16-bit DACs, one per channel. The synthesizer has one DAC per output channel as well. For speed reasons, the DAC width is more likely to be in the 8- to 12-bit range. Most CD players have asequencer that lets you play back the tracks in any order. Amemory-based synthesizer has amemory sequencer that allows playing back the “tracks” in any order. An example of aRAM-based synthesizer is the HP8770A waveform synthesizer (see Ref. [7]). Components in aRAM-Based Synthesizer

fi

The RAM stores the sampled data. It is akey component. The RAM must be clocked at F,c l o c k , s o I t s a c c e s s t i m e will be alimiting factor in useful bandwidth. Asuccessful way to increase the RAM output data rate is to form a

D I R E C T D I G I TA L S Y N T H E S I S T E C H N I Q U E S 2 1 7

parallel RAM array and multiplex the individual RAMs to form avery high speed data path. The size of the RAM is amajor design variable, as we will see.

The RAM is useless unless it is addressed by asequencer. In the simplest sense, asequencer scans awave

segment of data in the RAM. Awave segment is de ned as ablock of sampled waveform data. The simplest sequencer is an address counter with stop and start address parameters. Amore sophisticated sequencer has a mini-program that directs the addressing. In this type, several levels of looping are allowed. In many waveforms there are wave segments that are repeated often. These segments may be scanned by the sequencer to form a complex nal output. In effect, the sequencer allows for the RAM data to be compressed. Asimple example will shed some light on the method. It is desired to synthesize an NSTC color bar test pattern. This pattern has redundancy. Many horizontal lines have the same color. Abrute force RAM lookup would require about 525,000 points with aI4,317,816-Hz clock (this is exactly four times the color burst frequency). Careful analysis of the signal reveals that there are many wave segments that repeat. By loading the RAM with only the nonredundant data, the RAM size need only be 20K addresses. So. the sequencer has given us adata compression ratio of about 26:1.

Another component in aRAM-based synthesizer is the waveform development environment. The user needs a methodology to compose the desired waveform. For simple waves, like pure carriers or simple AM carriers, the user may choose to write dedicated software routines to calculate the sampled data. Amore general solution is to

provide the user with awaveform design language. Using this language, the user may create any waveform within the limits of creativity, the synthesizer’s bandwidth, and amplitude resolution. An example of this is the waveform generation language (WGL) that is acompanion product of the HP8770A waveform synthesizer [5].

Understanding the Design Variables in RAM Synthesis The output spectral purity is limited by the DAC bit width. For random data, the SNR of the system is nearly 6.02K. However, since the DAC is nonideal (it glitches), the actual limiting performance may come from the DAC produced spurious energy. Another source of spurious spectra comes from the digital data feed-through. The output picks up crosstalk from the digital section of the system. Besides the value of K. another design parameter is the size of the memory. Even though the RAM can contain the image of any arbitrary time, nite length, or waveform, insight into the method is gained by investigating the simple case of producing asingle-frequency tone. Let be the lowest frequency that may be produced. Thus F

F .c l o c k low

sequence length

(3-22)

This tone would be asingle cycle in Qpoints, assuming that the sequence length is Qpoints. In general, the single-tone output may be described by

Y(n) =sin[2tt(F/G)// +F„„.] where

Po f f I P

Q

(3-23)

any desired phase offset /th point in the sequence number of cycles of the desired tone in Qpoints number of sampled data points for Pcycles

Both Pand Qmust be integers. Also, ^oM =/"docL(^/0Hz

(3-24)

Note that 2%{P/Q)I is just another equivalent way to write mt with a> =2k(P/Q) and /the time index. The Q points of Y{n) are stored in the RAM. S o . i f Fc l o c k

is 100 MHz and a28-MHz tone is desired, then F^^^^ =(100 MHz) (P/Q). By inspection with P=1

fi

fi

fi

and Q=25, an of 28 MHz would be realized. This tone would only have 25/7 or about 3.5 points per cycle.

218 SPECIAL LOOPS

This is ne as long as the LPF is designed to remove the sampling energy at frequencies greater than and equal to 100-28 MHz.

By adjusting PIQ, many different tones may be generated. For this example, the user observes an analog output with the tone at exactly 28 MHz (actually the only error would be due to not being exactly at the desired frequency). In the tone, each cycle would be identical to all the others. However, the data feeding the DAC are composed of seven cycles of the desired tone before the sequence repeats. Each cycle has exactly the same fre¬ quency. The difference is that each cycle has adifferent distribution of sample points compared with any of the other six cycles. Try computing sin [2ic(7/25)/] for all 25 points to see this effect for yourself. As an aside, well-designed waveforms exhibit closure; that is, the last point in the sampled data is immediately followed by the rst point in the RAM stored sequence. This allows the address counter to return to the rst point in the sequence immediately after the last, and the nal output has no discontinuity. If either Por 2is not an integer, closure will not be maintained. In this case, the spectrum will be salted with unwanted spurious signals. Again, try an example for yourself using asimple software loop and you will see that closure will not be obtained. Given that the problem is to nd Pand Qfor any desired some interesting results surface. It turns out that for some maximum value of there is asolution set of PIQ such that there is no better t to the desired tone. It is true that there will usually be an error in the resulting frequency, but it may be made very small by choosing 2max large enough. The analysis is complicated by the fact that Pand Qmay only be whole numbers. To make matters worse, many combinations of PIQ yield identical frequencies. For example, for P=21 and Q=75, we obtain the same 28-MHz tone as with P=7and Q—25. Only for Pand Qrelatively prime (no common factors) is (obtained with a minimum value of Q. Naturally, we want Qto be small since it conserves memory space. The problem of nding Pand Qis the same one mathematicians face when asked to nd the best rational approximation to afraction number like O.dddddddd. With the help of Euler’s method of continued fractions, P

and Qmay be found given the desired fraction (PIQ) of Pj.,„^j. that is to be synthesized. For an excellent study in this area, refer to Ref. [16].

Some results of solving for Pand Qmay be summarized. The frequency resolution of aRAM-based synthesizer is not aconstant. In fact, there is no simple expression that you may use to nd the exact resolution versus frequency. However, atypical or expected resolution may be expressed as

fres =^c,ock(2ttV3eLx)

(3-25)

This odd expression may be derived by nding how many pairs of relatively prime PIQ fractions (with PIQ< 1)

are available given an upper limit on Qof G,^^x

analysis, Euler’s totient function is used to nd the sum of

the pairs. On average, aunique tone will be found at aspacing of For Q^^^ >32, as computed is accurate to

DAC.RSET AMPLITUDE (A)

2 D R C T L D R H O L D

D D S

SHIFT

K E Y I N G ■♦ i

A D I G r TA L R A M P

- 7 -

G E N E R AT O R

DROVERQ*

I 3

PSPiO]

Acos (urt +6)

PHASE (9) D A T A ROUTE

D A C 12-BlT

FREQUENCY (w)

A N D

A O U T

Asln (wt 9)

PARTITION C O N T R O L C L O C K

INTERNAL PROGRAMMING REGISTERS

l/0_UPDATE Lt

' A O U T

S Y S C L K

REF.CLK p INTERNAL AND

CLOCK

TIMING P L L

CONTROL

DO TO D31

I

X 4

FO TO F3 7*

POWER¬ D O W N C O N T R O L

(SreTckl

P

I MULTICHIP SYNCHRONIZATION

SYNC_CI.K

o i r

i

Q .

U J

>

a. o o
! tD>MPPtGE}

Bias/M Pi current API current

/j

T e n s

To n s

T u n e

voltage

}

B i a s

c o n t r o l

enable-

I IMSB

RTbli) Pulse rertxjve' enable

-

P u l s e

2MSD

9 ' s M 5 D C u r r e n t

I

Pulse remove trig. 2,4, 8

s o u r c e

too kHz reference,

b i t s P r e s e t t a b l e -

[TCP

P r e s e t ● r 5 - 4 - b i t U n i t s I

T e n s — C L K

GEHOCSH

from

H u n d s

.ilS.

1

PCLK

ref.

del u

I -15

e n a b l e

e n a b l e

NCLK

'Sar^pie API

reset

Te n s C L K

la

Figure 3-37 Block diagram of afractional Nloop.

r o