BETA RATAYI PRENTICE HALL COMMUNICATIONS ENGINEERING AND EMERGING TECHNOLOGIES SERIES UO PCE eee a leRF MICROELECTRONICS Behzad Razavi University of Californi: , Los Angeles To join a Prentice Hall PTR Internet mailing list, point to http:/www.prenhall.com/mail_lists PRENTICE HALL PTR Upper Saddle River, NJ. 07458Library of Congress Cataloging-in-Publication Daa Razavt, Bewzan, RF microelectronics { Behzad Razavi D Prentice Hall communications engineering and emerging technologies series ; 2) Tnelicies I references and index. 1, Radio circuits-Design and construction 2. Integrated cireuits-Design and construction I. Title. IL. Series, TRES6.R39 1998 97-3056 621.384'12-de2t cP Acquisitions Editor: Russ Hall Editorial Assistant: Maureen Diana Editorial/Production Supervision: Kerry Reardon Manufacturing Manager: Alexis R. Heydr Cover Design: Alamini Design Composition: PreT eX, Inc. © 1998 Prentice Hall PTR Prentice-Hall, Inc. ‘A Simon & Schuster Company ‘Upper Saddle River NI 07458 Prentice Hall books are widely used by corporations and government agencies for training, ‘marketing, and resale, The publisher offers discounts on this book when ordered in bulk ‘quantities. For more information contact Corporate Sales Department Phone: 800-382-3419 Fax: 201-236-7141 E-Mail: [email protected] cor write: Prentice Hall PTR. One Lake Street Upper Saddle River, NI 07458 All rights reserved. No part of this book may be reproduced, in any form or by any means, ‘without permission in writing from the publisher. Printed in the United States of America wok 7654321 ISBN O-13-487571-5 Prentice-Hall International (UK) Limited, London Prentice-Hall of Australia Pty. Limited, Sydney Prentice-Hall Canada Inc., Toronto Prentice-Hall Hispanoamericana, SA., Mexico Prentice-Hall of India Private Limited, New Delhi Prentice-Hall of Japan, Inc., Tokyo Simon & Schuster Asia Pte. Ltd., Singapore Ecitora Prentice-Hall do Brasil, Lida.. Rio de JaneiroContents Preface > 1 _ Introduction to RF and Wireless Technology 1 1.1 Complexity Comparison 2 1.2 Design Bottleneck 4 1.3 Applications 6 14 Analog and Digital Systems 7 15 Choice of Technology 9 References 10 P 2 Basic Concepts in RF Design u 2.1 Nonlinearity and Time Variance 11 2.1.1 Effects of Nonlinearity 14 2.1.2. Cascaded Nonlinear Stages 22 2.2 Intersymbol Interference 25 2.3. Random Processes and Noise 28 23.1 Random Processes 29 232 Noise 37 24 Sensitivity and Dynamic Range 48 2.5 Passive Impedance Transformation 50 References 53 PB ___ 3 Modulation and Detection 34 3.1 General Considerations 54 3.2 Analog Modulation 57 3.2.1 Amplitude Modulation 57 3.2.2. Phase and Frequency Modulation 58 viiviii 33 34 35 Contents Digital Modulation 63 3.3.1 Basic Concepts 64 3.3.2 Binary Modulation 74 33.3 Quadrature Modulation 81 Power Efficiency of Modulation Schemes 90 3.4.1 Constant- and Variable-Envelope Signals 90 34.2 Spectral Regrowth 91 Noncoherent Detection 93 References 96 4 Multiple Access Techniques and Wireless Standards 98 41 42 43 5 52 53 54 Mobile RF Communications 98 Multiple Access Techniques 103 42.1 Time- and Frequency-Division Duplexing 103 42.2. Frequency-Division Multiple Access 105 42.3 Time-Division Multiple Access 105 4.24 Code-Division Multiple Access 107 Wireless Standards 110 43.1 Advanced Mobile Phone Service 111 43.2 North American Digital Standard 112 4.3.3. Global System for Mobile Communication 113 434 QualeommCDMA 114 435 Digital European Cordless Telephone 116 References 117 5 Transceiver Architectures 118 General Considerations 118 Receiver Architectures 122 5.2.1 Heterodyne Receivers 122 5.2.2 Homodyne Receivers 129 5.2.3. Image-Reject Receivers 138 5.24 Digital-IF Receivers 146 52.5 Subsampling Receivers 147 ‘Transmitter Architectures 149 53.1 Direct-Conversion Transmitters 152 3.2 ‘Two-Step Transmitters 154 ‘Transceiver Performance Tests 155Contents ix 55 Case Studies 157 5.1 Motorola’s FM Receiver 157 5.5.2 Philips’ Pager Receiver 158 5.5.3 " DECT Transceiver 159 5.5.4 Lucent Technologies’ GSM Transceiver 161 5.5.5 Philips’ GSM Transceiver 162 References 163 6 Low-Noise Amplifiers and Mixers 166 6.1 62 TA 12 13 14 15 Low-Noise Amplifiers 166 6.1.1 General Considerations 166 6.1.2 Input Matching 170 6.1.3 Bipolar LNAs 173 6.1.4 CMOSLNAs 178 Downconversion Mixers 180 62.1 General Considerations 180 6.2.2 Bipolar Mixers 188 623 CMOS Mixers 192 624 Noise in Mixers 194 Cascaded Stages Revisited 200 References 204 7_Oscillators 206 General Considerations 206 Basic LC Oscillator Topologies 209 Voltage-Controlled Oscillators 212 Phase Noise 214 74.1 Effect of Phase Noise in RF Communications 215 7.4.2 Q ofan Oscillator 217 7.43 Phase Noise Mechanisms 220 7.44 Noise-Power Trade-off 224 7.4.5 Bifect of Frequency Division and Multiplication on Phase Noise 224 7.4.6 Oscillator Pulling and Pushing 225 Bipolar and CMOS LC Oscillators 227 7.5.1 Negative-G», Oscillators 227 752. Interpolative Oscillators 23116 1d 18 79 Contents Monolithic Inductors 233 Resonatorless VCOs 235 Quadrature Signal Generation 236 78.1 RC-CR Network 236 7.8.2 Havens’ Technique 239 783 Frequency Division 243 Single-Sideband Generation 243 References 244 8 Frequency Synthesizers 247 81 82 83 84 General Considerations 247 Phase-Locked Loops 249 8.2.1 Basic Concepts 249 8.2.2 Basic PLL 252 8.2.3 Charge-Pump PLLs 258 8.24 ‘Typeland Type II PLLs 265 825 Noise inPLLs 266 82.6 Phase Noise at Input 266 827 Phase Noise of VCO 267 828 Frequency Multiplication 269 RF Synthesizer Architectures 269 83.1 Integer-N Architecture 270 83.2. Fractional-N Architecture 277 833 Dual-Loop Architectures 284 834 Direct Digital Synthesis 285 Frequency Dividers 290 84.1 Divide-by-Two Circuits 290 84.2 Dual-Modulus Dividers 293 References 296 9 Power Amplifiers 298 General Considerations 298 9.1.1 Linear and Nonlinear PAs 301 Classification of Power Amplifiers 301 92.1 Class A and BPAs 302 9.2.2 ClassC PAs 305Contents xi 93 94 95 9.6 High-Efficiency Power Amplifiers 306 Large-Signal Impedance Matching 310 Linearization Techniques 313 95.1 Feedforward 314 95.2 Feedback 315 953 Envelope Elimination and Restoration 317 954 LINC 318 Design Examples 320 References 323Preface ‘The annual worldwide sales of cellular phones has exceeded $2.5B. With 4.5 mil- lion customers, home satellite networks comprise a $2.5B industry, The global positioning system is expected to become a $5B market by the year 2000. In Europe, the sales of equipment and services for mobile communications will reach $30B by 1998, The statistics are overwhelming. The radio frequency (RF) and wireless market has suddenly expanded to unimaginable dimensions. Devices such as pagers, cellular and cordless phones, cable modems, and RF identification tags are rapidly penetrating all aspects of our lives, evolving from luxury items to indispensable tools. Semiconductor and system companies, small and large, analog and digital, have seen the statistics and are striving to capture their own market share by introducing various RF produets, RF design is unique in that it draws upon many disciplines unrelated to integrated cireuits (ICs). The RF knowledge base has grown for almost a century, creating a seemingly endless body of literature for the novice. This book deals with the analysis and design of RF integrated circuits and systems, Providing a systematic treatment of RF electronics in a tutorial language, the book begins with the necessary background knowledge from microwave and communication theory and leads the reader to the design of RF transceivers and circuits. The text emphasizes both architecture and circuit level issues with respect to monolithic implementation in VLSI technologies. The primary focus is on bipolar and CMOS design, but most of the concepts can be applied to other technologies as well. The reader is assumed to have a basic understanding of analog IC design and the theory of signals and systems. The book consists of nine chapters. Chapter I gives a general introduction, posing questions and providing motivation for subsequent chapters. Chapter 2 describes basic concepts in RF and microwave design, emphasizing the effects of nonlinearity and noise. Chapters 3 and 4 take the reader to the commmnication system level, giving an overview of modulation, detection, multiple access techniques, and wireless standards, While initially appearing to be unnecessary, this material is in fact essential to the concurrent design of RF circuits and systems. Chapter S deals with transceiver architectures, presenting various receiver and transmitter topologies along with their merits and drawbacks. This chapter xixiv Preface also includes a number of case studies that exemplify the approaches taken in actual RF products. Chapters 6 through 9 address the design of RF building blocks: low-noise amplifiers and mixers, oscillators, frequency synthesizers, and power amplifiers, with particular attention to minimizing the number of off-chip components. An ~ important goal of these chapters is to demonstrate how the system requirements define the parameters of the circuits and how the performance of each circuit impacts that of the overall transceiver. Ihave taught approximately 80% of the material in this book in a 4-unit graduate course at UCLA. Chapters 3, 4, 8, and 9 had to be shortened in a ten- week quarter, but in a semester system they can be covered more thoroughly. Much of my RF design knowledge comes from interactions with col- leagues. Helen Kim, Ting-Ping Liu, and Dan Avidor of Bell Laboratories, and David Su and Andrew Gzegorek of Hewlett-Packard Laboratories have con- tributed to the material in this book in many ways. The text was also reviewed by anumber of experts: Stefan Heinen (Siemens), Bart Jansen (Hewlett-Packard), Ting-Ping Liu (Bell Labs), John Long (University of Toronto), Tadao Nak- agawa (NTT), Gitty Nasserbakht (Texas Instruments), Ted Rappaport (Vir- ginia Tech), Tirdad Sowlati (Gennum), Trudy Stetzler (Bell Labs), David Su (Hewlett-Packard), and Rick Wesel (UCLA). In addition, a number of UCLA students, including Farbod Behbahani, Hooman Darabi, John Lecte, and Ja- cob Rael, “test drove” various chapters and provided useful feedback. I am indebted to all of the above for their kind assistance. Twould also like to thank the staff at Prentice Hall, particularly Russ Hall, Maureen Diana, and Kerry Reardon for their support. Behzad Razavi July 1997INTRODUCTION TO RF AND WIRELESS TECHNOLOGY “This phone uses GFSK modulation in DECT-TDMA/TDD with zero-IF 1/Q detection,” said the engineer. “How can we modify it to work with DCS1800 as well?” asked the manager. “We would need to add a duplexer, at least one more LO with SSB mixing, and probably two SAW filters,” replied the engineer . ‘Telephones have gotten much more complicated than they used to be. So have RF circuits. The nonspecialist who uses a cellular phone (cellphone) to call home probably does not know that hundreds of scientists and engineers have worked for almost a century to make wireless technology affordable, Nor does he know that there is more computing power in the phone than in some of the early personal computers. Wireless technology came to existence when, in 1901, Guglielmo Marconi successfully transmitted radio signals across the Atlantic Ocean. The conse- quences and prospects of this demonstration were simply overwhelming; the possibility of replacing telegraph and telephone communications with wave transmission through the “ether” portrayed an exciting future. However, while two-way wireless communication did materialize in military applications, wire- less transmission in daily life remained limited to one-way radio and television broadcasting by large, expensive stations. Ordinary two-way phone conversa- tions would still go over wires for many decades. ‘The invention of the transistor, the development of Shannon’s informa- tion theory, and the conception of the cellular system—all at Bell Laboratories— paved the way for affordable mobile communications, as originally imple- mented in car phones and eventually realized in portable cellular phones. But why the sudden surge in wireless electronics? Market surveys show that in the United States more than 28,000 people join the cellular phone sys- tem every day, motivating competitive manufacturers to provide phone sets with increasingly higher performance and lower cost. In fact, the present goal 12 Chap. I Introduction to RF and Wireless Technology is to reduce both the power consumption and price of cellphones by 30% every year—although it is not clear for how long this rate can be sustained. A more glorious prospect, however, lies in the power of two-way wireless communica- tion when it is introduced in other facets of our lives: home phones, computers, facsimile, and television. While an immediate objective of the wireless industry is to combine cordless and cellular phones so as to allow seamless communica- tions virtually everywhere, the long-term plan is to produce an “omnipotent” wireless terminal that can handle voice, data, and video as well as provide com- puting power. Other luxury items such as the global positioning system (GPS) are also likely to become available through this terminal sometime in the future, Personal communication services (PCS) are almost here. 1.1 COMPLEXITY COMPARISON To gain an early perspective on modern RF design, first consider the sys- tem depicted in Fig. 1.1, a simple frequency modulation (FM) transceiver. In Fig. 1.1(a), Q1 operates as both an oscillator and a frequency modulator; that ) Figure 1.1 (a) FM transmitter, (b) FM receiver.3 Sec, 1.1 Complexity Comparison oO puegeseg 1 puegaseg [1] suoydyao & 30 wones ay ZL aan’ Teziseuuts, Aouanbat4 ZHIN OOF yep, aiqewuesBord4 Chap. 1 Introduction to RF and Wireless Technology is, the audio signal produced by the microphone varies the bias voltage across the varactor diode Dj, thereby modulating the frequency of oscillation. In the receive path, Q; serves as both an oscillator and a demodulator, generating the audio signal at node X. The result is then amplified and applied to the speaker. Now consider the circuit of Fig. 1.2, the RF section of a cellphone [1]. This circuit is orders of magnitude more complex than the FM circuits of Fig. 1.1, and we will postpone its analysis until Chapter 5. Why have RF designers gone from the circuit of Fig. 1.1 to that of Fig, 1.2? Whatis the thought process behind this evolution? Is all of this complexity really necessary? These questions will be answered in Chapters 2 through 5 as we develop our understanding of RF communication systems and architecture and circuit design issues. 1.2 DESIGN BOTTLENECK ‘Today’s pocket phones contain more than one million transistors, with only a small fraction operating in the RF range and the rest performing low-frequency “baseband” analog and digital signal processing (Fig. 1.3). In other words, the baseband section is, in terms of the number of devices, yet several orders of magnitude more complex than the circuit of Fig. 1.2. The definition of RF and _ baseband will become clear later, but here we note that the RF section is still the design bottleneck of the entire system. This is so for three reasons. Baseband Section Figure 13 RF and baseband processing in a transceiver, Multidisciplinary Field _ In contrast to other types of analog and mixed- signal circuits, RF systems demand a good understanding of many areas that are not directly related to integrated circuits (ICs), Shown in Fig. 1.4, most of these areas have been studied extensively for more than half a century, making it difficult for an IC designer to acquire the necessary knowledge in a reasonable amount of time. Even at present, the literature pertaining to RF design appears in more than 30 journals and conferences. Owing to thisissue, traditional wireless system design has been carried out at somewhat disjointed levels of abstraction: communication theorists create the modulation scheme and baseband signal processing; RF system experts plan the transceiver architecture; IC designers develop each of the building blocks; and manufacturers “glue” the ICs and other external components together. In fact, architectures are often planned according to the available off-the-shelf components, and ICs are designed to serve as many architectures as possible, leading to a great deal of redundancy at both system and circuit levels.Sec. 1.2 Design Bottleneck 5 ‘Communication Microwave ae Random ‘Theory Si | ae ‘Signals = A re Propagation Design Architectures: sae TN ont \.o Figure 1.4 Disciplines required in RF design. As the industry moves toward higher integration and lower cost, RF and wireless design demands increasingly more “concurrent engineering,” thereby requiring IC designers to have sufficient knowledge of all the disciplines shown in Fig. 1.4. RE Design Hexagon RF circuits must process analog signals with a wide dynamic range at high frequencies. It is interesting to note that the signals must be treated as analog even if the modulation is digital or the amplitude carries no information. (This will be clarified in later chapters) The trade-offs involved in the design of such circuits can be summarized in the “RF design hexagon” shown in Fig, 1.5, where almost any two of the six parameters trade with each other to some extent. We will sce these trade-offs in Chapters 6 to 9, but our observation here is that, while digital circuits directly benefit from advances in IC technologies, RF circuits donot as much. This issue is exacerbated by the fact that RF circuits often require external components—for example, inductors— that are difficult to bring onto the chip even in modern IC processes. Noise = Power Linearity Frequency Supply Figure L5 RF design hexagon, Design Tools Computer-aided analysis and synthesis tools for RF ICs are still in their infancy, forcing the designer to rely on experience, intuition, or inefficient simulation techniques to predict the performance. For example, nonlinearity, time variance, and noise in RF circuits usually require studying the spectrum of signals, but the standard ac analysis available in SPICE uses6 Chap. 1 Introduction to RF and Wireless Technology only linear, time-invariant models. Thus, circuits are simulated in the time domain so as to include nonlinear or time-variant effects, and the resulting waveforms are subsequently transformed to the frequency domain to obtain the spectrum, The difficulty is that the time-domain simulation must be run for a long period to resolve closely spaced frequency components. In addition, spectral averaging techniques may be necessary if random noise is used in the time-domain analysis. Another issue in simulating RF circuits relates to external components that cannot be modeled by typical devices in SPICE. For example, surface acoustic wave (SAW) filters, used in both the receive and the transmit paths, exhibit input and output impedances that can be characterized only by scat- tering (S) parameters—essentially a table of numbers. Modeling such circuit with RLC networks provides a first-order approximation, but it may not predict effects such as instability and impedance mismatch. 1.3 APPLICATIONS In addition to familiar wireless products such as pagers and cellular phones, RF technology has created many other markets that display a great potential for rapid growth, each presenting its own set of challenges to RF designers. WLANs Communication among people or pieces of equipment in a crowded area can be realized through a wireless local area network (WLAN), Using frequency bands around 900 MHz and 2.4 GHz, WLAN transceivers can provide mobile connectivity in offices, hospitals, factories, etc., obviating the need for cumbersome wired networks, Portability and reconfigurability are prominent features of WLANs. GPS The use of GPS to determine one’s location as well as obtain directions becomes attractive to the consumer market as the cost and power dissipation of GPS receivers drop. Operating in the 1.5-GHz range, such sys- tems are under consideration by automobile manufacturers, but they may be available as low-cost hand-held products sometime in the near future. RF IDs __ RF identification systems, simply called “RF IDs,” are small, low-cost tags that can be attached to objects or persons so as to track their position. Applications range from luggage in airports to troops in military operations. Low power consumption is especially critical here as the tag’s lifetime may be determined by that of a single small battery. RF ID products in the 900-MHz and 2.4-GHz range have recently appeared in the market. Home Satellite Network The programs and services available through satellite television have attracted many consumers to home satellite networks, Operating in the 10-GHz range, these networks require the addition of a dish antenna and a receiver to a television set and directly compete with cable TV.See. 1.4 Analog and Digital Systems 7 1.4 ANALOG AND DIGITAL SYSTEMS To gain a general, preliminary view of wireless transceivers, first consider the simple “analog” architecture shown in Fig. 1.6. In the transmit path, the signal generated by the microphone modulates a high-frequency carrier, and the re- sult is amplified and “buffered” so as to drive the antenna, In the receive path, the signal is amplified by a low-noise amplifier (LNA), the spectrum is trans- lated to a lower frequency by a “downconverter” (usually a mixer) to facilitate subsequent demodulation, and the demodulated output is amplified to drive the speaker. Low-Noise Audio Amplifier Amplifier Carrier O) Figure 1.6 Block diagram of a generic analog RF system: (a) transmitter, (b) receiver. Now, consider the “digital” transceiver shown in Fig. 1.7. Here, the voice signal is first digitized by an analog-to-digital converter (ADC) and compressed to reduce the bit rate and hence the required bandwidth [Fig. 1.7(a)]. Next, the data undergoes “coding” and “interleaving” [2]. ‘These two functions format the data such that the receiver can detect and minimize errors by performing the reverse operations. Since rectangular pulses are usually not optimum for modulation, the data is then “shaped” before it is applied to the modulator and the power amplifier (PA). In the receive path [Fig. 1.7(b)], the signal is ampli- fied, downconverted, and digitized. Subsequently, demodulation, equalization, decoding and deinterleaving, and decompression are performed in the digital domain, The resulting data is then converted to analog form by a digital-to- analog converter (DAC), amplified, and applied to the speaker.Chap. | Introduction to RF and Wireless Technology srastaaas (q) ‘sonmuisuen (2) aioisés 4 1etSip apaUN8 w yo uesep og "7 amy uo|ssasdwoseg 80100, souniduny orpny Biiinsopiaitiaa aznenba 4oveinpoweg @s!0N-mo7, @ soqsed uojssosduiog aaah 49ModSec. 1.5 Choice of Technology 9 In the simplest case, the performance of these transceivers can be quan- tified in terms of the maximum distance across which they operate while pro- viding satisfactory reception. This is determined by both the power delivered to the antenna and the sensitivity of the receiver, particularly the noise of the LNA.'However, in a realistic environment, many different transceivers operate simultaneously, often in close proximity, thereby interfering with each other. Furthermore, the communication medium is quite complex: reflections from buildings and other obstacles can result in destructive interference at a given point, suppressing the received signal strength to undetectable levels. In such an environment, signal processing in the digital transceiver achieves a higher performance than that of the analog system. Looking at the two transceivers of Figs. 1.6 and 1.7, the reader may won- der which parts are related to “RF electronics.” As we willsee in other chapters, the exact definition of RF depends on the system, but we tentatively consider the signal to be in the RF domain if it is analog and its spectrum is not centered around zero frequency. ‘This indicates that in Fig. 1.6 all the building blocks ex- cept for the audio amplifier are RF circuits. Similarly, in Fig. 1.7, the modulator, the PA, the LNA, and the downconverter operate in the RF range. Modern RF electronics, however, includes much more than the design of such building blocks. For example, the trade-offs at the system level must be carefully studied in each case [3, 4]. ‘This book primarily deals with the design of RF integrated circuits, but, following the premise of Fig. 1.4, it provides the reader with background in other related areas as well. As will be seen throughout the book, RF ICs, RF architectures, and wireless systems are strongly coupled, and the reader will frequently need to utilize this background knowledge. 1.5 CHOICE OF TECHNOLOGY The viable IC technology for RF circuits continues to change. Performance, cost, and time to market are three critical factors influencing the choice of technologies in the competitive RF industry. In addition, issues such as level of integration, form factor, and prior (successful) experience play an important role in the decisions made by the designers. At present, GaAs and silicon bipolar and BiCMOS technologies consti- tute the major section of the RF market. Usually viewed as a low-yield, high- power, high-cost option, GaAs field-effect and heterojunction devices nonethe- less have maintained a strong presence in RF produets [5], especially in power amplifiers and front-end switches. While GaAs processes offer useful features such as higher (breakdown voltage) (cutoff frequency) product, semi-insulating substrate, and high-quality inductors and capacitors, silicon devices ina VLSI technology can potentially ' The type of modulation is also an important factor, but we ignore it for the moment10 Chap. 1 Introduction to RF and Wireless Technology provide both higher levels of integration and lower overall cost, as demon- strated in complex circuits such as frequency synthesizers. In fact, all building blocks of typical transceivers are available in silicon bipolar technologies from many manufacturers, ‘The third contender is CMOS technology. Supported by the enormous momentum of the digital market, CMOS devices have achieved high transit fre- quencies, e.g,, tens of gigahertz in the 0.35-jam generation. As we will see in this, book, “RF CMOS” has suddenly become the topic of active research. CMOS, technology must nevertheless resolve a number of practical issues: substrate coupling of signals that differ in amplitude by 100 dB, parameter variation with temperature and process, and device modeling for RF operation. REFERENCES [1] C. Marshall et al., “2.7 V GSM Transceiver ICs With On-Chip Filtering,” ISSCC Dig. Tech. Papers, pp. 148-149, February 1995. [2] B. Sklar, Digital Communications, Englewood Cliffs, NJ: Prentice Hall, 1988. [3] K. Feher, Wireless Digital Communications, Englewood Cliffs, NJ: Pren- tice Hall, 1995. ; [4] R. Steele, ed. Mobile Radio Communications, Piscataway, NJ: IEEE Press, 1992. IS] R. Schneiderman, “GaAs Continues to Gain in Wireless Applications,” Wireless Systems Design, pp. 14-16, March 1997.2 BASIC CONCEPTS IN RF DESIGN RF designers draw upon many concepts that originate from the theory of sig- nals and systems. In this chapter, we describe these concepts and define the terminology used in RF electronics so as to prepare the reader for the material in the following chapters. Beginning with nonlinear systems, we describe effects such as harmonic distortion, gain compression, cross modulation, and intermodulation. We then briefly study intersymbol interference and Nyquist signaling, review random processes and noise, and introduce approaches to representing noise in circuits. Finally, we describe passive impedance transformation. 2.1 NONLINEARITY AND TIME VARIANCE A system is linear if its output can be expressed as a linear combination (super- position) of responses to individual inputs, More accurately, if for inputs x; (r) and x2(r) AO > nO, BO > yt), QQ) where the arrow denotes the operation of the system, then axy(t) + x(t) + ayy(t) + by2(t), (2.2) for all values of the constants a and . Any system that does not satisfy this condition is nonlinear. Note that according to this definition, we consider a system nonlinear if it has nonzero initial conditions or finite “offsets,” n12 Chap.2 Basic Concepts in RF Design A system is time invariant if a time shift in its input results in the same. time shift in its output. That is, if x(¢) > y(t), then x(t —t) > y(t — 7), for all values of 7. A system is time variant if it does not satisfy this condition. While nonlinearity and time variance are intuitively obvious concepts, they may be confused with each other in some cases, For example, consider the switching circuit shown in Fig. 2.1(a). The control terminal of the switch is driven by vini(t) = A; coset and the input terminal by vin2(t) = Az cos wot. We assume the switch is onif vin > Oand off otherwise. Is thissystem nonlinear or time variant? If, as shown in Fig. 2.1(b), the path of interest is from vjq) to Vout, (While viq2 is part of the system and still equal to Az cos zr), then the system is nonlinear because the control is only sensitive to the polarity of vin, and time variant because vay: also depends on viq2. On the other hand, if, as shown in Fig. 2.1(c), the path of interest is from vin2 10 Yass (While vint is part of the system and still equal to A; cos @1), then the system is linear [Eq. (2.2)] but time variant. Thus, general statements such as “switches are nonlinear” are ambiguous. As we will see in Chapter 6, these distinctions are critical in the design of mixers. Vout ©) Figure2.1 (a) Simple switching circuit, (b) nonlinear time-variant system, (©) linear time-variant system, Another interesting result of the above observation is that a linear system can generate frequency components that do not exist in the input signal. ‘This, is possible if the system is time variant, for example, Fig. 2.1(c). Since in this,Sec.2.1 Nonlinearity and Time Variance 13 circuit, Yoy: can be considered as the product of vig and a square wave toggling between 0 and 1, the output spectrum is given by Vout) 0 where + denotes convolution, 4(.) is the Dirac delta function, and T; = ‘Thus, the output consists of vertically scaled replicas of Vino) shifted by n/T; A system is called “memoryless” if its output does not depend on the past values of its input. For a memoryless linear system, yQ) = ax(r), (2.5) where a is a function of time if the system is time variant [e.g., Fig, 2.1(c)]. For a memoryless nonlinear system, the input-output relationship can be approxi- mated with a polynomial, YE) = a + x(t) + arx7(1) + a(N Fo, (26) where ce; are in general functions of time if the system is time variant. Fig. 2.2(a) illustrates an example where the input signal is applied to the base of Q; while Q> and Q3 are periodically switched by means of a square wave. For ideal bipolar transistors, the circuit can be viewed as in Fig, 2.2(b) and Your(t) = (1 exp #) sR, 27) Vr where [5; represents the saturation current of Q;, Vp = kT/q, and s(t) isa square wave toggling between —1 and +1. Figure 2.2 (a) Switching differential pair with tail current source driven by a signal, (6) equivalent circuit of (a)14 Chap. 2 Basic Concepts in RF Design The system described by (2.6) has “odd symmetry” ifits response to —x(¢) is the negative of that to x(t). This occurs if a; = 0 for even j. A circuit having, odd symmetry is called differential or “balanced.” For example, the bipolar differential pair of Fig. 2.3 exhibits the following input-output characteristic: Yin Vout = Rigg tanh . " cals (28) which is an odd function. @ O} Figure 2.3 Bipolar differential pair along with its input-output characteristic. A system is called “dynamic” if its output depends on the past values of, its input(s) or output(s). Fora linear, time-invariant, dynamic system, yO) = AD * x(0), (2.9) where f(r) denotes the impulse response. If a dynamic system is linear but time variant, its impulse response depends on the time origin; if 8() > h(), then 5(t — 1) + h(t, t). Thus, y(t) = A(t, t) * x(t). (2.10) Finally, if a system is both nonlinear and dynamic, then its impulse re- sponse can be approximated with a Volterra series [1, 2], a topic beyond the scope of this book. 2.1.1 Effects of Nonlinearity ‘While many analog and RF circuits can be approximated with a linear model to obtain their response to small signals, nonlinearities often lead to interesting and important phenomena, For simplicity, we limit our analysis to memoryless, time-variant systems and assume yt) © anx(t) + ax) + aax3(t). (en)Sec. 2.1 Nonlinearity and Time Variance 18 ‘The reader is cautioned, however, that the effect of storage elementsand higher- order nonlinear terms must be carefully examined to ensure (2.11) isa plausible representation. Harmonies _ If a sinusoid is applied to a nonlinear system, the output generally exhibits frequency components that are integer multiples of the input. frequency. In Eq. (2.11), if x(t) = A cosr, then y(t) = aA coset + a2A? cos’ wt + a3A> cost (2.12) aA? a = eA coset + S—(1 + cos2ext) + —[—Geosear + cos 30) (2.13) aA? a3 A3 3a3A% Az 4 (wa += ) cose + SE cos tut + = cost. (2.14) In Eq. (2.14), the term with the input frequency is called the “fundamental” and the higher-order terms the “harmonics.” From the above expansion, we can make two observations. First, even- order harmonics result from with even j and vanish if the system has odd symmetry, i, if itis fully differential. In reality, however, mismatches corrupt the symmetry, yielding finite even-order harmonics. Second, in (2.14) the am- plitude of the th harmonic consists of a term proportional to A” and other terms proportional to higher powers of A. Neglecting the latter for small A, we can assume the mth harmonic grows approximately in proportion to A”. Gain Compression ‘The small-signal gain of a circuit is usually obtained with the assumption that harmonics are negligible. For example, if in (2.14), A is much greater than all the other factors that contain A, then the small- nal gain is equal to a. This quantity can be seen in the familiar differential pair of Fig. 2.3 to be equal to Vout Yin 2Vr However, as the signal amplitude increases, the gain begins to vary. In fact, nonlinearity can be viewed as variation of the small-signal gain with the input level. This is evident from the term 3a3A3/4 added to @1 A in (2.14), as well as the input-output characteristic shown in Fig. 2.3. In most circuits of interest, the output is a “compressive” or “saturating” function of the input; that is, the gain approaches zero for sufficiently high input levels. In (2.14) this occurs if @3 < 0. Written as a) + 3a3A2/4, the gain is therefore a decreasing function of A. In RF circuits, this effect is quantified by the “1-dB compression point,” defined as the input signal level that causes the (2.15)16 — Chap.2 Basic Concepts in RF Design small-signal gain to drop by 1 dB. If plotted on a log-log scale as a function of the input level, the output level falls below its ideal value by 1 dB at the 1-dB compression point (Fig. 2.4), 20 log Aout 148 At-a8 20 log Ain Figure 24 Definition of the 1-4B compression point ‘To calculate the 1-dB compression point, we can write from (2.14) 3 2log lay + ZasAj-aal = 20logiay| — 14B (2.16) That is, Avap = ,/0.145 (2.17) a ‘A measure of the maximum input range of the circuit, the 1-dB com- pression point occurs around —20 to —25 dBm (63.2 to 35.6 mV pp in a 50-2 system) in typical front-end RF amplifiers. Desensitization and Blocking Circuits with compressive characteristics exhibit an interesting effect when they process a weak, desired signal along with a strong interferer. Since a large signal tends to reduce the “average” gain of the circuit, the weak signal may experience a vanishingly small gain. Called “desensitization,” this effect can be analyzed for the characteristics of (2.11) by assuming x(t) = A; cosqit + Az cosayt. The output is xt) = (aoa + jaa + Suda) cosayt +--+, (2.18) which, for A; < Az, reduces to x0) = (« + jos) Ay cosayt +++. (2.19) Thus, the gain for the desired signal is equal to (a; + 3a3A3/2), a decreasing, function of Aa if 3 < 0. For sufficiently large Az, the gain drops to zero, and we say the signal is “blocked,” In RF design, the term “blocking signal” usuallySec.2.1 Nonlinearity and Time Variance 17 refers to interferers that desensitize a circuit even if the gain does not fall to zero. Many RF receivers must be able to withstand blocking signals 60 to 70 4B greater than the wanted signal. Cross Modulation Another phenomenon that occurs when a weak sig- nal and a strong interferer pass through a nonlinear system is the transfer of modulation (or noise) on the amplitude of the interferer to the amplitude of the weak signal. Called “cross modulation,” this effect is evident from Eq. (2.19), where variations in A affect the amplitude of the output component at @. For example, if the amplitude of the interferer is modulated by a sinusoid Ap(1 + m cos mt) cosant, where m is the modulation index and less than unity, then (2.19) assumes the following form: 2? a) As tS apdy dg (1 tl co 7 (14 2 + cos2y y= Jonas + Sara? Ft 008 20m + 2moosiret) Jomont +e, (2.20) Thus, the desired signal at the output contains amplitude modulation at «», and 2p. A common case of cross modulation arises in amplifiers that must simul: taneously process many independent signal channels, ¢.g,, in cable television transmitters. Nonlinearities in the amplifier corrupt each signal with the am- plitude variations in other channels. Intermodulation While harmonic distortion is often used to describe nonlinearities of analog circuits, certain cases require other measures of non- linear behavior. For example, suppose the nonlinearity of an active low-pass filteris to be evaluated. If, as depicted in Fig. 2.5, the input sinusoid frequency is chosen such that its harmonics fall out of the passband, then the output distor- tion appears quitesmall evenif the input stage of the filter introduces substantial nonlinearity. Thus, another type of test is required here, Commonly used is the “intermodulation distortion” in a “two-tone” test. la Go| o Figure 25 Harmonic distortion in a low-pass filter. When two signals with different frequencies are applied to a nonlinar sys- tem, the output in general exhibits some components that are not harmonics of the input frequencies. Called intermodulation (IM), this phenomenon arises8 Chap.2 Basic Concepts in RE Design from “mixing” (multiplication) of the two signals when their sum is raised to a power greater than unity. To understand how Eq. (2.11) leads to intermodula- tion, assume x(t) = Ay cos@if + Ap cos ant. Thus, y(0) = a (Aj cost + Az coset) + a2(A1 cosait + Az cost)? + @3(A; cos at + Ar cosant). (2.21) Expanding the left side and discarding de terms and harmonies, we obtain the following intermodulation products: @ = @ $y : aA Az COs(W; + Wy)t + aA, AQ COS(@ — 2) (2.22) = 20 + w= 3A? cosc2. + an)t + Sei con, — a)t (2.23) = Bang hy 2 EB coatoy ene + 304A ost20n = o)t (2.24) and these fundamental components Rog. 3 Z @ =m, : [aA + qa os pasa A; cosat + («vas + FasA} + jeasat) 60s wat. (2.28) Of particular interest are the third-order IM products at 2001 —o and 20 —o1, illustrated in Fig. 2.6. The key point here is that if the difference between a and ©» is small, the components at 2a; — a and 22 — @ appear in the vicinity of «@ and 2, thus revealing nonlinearities even in cases such as the LPF of Fig. 2.5. Ina typical two-tone test, Ay = Ap = A, and the ratio of the amplitude of the output third-order products to aA defines the IM distortion, For example, if A = 1 Vpp, and 303A3/4 = 10 mV>p, then we say the IM components are at —40 dBc, where the letter “c” means “with respect to the carrier.” 0 2 OO > 20-02 20-0, Figure 26 _Intermodulation in a nonlinear system,Sec. 2.1 Nonlinearity and Time Variance 19 Intermodulation is a troublesome effect in RF systems. As shown in Fig. 2.7, if a weak signal accompanied by two strong interferers experiences third-order nonlinearity, then one of the IM products falls in the band of in- terest, corrupting the desired component. While operating on the amplitude of the signals, this effect degrades the performance even if the modulation is on the phase (because zero-crossing points are still affected.) Note that this phenomenon cannot be directly quantified by harmonic distortion. Interferers n Low-tose Desired Amplifier Channel © © Figure 2.7 Corruption of a signal due to intermodulation between two interferers. ‘The corruption of signals due to third-order intermodulation of two nearby interferers is so common and so critical that a performance metric has been de- fined to characterize this behavior. Called the “third intercept point” (J Ps), this parameter is measured by a two-tone test in which A is chosen to be sul ficiently small so that higher-order nonlinear terms are negligible and the g: is relatively constant and equal to a;. From (2.23), (2.24), and (2.25), we note that as A increases, the fundamentals increase in proportion to A, whereas the third-order IM products increase in proportion to A? [Fig. 2.8(a)]. Plotted on a logarithmic scale [Fig. 2.8(b)], the magnitude of the IM products grows at three times the rate at which the main components increase. The third-order intercept point is defined to be at the intersection of the two lines. The hori- zontal coordinate of this point is called the input P; (71 P;), and the vertical coordinate is called the output J P; (O1 Ps). 20log(o1A) ona J 2 Sosa A UP, 20 log A f@ ) Figure 2.8 Growth of output components in an intermodulation test.20 Chap.2 Basie Concepts in RF Design It is important to appreciate the advantage of I P; over a simple IM mea- surement. If the magnitude of IM products (normalized to that of the carrier) is used as a measure of linearity, then the input amplitude with which the test is performed must be specified. ‘The third intercept point, on the other hand, is a unique quantity that by itself can serve as a means of comparing the linearity of different circuits. From the input-output characteristic of Eq. (2.11), we can derive a simple expression for / Ps. Let x(t) = Acoswit + Acosant. Then, 9 9 yt) = (a: + jos?) Acosayt + (« + zona?) Acos wot 33 3 + FasA¥ cos(2ax — an)i + jor cos(2a, — wi) tees. (2.26) Ifa; >> 9a3A?/4, the input level for which the output components at #; and cy have the same amplitude as those at 2c — an and 2c — a is given by 3 . lanlAres = [les|Ai ps 27) ‘Thus, the input 7 Ps is (2.28) and the output / P3 is equal to @Ayp3- ‘The parameter / P; characterizes only third-order nonlinearities. In prac- tice, if the input level is increased to reach the intercept point, the assumption @ > 9a3A?/4 no longer holds, the gain drops, and higher-order IM products become significant. In fact, in many circuits the J P3 is beyond the allowable input range, sometimes even higher than the supply voltage. Thus, the practical method of obtaining the J P; is to measure the characteristic of Fig. 2.8(b) for small input amplitudes and use linear extrapolation on a logarithmic scale to find the intercept point. A quick method of measuring the 1 P; is as follows. Let us denote the input level at each frequency by Ajq, the amplitude of the output components at @ and @) by Aqin2.and the amplitude of the M3 products by Ay. Then. from (2.26), we have Avior _ _|eilAn Aims 3|a3|A3,/4 4la| 1 = Gaal aE" (2.30) (2.29)Sec. 2.1 Nonlinearity and Time Variance 21 which, in conjunction with (2.28), reduces to (2.31) Consequently, 2log Autor — 2log Arm; = 20log A} p; — 2log AZ, (232) and 1 20log Arps 720 log Ann w2 — 20log Arya) + 20log Ain 2. ‘Thus, if all the signal levels are expressed in dBm, the input third intercept point is equal to half the difference between the magnitudes of the fundamentals and the IM; products at the ouiput plus the corresponding input level [Fig. 2.9(a)] ‘The key point here is that 1 P; can be measured with only one input level, obviating the need for extrapolation. Main Signal Power 200g Ain @ (b) Figure 2.9 (a) Calculation of 1, without extrapolation, (b) graphical interpretation of (a). Shown in Fig. 2.9(b) is a geometric interpretation of the above relation- ship. Since line L; has a slope equal to unity and line Lz a slope equal to 3, an input increment AP/2 yields an equal increment in L; and an increment equal to 3A P/2 in La, reducing the difference between the two lines to zero. ‘The above approach provides an estimate of / Py in initial phases of the design or characterization, The actual value of / P;, however, must still be obtained through accurate extrapolation to ensure that all nonlinear and frequency-dependent effects are taken into account.22 Chap. 2 Basic Concepts in RF Design Another measurement method encountered in the literature is to apply a single tone, plot the third harmonic magnitude versus the input level, and obtain the intercept point by extrapolation. From the example of Fig. 2.5, we note that this technique does not yield a correct value for I Ps. ‘To gain a better feeling about the required linearity in typical RF systems, let us calculate the amount of corruption that a 1-1 V;ms signal experiences by two 1-mV,m, interferers in an amplifier having an J Ps of 70 MV ms (% —10 dBm) (Fig. 2.10). Neglecting desensitization and cross modulation, we can write Aizen, Aint (2.34) Asien Ainuin ee) where A,jz denotes the signal amplitude and Ajn: the interferer amplitude. It follows from (2.31) that Asig.out 2.38 AlM3.out 225) where Agig,in = 1 Vrms, Arps = 70 MVyms, and Aintin = 1 MV ems. Thus, the ratio is equal to 4.9 © 13.8 dB. <1 Vins . 4 [Vane 4P,=-10 dm Ore 6 £8 fo 204-2 2W_~ Oy Figure 210 Example of achievable SIV R in the presence of large interferers. Itis also instructive to find the relationship between the 1-dB compression point and the input / P; for a third-order nonlinearity. From (2.17) and (2.28), we conclude that the two are related by (2:36) (237) 2.1.2 Caseaded Nonlinear Stages Since in RF systems, signals are processed by cascaded stages, it is important to know how the nonlinearity of each stage is referred to the input of the cascad In particular, itis desirable to calculate an overall input third intercept point in terms of the / P; and gain of the individual stages.Sec. 2.1 Nonlinearity and Time Variance 23 Consider two nonlinear stages in cascade (Fig, 2.11). If the input-output characteristics of the two stages are expressed, respectively, as vile) = ay x(t) + anx*(t) + a5x3(0) (2.38) yet) = Biyi(t) + Boyf(r) + Baz), (2.39) then y(t) = Pile) + @2x7(t) + a32°(0)] + Brfonx(t) + a2x7(t) + ogx3(0)P + Bsloax() + a227(1) + ax (O)P. (2.40) Considering only the first- and third-order terms, we have yolt) = cBix(t) + (Bi + 2orcerBe + @2Bs)x%(t) +o. (241) Thus, from (2.28) ———— A = (2.42) ” 3 asf, + 2ayanfh + af Interestingly, proper choice of the values and signs of the termsin the denomina- tor can yield an arbitrarily high J P;. In practice, however, other considerations such as noise, gain, and active device characteristics may not permit this choice. Asa worst-case estimate, we add the absolute values of the three terms in the denominator. UP, UP 5 Figure 211 Cascaded nonlinear stages. Equation (2.42) can be simplified if the two sides are inverted and squared: 1 _ 3 JasBil + arorbal + la? (243) Aip; 4 |i] " 1 Bay = ‘ (2.44) jes 2B” Apa where Arps, and Ajps,2 represent the input 7 Ps points of the first and second stages, respectively. Note that Ajp3, Arps... and Ay ps,2 are voltage quantities rather than power quantities.24 Chap.2 Basic Concepts in RF Design From the above result, we note that as on increases, the overall / P; de- creases. This is because with higher gain in the first stage, the second stage senses larger input levels, thereby producing much greater M5 products. (Re- call that / My products grow with the third power of the input amplitude.) To gain more insight, we assume x(t) = A cose! + A coset and iden- tify the various IM products. Referring to Fig. 2.12, we make the following observations. '(1) The fundamental input components are amplified by approxi- mately ot in the first stage and f; in the second. Thus the output fundamentals are @)8)A(cos af + cosw2t). (2) The 1M; products generated by the first stage, namely, (33/4) A*[cos(2«1 —a2)t +c0s(2e2 —«)t], are also amplified by #; when they appear at the output of the second stage. (3) The second stage senses a A(cos «)f + cos «9t) at its input and hence generates these J M3 prod- ucts: (3683/4) (a1 A)? cos(2@; — @2)t + (38s/4)(@1 A)? cos(2a» — @)t. (4) The second-order nonlinearity in y (1) generates components at «; ~ 2, 201, and 2. Upon experiencing a similar nonlinearity in the second stage, such com- ponents are translated to 2c; — c» and 2w> — «;. More specifically, as shown in Fig. 2.12, y2(r) contains terms such as 282[a A cos yf - aA? cas(e; — a»)t] and 28>(a A cos wf -0.Sr2A* cos 211). The resulting third-order IM products can be expressed as (3; a2A7/2)[cos(2e — a)t + cos(2w2 — ant]. From these observations, we can write ya(t) = a8) A(cosayt + cos ant) Bars, 3a}, Ber 4 7“ (= 2 at + scahs A [cosa — an)t + cos(2m2 — a)t] +++, (2.45) obtaining the same J P; as above. In many RF systems, each stage in a cascade has a narrow frequency band. Thus, the components described in the fourth observation above fall out of the band and are heavily attenuated. Consequently, the second term on the right-hand side of (2.44) becomes negligible, giving 1 1 a? = +a (2.46) Alps Atos, Alesa This equation readily gives a general expression for three or more stages: tg too a 2 wa 2 Alps Alps Alps where Aj ps3 denotes the input / P; of the third stage. Thus, if each stage in a cascade has a gain greater than unity, the nonlinearity of the latter stages becomes increasingly more critical because the J P; of each stage is effectively +o tee, (2.47) ‘The spectrum of A cos ct consists of two impulses, each with a weight A/2. We drop the factor 1/2 in the figures for simplicity. The end result is still correctSec. 2.2 Intersymbol Interference 25 IPs OP ay x(t) ¥ft) A “Yt On BiA _ Ho | =o Os 7 ee 5 ee Sasa i Gosia © o 2e,- 02 20,- 0; i 204-2 202-01 3 3 Ba(onay ® yA 20,- 02 20,70; 2A ty 012 B25, o o 7-0; &] Wy 2a4-O2 205-0; oA Lara? SeareraBa4 A fey tt, Oy Og 2-2 202-0; 20, 20, Figure 2.12 _Intermodulation mechanisms in cascade of two nonlinear stages scaled down by the total gain preceding that stage. We should emphasize that (2.47) is merely an approximation. In practice, more precise calculations or simulations must be performed to predict the overall 1 P3. 2.2 INTERSYMBOL INTERFERENCE Linear time-invariant systems can also “distort” a signal if they do not have sufficient bandwidth. Attenuation of high-frequency components of a peri- odic square wave in a low-pass filter is a familiar example of such behavior [Fig. 2.13(a)]. However, limited bandwidth has a more detrimental effect on random bit streams, To understand the issue, first recall that if a single ideal rectangular pulse is applied to a low-pass filter, the output exhibits an exponen- tial tail that becomes more significant as the filter bandwidth decreases. Thi occurs fundamentally because a signal cannot be both time limited and band- width limited: when the time-limited pulse passes through the band-limited system, the output must extend to infinity in the time domain,26 Chap. 2 Basic Concepts in RF Design Vin wr Mout @ ob) Figure 2.13 Response of a low-pass filter to (a) a periodic square wave, (b) a random sequence of ONEs and ZEROs ‘Now suppose the output of a digital system consists of a random sequence of ONEs and ZEROs, each ONE represented by an ideal rectangular pulse and each ZERO by the absence of such a pulse. If this sequence is applied to a low- pass filter, the output can be obtained as the superposition of the responses to each input bit [Fig. 2.13(b)]. We note that each bit level is corrupted by decaying tails created by previous bits. Called “intersymbol interference” (ISI), this phenomenon leads to higher error rate in the detection of random waveforms that are transmitted through band-limited channels. ‘The problem of IST is particularly troublesome in wireless communica- tions because the bandwidth allocated to each channel is fairly narrow. Methods of reducing ISI include pulse shaping (“Nyquist signaling”) in the transmitter and “equalization” in the receiver. We briefly describe Nyquist signaling here and refer the reader to the extensive literature on equalization for ISI mitiga- tion (3, 4]. In order to reduce ISI, the pulse shape can be chosen such that it is less susceptible to interference with its shifted replicas. In Nyquist signaling, each pulse is allowed to overlap with past and future pulses, but the shape is selected such that ISI is zero at certain points in time, Illustrated in Fig. 2.14, the idea is that all other pulses go through zero at the point when the present pulse reaches its peak. Thus, if the bit stream is sampled at ¢ = KTs, no IST exist A simple calculation leads to a basic condition for Nyquist signals. For a pulse shape, p(t), to introduce zero ISI, we have PTs) = 1 ifk =1 (2.48) =0 itk #0 (2.49)Sec. 2.2 Intersymbol Interference 27 Figure 2.14 Pulse shape with no ISI. Using a train of impulses to sample this pulse, we obtain PO) 28 = kTs) = 8). (2.50) ‘Taking the Fourier transform of both sides, we have PD EDS (*-#) 1 (2.51) That is, (2.52) Originally proposed by Nyquist and shown in Fig. 2.15, this result indicates that the shifted replicas of Pf) must add up to a flat spectrum. For example, a sinc waveform satisfies this condition because its Fourier transform is a rectangular box. ale Ts f Figure 2.15 Nyquist’s condition for the spectrum of a pulse shape that gives no ISI. A sine pulse shape, however, introduces difficulties in the design of the system. The filter required to produce the rectangular spectrum becomes quite complex if a sharp cutoff is necessary. Furthermore, the substantial signal energy near the edge of the spectrum complicates the filtering requirements in both the transmit and receive paths. In addition, the sinc waveform decays slowly with time, introducing considerable ISI in the presence of timing errors in the sampling command. A pulse shape often employed in Nyquist signaling is related to a “raised cosine” spectrum. Shown in Fig. 2.16, the time- and frequency-domain expres- sions of this function are, respectively, (a) = SMBHITS) costrat/Ts) i at/Ts 1 — 402/72 (2.53)28 Chap.2 Basic Concepts in RF Design Pit) 0 Ts 27s @ ) Figure 2.16 Raised-cosine pulse: (a) in time domain, (b) in frequency domain, and 1- P(=Ts O= lim = n(t)dt. (2.55) rat Jory This notion of dc component of a random signal is called the “time average.” Another definition of the average value is based on simultaneous sam- pling of all the waveforms in an ensemble (Fig. 2.19). Here, we compute the average by adding the sampled values and normalizing the sum to the numberSec.2.3 Random Processes and Noise. 31 Vn2 Figure 2.19 Averaging over sample functions. of waveforms, Called the “ensemble average” or “statistical averag is formally defined as » this type nit) [ners (2.56) where P,(n) is the probability density function of the process From the above definitions arise two questions. First, is the time average measured today equal to that measured tomorrow? Not necessarily. A process whose statistical properties are invariant to a time shift is called “stationary” (more accurately “strict-sense stationary”). Thus, the concept of time average is useful for stationary processes, e.g., noise voltage of a resistor held at a constant, temperature, Fortunately, most of the random phenomena in RF systems can be considered stationary. ‘The second question is: Is the time average of a stationary process equal tothe ensemble average? Not always, but for most random processes of interest in this book, we can assume so, thus avoiding ensemble averages. The time and ensemble averages defined above are of first order. Higher order averages can also be defined. Of particular interest are second-order averages, for they represent the power of signals. In the time domain, = 1 pan lim = t)dt, (2.57) fing [eo 57) which is also called the “mean square” power (with respect to a 1-@ resistor) if n(¢) isa voltage quantity. The second-order ensemble average is n(t) | * 12 (t) Py dn (2.58) For our purposes, < 17(t) > = n?(t). Probability Density Function When considering a random signal in the time domain, we usually need to know how often its amplitude is between certain limits, For example, if a binary data sequence is corrupted by additive noise (Fig. 2.20), it is important to find the probability that a logical ONE is32 Chap. 2 Basic Concepts in RF Design interpreted as a ZERO and vice versa, that is, how often the noise amplitude exceeds half of the signal amplitude. The amplitude statistics of a random signal x(0) is characterized by the probability density function, P,(x), defined as P.(x)dx = probability ofx < X < x + dx, (2.59) where X is the measured value of x(t) at some point in time. To estimate the PDF, we sample x(#) at many points (for many functions in the ensemble), construct bins of small width, choose the bin height equal to the number of samples whose value falls between the two edges of the bin, and normalize the bin heights to the total number of samples. Note that the PDF provides no information as to how fast the random signal varies in the time domain t Figure 2.20 Binary signal corrupted by noise. An important example of PDFs is the Gaussian (or normal) distribution. ‘The central limit theorem states that if many independent random processes with arbitrary PDFs are added, the PDF of the sum approaches a Gaussian distribution. Itis therefore not surprising that many natural phenomena exhibit Gaussian statistics. For example, since the noise of a resistor results from random “walk” of a very large number of electrons, each having relatively independent statistics, the overall amplitude follows a Gaussian PDF. ‘The Gaussian PDF is defined as 1 —@ — m)? 260 odin? 2 26) where @ and m are the standard deviation and mean of the distribution, re- spectively. From the PDF of the amplitude ofa random signal, we can also answer the following question: If a large number of samples are taken, what percentage will fall between x; and x2? This is given by the area under P,(x) from x; to xp, and for a Gaussian PDI m4 P(t; oo. As shown in Fig. 2.21, this condition is violated by two classes of signals: periodic waveforms and random signals, In most cases, however, these waveforms have a finite power: 1 pa P= jim = Ix(t)|’"dt < 00. 2.65) ting [oot (265) For periodic signals with P < 00, the Fourier transform can still be defined by representing each component of the Fourier series with an impulse in the frequency domain. For random signals, on the other hand, this is generally not possible because a frequency impulse indicates the existence of a deterministic sinusoidal component. Another practical problem is that even if we somehow define a Fourier transform for a random (stationary or nonstationary) process, the result itself is also a random process [5]. : SS ANANANMA t - t Sughd te ? t 3 The definition of energy can be visualized if.x() is a voltage applied across a 1-2 resistor.34 © Chap.2 Basic Concepts in RF Design From the above discussion we infer that frequency-domain characteristics of random signals are embodied in a function different from a direct Fourier transform. The power spectral density (PSD) (also called the “spectral density” orsimply the “spectrum”) is such a function. Before giving a formal definition of PSD, we describe its meaning from an intuitive point of view [6]. The spectral density, S-(f), of a random signal x(t) shows how much power the signal carries in a unit bandwidth around frequency f, As illustrated in Fig, 2.22, if we apply the signal to a bandpass filter with a 1-Hz bandwidth centered at f and measure the average output power over a sufficiently long time (on the order of 1 s), we obtain an estimate of S,(f). If this measurement is performed for each value of f, the overall spectrum of the signal is obtained. This is in fact the principle of operation of spectrum analyzers.* Band-Pass Filters, Power Meters Figure 222 Measurement of spectrum. The formal definition of the PSD is as follows [3]: S61) = jim, ver (2.66) where Xr(f) -[ x(t) exp(—j2mft)dt. (2.67) 0 4 Building a low-loss BPF with 1-Hz bandwidth and a center frequeney of, say, 1 GEIz is im- practical, Thus, actual spectrum analyzers both translate the spectrum to a lower center frequency and measure the power in a band wider than 1 Hz.Sec.23 Random Processes and Noise 35 ‘The definition can be understood with the aid of a corresponding computational algorithm (Fig. 2.23): (1) truncate x(t) to a relatively long interval (0, 7], (2) calculate the Fourier transform of the result and hence |Xy(f)|?, (3) repeat steps 1 and 2 for many sample functions of x(t) (e.g, for many noise voltage waveforms measured across a resistor), and (4) take the average of all |Xr()/? functions to arrive at [X7(f)/? and normalize the result to T. This algorithm proves useful in time-domain simulations incorporating random noise wave forms. 4x, (n/? x(t) s wat Nien marrage, ot. AL ; va (f)|? erates oath, Figure 2.23. Algorithm for PSD estimation. Since Sy(f) is an even function of f for real x(/) [3], as depicted in Fig. 2.24(a) the total power carried by x(t) in the frequency range [ /; fs] is equal to “fi phe a f SPdf t+ fo SA Pdf = f 2S.( Pdf. (2.68) fh fi A In fact, the right-hand side integral is the quantity measured by a spectrum analyzer; ie., the negative-frequency part of the spectrum is folded around the vertical axis and is added to the positive-frequency part [Fig. 2.24(b)]. W call the representation of Fig. 2.24(a) the “two-sided” spectrum and that of . 2.24(b) the “one-sided” spectrum @ ) Figure 2.24 (a) Two-sided and (b) one-sided spectra,36 Chap. 2 Basic Concepts in RF Design In graphical analysis of frequency-domain operations, itis generally sim- pler to use a two-sided spectrum, whereas actual noise calculations are more casily carried out with a one-sided spectrum, Nevertheless, these two repre- sentations bear no fundamental difference—though they can cause confusion. Asan example of $,(/), we consider the thermal noise voltage across a resistor of value R. The two-sided PSD is Si(f) = 2kTR, (2.69) where & is the Boltzmann constant and equal to 1.38 x 10-* J/K and T is the absolute temperature. Such a flat spectrum is called “white” because it contains the same level of power at all frequencies. Equation (2,69) raises two interesting questions, First, is the total noise power of a resistor [the area under S,(f)] infinite? In reality, S, () is flat for only | f| < 100 GHz, dropping beyond this frequency such that the total power remains finite [3]. Second, is the dimension of 2k7 R power per unit bandwidth (W/Hz)? No, the actual dimension is mean square voltage per unit bandwidth. We tacitly assume that this voltage is applied across a 1-& resistor to generate a power of 2kTR in a 1-Hz bandwidth. In circuit noise calculations, we often write Ve = 4kTR- Af, (2.70) where V7 is the mean square noise voltage generated by resistor R in a band- width Af. Called the “spot noise” for Af = 1 Hz, V2 is measured in V?/Hz. ‘To summarize the concepts of PDF and PSD, we note that the former is a statistical indication of how often the amplitude of a random process falls in a given range of values while the latter shows how much power the signal is expected to contain in a small frequency interval. In general, the PDF and PSD bear no relationship: thermal noise has a Gaussian PDF and a white PSD, whereas flicker (1//) noise has the same type of PDF but a PSD proportional tol/f. Random Signalsin Linear Systems _ The principal reason for defining the power spectral density function is that it allows many of the frequency-domain operations used with deterministic signals to be applied to random processes as well, It can be shown that if a signal with spectral density S,(f) is applied toa linear time-invariant system with transfer function H(s) (Fig, 2.25), then ‘the output spectrum is Sf) = SCMIACP, Qn where H(f) = H(s = j2rf) [3]. This agrees with our intuition that the spectrum of the signal is shaped by the transfer function of the system. It can also be shown that if x(#) is Gaussian, so is y(t) [3]