Behzad Razavi
`
`s wireless products such as cellular phones become an
`everyday part of people’s lives, the need for higher
`performance at lower costs becomes even more impor-
`tant. Overcoming the challenges involved in the design
`of radio-frequency (RF) transceivers can help meet this need. This
`article provides an overview of RF electronics in portable transceivers
`and describes design issues as well as current work toward achieving
`both high performance and low cost. To understand the implications
`in the design of RF integrated circuits (ICs) we look at the properties
`of the mobile communications environment. We then study receiver
`and transmitter architectures and their viability in present IC technolo-
`gies. An example of an RF transceiver is given and the design of
`transceiver building blocks is discussed. We conclude by looking at
`future directions in RF design.
`
`Wireless Communication Development
`Wireless technology came to existence in 1901 when Guglielmo
`Marconi successfully transmitted radio signals across the Atlantic
`Ocean. The consequences and prospects of this demonstration were
`simply overwhelming; the possibility of replacing telegraph and tele- 5
`phone communications with wave transmission through the “ether”
`portrayed an exciting future. However, while two-way wireless corn-
`2
`munication did soon materialize in the military, wireless transmission
`in daily life remained limited to one-way radio and television broad- 5
`casting by large, expensive stations. Ordinary, two-way phone con- $
`versations would still go over wires for many decades. The invention
`of the transistor, the development of Shannon’s information theory,
`and the conception of the cellular system - all at Bell Laboratories
`
`.$
`
`8755-3996/96/$5.0001996EEE
`
`Circuits & Devices
`
`TCL EXHIBIT 1037
`Page 1 of 14
`
`

`
`- paved the way for affordable mobile
`communications, as originally implemented
`in car phones and eventually realized in
`portable cellular phones (cell phones).
`But, why the sudden surge in wireless
`electronics? Market surveys show that in the
`United States more than 20,000 people join
`the cellular phone system every day, moti-
`vating competitive manufacturers to provide
`phone sets with increasingly higher perform-
`ance and lower cost. In fact, the present goal
`is to reduce both the power consumption and
`price of cell phones 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 wire-
`less communication when it is introduced in
`other facets of our lives: home phones, com-
`puters, facsimile, and television.
`While an immediate objective of the
`wireless industry is to combine cordless and
`cellular phones to allow seamless commu-
`nications virtually everywhere, the long-
`term plan is to produce an “omnipotent”
`wireless terminal that can handle voice, data,
`and video as well as provide computing
`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 communi-
`cation services (PCS) are almost here.
`Today’s pocket phones contain more
`than one million transistors, with only a very
`small fraction operating in the RF range and
`the rest performing low-frequency baseband
`signal processing. However, the RF section
`is still the design bottleneck of the entire
`system. This is primarily for three reasons.
`First, while digital circuits directly benefit
`from advances in integrated-circuit (IC)
`technologies, RF (analog) circuits do not
`benefit as much because they suffer from
`many more trade-offs and often require ex-
`ternal components (such as inductors) that
`are difficult to bring onto the chip even in
`modern fabrication processes. Second, in
`contrast to other types of analog circuits,
`proper RF design demands a solid under-
`standing of many areas that are not directly
`related to integrated circuits, e.g., micro-
`wave theory, communication theory, analog
`and digital modulation, transceiver architec-
`tures, etc. Each of these disciplines has been
`under development for many decades, mak-
`ing it difficult for an IC designer to acquire
`the necessary knowledge in a short time.
`Third, computer-aided analysis and synthe-
`sis tools for RF are still in their infancy,
`
`September 1996
`
`1. Simple RF front end.
`
`2. Effect of third-order nonlinearity in LNA.
`
`3. Definition of third-order intercept point.
`
`13
`
`TCL EXHIBIT 1037
`Page 2 of 14
`
`

`
`ple, we note that if the inpuvoutput static
`characteristic of the LNA is approximated
`as y(t) = aln(t) +. ap2(t) + a3x3(t) and x(t)
`= Aicosai t + A~cosa2t, then the cubic term
`yields components at 201 - 02 and 2w2 -
`mi, either of which may fall in the band. The
`standard approach to quantifying this effect
`is to choose A1 =A2 and, using extrapolation,
`calculate the input power that results in
`equal magnitudes for the fundamental com-
`ponents and the intermodulation products
`(Fig. 3). Such value of input power is called
`the “third-order intercept point” (IPS). It is
`interesting to note that this type of nonlinear-
`ity is important even if the signal carries
`
`conversion
`
`information in its phase or frequency rather
`than in its amplitude.
`Another important issue in the design of
`wireless receivers is the dynamic range of
`the input signal. Typically around 100 dB (a
`factor of 100,000 for voltage quantities), the
`dynamic rage is limited by a lower bound
`due to noise and an upper bound due to
`nonlinearities and saturation. The minimum
`detectable signal in today’s handsets is in the
`vicinity of -110 dBm (e0.71 pV,,
`in a
`50-!2 system), thus demanding very low
`noise in the receive path. For the upper
`bound, the receiver must achieve a high
`
`rare species.
`
`Wireless Environment
`The wireless communications environment,
`especially in urban areas, i s often called
`“hostile” because it imposes severe con-
`straints upon the transceiver design. Perhaps
`the most important constraint is the limited
`spectrum allocated by regulatory organiza-
`tions to wireless users. From Shannon’s
`theorem, this translates to a limited rate of
`information, mandating the use of sophisti-
`cated techniques such as coding, compres-
`sion, and bandwidth-efficient modulation,
`even for voice signals.
`The narrow bandwidth available to each
`user also impacts the design of the RF front
`end. As depicted in Fig. 1, the transmitter
`must employ narrowband amplification and
`filtering to avoid “leakage” to adjacent
`bands, and the receiver must be able to proc-
`ess the desired channel while sufficiently
`rejecting strong neighboring channels. To
`gain a better feeling about the latter issue,
`we note that if the front-end bandpass filter
`(BPF) in a 900-MHz receiver i s to provide
`60 dB of rejection at 45 kHz from the center
`of the channel, then the equivalent Q of the
`filter is on the order of lo’, a value difficult
`to achieve even in surface acoustic wave
`(SAW) filters. Since typical filters exhibit a
`trade-off between the loss and the Q and
`since in receiving very small signals the loss
`must be minimized, the out-of-channel re-
`jection of the front-end filters is usually
`insufficient, requiring further filtering in the
`following stages (typically at lower center
`frequencies). This will be clarified later in
`this article.
`The existence of large unwanted signals
`in the vicinity of the band of interest even
`after filtering creates difficulties in the de-
`sign of the following circuits, in particular
`the front-end low-noise amplifier (LNA).
`As shown in Fig. 2, if the LNA exhibits
`nonlinearity, then the “intermodulation 7. Effect of second-order distortion.
`
`6. Lo leakage to input.
`
`14
`
`Circuits & Devices
`
`TCL EXHIBIT 1037
`Page 3 of 14
`
`

`
`linearity so as to minimize intermodulation
`products. Also, saturation effects at high
`input levels often mandate the use of gain
`control in various parts of receivers.
`
`Receiver Architectures
`Complexity, cost, power dissipation, and the
`number of external components have been
`the primary criteria in selecting receiver ar- ‘
`chitectures. As IC technologies evolve, ar-
`chitectures that once seemed impractical
`may return because, when they are imple-
`mented in today’s advanced processes, their
`advantages outweigh their drawbacks.
`
`Homodyne Architecture
`Also called “direct conversion” architec-
`ture, the homodyne receiver is the natural
`topology for downconverting a signal from
`RF to baseband. The idea is simply to mix
`the RF signal with a local oscillator (LO)
`output and low-pass filter the result such that
`the center of the band of interest is translated
`directly to zero frequency (Fig. 4). Because
`of its typically high noise, the mixer is usu-
`ally preceded by an LNA. Also, in phase and 8. Heterodyne architecture.
`frequency modulation schemes, the W sig-
`nal is mixed with both the LO output and its
`quadrature so as to provide phase informa-
`tion (Fig. 5).
`The simplicity of the homodyne archi-
`tecture makes it attractive for compact, effi-
`cient implementation of RF receivers [ 1,2].
`However, several issues have impeded its
`widespread use, We briefly describe these
`issues and their impact on the design of
`related ICs.
`DC Offsets. Since in a homodyne re-
`ceiver the downconverted band extends to
`the vicinity of the zero frequency, extrane-
`ous offset voltages can corrupt the signal
`and, more importantly, saturate the follow-
`ing stages. To understand the origin and
`impact of offsets, consider the more realistic
`circuit shown in Fig. 6. Here, the mixer is
`followed by a low-pass filter, a post-ampli-
`fier, and an analog-to-digital converter
`(ADC). We make two observations: (1) The
`isolation between the LO and RF ports of the
`mixer is not perfect; due to capacitive cou-
`pling and, if the LO signal is supplied exter-
`nally, bond wire coupling, afinite amount of
`feedthrough exists from the LO port to
`points A and B. This effect is called “LO
`leakage.” The leakage signal appearing at
`the input of the LNA is amplified and mixed
`with the LO signal, thus producing a DC
`component at point C. This phenomenon is
`
`September 1996
`
`15
`
`TCL EXHIBIT 1037
`Page 4 of 14
`
`

`
`called “self-mixing.” (2) The total gain from
`the antenna to point X is typically around
`100 dB so that the microvolt input signal
`reaches a level that can be digitized by a
`low-cost ADC. Of this gain, approximately
`
`25 to 30 dB is contributed by the LNNmixer
`combination.
`With the above observations and noting
`that the LO power i s typically around 0 dBm
`(approximately 0.6 V,,),
`and the LO leak-
`
`+i
`
`+i
`
`Desired
`Image Channel
`
`I
`
`sincLLot
`coswiot
`
`-Lo
`
`0
`
`-c w
`
`10. Image rejection using single-sideband mixing.
`
`i j +j
`
`-a1
`
`0
`
`o w
`
`11. Weaver architecture
`
`12. Direct conversion transmitter.
`
`16
`
`age to point A on the order of -60 dB, we
`infer that the DC component at the output of
`the mixer due to self-mixing is roughly
`equal to 0 dBm -60 dB + 30 dB = -30 dBm,
`corresponding to a level of 10 mV. We also
`note that the signal level at this point can be
`as low as 25pVms. Thus, if directly ampli-
`fied by the remaining gain of 70 dB, the DC
`component saturates the following circuits,
`prohibiting the amplification of the desired
`signal.
`While high-pass filtering (i.e., AC cou-
`pling) may seem the solution here, in most
`of today’s modulation schemes the spectrum
`contains information at frequencies as low
`as a few tens of hertz, mandating a very low
`corner frequency in the filter. In addition to
`difficulties in implementing such a filter in
`IC form, a more fundamental problem is its
`slow response, an important issue if the off-
`set varies quickly. This occurs, for example,
`when a car moves at a high speed and the LO
`leakage reflections from the surrounding ob-
`jects change the offset rapidly.
`For these reasons, homodyne receivers
`require sophisticated offset-cancellation
`techniques. In [3], for example, the offset in
`the analog signal path is reduced by feeding
`information from the baseband digital signal
`processor (DSP). Alternatively, modulation
`schemes can be sought that contain negli-
`gible energy below a few kilohertz [A].
`Even-Order Distortion. While third-
`order mixing was considered as a source of
`interference in Fig. 2, even-order distortion
`also becomes problematic in homodyne
`downconversion. As depicted in Fig. 7, if
`two strong interferers close to the channel of
`interest experience a nonlinearity such as
`y(t) = ai x(t) + a2 x2(t), then they are trans-
`lated to a low frequency before the mixing
`operation and the result passes through the
`mixer with finite attenuation. This is be-
`cause, in the presence of mismatches that
`degrade the symmetry of the mixer, the mix-
`ing operation can be viewed as x(t)(a + A
`cos at), indicating that a fraction of x(t)
`appears at the output without frequency
`translation. A similar effect occurs if the LO
`output duty cycle deviates from 50%. An-
`other issue is that the second harmonic of the
`input signal (due to the square term in the
`above equation) is mixed with the second
`harmonic of the LO output, thereby appear-
`ing in the baseband and interfering with the
`actual signal [5]. For these reasons, even-or-
`
`Circuits & Devices
`
`TCL EXHIBIT 1037
`Page 5 of 14
`
`

`
`der intermodulation corrupts the baseband
`signal.
`I-Q Mismatch. As mentioned above,
`in most phase and frequency modulation
`schemes the downconversion path must
`use quadrature mixing. The required I and Q
`phases of the LO raise an issue related to the
`mismatches between these two signals. If
`the amplitudes of the I and Q outputs are
`not equal or their phase difference deviates
`from go", the error rate in detecting the
`baseband signal rises. The task of generating
`I and Q phases with precise matching is
`discussed later.
`A second-order effect arises from the
`mismatches between the two mixers them-
`selves. Since the mixers process high-fre-
`quency signals here, their phase and gain are
`sensitive to parasitics and hence susceptible
`to mismatches.
`LO Leakage. In addition to introducing
`DC offsets, leakage of the LO signal to the
`antenna and radiation therefrom creates in-
`terference in the band of other receivers. The
`design of the wireless infrastructure and the
`regulations of the Federal Communications
`Commission (FCC) impose upper bounds
`on the amount of LO radiation, typically
`between -60 to -80 dBm.
`Flicker Noise. Owing to the limited gain
`provided by the LNA and the mixer, the
`downconverted signal is relatively small and
`quite sensitive to noise. Since device flicker
`noise becomes significant at low frequen-
`
`LNA
`
`Duplexer
`Filter
`
`15. Representative RF transceiver.
`
`September 1996
`
`13. Disturbance of VCO by PA in direct conversion transmitter.
`
`PA
`
`vcop
`U -L
`
`Phase
`Splitter
`
`+
`
`14. Alternative transmitter architectures. (a) Two-step conversion, (b) Offset VCO.
`
`Oversampled
`I
`
`Oversampled
`Q
`
`17
`
`TCL EXHIBIT 1037
`Page 6 of 14
`
`

`
`cies, amplification of the baseband signal
`with low noise is an important issue.
`
`Heterodoyne Architectures
`The design issues mentioned above for the
`homodyne receiver have motivated the in-
`vention of other architectures. Most com-
`monly used is the heterodyne topology. (In
`this article, we do not make a distinction
`between “heterodyne” and “superhetero-
`dyne.”) Illustrated in Fig. 8 in a simple form,
`a heterodyne receiver first downconverts the
`input to an “intermediate frequency” (IF).
`The resulting signal is subsequently band-
`pass filtered, amplified, and downconverted
`again. In the case of digital modulation, the
`last downconversion generates both I and Q
`phases of the signal.
`The heterodyne architecture alleviates
`two of the homodyne reception issues by
`avoiding them at high frequencies or low
`signal levels. The effect of DC offsets of the
`first few stages is removed by bandpass
`filtering, and that of the last stage is sup-
`pressed by the total gain in the preceding
`stages. Also, I and Q mismatches occur at
`much lower frequencies and are therefore
`easier to control and correct. As for the LO
`leakage, since 0~01 is out of the band of
`interest, it is suppressed by the front-end
`BPF and its radiation from the antenna is less
`objectionable.
`Perhaps the most important feature of the
`heterodyne receiver is its selectivity, i.e., the
`capability to process and select small signals
`in the presence of strong interferers. While
`selecting a 30-kHz channel at a center fre-
`quency of 900 MHz requires prohibitively
`large Qs, in Fig. 8 bandpass filtering is per-
`formed at progressively lower center fre-
`quencies. For example, the third BPF may
`operate at a center frequency of 400 kHz,
`thereby providing high selectivity for a 30-
`kHz channel. In other words, the filters have
`much more relaxed requirements.
`Despite the above merits, heterodyning
`entails a number of drawbacks. The most
`significant problem is the “image fre-
`quency.” Since a simple mixer does not pre-
`serve the polarity of the difference between
`its input frequencies, it translates the bands
`both above and below the carrier to the same
`frequency [Fig. 9(a)]. Thus, the mixing op-
`eration must be preceded by an “image re-
`ject” filter [Fig. 9(b)], usually apassive one.
`The issue of image rejection leads to an
`interesting trade-off among three parame-
`ters: the amount of image noise, the spacing
`
`18
`
`between the band and the image (= 2 IF),
`and the loss of the filter. To minimize the
`image noise, we can either increase the IF
`(so that the filter provides more attenuation
`at the image frequency) or tolerate greater
`loss in the filter while increasing its Q. Since
`the LNA gain is typically less than 15 dB,
`the filter loss should not exceed a few dB,
`and the trade-off reduces to one between the
`image noise and the value of IF.
`How high can the IF be? Recall from Fig.
`8 that the filter following the first mixer must
`select the band. As the IF and, hence, the
`center frequency of this filter increase, so
`does the required Q, thereby imposing a
`fundamental trade-off between image rejec-
`tion and channel selection. For the 900-MHz
`and 1.8-GHz bands, typical IFS range from
`70 MHz to 200 MHz.
`Another drawback of the heterodyne ar-
`chitecture is that the LNA must drive a 5 0 4
`impedance because the image-reject filter
`cannot be integrated and is therefore placed
`off-chip. This adds another dimension to the
`trade-offs among noise, linearity, gain, and
`power dissipation of the amplifier, further
`complicating the design. The image-reject
`and channel-select filters are typically ex-
`pensive and bulky, making the heterodyne
`approach less attractive for small, low-cost
`wireless terminals. Nevertheless, hetero-
`dyning has been the dominant choice for
`many decades [6,7].
`
`Image-Reject Architectures
`The issues related to the image-reject filter
`have motivated RF designers to seek other
`techniques of rejecting the image in a het-
`erodyne receiver. One such technique origi-
`nates from a single-sideband modulator
`introduced by Ralph Hartley in 1928 [&I.
`Illustrated in Fig. 10, Hartley’s circuit mixes
`the RF input with the quadrature outputs of
`the local oscillator, low-pass filters the re-
`sulting signals, and shifts one by 90” before
`adding them together. The reader can easily
`verify that if the input is equal to
`Awcoswwt+Aposcofl, where c.11 is the im-
`age frequency, then the output is propor-
`tional to ARFCOS(WLO--ORF)~. As a more
`general case, we consider the input spectrum
`shown in Fig. 10 and note that mixing with
`sinmot and cosoLot yields the spectra of
`Fig. 10 at nodes A and B, respectively (the
`factor *j in these spectra is to indicate con-
`volution with 2jj6(atw~0)/2 (spectrum of
`sinoLot)). Since a phase shift of +90” in the
`
`signal at A corresponds to multiplication by
`+j and inverting the positive frequencies, we
`obtain the four spectra at nodes B and C as
`the inputs to the adder. The output is there-
`fore free from the image.
`The principal drawback of image-reject
`mixers is their sensitivity to mismatches. For
`example, if the phase difference between the
`LO quadrature phases deviates from 90”, the
`cancellations shown in Fig. 10 are imperfect
`and some image noise corrupts the down-
`
`Frequency
`
`
`
`
`Noise -Power
`
`/
`\
`
`Linearity
`
`\
`-
`/
`
`Supply
`Voltage
`
`Gain
`
`16. RF design hexagon.
`
`17. Low-noise amplifier.
`
`I
`
`vcc
`
`18. Gilbert mixer.
`
`Circuits & Devices
`
`TCL EXHIBIT 1037
`Page 7 of 14
`
`

`
`Vout2 vouta
`
`Transmitter Architectures
`In contrast to the variety of approaches in-
`vented for RF reception, transmitter archi-
`tectures are found in only a few forms. This
`is because issues such as image rejection and
`band selectivity are more relaxed in trans-
`mitters, leaving the output power amplifier
`(PA) design as the primary challenge.
`A simple direct conversion transmitter is
`shown in Fig. 12. Here, the baseband signal
`is mixed with the LO output and the result
`is bandpass filtered and applied to the PA. A
`matching network is usually interposed be-
`tween the PA and the antenna to allow maxi-
`mum power transfer and filter out-of-band
`components that result from nonlinearities
`in the amplifier. Note that since the base-
`band signal is produced in the transmitter
`and is therefore sufficiently strong, the noise
`of the mixers is much less critical here than
`in receivers.
`Direct conversion architectures suffer
`from an important drawback disturbance of
`the transmit local oscillator by the output
`PA. Illustrated in Fig. 13, this issue arises
`because the PA output is a modulated wave-
`form with high power and a spectrum cen-
`t h e voltage-controlled
`tered around
`oscillator VCO frequency. Thus, despite
`various shielding techniques that attempt to
`isolate the VCO, the “noisy” output of the
`PA still corrupts the oscillator spectrum.
`(The actual mechanism of this corruption is
`
`called “injection pulling” or “injection lock-
`ing.” When disturbed by a close interferer at
`frequency mz, an oscillator operating at WO
`tends to shift to oz.) This problem worsens
`if the PA is turned on and off periodically to
`save power.
`The above difficulty is alleviated if the
`PA output spectrum is sufficiently higher or
`lower than the VCO frequency. For exam-
`ple, as shown in Fig. 14(a), the upconversion
`can be performed in two steps, generating a
`final spectrum that differs from w by (UI
`[ 111. Alternatively, the VCO frequency can
`be “offset” by adding or subtracting the out-
`put frequency of another oscillator (Fig.
`14(b)) [7]. Note that in both cases, some
`filtering is required to reject unwanted parts
`of the spectrum.
`The most difficult part of transmitters to
`design is the PA, mainly because of severe
`trade-offs among its efficiency, linearity,
`and supply voltage. In typical PA topolo-
`gies, the efficiency drops as the circuit is
`designed for higher linearity or lower supply
`voltage. For a typical peak output power of
`1 W, an efficiency of 50% means that an
`additional 1 W is wasted, which is a substan-
`tial amount with respect to the power dissi-
`pation of the rest of a portable phone.
`The reader may wonder why the linearity
`of the PA is important if only the phase of
`the carrier is modulated. Indeed in analog
`
`Yo
`
`19. Single-balanced mixer.
`
`converted signal [5]. For typical matching in
`IC technologies, the image is rejected by
`about 30 to 40 dB [9]. Another important
`issue is the higher power dissipation andor
`noise due to the use of two high-frequency
`mixers. Also, circuits that shift the down-
`converted signal by 90” generally suffer
`from trade-offs among linearity, noise, and
`power dissipation.
`
`Weaver Architecture
`In our discussion of images, we noted that
`any frequency translation leads to corrup-
`tion of the signal by the image, except when
`a symmetric band is brought down to zero
`frequency (homodyne). The Weaver tech-
`nique allows an arbitrary translation of the
`signal band without image interference [lo].
`Illustrated in Fig. 11, this approach d o m -
`converts the signal in two steps. In the first step,
`the input is mixed with the quadrature phases
`of the fiist local oscillator and the result is
`low-pass filtered, yielding the spectra at nodes
`A and B. In the second step, these signals are
`translated to zero frequency and added to-
`gether, thereby effecting image cancellation.
`The important advantage of the Weaver
`architecture is that it does not require high-Q
`bandpass filters. Even though the LPFs shown
`in Fig. 1 l(a) must be preceded by capacitive
`coupling to eliminate DC offsets (similar to
`homodyne) and, as such, the combination is a
`bandpass filter, the out-of-band rejection of
`these fdters is quite relaxed. Note that some
`amplification is necessary before the second
`set of mixers to reduce the effect of their noise.
`The Weaver method suffers from the same
`drawback as the image-reject mixer: incom-
`plete cancellation of the image in the pres-
`ence of mismatches.
`
`September 1996
`
`Ideal Oscillator
`
`Actual Oscillator
`i
`
`WO
`
`w
`
`WO
`
`20. Phase noise and sidebands in the output of oscillators.
`
`21. Pulse swallow synthesizer.
`
`fOUT
`
`-
`
`19
`
`TCL EXHIBIT 1037
`Page 8 of 14
`
`

`
`FM systems, the linearity is not critical and
`the efficiency trades only with the supply
`voltage, usually approaching 60% at the
`peak output power. On the other hand, in
`digital modulation schemes such as quadra-
`ture phase shift keying (QPSK) the situation
`is more complicated. Since a QPSK signal
`has a relatively wide spectrum, it usually
`undergoes bandpass filtering to limit its
`bandwidth to that of one channel. The result-
`ing signal, however, does not have a con-
`stant envelope, i.e., it exhibits some
`amplitude modulation. Now, if this signal
`experiences nonlinear amplification, its
`spectrum widens, spilling into adjacent
`channels and defeating the purpose of band-
`pass filtering.
`
`In order to resolve this issue, RF system
`designers have employed two different
`strategies. First, they have found digital
`modulation schemes in which the envelope
`of the signal remains constant after filtering
`and, hence, the spectrum does not widen in
`the presence of PA nonlinearities. These
`schemes are known as “continuous phase
`modulation,” where the phase of the carrier
`varies smoothly from one bit to the next.
`Second, they have devised feedback and
`feedforward circuit techniques to improve
`the linearity of PAS with negligible degrada-
`tion in efficiency [13, 14,211.
`
`Overall System
`With the above discussion of transceiver
`architectures, we can now consider a more
`
`Pulse
`Remover
`
`X
`
`Y
`
`Channel
`Select
`
`f
`
`e
`
`r
`
`e
`
`n
`
`c
`
`e
`
`e
`
`
`
`f
`
`R
`
`e
`
`Input
`
`LPF OutDut!
`
`vco
`output
`
`22. (a) Fractional-N synthesizer, (b) problem of reference sidebands.
`
`20
`
`complete system. Shown in Fig. 1.5 is a
`transceiver with heterodyning in the receive
`path and direct conversion in the transmit
`path. The transmit VCO may employ the
`offset technique of Fig. 14(b) to avoid injec-
`tion pulling.
`In most mobile phone systems, the trans-
`mit and receive bands are different, with the
`translation performed at the base station. In
`a full-duplex system (where reception and
`transmission occur simultaneously through
`a single antenna), this is necessary because
`the two paths must be somehow separated.
`With two different bands, this is accom-
`plished by a narrowband front-end filter,
`called the “duplexer.” This filter also sup-
`presses out-of-band noise and interference
`in the receive path.
`In Fig. 1.5, the receive and transmit
`LOS are embedded in a frequency synthe-
`sizer. When initiating a call, a mobile unit
`is assigned two communication channels
`(for receive and transmit) by the base sta-
`tion. The synthesizer selects the proper
`carrier frequency for each channel accord-
`ing to a digital input. The important issues
`here are how “pure” the synthesizer output
`is, and how fast can it can switch the LO
`frequency from one channel to another.
`We return to these issues in the section on
`frequency synthesizers.
`In the receive path, the downconverted
`signal is applied to an ADC. The ADC is
`necessary even if the information lies in the
`phase (or frequency), because baseband op-
`erations such as equalization, matched filter-
`ing, and despreading are performed with
`higher precision in the digital domain than
`in the analog domain. Digital signal proces-
`sors have thus become an integral part o f
`wireless transceivers,
`In the transmitter, the digitized voice un-
`dergoes compression and coding. The re-
`sulting stream of ONES and ZEROS is
`subsequently oversampled and subdivided
`into multi-bit words, which are then applied
`to two digital-to-analog converters (DACs)
`(Fig. 1.5). This operation takes place for an
`interesting reason. In digital modulation
`schemes, the ideal pulse shape for each bit
`produced in the baseband is quite different
`from a rectangular function. For example, as
`mentioned earlier, the modulated carrier
`may need to be such that its envelope re-
`mains constant after filtering. Thus, it is
`usually necessary to convert the rectangular
`pulses to another shape. Furthermore, the
`bandpass filtering required after modulation
`
`Circuits & Devices
`
`TCL EXHIBIT 1037
`Page 9 of 14
`
`

`
`The very low noise required of the LNA
`usually mandates the use of only one active
`device at the input without any (high-fre-
`quency) resistive feedback. In order to pro-
`vide sufficient gain while driving 50 Q,
`LNAs typically employ more than one stage.
`An interesting example is shown in Fig. 17
`[15], where the first stage utilizes a bond-
`wire inductance of 1.5 nH to degenerate the
`common-emitter amplifier without intro-
`ducing additional noise. This technique both
`linearizes the LNA and makes it possible to
`achieve a 5042 input impedance. Bias volt-
`ages v b i and vb2 and the low-frequency
`feedback amplifier A1 are chosen so as to
`stabilize the gain against temperature and
`supply variations. The circuit exhibits a
`noise figure of 2.2 dB, an ZP3 of -10 dBm,
`and a gain of 16 dB at 900 MHz.
`The issue of linearity becomes more sig-
`nificant in mixers because they must handle
`signals that are amplified by the LNA. While
`it may seem that the issue of noise is relaxed
`by the same factor, in practice, (active) mix-
`ers exhibit much higher noise simply be-
`cause they employ more devices in the
`signal path than do LNAs and suffer from
`various noise frequency folding effects. As
`an example, consider the Gilbert cell mixer
`shown in Fig. 18. Since the LNA output is
`usually single-ended, the base of Q2 is con-
`nected to a reference voltage. The RF input
`stage is resistively degenerated to provide
`sufficient linearity, but at the cost of higher
`input noise. Now consider the four switch-
`ing devices e 3 - Q ~ . During switching, all of
`these devices are on for part of the period (if
`the LO waveform is not an ideal rectangular
`signal), thereby contributing both shot noise
`and base resistance thermal noise to the out-
`put. Furthermore, even when the switching
`is complete, the devices that are on (e.g., Q3
`and Q4) continue to introduce noise because
`the capacitance at nodes X and Y is quite
`large. For these reasons, the noise figure of
`a Gilbert cell with reasonable linearity usu-
`ally exceeds 10 dB.
`Shown in Fig. 19 is a simpler mixer with
`single-ended RF input [15]. This circuit
`achieves a noise figure of 15.8 dB with an
`IP3 of +6 dBm. An interesting point should
`be mentioned regarding the noise behavior
`of this circuit with differential or single-
`ended outputs. We note that if the output is
`taken from X with respect to ground, then
`the mixer operation can be viewed as multi-
`plication of the input signal by a square wave
`toggling between 0 and A, where A is the
`
`21
`
`23. Direct digital synthesizer.
`
`”‘‘ Single-Sideband -
`
`hut
`
`Mixer
`
`24. Phase-locked synthesizer with DDS offset mixing.
`
`can be equivalently performed as low-pass
`filtering on the baseband signal.
`
`Building Blocks
`RF architectures impose severe require-
`ments upon the performance of their con-
`stituent circuits. The very small signal
`amplitude received by the antenna in the
`presence of large interferers mandates both
`careful allocation of noise and linearity to
`various stages and sufficient suppression of
`spurious components generated in the fre-
`quency