RF Transmitter Architectures and Circuits
`
`Behzad Razavi
`Electrical Engineering Department
`University of California, Los Angeles
`
`today's mobile communications, all three parameters are im(cid:173)
`portant because they determine the capacity, the talk time,
`and the maximum range, respectively. The design of trans(cid:173)
`mitters deals with primarily the trade-off between bandwidth
`efficiency and power efficiency, an issue arising from the prop(cid:173)
`erties of "constant-envelope" (or "nonlinear") and "variable(cid:173)
`envelope" (or "linear") modulation schemes.
`A modulated signal x(t) = Acos[wct + ¢(t)] has a con(cid:173)
`stant envelope if A does not vary with time. Such a wave(cid:173)
`form carries information in only the zero-crossing points and
`can therefore be processed by a nonlinear power amplifier
`with high power efficiency. A simple example, used in pag(cid:173)
`ing applications, is binary frequency shift keying (BFSK),
`whereby the baseband rectangular pulses are directly applied
`to a voltage-controlled oscillator (VCO) [Fig. l(a)]. In this
`case, ¢(t) = Kvco f XBB(t)dt, where Kvco is the gain of
`xBB(t) 0
`
`Baseband Data
`
`(a)
`
`Abstract
`This paper describes the design of RF transmitters for
`wireless applications. Following a review of co111stant- and
`variable-envelope modulation, general issues regarding the
`baseband/RF interface and the power amplifi1er/antenna
`interface are introduced. Various transmitter architec(cid:173)
`tures are then presented and the design of upconversion
`mixers and power amplifiers is studied. Examples of state
`of the art are also described.
`
`I. INTRODUCTION
`
`The design of RF transmitters for wireless applications en(cid:173)
`tails many challenges at both architecture and circuit levels.
`The number of off-chip components, the restrictions on un(cid:173)
`wanted emissions, and the trade-offs between the output power,
`the efficiency, and the required linearity directly impact the
`choice of the transmitter topology and the implementation
`of each circuit block. Furthermore, the disturbance of the
`transceiver's oscillators and receive path by the transmit path
`influences the frequency planning and limits the level of inte(cid:173)
`gration.
`This paper provides an overview of RF transmitter design for
`wireless systems with emphasis on high levels of integration.
`Focusing on mobile units, the paper begins with a review of
`constant- and variable-envelope modulation and a study of
`general transmitter design issues in Section IL Transmitter
`architectures are described in Section III and the design of
`building blocks in Section IV. Examples of the state of the art
`are presented in Section V.
`
`II. GENERAL CONSIDERATIONS
`
`An RF transmitter performs modulation, upconversion, and
`power amplification, with the first two functions combined in
`some cases. Transmitter design requires a solid understand(cid:173)
`ing of modulation schemes because of their influence on the
`choice of such building blocks as upconversion mixers, oscil(cid:173)
`lators, and power amplifiers (PAs). In this section, we briefly
`describe two commonly-used modulation formats and their
`design implications.
`
`A. Constant- and Variable-Envelope Modulation
`Modulation formats generally exhibit trade-offs between
`bandwidth efficiency, power efficiency, and detectability. In
`
`Baseband Data
`
`xBB(t)O
`
`Filter )//
`
`(t)
`XGMSK
`
`t
`
`Gaussian ..£\...II
`~ amnnonnn ..
`~=-~ ~ mvuvuuvu t
`
`(b)
`
`Fig. 1. Generation of (a) FSK and (b) GMSK signals.
`the VCO and XBB(t) the baseband signal.
`While exhibiting a constant envelope, BFSK signals oc(cid:173)
`cupy a relatively wide spectrum, partly owing to the abrupt
`frequency jumps introduced by the sharp edges of the base(cid:173)
`band pulses. lfthe frequency changes more smoothly from
`one bit to the next, then the required bandwidth decreases. To
`this end, Gaussian minimum shift keying (GMSK) alters the
`shape of the baseband pulses so as to vary the frequency grad(cid:173)
`ually. As shown in Fig. 1 (b ), the Fectangular pulses are first
`applied to a Gaussian filter, thereby generating smooth edges
`at the input of the frequency modulator. The resulting output
`is expressed as
`
`XGMSK(t) = Acos[wct + Kvco J XBB(t) * h(t)dt], (1)
`
`where h( t) is the impulse response of the Gaussian filter.
`Used in standards such as GSM, DECT, HIPERLAN, and
`
`0-7803-5443-5/99/$10.00©1999 IEEE
`
`10.1.1
`
`197
`IEEE 1999 CUSTOM INTEGRATED CIRCUITS CONFERENCE
`
`TCL EXHIBIT 1086
`Page 1 of 8
`
`

`
`the frequency-hopped version of IEEE 802.11, GMSK modu(cid:173)
`lation lends itself to nonlinear amplification while consuming a
`moderate bandwidth. Note, however, that the Gaussian shaped
`pulses in Fig. l(b) suffer from overlap in the time domain, in(cid:173)
`troducing intersymbol interference.
`The conceptual modulation method of Fig. l(b) is indeed
`employed in some transmitters, e.g., in DECT. However, if the
`amplitude of the baseband signal applied to the VCO or the gain
`of the VCO are poorly controlled, so is the bandwidth of the
`modulated signal. For this reason, in high-precision systems
`such as GSM, the waveform in (1) is rewritten as xaM sK(t) =
`AcoswctcosO- sinwctsinO, where 0 = Kvco J XBB(t) *
`h ( t) dt, and cos 0 and sin 0 are generated by accurate mixed(cid:173)
`signal techniques [l, 2].
`Variable-envelope signals are also used in many commu(cid:173)
`nication systems. Such signals can be expressed as x(t) =
`A(t) cos[wct + <f>(t)], where A(t) is the envelope. The varia(cid:173)
`tion of A(t) with time is undesirable but typically inevitable in
`linear modulation schemes designed to occupy a small band(cid:173)
`width. Consider a quadrature phase shift keying (QPSK)
`waveform with abrupt phase jumps [Fig. 2(a)]. Expressed
`as x(t) = a(t) cos wet+ b(t) sin wet, where a(t) and b(t) are
`sequences of binary rectangular pulses, the signal can be gen-
`
`o o o o o no o oo ~
`v v w v vu v v u v t
`
`180°
`
`180°
`
`90°
`
`a(t).fl_flf
`
`b(t) Jlflf
`
`(a)
`
`slnroct
`
`(b)
`
`Fig. 2. (a) QPSK waveform, (b) generation of QPSK signal from baseband
`streams.
`erated as shown in Fig. 2(b ), indicating the output spectrum
`is of the form [sin{ 7rTs !)2)/ ( 7rTs /) 2
`, where Ts denotes the
`width of one pulse. Such a waveform can be amplified nonlin(cid:173)
`early without corruption of the information, but it consumes
`substantial bandwidth.
`In order to tighten the spectrum of QPSK waveforms, the
`baseband pulses are altered from a rectangular form to a shape
`exhibiting a more compact spectrum. A popular format is
`"raised-cosi.ne" filtering, where each bit is represented by a
`sine-like waveform (Fig. 3). The actual expressions for the
`time-domain and frequency-domain representations of raised(cid:173)
`cosine signals can be found in [2], but we point out that the
`resulting spectrum approaches a "box" shape, providing high
`bandwidth efficiency.
`Used in standards such as IS-54, IS-95, and the direct(cid:173)
`sequence spread-spectrum version of IEEE 802.11, QPSK
`modulation with raised-cosine filtering 1 occupies minimal band(cid:173)
`width but suffers from a variable envelope.2 As illustrated in
`
`1 In reality, the raised-cosine filter is decomposed into two root raised(cid:173)
`cosine filters, one placed in the transmitter and the other in the receiver [2].
`20ffset QPSK and 1r / 4-QPSK exhibit less ripple.
`
`198
`
`ONE LL-
`
`ONE
`
`I
`I
`
`I v
`
`,
`
`Raised-Cosine
`Fiiter
`
`~ ~~ M ..
`....,,_,...
`';o,j V,'
`\>.$
`
`~\~,,,,_.
`•
`
`t
`
`l
`l
`l
`
`I
`I
`I
`
`'-'
`ZERO
`Fig. 3. Shaping baseband pulses to obtain a compact spectrum.
`
`Fig. 4, at every phase transition point the waveform experi-
`
`180°
`
`180°
`
`90°
`
`~
`
`QPSKwlth
`Square
`Baseband
`Pulses
`QPSKwlth ~
`Raised-Cosine
`Baseband
`Pulses
`
`f
`t
`Fig. 4. Variation of envelope with raised-cosine baseband pulses.
`ences a "ripple" in the envelope whose magnitude is propor(cid:173)
`tional to the change in phase. If such a signal is applied to a
`nonlinear PA, then the output exhibits significant power in ad(cid:173)
`jacent channels (Fig. 5) (an effect called "spectral regrowth")
`[2], possibly violating emission mask requirements of the stan-
`
`Variable-Envelope
`Signal
`
`f {: ,JJ.:
`..IL_
`
`Adjacent Channel Power
`Fig. 5. Generation of adjacent power in a nonlinear PA.
`dard. For this reason, the above standards incorporate linear
`- inevitably inefficient - power amplifiers while conferring a
`high capacity.
`
`B. Baseband/RF Interface
`The baseband/RF interface in a transmitter performs the
`desired modulation and upconversion. Since square baseband
`pulses are generally ill-suited to bandwidth-efficient communi(cid:173)
`cations, first the proper pulse shape is created by table-lookup
`or other waveform generation techniques. For example, as
`illustrated in Fig. 6, each baseband pulse is oversampled,
`mapped to the desired shape, and constructed in the analog
`domain by means of a digital-to-analog converter (DAC).
`The baseband/RF interface typically assumes one of two
`forms. For nonlinear modulation, the interface can be realized
`as shown in Fig. 7, where a VCO converts the input shaped
`pulses to frequency. In order to define the output center fre-
`10.1.2
`
`TCL EXHIBIT 1086
`Page 2 of 8
`
`

`
`~.It
`
`Address
`Generator
`
`Baseband r----i
`_A
`
`Data
`
`0
`
`.. ~ r;;;l--r~~-L .
`
`·~~
`
`Fig. 6. Baseband pulse shaping.
`
`~Modulated
`Xee<t> ~ Signal
`
`t
`
`Fig. 7. Baseband/RF interface in for nonlinear modulation.
`
`quency accurately, the VCO is first placed in a :synthesizer
`loop.
`Another topology, suited to both linear and nonlinear modu(cid:173)
`lation, is the quadrature upconverter depicted in Fig. 8. If sin(}
`and cos(} are generated according to our derivations in Sec-
`
`sin roe t
`
`Fig. 8. Quadrature upconverter for linear and nonlinear modulation.
`tion II.A, then the output is a phase- or frequency-modulated
`signal. For linear modulation formats such as QPSK with
`raised-cosine filtering, we can write
`
`n
`
`XQPSK = L anp(t-nTs) cos wet+ L bnp(t-nTs) sin wet,
`(2)
`where an, bn = ± 1 and p( t) denotes the desired pulse shape.
`Thus, as shown in Fig. 9, the single stream of rectangular
`pulses baseband pulses is first decomposed into two streams
`at half the rate and subsequently shaped and applied to the
`quadrature modulator.
`
`n
`
`C. PA/Antenna Interface
`RF transmitters employ a carefully designed amplifier and
`matching network to deliver the required power to the antenna.
`Fig. 10 depicts a typical configuration, where a duplexer filter
`separates the transmit (TX) and receive (RX) bands [in the
`case of frequency-domain duplexing (FDD)]. If the TX and
`RX bands coincide, the duplexer is replaced by an RF switch
`to perform time-domain duplexing (TDD). The duplexer filter
`suffers from a loss of2 to 3 dB, thereby dissipating 30% to 50%
`of the PA output in the form of heat. For TDD switches, on the
`other hand, the loss is in the range of 0.5 to 1 dB. While this
`issue makes TDD more attractive in power-conscilous applica-
`
`a(t)
`
`Pulse
`Shaper
`
`Serial-to-
`Parallel
`Converter
`
`Pulse
`Shaper
`
`b(t)
`
`Fig. 9. QPSK modulation with pulse shaping.
`
`sincoct
`
`XQPSK (t)
`
`Fig. 10. PNantenna interface.
`
`tions, FDD is more widely used in stringent cellular systems
`so as to ensure that the high-power signals transmitted by the
`users do not fall in their own receive band.
`In addition to substantial loss of power, duplexers also intro(cid:173)
`duce feedthrough from the transmit band to the receive band.
`Illustrated in Fig. 11, this effect arises simply because of the
`finite attenuation of the TX filter in the receive band - only
`LNA
`
`RX
`TX
`Band Band
`,_
`'
`I VI I
`Feedthrough
`
`'\.
`
`_;:::
`
`~/
`
`10 dB/div.
`
`...
`
`- ... ·~
`
`20 MHz/div.
`
`Fig. 11. Feedthrough from TX path to RX path.
`45 to 50 dB at the edge. (In practice, the leakage through the
`duplexer package or the printed circuit board may exacerbate
`the issue.) The feedthrough of the PA output to the LNA input
`creates two difficulties. ( 1) The large signal generated by the
`PA heavily desensitizes the receiver. For example, if the PA
`delivers +30 dBm and the feedthrough is -50 dB, then the
`receiver experiences an input of -20 dBm, a level comparable
`with the 1-dB compression point of many receiver designs. (2)
`The thermal noise at the output of the PA raises the input noise
`floor of the receiver. For example, if the PA thermal noise is
`-120 dBm/Hz and the leakage in the middle of the RX band
`is -60 dB, then the thermal noise introduced in the receive
`band reaches -180 dBm, only 6 dB higher than the available
`thermal noise power from the antenna. This issue requires that
`the entire TX path be designed for low noise.
`The duplexer feedthrough makes it desirable to avoid op(cid:173)
`erating the TX and RX paths simultaneously. For example,
`in the digital mode of IS-54 and in GSM, the TX and RX
`
`10.1.3
`
`199
`
`TCL EXHIBIT 1086
`Page 3 of 8
`
`

`
`time slots are offset such that they have no overlap (Fig. 12).
`Note that this does not introduce any "dead" time because in a
`TX
`
`users stand in proximity, then the thermal noise transmitted by
`one may corrupt the signal received by the other (Fig. 15).
`User 1 TX
`
`RX
`r-1
`
`Fig. 12. Time offset between TX and RX time slots in a TDMA system.
`time-division multiple access (TDMA) system, each user finds
`access to the network for only part of the time. Such an offset
`also allows sharing some of the transceiver's components, e.g.,
`oscillators and synthesizers, between the TX and RX paths. (In
`principle, much of the supply current can also be time-shared
`between the two paths.)
`Simultaneous operation of the transmitter and the receiver is
`nonetheless necessary in some applications. For example, in
`the relatively old "analog" standard Advanced Mobile Phone
`System (AMPS), neither of the two paths can be turned off
`simply because of the lack of digital storage of the signal.
`Also, in direct-sequence spread-spectrum communication sys(cid:173)
`tems such as IS-95, continuous monitor and control of the
`transmitted power requires that the mobile unit's RX and TX
`paths operate simultaneously.
`The signal transmitted by the antenna must comply with
`various emission regulations so that it does not corrupt other
`users' communication. GSM, for example, enforces the mask
`shown in Fig. 13, constraining both the modulation index
`
`10 dB/div.
`
`, ............. \
`I//
`f/....
`, ...
`//
`~
`
`\1
`'•:::\
`
`...,
`... :\
`'
`
`100 kHz/div.
`Fig. 13. GSM emission mask.
`of the GMSK waveform and the design of the TX building
`blocks (Section IV). In standards using linear modulation, e.g.,
`in IS-95 and the digital mode of IS-54, the adjacent-channel
`power ratio (ACPR) is specified. Illustrated in Fig. 14, ACPR
`requirements necessitate adequate linearity in the power am-
`
`~~~~-U-s~e-~1..._RX-t~
`
`Fig. 15. Effect of TX !henna! noise in RX band of another user.
`
`Ill. TRANSMITIER ARCHITECTURES
`
`The choice of a transmitter architecture is determined by
`two important factors: wanted and unwanted emission re(cid:173)
`quirements and the number of oscillators and external filters.
`In general, the architecture and frequency planning of the trans(cid:173)
`mitter must be selected in conjunction with those of the receiver
`so as to allow sharing hardware and possibly power.
`
`A. Direct-Conversion Architecture
`
`In direct-conversion transmitters, the output carrier fre(cid:173)
`quency is equal to the LO frequency, and modulation and
`upconversion occur in the same circuit (Fig. 16). The sim(cid:173)
`plicity of the topology makes it attractive for high levels of
`
`Baseband
`I
`
`Baseband
`Q
`
`Matching
`Network
`
`Fig. 16. Direct-conversion transmitter.
`
`integration.
`The direct-conversion architecture nonetheless suffers from
`an important drawback: disturbance of the local oscillator
`by the power amplifier output. Illustrated in Fig. 17, this
`issue arises because the PA output is a modulated waveform
`
`,_
`
`a -
`
`--,
`
`30 kHz 30 kHz
`
`Fig. 17. LO pulling by PA.
`having a high power and a spectrum centered around the LO
`frequency. Despite various shielding techniques employed to
`isolate the VCO, the "noisy" output of the PA still corrupts the
`oscillator spectrum. This corruption occurs through "injection
`Fig. 14. ACPRin IS-54 and IS-95.
`pulling" or "injection locking" [3], whereby the frequency of
`an oscillator tends to shift towards the frequency of an external
`plifier and upconversion mixers. Similar regulations apply to
`stimulus. As shown in Fig. 18, if the frequency of the injected
`spurs and harmonics at the output of the transmitter.
`It is interesting to note that, despite the use of offset TX and
`noise is close to the oscillator natural frequency, then the LO
`output is disturbed increasingly as the noise magnitude rises,
`RX time slots, GSM still requires a very low thermal noise
`eventually "locking" to the noise frequency. In practice, noise
`emission in the receive band. This is because if two mobile
`10.1.4
`
`200
`
`TCL EXHIBIT 1086
`Page 4 of 8
`
`

`
`Higher Injection Level
`
`Fig. 18. Injection pulling as the magnitude of the injected noise increases.
`
`levels as low as 40 dB below the oscillation level may create
`tremendous disturbance.
`The phenomenon of LO pulling is alleviated if the PA output
`spectrum is sufficiently far from the oscillator frequency. For
`quadrature upconversion, this can be accomplished by "offset(cid:173)
`ting" the LO frequency, that is, by adding or subltracting the
`output frequency of another oscillator [4]. Fig. 19 shows an
`example where the output signals of VC01 and VC02 are
`
`Fig. 19. Direct-conversion transmitter with offset LO.
`mixed and the result is filtered such that the carrier frequency
`is equal to w1 + w2, far from either w1 or w2.
`The selectivity of the first bandpass filter, BP F1, in Fig. 19
`impacts the quality of the transmitted signal. Owing to nonlin(cid:173)
`earities in the offset mixer, many spurs of the form mw1 ± nw2
`appear at the input of BP F1• If not adequately suppressed by
`the filter, such components degrade the quadraturn generation
`of the carrier phases as well as create spurs in the upconverted
`signal.
`
`B. Two-Step Architecture
`Another approach to circumventing the problem of LO
`pulling in transmitters is to upconvert the baseband signal
`in two (or more) steps so that the PA output spectrum is far
`from the frequency of the VCOs. As an example, consider
`the circuit shown in Fig. 20. Here, the baseband I and Q
`
`The first BPF suppresses the harmonics of the IF signal while
`the second removes the unwanted sideband centered around
`w, -w2.
`An advantage of two-step upconversion over the direct ap(cid:173)
`proach is that since quadrature modulation is performed at
`lower frequencies, I and Q matching is superior, leading to
`less cross-talk between the two bit streams (Section IV.A).
`Also, a channel filter may be used at the first IF to limit the
`transmitted noise and spurs in adjacent channels.
`The difficulty in two-step transmitters is that the bandpass
`filter following the second upconversion must reject the the un(cid:173)
`wanted sideband by a large factor, typically 50 to 60 dB. This
`is because the simple upconversion mixing operation produces
`both the wanted and the unwanted sidebands with equal mag(cid:173)
`nitudes. Owing to the higher center frequency, this filter is
`typically a passive, relatively expensive off-chip device.
`
`C. Offset-PLL Architecture
`A transmitter topology suited to systems using constant(cid:173)
`envelope modulation is the offset-PLL architecture [5, 6]. This
`technique has been invented to meet the stringent GSM re(cid:173)
`quirements for the thermal noise in the receive band (Fig. 15).
`First, consider the circuit shown in Fig. 21, where a quadra(cid:173)
`ture modulator is followed by a phase-locked loop. Here, the
`
`Fig. 21. Upconversion of a constant-envelope signal by means of a PLL.
`PLL operates as a narrowband filter centered around f o, sup(cid:173)
`pressing the out-of-band noise generated by the modulator. If
`the loop bandwidth of the PLL is chosen properly, the phase
`information in x 1 (t) is transferred to x2(t) faithfully while the
`output noise at large frequency offsets is determined by that
`of the VCO. Note that the VCO phase noise exists in other
`architectures as well, and the advantage of the circuit of Fig.
`21 is that it suppresses the noise contributed by other sources.
`In practice, it is difficult to operate the phase detector of
`Fig. 21 at high frequencies. If a divide-by-N is inserted
`in the feedback path (Fig. 22), then the phase modulation
`of x 1(t) is "amplified" by a factor of N when it appears in
`
`PA
`
`'t
`
`......l..""'----"Cf\"".: '""'' ........
`
`2C01 co
`
`C01 I 2C01 co
`~ro~,~
`: cosco2 t ffi A I
`f ~ "(;)
`____;:_
`C01-C02 i C01+C02
`002
`Fig. 20. Two-step transmitter.
`channels undergo quadrature modulation at a lower frequency,
`w1 [called the intermediate frequency (IF)], and the result is
`upconverted to w1 + w2 by mixing and band-pass filtering.
`
`Fig. 22. Upconversion of a constant-envelope signal by means of a PLL with
`feedback divider.
`x 2(t), requiring "finer" phase modulation in sin() and cos 0.
`More importantly, since the frequency steps in ho1 are also
`amplified by N, the synthesizer producing hot must provide
`small channel spacing, suffering from a long settling time. To
`
`10.1.5
`
`201
`
`TCL EXHIBIT 1086
`Page 5 of 8
`
`

`
`alleviate these issues, as shown in Fig. 23 [5], the PLL can
`incorporate an offset mixer driven by another oscillator so as
`
`cose
`
`sine
`
`fL02
`
`Fig. 23. Upconversion using offset PLL.
`to lower the frequency presented to the PD. Note that in this
`case, sin 0 and cos 0 are upconverted to an IF signal such that
`h02 ± ho1 = fo.
`In the offset-PLL architecture, the quadrature upconverter
`can be embedded inside the loop. Shown in Fig. 24 [6], the
`loop senses a constant frequency, !REF• thus minimizing the
`
`fL02
`
`Fig. 24. Offset PLL including quadrature upconversion.
`phase variation of x 1 (t) and hence modulating the phase of
`x2(t) according to the baseband waveforms.
`The offset-PLL architecture requires two high-frequency
`VCOs and a relatively selective filter (at the output of the offset
`mixer), and as such is more complex than other topologies de(cid:173)
`scribed above. Nevertheless, the low out-of-band noise offered
`by this configuration allows operating the transceiver without a
`duplexer, saving several tens of percent of the power delivered
`by the PA. With no expensive, bulky duplexers, offset-PLL
`transmitters are well-suited to low-cost, high-performance sys(cid:173)
`tems using constant-envelope modulation, but the issue of
`VCO pulling by the PA may be of concern here.
`
`IV. BUILDING BtOCKS
`
`A. Upconversion Mixers
`The upconversion mixers in a quadrature modulator can
`easily be realized as Gilbert cells, with their outputs added in
`the current domain (Fig. 25). Interestingly, both the linearity
`of the baseband ports and the phase and gain matching of
`the mixers impact the quality of the modulated signal. We
`consider the effect of nonlinearity in GMSK, e.g., the GMSK
`waveform of Section II.A with the assumption that cos 0 and
`sin 0 experience third-order distortion. The resulting signal
`can then be expressed as
`
`XGMsK(t) = Acoswct[cos 0 +a cos(30)]
`-A sinwct[sinO +a sin(30)],
`
`(3)
`
`Fig. 25. I/Q upconverter using Gilbert cells.
`
`where a represents the amount of third-order nonlinearity.
`Grouping the terms in (3), we obtain
`
`XGMSK(t) = Acos[wct + I<vco J XBB(t) * h(t)dt]
`+aAcos[wct + 3I<vco j XBB(t) * h(t)dt].
`
`(4)
`
`Equation (4) reveals that third-order distortion gives rise to a
`component centered around We but with a modulation index
`three times that of the ideal GMSK signal. Invoking Carson's
`rule [7], we note that the second term occupies roughly three
`times the bandwidth, raising the power transmitted in adjacent
`channels. Fig. 26 shows the simulated spectra of the two com-
`
`(4).
`
`(Horiz. scale
`
`Fig. 26. Simulated spectra of th'e two terms in Eq.
`normalized to bit rate, vert. scale: 5 dB/div.)
`ponents in Eq. (4) with a = 1, indicating that the unwanted
`signal indeed consumes a wider band. For this reason, a must
`be small enough that the transmission mask is not violated.
`Thus, the baseband port of the mixers typically incorporates
`resistive degeneration to achieve sufficient linearity.
`The matching of upconversion mixers is also important.
`Similar to the UQ mismatch effect in direct-conversion re(cid:173)
`ceivers [2], this imperfection leads to cross-talk between the
`two data streams modulated on the quadrature phases of the
`carrier. A common approach to quantifying the UQ mis(cid:173)
`match in a transmitter is to apply two signals Vo sinwint
`and Vo cos Wint to the I and Q input terminals and examine
`the spectrum produced by the adder. In the ideal case, the
`output in the band of interest is simply given by Vout(t) =
`VosinwintsinwLot+ VocoswintcoswLot = Vocos(wLo -
`Win)t. On the other hand, in the presence of a gain mis(cid:173)
`match of e and phase imbalance of 0, an unwanted sideband at
`
`202
`
`10.1.6
`
`TCL EXHIBIT 1086
`Page 6 of 8
`
`

`
`w LO + Win appears at the output. The power of the sideband
`at WLO +win divided by that of the sideband at <.vLo - Win
`serves as a measure of the l/Q imbalance [2]. In practice, the
`crosstalk between the two data streams becomes negligible if
`the above test yields an unwanted sideband about 30 dB below
`the desired signal.
`
`B. Power Amplifiers
`Power amplifiers are typically the most power-hungry build(cid:173)
`ing blocks of RF transceivers. The design of PAs,, especially
`for linear, low-voltage operation, remains a difficult problem,
`still defying an elegant solution. In practice, PA design has in(cid:173)
`volved a substantial amount of trial and error - one reason why
`discrete or hybrid implementations of this circuit are favored.
`Suppose a PA is to deliver 1 W to a 50-0. antenna while
`operating with a 3-V supply. Then, in order to generate a 20-
`V pp swing across the antenna, the circuit must incorporate a
`"matching network," e.g., a transformer. Shown in Fig. 27 is a
`simple example indicating that the peak current in ithe primary
`
`Fig. 27. Simple PA with matching network.
`of the transformer exceeds 800 mA. Furthermore, M 1 must
`sink the currents flowing through both the radip-frequency
`choke (RFC) and the transformer when Vx ~ 0, a peak value
`of 1.6 A. With realistic efficiencies of 30 to 50%, the peak
`current through M 1 may need to be even higher.
`The enormous currents in the output device and the matching
`network are one of the difficulties in the design of power
`If the peak current
`amplifiers and especially the package.
`through the output transistor is several amperes, then the slew
`rate at 900 MHz is on the order of 10 A/ns. Thus, even
`a parasitic inductance of 10 pH causes a 100-mV reduction
`in the voltage swing. Furthermore, parasitic inductances can
`introduce various resonances and even instability in the circuit.
`Similarly, a series resistance of a few tens of milliohms in
`the transistor, the RFC, or the matching network may result
`in a considerable loss. For these reasons, many layout and
`packaging issues that are usually unimportant in other analog
`and RF circuits become crucial in power amplifiers.
`In addition to output power, efficiency, and linearity, other
`parameters such as the required supply voltage, spurs and har(cid:173)
`monics, and power control are also critical in many appli(cid:173)
`cations. For example, in IS-95, the output power must be
`adjustable in 1-dB steps.
`PA Classes. Among many classes of PAs, only a few have
`proved practical in mobile communications. The limited bat(cid:173)
`tery voltage and energy constrain the choice of PA topologies
`in portable systems.
`
`The operation of PAs in classes A, B, and C can be viewed in
`terms of the "conduction angle," i.e., the percentage of the pe(cid:173)
`riod for during which the PA is on. As shown in Fig. 28(a), as
`the conduction angle,(}, decreases, the circuit becomes increas-
`
`Clasa A
`Bias { \ { \
`
`,,
`
`100%
`
`Current V
`0----C\ (\.
`,_.
`··~::':" tzrnmJ .
`
`Class B
`
`'.._:
`
`t
`
`(a)
`
`ClassC
`t:\
`0
`,'
`~
`'
`I
`
`..
`' t
`
`o..._ __ 1a~o'""0,....--3~so-:r-1e~
`
`o
`
`1
`1ao0
`
`1
`360° e
`
`(b)
`Fig. 28. (a) Signal current waveforms for classes A, B, and C, (b) efficiency
`and output power vs. conduction angle.
`ingly more nonlinear and the output power drops. However,
`the efficiency, 77, rises. In fact, it can be proved that 77 and the
`output power vary as illustrated in Fig. 28(b) [8], suggesting
`that true class C operation is not suited to portable transceivers,
`where the efficiency at the maximum power available from the
`supply is of greatest interest.
`In modern PA design, two other classes, E and F, have
`emerged as viable choices for high efficiency. Also known as
`"switching PAs" because their output transistor operates as a
`switch rather than a current source, class E and F amplifiers
`are well-suited to constant-envelope modulation systems. The
`high efficiency is obtained by ensuring that the output transistor
`sustains a small voltage when it carries substantial current and
`vice versa, a key property that does not exist in class C stages.
`Fig. 29 shows a class E amplifier [9], where the network
`consisting of C1, C2, L1> and RL is designed such that thedrain
`
`C2
`
`L1
`
`v
`101
`x(\ (\'
`
`~Vout J U .\. •
`M1':' J 1
`
`V:
`lno---1
`
`C
`
`R
`
`':'
`
`l
`
`t
`
`Fig. 29. Class E stage.
`current !DI and drain-source voltage Vx of M1 exhibit negli(cid:173)
`gible overlap in the time domain. Thus, even though the gate
`and drain voltages of M1 suffer from a finite transition time,
`the power consumed by the transistor is small. Class E stages,
`in theory, achieve an efficiency of 100% while deliveringfull
`power.
`In a class F stage, e.g., Fig. 30, the load network provides
`a high impedance at the second or third harmonic, creating
`an approximation of a rectangular waveform at the drain of
`the transistor. Thus, the overlap between the In and Vns
`waveforms can be minimized, leading to efficiencies greater
`than 85% [10].
`
`10.1.7
`
`203
`
`TCL EXHIBIT 1086
`Page 7 of 8
`
`

`
`Fig. 30. Class F stage.
`
`V. DESIGN EXAMPLES
`
`Fig. 31 depicts a two-stage 900-MHz PA [11]. The driver
`stage is a class F circuit incorporating two tanks in series, one
`
`- - . - - - -....... -Voo
`
`a 450-MHz LO so as to generate the quadrature phases of the IF
`signal with no RC-CR network. The IF signal is subsequently
`routed to one of two single-sideband mixers driven by a 1350-
`MHz LO, producing either 900 MHz or 1800 MHz. Fabricated
`in a 0.6-µm CMOS technology, the transmitter exhibits spurs
`40 dB below the carrier while drawing 70 mW from a 3-V
`supply.
`
`1800 MHz
`
`900MHz
`
`Class E
`Load
`
`Matching
`Network
`
`'. ................ ~ifl
`T ~I
`!JRL
`!ii~
`: L.~---•H••••••·.i t!.H•~---··········_j ~
`
`vb
`
`Fig. 31. Two-stage GaAs PA.
`tuned to the first harmonic and the other to the third, thereby
`producing relatively sharp edges at the input of the second
`stage and switching M 2 rapidly. The output stage is a class E
`amplifier designed for high efficiency. Fabricated in an 0.8-
`µm GaAs technology and operating from a 2.5-V supply, the
`PA delivers an output power of 250 mW with an efficiency of
`50%.
`A 1.9-GHz CMOS class EPA is shown in Fig. 32 (12].
`Configured in fully differential form so as to lower the cur-
`
`- . - - - - - . - - - - -....... --...,... Voo
`
`Fig. 32. Two-stage PA using injection-Jocked oscillators.
`rent drawn by each transistor and reduce the substrate noise,
`the circuit is in fact designed as two cascaded oscillators.
`The idea is that part of the input capacitance of each stage
`is driven by the stage itself, increasing the overall efficiency.
`The two oscillators are injection-locked to the input signal -
`a constant-envelope waveform. Implemented in a 0.35-µm
`CMOS technology and operating from a 2-V supply, the PA
`delivers 1 W with 48% efficiency. A point of concern in this
`design may be injection pulling of the output oscillator by a
`strong adjacent-channel interferer produced by a nearby user
`in the transmit band and received by the antenna.
`A dual-band CMOS transmitter is shown in Fig. 33 [13].
`Designed for operation in the 900-MHz and 1.8-GHz bands,
`the circuit incorporates two quadrature upconverters driven by
`
`Fig. 33. Simplified architecure of a dual-band transmitter.
`
`REFERENCES
`
`[I] K. Feher, Wireless Digital Communications, Upper Saddle
`River, New Jersey: Prentice-Hall, 1995.
`[2] B. Razavi, RF M