`
`1399
`
`Direct-Conversion Radio Transceivers
`for Digital Communications
`
`Asad A. Abidi, Senior Member, IEEE
`
`Abstract—Direct-conversion is an alternative wireless receiver
`architecture to the well-established superheterodyne, particularly
`for highly integrated, low-power terminals. Its fundamental ad-
`vantage is that the received signal is amplified and filtered at
`baseband rather than at some high intermediate frequency. This
`means lower current drain in the amplifiers and active filters
`and a simpler task of image-rejection. There is considerable
`interest to use it in digital cellular telephones and miniature
`radio messaging systems. This paper briefly covers case studies
`in the use of direct-conversion receivers and transmitters and
`summarizes some of the key problems in their implementations.
`Solutions to these problemsarise not only from more appropriate
`circuit design but also from exploiting system characteristics,
`such as the modulation format in the system. Baseband digital
`signal processing must be coupled to the analog front-end to make
`direct-conversion transceivers a practical reality,
`
`I.
`
`INTRODUCTION
`
`HE CURRENT interest in portable wireless communica-
`tions devices is prompting research into new IC technolo-
`gies, circuit configurations and transceiver architectures. Low-
`power miniature radio transceivers are sought to communicate
`digital data in cellular telephones, wireless networks, and
`radio messaging systems. While transistor technology scaling
`and improved circuit techniques will contribute evolutionary
`advances towards this goal, architectural innovations in the
`transceiver may lead to revolutionary improvements [1]. It
`is in this context that
`there is a resurgence of interest in
`direct-conversion.
`The superheterodyne receiver, which Armstrong introduced
`in 1918 [2], is generally thought to be the receiver of choice
`owing to its high selectivity and sensitivity. Something like
`98% of radio receivers use this architecture. In a superhetero-
`dynereceiver, the input signal is first amplified at RF in a tuned
`stage, then converted by an offset-frequency local oscillator
`to a lower intermediate frequency (IF), and substantially
`amplified in a tuned IF “strip” containing highly-selective
`passive bandpass filters. The role of the various filters is -
`illustrated by the typical frequency plan of a superheterodyne
`receiver (Fig. 1). The IF must be sufficiently high so that the
`image channellies in the stopband of the RF preselection filter
`or the antenna, otherwise the IF filter will pass this channel
`unattenuated in its own image passband. These considerations
`determine the familiar intermediate frequencies used in radio
`and TV reccivers.
`
`
`
`x” RF Spectrum
`_gieoniaemn
`nena, |
`#
`‘
`/:
`%
`
`< QxIF
`
`RF Preselect Filter Response
`
`Downconvert <>
`
`
`
` IF Filter
`Response
`
`-F
`
`+HIF
`
`IF Spectrum
`
`(after filter)
`
`ll
`
`OHz
`
`4
`
`Fig. 1. Frequency plan of a superheterodyne receiver. Choice of IF is
`governed by width of preselectfilter passband. The RF preselect and IF filters
`work together to select the desired channel.
`
`Manuscript received May 22, 1995; revised August 29, 1995.
`The author is with the Electrical Engineering Department, University of
`California, Los Angeles, CA 90095 USA.
`. IEEE Log Number 9415818.
`
`In a standard broadcast FM receiver, for instance, the 10.7
`MHzIF guarantees that the image channel lies outside the
`20 MHz wide FM band. Therefore, even if the preselect
`filter inadequately suppresses the image, which is assumed not
`to be an FM signal, the subsequent frequency-discriminating
`detector will
`inherently tend to reject
`it. This relaxes the
`selectivity of the 100 MHz preselect filter, which may be
`constructed with either a single or ganged collection. of LC-
`tuned circuits. Ceramic filters are an enabling technology
`for the 10:7 MHz FM IF. These small and cheap filters
`0018-9200/95$04.00 © 1995 IEEE
`
`1
`
`SAMSUNG 1024
`
`SAMSUNG 1024
`
`1
`
`
`
`1400
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 30, NO.'12, DECEMBER 1995
`
`ory
`2.
`ea
`
`pallseq
`
`{a)
`
`Image
`
` 100
`
`
`
`| Gain Mismatch, dB
`
`whQoQ.
`
`hyoO
`
`ooO
`
`eloS
`
`ona
`
`
`
`offer a narrow passband and good stopband attenuation, and
`they are widely used. The traditional 43.5 MHz IF in a TV
`receiver cannot suppress the image across the entire VHF,
`hyperband, and UHF bands,
`so as the receiver is tuned
`across various sub-bands, one of an array of RF narrowband
`filters is switched into the RF front-end [3] to suppress the
`image channel. Many analog cellular telephones use a 90
`MHz IF. Amplification andfiltering at these high intermediate
`frequencies between 10-100 MHz comesat the price of power
`dissipation because transistors must be biased at large currents
`to drive the parasitics and the low characteristic impedance of
`the passive IF filters. Purther, the IF strip may require a large
`number of off-chip passive components, which add to receiver
`size. Although these are not serious problems for tabletop
`receivers—the easy alignment of a superheterodyne, resulting
`in a high selectivity, was always one of its strengths—they
`may become limitations in miniature,
`low power transceiv-
`ers,
`Wireless receivers must often handle very weak channels
`existing side-by-side with very strong channels in the same
`band. Thus, in addition to a minimum stopband attenuation
`to suppress the tmage channel,
`the filter must also have a
`wide dynamic range,
`that
`is,
`the ability to handle strong
`signals without distortion while remaining sensitive to weak
`signals above the intrinsic noise level in the passband. In
`this respect, passive filters are almost always superior, as the
`small-signal handling of active bandpass filters is limited by
`a fundamentally higher noise level [4], and nonlinearity in
`the active device tends to distort large signals. The dynamic
`range of active filters may only be increased at the expense of
`capacitor size and power dissipation.
`Although most often a passive RF preselectfilter attenuates
`the image channel,
`it may also be suppressed by selective
`signal cancellation. Here, the entire RF spectrum is downcon-
`verted to an IF in two identical mixers driven by quadrature
`phases of a local oscillator (LO). The downconverted spectra
`in the two branches are subjected to a 90° phase-shift relative
`to one another and then added (Fig. 2(a)). With appropriate
`design,
`this arrangement downconverts the desired channel
`to IF with, the same phase in the two branches and the
`image channel
`to -IF but antiphase in the two branches.
`After addition,
`the desired channel appears at
`the output
`with double strength, while the image channel subtracts and
`disappears. This image-reject downconverter is the dual of
`Weaver's celebrated phasing methodof sideband selection
`[5], which is discussed later in the section on transmitters.
`In practice, departures from quadrature in the two LO signals
`and gain mismatch in the two branches will limit the extent
`- of signal cancellation (Fig. 2(b)). When used in a receiver,
`the effectiveness of this image-suppression method is fur-
`ther limited by the wide dynamic range of radio signals.
`If the unwanted image channel
`is much stronger than the
`desired one, then after imperfect. signal cancellation, it may
`only be suppressed to a comparable level
`to the desired
`channel, resulting in an intolerably large interference. Nev-
`ertheless, this type of downconverter is used, for instance, in
`a single-chip broadcast FM receiver with a low IP of 150 kHz
`[6].
`
` UnwantedSuppression,dBImage
`
`
`
`
`
`on°
`
`10
`Phase Error in Quadrature, deg
`
`(b)
`
`(a) The image-canceling downconversion mixer. The desired signal
`Fig. 2.
`appears in-phase in the two branches and the undesired signal anti-phase.
`Allpassfilters may be used to synthesize 90° phase shift. (b) Unwanted signal
`suppression as a function of errors in phase from ideal quadrature, with gain |
`mismatch in the two. branches as a parameter:
`
`IL, THE DIRECT-CONVERSION. ARCHITECTURE
`Suppose that the IF in a superheterodyne. is reduced to |
`zero. The LO will
`then translate the center of the desired
`channel to 0 Hz, and the portion of the channel translated
`to the negative frequency half-axis becomes the image to the
`other half of the same channeltranslated to the positive fre-
`quency half-axis (Fig. 3). The downconverted signal must be
`reconstituted by a phasing method of the type described above,
`otherwise the negative-frequency half-channel will fold over
`and superpose on to the positive-frequency half-channel. Zero-
`IF, therefore, mandates quadrature downconversion into two
`arms and a vector-detection scheme. However,
`this scheme
`does not suffer from the strong-image problem when the
`image-reject downconverteris used in a nonzero IF heterodyne
`receiver, and the typical gain mismatches and phase errors in
`the two branches cause only asmall loss in detected SNR.
`A lowpass filter, which -is in effect a bandpass centered at
`de when the negative frequency axis is included, may be
`used to select the desired channel and toreject all adjacent
`channels. Therefore, RF preselection may in principle be
`
`|
`
`2
`
`
`
`ABIDI: DIRECT-CONVERSION RADIO TRANSCEIVERS
`
`1401.
`
` ra
`
`a
`va gs
`aege
`
`
`
`Limiting
`Lowpass
`Channel
`Tone
`Filters
`
`
`
`Detector
`
`Fig. 3.
`
`Spectrum before and after direct-conversion.
`
`eliminated because there is no image channel. In practice, it
`is still required to suppress strong out-of-band signals that
`may create large intermodulation distortion in the front-end
`prior to baseband channel selection and to avoid harmonic
`downconversion. There is also the advantage that if a high-
`orderactivefilter is used for channelselection,it will dissipate
`lower power and occupy a smallerchip area at a given dynamic
`range than an active bandpassfilter with the sameselectivity
`centered at a high IF [4]. All amplification past the front-end
`is also at baseband, and therefore consumes a small power.
`This zero-IF scheme is also called direct-conversion. When
`the local oscillator is synchronized in phase with the incoming
`carrier frequency, the receiver is called a homodyne.
`As early as 1924, radio pioneers had considered use of
`homodynearchitectures for single vacuum-tube receivers, but
`it was a homodyne measuring instrument for carrier-based
`telephony built
`in 1947 that
`first employed a high-order
`lowpassfilter for channel-selection [7]. Thereafter, the concept
`lay dormant, until it was revived in 1980 in the radio-paging
`receiver, the first miniature digital wireless device for personal
`communication to attain widespread consumeruse.
`
`TI. DIRECT-CONVERSION FSK RECEIVERS
`
`Digital data in broadcast paging modulates the carrier by
`frequency-shift keying (FSK). The carrier frequencies may lie
`in the 400 MHz or the 900 MHz bands and binary data at 512
`b/s or 1.2 kb/s rates shifts the carrier frequency by £4.5 kHz.
`This large modulation index results in a spectrum with two
`lobes symmetrically offset around the carrier (Fig. 3). Vance
`at ITT Standard Telecommunications Labs was the first to
`apply direct-conversion to this signal spectrum with a single-
`chip paging receiver [8], thus establishing a key concept for
`small, light paging receivers. Not all pagers today, though, use
`direct-conversion; some continue to use the superheterodyne
`implementation for higher performance [9].
`Following a single-stage of RF amplification in this simple
`FSK receiver (Fig. 4), a local oscillator (LO) tuned to the
`incoming carrier downconverts the center of the desired paging
`channel to de. In fact, quadrature phases of the LO downcon-
`vert the signal into two branches, labeled J and @, enabling
`the detector to discriminate the signal at positive and negative
`frequencies(i.e., data 1’s and 0’s). A high-order lowpass filter
`in each branch with a cut-off at about 10 kHz selects the
`
`Fig. 4. Block diagram of a direct-conversion FSK receiver as may be used
`for radiopaging signals.
`:
`
`desired channel, while all other channels fall into the filter
`stopband. This may be integrated as a low-poweractivefilter.
`A single-chip paging receiver from Philips uses a tenth-order
`continuous-time lowpassfilter [10]. The data is encoded in the
`zero-crossings of the downconverted andfiltered signal, which
`a limiter then amplifies to logic levels, eliminating the need for
`AGC. However, de offsets directly add to the downconverted
`signal and may beso large as to disable zero-crossings in the
`limiter output, causing the receiver to fail to detect data. This
`problem is overcome by capacitively coupling the baseband
`signal path into the limiter to null these offsets. Some of the
`consequences of.capacitive coupling are discussed in Section
`VI.
`The detector is, in principle, only a flip-flop, driven at the
`D input by the J branch limiter and at the CK input by the Q
`branch limiter. The flip-flop output attains one steady-state if
`transitions at the CK input lead the transitions at the D-input
`and the other state if they lag. This corresponds exactly to a
`positive or negative frequency shift of the carrier, and thus,
`the data. Although this simple detector is found in many FSK
`receivers, it is susceptible to upset from a single noise impulse
`at the input. A more sophisticated detector oversamples the
`limiter output at a multiple of the data rate, thereby morefinely
`quantizing the zero-crossing instant, and correlates this with
`quadrature phases of the expected frequency shift (4.5 kHz in
`the paging channel). The correlated output from the J and Q
`channels is integrated over a bit period, and the bit decision
`is made depending on which of the integrators first crosses a
`preset threshold. Correlation reduces the noise bandwidth. A
`one-bit implementation of this correlation detector in a spread-
`spectrum FSK receiver shows that it
`is very compact and
`dissipates a small power [11].
`There are now many low-power, single-chip bipolar IC’s
`implementing direct-conversion paging receivers [10], some
`operating at a supply as low as 1.1 V [12]. The on-chip
`capacitors required by the two active lowpass filters and ac
`coupling after downconversion occupy a large portion of the
`total die area. Aside from a reference crystal,
`the circuits
`only need some miniature off-chip inductors for the tuned RF
`amplifier loads and sometimes for the quadrature phase-shift
`network. The complete pager, including the microprocessor
`
`3
`
`
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 30, NO. 12, DECEMBER 1995
`
`Asthe LO frequency in a direct-upconverter is centered in
`the transmit band, energy at this frequency may be spuriously
`radiated through parasitic unbalanced coupling into the power
`amplifier or antenna. For instance, the single-ended signal
`produced by an on-chip oscillator circuit tuned with an off-chip
`resonator may couple to the power amplifier input across pins
`of the RF package. Frequency-offset multi-step upconversion
`schemes, which are the dual of a heterodyne downconverter,
`have been proposed [14] to combat this: coupling problem.
`However, as LO phase-noise in a transmitter appears as noise
`added to the emitted signal, a process called reciprocal mixing
`in the radio literature, direct upconversion has the advantage
`over a frequency-offset scheme that only. one LO contributes
`noise. Other spurious output tones in a single-sideband trans-
`mitter may arise from parasitic remixing of the. modulated
`output with the baseband signal and by intermodulation dis-
`tortion in the output stage [15]. Balancedcircuit topologies,
`- on-chip LO’s that require no external resonators [1], [16], and
`the lowered transmit power levels required in microcells, are
`all expected to lessen the magnitude of these problems.
`
`V. DIRECT-CONVERSION RECEIVERS
`FOR DIGITAL CELLULAR TELEPHONES
`
`Designers of portable digital cellular telephones are very
`interested in low-power radio architectures. Several integrated
`receiver and transmitter IC’s.conformingto established stan-
`dards such as GSM and DECT have been developed in the
`past few years. This section summarizes some of their main
`features.
`.
`os
`All transmitters in these portable phones use direct’ up-
`conversion to produce a single-sideband output. In receivers,
`however, the superheterodyne architecture is more common.
`For instance, a 900 MHz bipolar IC GSM receiver from
`Siemens [17] downconverts the amplified RF signal from an
`off-chip low-noise amplifier to an IF’of 45-90’ MHz (Fig. 6).
`At this IF, the image lies in the stopband of the fixed RF
`preselect SAW filter. The amplified IF signal is sent to another
`off-chip SAW filter to reject adjacent channels. A quadrature
`mixer then downconverts the signal
`to: baseband, and the
`vector baseband signal is finally detected. This architecture
`is preserved in later generations of this transceiver operating
`up to 2 GHz for DECT use [18], [19]. Other recent GSM
`transceivers build on a similar single superheterodyne archi-
`tecture [20], in one case with a very high IF of 400 MHz
`{21]. Alcatel has publicized its use of direct-conversion in
`GSM and DECT receivers [22]~[24], although others [25]
`are exploring its possible use, and not alli companies using
`direct conversion have published their. experience. Alcatel’s
`RF front-end is a relatively small silicon bipolar IC (Fig. 7),
`and the remainder of the signal processing, including lowpass
`channel-select filtering, takes place at baseband in a mixed —
`analog-digital CMOS IC [26].
`Given the many decades of familiarity with the superhetero-
`dyne, there will likely remain some reluctance towards adopt-
`ing a new architecture until there is widespread experience
`in its effectiveness. However, direct-convetsion also suffers
`from some unique problems to which the superheterodyne is
`immune. These are discussed in the following section.
`
`image
`
`a a=
`
`1402
`
`c — Lo
`
`a1SEg
`
`to =
`3a
`
`Fig. 5. A direct-upconversion mixer using the phasing methodfor selecting
`one sideband. Either one of the sidebands may be selected by combining the
`branches with the appropriate sign.
`
`and display driver, may be a two-chip device with battery life
`in excess of six months.
`One may appreciate the simplicity of a direct-conversion
`state-of-the-art paging receiver by comparing its inventory of
`parts with a superheterodyne implementation of the samebuilt
`in a comparable technology [13]. The superheterodyne requires
`one more crystal, two trim capacitors, and a SAW filter, which
`together add a significant fraction to the total parts count, thus
`increasing the physical volume of the receiver and its power
`dissipation.
`
`TV. DIRECT-CONVERSION SINGLE-SIDEBAND SYNTHESIZERS
`For reasons of spectral efficiency,
`the transmitted signal
`in digital communications is usually single-sideband with
`suppressed-carrier. It would require an RF filter with a very
`sharp transition band to suppress one sideband on the modu-
`lated carrier whilepassing the other. The much more practical
`solution is the phasing method [5], where the modulated
`signal is first synthesized in quadrature at baseband, directly-
`upconverted into two branches by a quadrature LO centered at
`the carrier frequency, and added or subtracted to select either
`the upper or lower sideband (Fig. 5). The phasing method of
`sideband selection has been used for many years in single-
`sideband communication transceivers.
`-,
`The unwanted sideband is suppressed to an extent limited
`by the gain mismatch in the two upconversion branches and by
`departures from quadrature in the two LO outputs (Fig. 2(b)).
`The dc offsets in the branches produce an output tone at the
`LO frequency. The unwanted sideband and LO leakage are
`unavoidable spurious emissions in the transmitted spectrum.
`Although the two upconversion branches will match well
`on the same IC, a gain mismatch as small as 1% (0.1 dB)
`limits unwanted sideband suppression to about 45 dB. With
`this: gain mismatch, a phase-error of up to 1° is tolerable
`between the two LO outputs before the unwanted sideband
`growsfurther in relative amplitude. These mismatches may be
`trimmed at time of transceiver manufacture or self-calibrated
`with loopback modes that are activated during idle times to
`sense and suppress the unwanted sideband. Some trimming
`and adaptive methods are discussed in Section VIL.
`
`4
`
`
`
`ABIDJ: DIRECT-CONVERSION RADIO TRANSCEIVERS
`
`1403
`
` Bi shagersske ovetaasiaes sesteeapeansroe
`
`Lo/Signal
`
` e 4
`
`Sa” RE Spectrum
`
`Fig. 6. A superheterodyne receiver for a digital cellular telephone.
`
`be
`
`Baseband Spectrum
`(a)
`
`t
` gesitraaes:seu eatinidarineiOuE:ogise IOUTEEEEERS,
`
`7
`z
`:
`z
`Sueeenert oben ont marie big seutsi aodpao maui uominedines %
`
`
`
`
`Pole at 300 kHz
`[kicosbi)]
`
`
`LO [k;cosfG)|
` 650 pF
`To CMOS
`baseband IC
`0 dBm out
`
`
`
`
`Nd
`
`Fig. 7. A direct-conversion receiver and transmitter for a digital cellular
`telephone.
`
`VI. PROBLEMS IN DIRECT-CONVERSION RECEIVERS
`
`LO/Signal
`
`LO/Signal
`3rd Harmonic
`
`
`
`
`e®" RF Spectrum
`
`oc
`
`Downconverted Spectrum
`(b)
`
`LO/Signal
`are
`
`LO/Signal
`3rd Harmonic
`
`
`
`Among the problems in direct-conversion receivers, spuri-
`ous LO leakage is probably best known. This arises because
`the LO in a direct-conversion receiver is tuned exactly to the
`center of the LNA and antenna passbands. In receive mode,
`a small fraction of the LO energy may make its way back
`to the antenna through the mixer and LNA, owing to their
`finite reverse isolation, or couple into the antenna through
`IF 3*IF_=Downconverted Spectrum
`external
`leads, and then radiate out [27]. This becomes an
`(c)
`in-band interferer to other nearby receivers tuned to the same
`band, and for some of them it may even be stronger than
`the desired signal. Regulatory bodies such as the FCCstrictly
`limit the magnitude of this type of spurious LO emission. The
`problem is much less severe in a superheterodyne whose LO
`frequency usually lies outside the antenna passband. However,
`experimental studies [28] suggest that standard shielding in the
`receiver may control LO leakage to the pointthat it does not
`seriously handicap the use of direct-conversion.
`Distortion produced by strong signals in the downconver-
`sion mixer will cause the sensitivity of a direct-conversion
`receiver to degrade more rapidly than of the superheterodyne.
`Second-order distortion in a single-ended mixer will rectify the
`envelope of an amplitude-modulated RF input such as QPSK
`data to produce spurious baseband spectral energy centered at
`dc, which then adds to the desired downconverted signal [25],
`[27] (Fig. 8(a)). This is particularly serious if the envelope is
`that of a large unwanted signal lying in the preselect filter
`passband, which has not yet been rejected by the baseband
`channel-select filter. The most effective solution is to use
`
`
`
`a
`in
`nonlinearity
`by
`caused
`downconversion
`Spurious
`Fig. 8.
`direct-conversion receiver.
`(a) Second-order distortion detects envelope
`of near-band interferer at baseband, overlaying desired signal.
`(b) LO
`harmonics downconvert
`signal
`harmonics
`to
`baseband,
`resulting
`in
`interference,
`(c) whereas in a superheterodyne downconverted harmonic
`products lie in IF stopband.
`
`balanced circuits in the RF front-end, particularly the mixer,
`which will only create odd-order distortion.!
`However, even in a balanced circuit, the third harmonic of
`the desired signal may downconvert the third LO overtone
`to create spurious dc energy competing with the fundamental
`downconverted signal (Fig. 8(b)), whereas in a superhetero-
`dyne this downconverted component lies in the stopband of
`the IF filter (Pig. 8(c)). This is a small effect to the extent
`
`There is no fundamentalloss of noise figure in a balanced front-end. When
`the antenna signal drives a balanced low-noise amplifier through a power-
`splitting balun, the noise figure is exactly the same as directly driving the
`signal into a single-ended half circuit. However, the balancedcircuit drains
`twice the current of the single-ended half circuit.
`
`5
`
`
`
`1404
`
`IEEE JOURNAL OF SOLID-STATE CIRCUITS, VOL. 30, NO. 12, DECEMBER 1995
`
` LO self-mixing
`
`Some sources of dynamic offset in a direct-conversion receiver. The
`Fig. 9.
`LO may leak into the antenna,reflect off external objects, and downconvert
`to de, A strong interferer may leak into the LO port and downconvert itself
`to de, These de offsets vary with physical location.
`
`that both are third-order terms.. To suppress this harmonic
`downconversion, mixers with embedded overtone-rejection
`bandpass filters have been proposed [29]. If exactly the same
`immunity to spurious downconversion is sought as in a super-
`heterodyne, the direct-conversion receiver must use a balanced
`mixer with inherently greater linearity. Again, this is not an
`insurmountable obstacle; mixer linearity and related design
`issues are discussed in Section VII.
`Perhaps the most serious problem is that of de offset in
`the baseband section of the receiver following the mixer. This
`offset appears in the middle of the downconverted signal spec-
`trum, and may be larger than the signal itself, and much larger
`than thermal and flicker noise. For instance, a downconverted
`signal of a few hundreds of microvolts rms may compete with
`an offset of a few millivolts. Unless the offset is removed,
`the SNR at the detector input will be very low. Offset arises,
`first, from transistor mismatch in the signal path between the
`mixer and the J and @ inputs to the detector. To this will
`add other offsets peculiar to the wireless environment [24],
`[30]. As described above,
`the LO signal
`leaking from the
`antenna during receive mode mayreflect off an external object
`and self-downconvert to dc in the mixer (Fig. 9). Similarly, a
`large undesired’ near-channel interferer in the preselectfilter
`passband may leak into the LO port of the mixer and self-
`downconvert to de. These offsets are insidious because their
`magnitude changes with receiver location and orientation, and
`in a frequency-hopping receiver, with the instantaneous LO
`frequency. It is unlikely whether through better circuit design
`alone,
`the reverse isolation of mixers and LNA’s can be
`improved to the point that these offsets become on the order
`of thermal noise. Appropriate circuits must be built into the
`receiver to remove them.
`In the direct-conversion paging receiver, de offset is re-
`moved by capacitively coupling the baseband circuits up to,
`and including,
`the limiter. Owing to the high impedance
`levels of active filters,
`the coupling capacitors are small
`enough to be integrated on to the receiver IC [10],
`[31].
`However,
`this simple solution only works because of the
`specific spectral features of wideband-FSK modulation [10].
`'The paging spectrum peaks at +4.5 kHz and dips at dc by
`at least 25 dB relative to its level at the peaks (Fig. 10(a)).
`Simulations show that a first-order capacitive coupling, which
`produces a lower cutoff at 1 kHz in the channel-selectfilter
`preceding the limiting amplifier (Fig. 10(b)), causes no loss
`
`0
`
`&
`
`TTTTR" eae
`$s3 v=|”|a a bi
`
`© Frequency, itz
`(a)
`
`10 kHz
`
`| 10th-order
`
`1 kHz
`|
`
`
`
`Frequency
`
`(b)
`
`with Filter
`
`al KHZ, 10 kHz).
`
`(a) Downconverted spectrum of a paging channel carrying 500 b/s
`Fig. 10.
`data with 4.5 kHz carrier frequency keying. (b) A typical channel-selectfilter
`in the receiver, with a 1 kHz lower cutoff frequency to null de offsets.
`Channels are spaced apart by 25 kHz. (c) Receiver sensitivity simulation,
`comparing an ideal offset-free receiver containing 10 kHz brickwall filter,
`with receiver suffering from offset, but using the filter in (b).
`
`in receiver sensitivity (Fig. 10(c)) when the channel filter
`captures the valuable signal spectrum between 1-10 kHz.
`These are precisely the characteristics of the channel-select
`filter in a Philips single-chip paging receiver [10].
`A small frequency-error between the transmitter and re-
`‘ceiver LO’s will cause the signal to downconvert asymmet-
`rically around dc, and the capacitively-coupled filter may
`now place its notch in one ofthe two signal-bearing spectral
`lobes. For instance, a 5 ppm relative frequency-error at 900°
`MHz places the capacitive notch 4.5 kHz away from the
`center of the spectrum in the middle of:one of the lobes.
`This limits the acceptable frequency tolerance on ‘the crystal
`
`6
`
`
`
`ABIDI: DIRECT-CONVERSION RADIO TRANSCEIVERS
`
`1405
`
`"
`
`©-600-400-300-200-00 0 100200 ato400500
`
`Frequency, kHz
`(a)
`
` Spectral
`Density,dB
`BER et,|
`
`ToTPSS
`
`- = Ideal
`=h-= 1 Hz Notch
`—=@—5 Hz Notch
`—— 20 Hz Notch
`
`2
`
`4
`
`6
`SNR, dB
`
`(b)
`
`8
`
`10
`
`(a) Typical direct downconverted spectrum of an efficiently modu-
`Fig. 11.
`lated carrier, here a cosine-filtered BPSK.(b) The impactof a notch of varying
`widths at dc on receiver performance. At speech-quality BER of 1073, the
`receiver ceases to function if the notch width is 20 Hz.
`
`oscillator regulating the LO. In the absence of a sufficiently
`stable frequency reference, some form of automatic frequency-
`control must be used. The LO frequency error may be deduced
`from the long-term average of the frequency shifts keyed
`by pseudorandom data, which should ideally be zero after
`downconversion. This may be used as a correction signal on a
`varactor diode, or equivalent means of fine-tuning the LO. In
`a mostly digital implementation of a frequency-hopped FSK
`receiver, a numerically-controlled oscillator with resolution of
`a few hertz compensates frequencyerrors in the receiver clock
`as part of a frequency-acquisition loop [32].
`Wideband FSK is spectrally inefficient, but it is used in
`paging because of the very low data rates and small duty cycle
`of use. However, digital cellular systerms must fit continuous
`voiceband traffic at 200-1100 kb/s into 50-100 kHz wide
`channels, and to do so they use other spectrally-efficient
`modulation schemes. The GSM system, for imstance, em-
`ploys Gaussian-Filtered Minimum-Shift Keying (GMSK), a
`sort of optimized FSK. Other systems may use single-ended
`or differential Quadrature Phase-Shift Keying (QPSK). The
`spectra of all these modulation schemes peak at de (Fig. 11(a)).
`
` ewel Succ Approx Reg
`
`Fig. 12. Digital offset removal in a direct-conversion receiver. Offset is
`estimated from long-term average of digitized signal and subtracted from
`analog signal.
`
`Following downconversion of the received signal to zero-IF,
`offsets will now directly add to the peak of the spectrum.
`It
`is no longer practical
`to null
`the offsets by capacitive
`coupling of the basebandsignal path because signal energy will
`almost surely be lost from the spectral peak. Simulations on
`a representative 200 kHz wide spectrum (Fig. 11(b)) suggest
`that at a target bit-error rate of 1073, a 5 Hz notch at de
`causes about 0.2 dB loss in receiver sensitivity, yet this notch
`need only widen to 20 Hz when the receiver will cease to
`function. It would require impractically large capacitors to
`produce this narrow notch, and the phase-distortion due to the
`CR coupling, which these simulations do not model, would
`cause the receiver performanceto further deteriorate. Further-
`more, during burst-mode communications, the capacitor would
`produce intolerably long transients.
`DC offsets may be estimated and removeddigitally [33], an
`approach Alcatel takes in their GSM receiver [26] (Fig. 12).
`The baseband signal
`is digitized and averaged in an 8-b
`DSP over a programmable time window. In addition, a 10-b
`successive-approximation A/D converter measures the analog
`signal polarity. These measurements are weighted together in
`a DAC, which subtracts off an estimate of the offset from
`the baseband signal. The spectrum loss around dc is only .
`a few hertz, and digital filtering does not distort the group
`delay. The bandwidth of the feedback loop must be wide
`enough to accommodate the dynamic offsets described above.
`The settling time of the offset subtraction circuit may still
`cause loss of the first few symbols in a TDMA [30] or a
`frequency-hopping CDMA receiver.
`Finally, when offsets have been satisfactorily nulled, flicker
`noise at the mixer output addsto the baseband signal. The SNR
`is lower than in a high-IF superheterodyne where only thermal
`noise is present. A bipolar transistor front-end may be superior
`in this respect to an FET circuit, but it is also possible to
`use autozero or double-correlated sampling to suppress flicker
`noise in MOS opamp-based circuits.
`
`VII. COMPONENTS OF DIRECT-CONVERSION TRANSCEIVERS
`
`A. Carrier-Frequency Local Oscillators
`with Quadrature Outputs
`Direct-conversion transmitters and receivers need a local
`oscillator with quadrature outputs for vector modulation and
`demodulation,
`respectively. Whereas in a superheterodyne
`
`7
`
`
`
`1406
`
`receiver, the signal is quadrature-downconverted by a low fre-
`quency second LO, the greater challenge in direct-conversion
`is to produce accurate quadrature phases with good amplitude
`match at
`the much higher carrier-frequency. An error in
`quadrature of Jess than 1°