Nội dung

Adaptive Wireless Tranceivers L. Hanzo, C.H. Wong, M.S. Yee Copyright © 2002 John Wiley & Sons Ltd ISBNs: 0-470-84689-5 (Hardback); 0-470-84776-X (Electronic) L 1 3 1 Adaptive Multicarrier Modulation T. Keller and L. Hanzo’ 13.1 Introduction High data rate communications are limited not only by noise, --butespecially with increas(ISI) due to the ing symbolrates -often more significantly by the Inter Symbol Interference memory of the dispersive wireless communications channel [317]. Explicitly, this channel memory is caused by the dispersive channel impulse response (CIR) due to the differentlength propagation paths between the transmitting and the receiving antennae. This dispersion effect could theoretically be measured by transmitting an infinitely short impulse and “receiving” the CIR itself. On this basis, several measures of the effective duration of the impulse response can be calculated, one being the delay spread. The multipath propagation of the channel manifests itself by different echos of possibly different transmitted symbols overlapping at the receiver, which leads to error rate degradation. This effect occurs not only in wireless communications, but also over all types of electrical and optical wave-guides, although for these media the relative time differences are comparatively small, mostlydue to multi-mode transmission or incorrect electrical or optical termination at interfaces. In wireless communications systems the duration and the shape of the CIR depend heavily on the propagation environment of the communications system in question. While indoor wireless networks typically exhibit only short relative delays, outdoor networks, like the Global System of Mobile communications (GSM)[ 131 can face delay spreadsin the order of 15ps. As a general rule, the effects of IS1 on the transmission error statistics are negligible as long as the delay spreadis significantly shorter than the duration of one transmitted symbol. This implies that the symbol rate of communications systems is practically limited by the 535 536 CHAPTER 13. MULTICARRIER ADAPTIVE MODULATION channel’s memory. For higher symbol rates, there is typically significant deterioration of the system’s error rate performance. If symbol rates exceeding this limit are to be transmitted over the channel, mechanisms must be implemented in order to combat the effects of ISI. Channel equalization techniques [317] can be used to suppress the echoes caused by the channel. In order to perform this operation, the CIR must be estimated. Significant research efforts were invested into the development of such channelequalisers, and most wireless systems in operation use equalisers to combat ISI. There is, however, an alternative approach towards transmitting data over a multipath channel. Instead of attempting to cancel the effects of the channel’s echos, Orthogonal Frequency Division Multiplexing(OFDM) [317] modems employ a set of subcarriers in order to transmit information symbols in parallel - in so-called subchannels - over the channel. Since the system’s data throughputis the sum of all the parallel channels’ throughputs, the data rate per subchannel is only a fraction of the data rate of a conventional single-carrier system having the same throughput. This allows us to design a system supporting high data rates,while maintaining symbol durations much longer than the channel’s memory, thus circumventing the need for channel equalization. The outline of the chapter is as follows. Section 2 commences with a historical perspective on OFDM, highlighting the associated research issues with reference to the literature. Based on the above overview of the state-of-the-art, Section 3 characterizes the performance of OFDM over dispersive, wideband channels, while Section 4 quantifies the effects of synchronization errors on OFDM, leading onto Section 5, which highlights the rangeof synchronization solutions proposed by the research community at large. Again, commencing with a literature survey, the key topic of adaptive bit allocation over highly frequency-selective wireless channels is the subject of Section 6, while Section 7 is dedicated to the closely related subject of pre+qualization and channel coding. Our discourse is concluded in Section 8 with a wide-ranging throughput comparison of the schemes discussed in the chapter under the unified constraint of a fixed target bit error rate of l V 4 . 13.2 OrthogonalFrequencyDivisionMultiplexing 13.2.1 HistoricalPerspective Frequency Division Multiplexing (FDM) or multi-tone systems have been employed in military applications since the 196Os, for example by Bello [426], Zimmerman [427], Powers and Zimmerman [428], and others. Orthogonal Frequency Division Multiplexing (OFDM), which employs multiple carriers overlapping in the frequency domain, was pioneered by Chang [429,430].Saltzberg [43 l ] studied amultikarrier system employing orthogonaltimestaggered quadrature amplitude modulation (0-QAM) on the carriers. The use of the discrete Fourier transform (DFT) toreplace the banksof sinusoidal generators and the demodulators - suggested by Weinstein and Ebert [432] in 1971 - significantly reduces the implementation complexity of OFDM modems. This substantial implementational complexity reduction was attributable to the simple realization that the DFT uses a set of harmonically related sinusoidal and cosinusoidal basis functions, whose frequency is an integer multiple of the lowest non-zero frequency of the set, which is referred to as the basis frequency. These harmonically related frequencies can hence be used as the set of carriers 13.2. ORTHOGONAL FREOUENCY DIVISION MULTIPLEXING 537 required by the OFDM system. For a formal proof of this the interested reader is referred to [317]. In 1980, Hirosaki [433] suggested an equalization algorithm in order to suppress both inter-symbol and inter-subcarrier interference caused by the CIR or timing- and frequencyerrors. Simplified OFDM modemimplementationswerestudied by Peled [434] in 1980, while Hirosaki [435] introduced the Dm-based implementation of Saltzberg’s 0-QAM OFDM system. Kolb [436], SchiiBler [437], Preuss [438] and Riickriem [439] conducted further research into the application of OFDM. Kalet [62] introduced the concept of allocating more bits to subcarriers, which were for example near the centre of the transmission frequencyband and hence were less attenuated than those near the edge of the transmission band. However, since Kalet’s discussions were cast in the context of slowly varying channels, the concept of near-instantaneously adaptive transmission was not introduced at this early stage of OFDM research. This concept was often referred to as ’water-filling’ in the frequency domain. A few years later Cimini [440] provided early seminal results on the performance of OFDM modems in mobile communications channels. More recent advances in OFDM transmission are presentedin the impressive state-of-theart collection of works edited by Faze1 and Fettweis [441], including research by Fettweis et al., Rohling et al., Vandendorp, Huber et al., Lindner et al., Kammeyer et al., Meyr et al. [442,443], but the impressive individual contributions are too numerous to mention. While OFDM transmissions over mobile communications channels canalleviate the problem of multi-path propagation, recentresearch efforts have focussed on solvinga set of inherent difficulties regarding OFDM, namely on reducing the associated peak-to-mean-power ratio fluctuation, on time- and frequency synchronization and on mitigating the effects of co-channel interference sensitivity in multi-user environments. These issues are addressed below in more depth. 13.2.1.1 Peak-to-MeanPowerRatio It is plausible that the OFDM signal - which is the superposition of a high number of modulated subchannel signals - may exhibit a high instantaneous signal peak with respect to the average signal level. Furthermore, large signal amplitude swings are encountered, when the time-domain signal traverses from a low instantaneous power waveform to a high-power waveform. Similarly, the peak-to-mean power envelopefluctuates dramatically, when traversing the origin upon switching from one phasor to another. Both of these events may results in a high out-of-band (OOB) harmonic distortion power, unless the transmitter’s power amplifier exhibits an extremely highlinearity [3 171 across the entire signal dynamic range. This potentially contaminates the adjacent channels with adjacent channel interference. Practical amplifiers exhibit a finite amplitude range, in which they can be considered near-linear. In order to prevent severe clipping of the high OFDM signal peaks - which is the main source of OOB emissions - the power amplifier must not be driven into saturation and hence they are typically operated with a certain so-called backoff, creating a ’head-room’ for the signal peaks, which reduces the risk of amplifier saturation and OOB emmission. Two different families of solutions have been suggested in the literature, in order to mitigate these problems, either reducing the peak-to-mean power ratio, or improving the amplification stage of the transmitter. More explicitly, Shepherd [444], Jones [445], and Wulich [446] suggested different cod- 538 CHAPTER 13. MULTICARRIER ADAPTIVE MODULATION ing techniques which aim to minimise the peak power of the OFDM signal. According to their approach different data encoding or mapping schemes are employed before modulation. A simple example is concatenating a number of dummy bitsto a stringof information bits with the sole aimof mitigating the so-called Crest Factor (CF) or peak-to-mean signal envelope ratio. In a further attempt to mitigate the CF problem Muller [447], Pauli(4481, May [449] and Wulich [450] suggested different algorithms for post-processing the time-domain OFDM employed adaptive subcarsignal prior to amplification, while Schmidt and Kammeyer l] [45 rier allocation in order to reduce the Crest factor. Dinis and Gusmiio [452-454] researched the use of two-branch amplifiers, while the so-called clustered OFDM technique introduced by Daneshrad, Cimini and Carloni [455] operates with a set of parallel partial FFT processors with associated transmitting chains. More explicitly, clusteredOFDM allows a number of users on a demand basis, potentially of users to share a given bandwidth amongst a number supporting a peak data rate identical to thatof a single-user OFDM system. The bandwidth assigned to a particular user is typically constituted by a numberof subcarrier clusters, which are spread sufficiently far apart from each other, in order to provide frequency diversity. OFDM systems with increased robustness to nonlinear distortion have been proposed for example by Okada, Nishijima and Komaki [456]as well as by Dinis and Gusmiio [457]. 13.2.1.2 Synchronization Time and frequency synchronization between the transmitter and receiver areof crucial importance in terms of the performance of an OFDM link [458462]. A wide variety of techniques has been proposed for estimating and correcting both timing and carrier-frequency offsets at the OFDM receiver. Rough timing and frequency acquisition algorithms relying on known pilot symbols or pilot tones embedded into the OFDM symbols have been suggested by Claljen [442], Warner [463], Sari [464], Moose [465], as well as Briininghaus and Rohling [466]. Fine frequency and timing tracking algorithms exploiting the OFDM signal’s cyclic extension were publishedby Moose [465], Daffara [467] and Sandell [468]. 13.2.1.3 OFDM / CDMA Combining OFDM transmissions with Code Division Multiple Access (CDMA) allows us to exploit the wideband channel’s inherent frequency diversity by spreading each symbol across multiple subcarriers. This technique has been pioneeredby Yee, Linnartz and Fettweis [206], by Chouly, Brajal and Jourdan [469], as well asby Fettweis, Bahai and Anvari [470]. Faze1 and Papke C2071 investigated convolutional codingin conjunction with OFDMKDMA. Prasad and Hara [471] compared various methods of combining the two techniques, identifying three different structures, namely multi-carrier CDMA (MC-CDMA), multikcarrier direct-sequence CDMA (MC-DS-CDMA) and multi-tone CDMA (MT-CDMA). Like nonspread OFDM transmission,OFDMKDMA methods suffer from high peak-t+mean power ratios, which are dependenton the frequency-domain spreading scheme,as has been investigated by Choi, Kuan and Hanzo [220]. 13.2.1.4AdaptiveAntennas Combining adaptive antenna techniques with OFDM transmissionswas shown to be advantain cellular communications systems. Li, Cimini geous in suppressing co-channel interference 13.2. ORTHOGONAL FREQUENCY DIVISION MULTIPLEXING 539 and Sollenberger [472475], Kim, Choi and Cho [476] as well as Munster et al. [477] have investigated algorithms for multi-user channel estimation and interference suppression. The employment of adaptive antennas is always beneficial in terms of mitigating the effects of multi-user interference, since with the aid of beam-steering it becomes possible to focus the receiver’s antenna beam on the served user, while attenuating the co-channel interferers. This isof particularly high importancein conjunction with OFDM, which exhibits a high sensitivity against co-channel interference, potentially hampering its application in co-channel interference limitedmulti-user scenarios. 13.2.1.5 OFDM Applications Due to their implementational complexity, OFDM applications have been scarce until quite recently. Recently, however, OFDM has been adopted as the new European digital audio broadcasting (DAB) standard [478-482] as well as for Terrestrial DigitalVideo Broadcasting (DVB-T) system [464,483]. The hostile propagation environment of the terrestrial system requires concatenated Reed-Solomon[ 131 (RS) and rate compatible punctured convolutional of delivering highcoding [ 131 (RCPCC) combined with OFDM. These schemes are capable definition video at bitrates of up to 20 Mbits/s in slowly time-varying broadcast-mode distributive wireless scenarios. Recently a rangeof DVB system performance studies were also published in the literature [484-487], portraying the DVB-T system. For fixed-wire applications, OFDM is employedin the Asynchronous Digital Subscriber Line (ADSL)and High bit-rate Digital Subscriber Line (HDSL) systems [488-491] and it has also been suggested for power-line communications systems [492,493] due to its resilience to time-dispersive channelsand narrow-band interferers. 4th Framework AdMore recently, OFDM applications were studied within the European vanced Communications Technologies and Services (ACTS) programme [494]. Specifically, the Pan-European Median project investigated a 155 Mbit/s (Mbps) Wireless Asynchronous Transfer Mode(WATM) network [495498], while theMagic WAND group [499,500] developed a wireless Local Area Network (LAN). Hallmann and Rohling [501] presented a range of different OFDM-based systems that were applicableto the European Telecommunication Standardization Institute’s (ETSI) third-generation air interface [502]. Lastly, the recently standardized High PERformance Local Area Network standard known as HIPERLAN/2was designed for providing convenient wireless networking in indoor environments and also invoked OFDM. The wireless provision of high bit rate services appears a more attractive alternative than installing wireline based networks. The HIPERLAN standard specifies the air interface and the physical layer, in order to ensure the compatibility of different manufacturers’ equipment, while refraining from standardising the higher layer functions of the system. The HIPERLAN standard constitutes a member of the Broadband Radio Access Networks family often referred to as BRAN [503]- [504]. The BRANoffamily recommendations is constituted by the HIPERLANA and /2 systems operatingin the 5GHz frequency band. Further members of the family include the so-called HIPERACCESS standard contrived for fixed wireless broadband Point-to-multipoint access and the HIPERLINK recommendation designed for wireless broadband communicationsin the 17 GHz frequency band. The system’s parameters are summarisedin Table 13.1. CHAPTER 13. ADAPTIVE MULTICARRIER MODULATION 540 OFDM symbol Table 13.1: HIPERLANQ physical layer parameters [505]. i l 4 0' rl 4 'N-l 'N-l p 1 i 1 rem C.Ext. Figure 13.1: Schematic of N-subcarrier OFDM transmission system. 13.2.2 OFDM Modem Structure The principle of any Frequency Division Multiplexing (FDM) system is to split the information to be transmitted into N parallel streams, each of which modulates a carrier using an arbitrary modulation technique. The frequency spacing between adjacent carriers is A f , resulting in a total signal bandwidthof N . A f . The resulting N modulated and multiplexed N parallel receiver branches resignals are transmitted over the channel, and at the receiver cover the information. A multiplexer then recombines the N parallel information streams 13.2. ORTHOGONAL MULTIPLEXING FREQUENCY DIVISION 541 N+N g > 771 Figure 13.2: Stylized plot of N-subcarrier OFDM time domain signal with a cyclic extension of Ng samples. into a high-rate serial stream. The conceptually simplest implementation of an FDM modem is to employ N independent transmittedreceiver pairs, which is often prohibitive in terms of complexity and cost [435]. Weinstein [432] suggested the digital implementation of FDM subcarrier modulators/demodulators based on the Discrete Fourier Transform (DFT). The DFT and its more efficient implementation, the Fast Fourier Transform (FFT) are employed for the base-band OFDM modulation/demodulation process, as it can be seen in the schematic shown in Figure 13.1. The associated harmonically related frequencies can hence be used as the set of subchannel carriers required by the OFDM system. However, instead of carrying out the modulation / demodulation on a subcarrier by subcarrier basis, / as in Hirosaki's early proposal for example [433], all OFDM subchannels are modulated demodulated in a single inverse DFT (IDFT) / DFT step. For more detailed explanations and signal waveforms the interested reader is referred to [317]. The serial datastream is mapped to data symbols with a symbol rateof l/T,, employing a general phase and amplitude modulation scheme, and the resulting symbol stream is demulti. parallel data symbol rate isl/N.'T.q, plexed into a vector of N data symbols SOto S N - ~The i.e. the parallel symbol duration is N times longer than the serial symbol duration T,. Hence the the effects of the dispersive channel - which are imposed on the transmitted signal as convolution of the signal with the CIR - become less damaging, affecting only a fraction of the extended signalling pulse duration. The inverse FFT (IFFT) of the data symbol vector is computed and the coefficients SO to S N - 1 constitute an OFDM symbol, as seen in the figure. Since the harmonically related and modulated individual OFDM subcarriers can be conveniently visualised as the spectrum of the signal to be transmitted, it is the IFFT - rather than the FFT - which is invoked, in order to transform the signal's spectrum to the time-domain for transmission over the channel.The associated modulated signal samplesS, are the timedomain samples of the OFDM symbol and are transmitted sequentially over the channel at a symbol rate of l/Ts. At the receiver, a spectral decomposition of the received time-domain samples T , is computed employing an N-tap FFT, and the recovered data symbols R, are restored in serial order and demultiplexed, as seen in Figure 13.1. The underlying assumption in the context of OFDM upon invoking the IFFT for modulation is that although N frequency-domain samples produce N time-domain samples, both signals are assumedto be periodically repeated over an infinite time-domain and frequencydomain interval, respectively. In practice, however, it is sufficient to repeat the time-domain signal periodically for the duration of the channel's memory, i.e. for a duration that is com- 542 CHAPTER 13. ADAPTIVE MULTICARRIER MODULATION parable to the length of the CIR. This is namely the time interval required for the channel’s transient response to die down after exciting the channel with a time-domain OFDM symbol. Once the channel’s transient response time has elapsed, its output is constituted by its steadystate response constituted by the received time-domain OFDM symbol. In order to ensure that the received time-domain OFDM symbol is demodulated from the channel’s steady-state - rather than from its transient - response, each time-domain OFDM symbol is extended by the scxalled cyclic extension (C. Ext.in Figure 13.1) or guard interval of Ng samples duration, in order to overcome the inter-OFDM symbol interference due to the channel’s memory. The signal samples received during the guard interval are discarded at the receiver and the N-sample receivedtime-domain OFDM symbol is deemed to follow the guard interval of Ng samples duration. The demodulated OFDM symbol is then generated from the remaining N samples upon invoking the IFFT. We note, however that since the transmitted time-domain signal was windowed to the finite duration of N + Ng samples, the corresponding transmitted frequency-domain signal is convolved with the sinc-shaped frequency-domain transfer function of the rectangular time-domainwindow function. As a results of this frequency-domain convolution, the originally pure line-spectrumof the IFlT’s output generates a sinc-shaped subchannel spectrum centred on each OFDM sub-carrier. The samples of the cyclic extension are copied from the end of the time-domain OFDM ( s N - N , - ~ , . . . , S N - ~ ,so,. . . , S N - ~ ) symbol, generating the transmitted time domain signal depicted in Figure 13.2. At the receiver, the samples of the cyclic extension are discarded. in time dispersive environments reduces the efficiency Clearly, the need for a cyclic extension of OFDM transmissionsby a factor ofN / ( N N g ) .Since the durationNg of the necessary cyclic extension depends only on the channel’s memory, OFDM transmissions employing a high number of carriers N are desirable for efficient operation. Typically a guard interval length of not more than 10% of the OFDM symbol’s duration is employed. Again, for further details concerning the operationof OFDM modems please refer to [205,3 17,5061. + 13.2.3 Modulation in the Frequency Domain Modulation of the OFDM subcarriers is analogous to the modulation in conventional serial systems. The modulation schemes of the subcarriers are generally Quadrature Amplitude Modulation (QAM) or Phase Shift Keying (PSK) [317] in conjunction with both coherent and non-coherent detection. Differentially coded Star-QAM (DSQAM) [3 171 can also be employed. If coherently detected modulation schemes are employed, then the reference phase of the OFDM symbolmust be known, which canbe acquired with the aidof pilot tones [ 191 embedded in the spectrum of the OFDM symbol, as will be discussed in Section 13.3. For differential detection the knowledge of the absolute subcarrier phase is not necessary, and differentially coded signalling can be invoked either between neighbouring subcarriers or between the same subcarriersof consecutive OFDM symbols. 13.3. OFDM TRANSMISSION OVER FREQUENCY SELECTIVE CHANNELS 543 13.3 OFDM Transmission over Frequency Selective Channels 13.3.1SystemParameters Based on the above advances in the field of OFDM modems, below we will characterize the expected performanceof OFDM modems using the example of high-rate Wireless AsynchronousTransferMode (WATM) systems[495-497,499,500]. Specifically,thesystem parameters used in characterizing the performance of various OFDM algorithms closely followed the specifications of the Advanced Communications Technologies and Services (ACTS) Median system [495498], which is a proposed wireless extension to fixed-wire ATM-type networks. In the Median system, the OFDM FFT length is 512, and each symbol is padded with a cyclic prefix of length 64. The sampling rate of the Median system is 225 Msamplesh, and the carrier frequency is 60 GHz. The uncoded target data rate of the Median system is 155Mbps. OFDM modems were originally conceived in order to transmit data reliably in timedispersive or frequency-selective channels without the need for a complex time-domain of QAM channel equaliser. In this chapter the techniques employed for the transmission OFDM signals over a time-dispersive channel are discussed and channel estimation methods are investigated [317]. 13.3.2 The ChannelModel The channel model assumed in this chapter is thatof a Finite Impulse Response (FIR)filter with time-varying tap values. Every propagation path i is characterized by a fixed delay ~i and atime-varying amplitude Ai ( t )= ai .gi ( t ) ,which is the productof a complex amplitude ai and a Rayleigh fading process gi ( t ) .The Rayleigh processesgi are independent from each f;. other, but they all exhibit the same normalized Doppler frequency The ensemble of the p propagation paths constitutes the impulse response P h(t,T ) = C Ai(t). S(T i=l P - ~ i =) C a, . g i ( t ) . S(T -~ i ) , (13.1) i=l which is convolved with the transmitted signal. The channel model employedin this chapter is the worst-case operating environment for an indoor wireless ATM network similar to that of the ACTS Median system [495-498]. We assumed a vehicular velocity of about 50 k m h or 13.9 d s , resulting in a normalized Doppler frequency off; = 1.235.lop5. We note here that the normalized Doppler frequency in this chapter was related to the OFDM symbol duration, rather than to the time4omain signal’s sample duration. This relationshipwill be formally defined in Equation 13.5, hence suffice to say here that the normalized Doppler frequency in this sense is typically5 12 times lower, than the conventional normalized Doppler frequency due to having 512 samplesper PFDM symbol. The significance of this will become more clear in the context of adaptive OFDM schemes, where the predictability of the channel’s frequency-domain transfer function between consecutive OFDM symbols depends explicitly on the duration of the symbol. 544 CHAPTER 13. MULTICARRIER ADAPTIVE MODULATION Time Delay [ns] 0 10, 25 50 75 100 125 150 175 200 225 250 275300 __L_? a0 E 0.4 0.2 2 01 0.0 0 10 20 30 40 50 60 70 80 ~..”.,,.,, ~,Path Length Difference [m] 90 a -1 0 512 384 256 128 (a) channel impulse response Subcarrier index n 576 (b) channel frequency response Figure 13.3: WATM channel: (a) impulse response (b) frequency domain channel transfer function H ( n ) experienced by a specific OFDM symbol. The vehicular velocity of 50 km/h constitutes the highest possible speed of for example an indoor fork-truck in a warehouse environment. Again, this worst-case speed was employed in order to provide performance results characterizing the worst possible scenario in the context of adaptive OFDM transceivers, which are sensitive to rapid CIR or transfer function variations. This issue will become more explicit during our further discourse. The impulse response was determined by simple ray-tracing in a warehouse-type environment, and is shown in Figure 13.3(a), where each CIR tap corresponds to a specifically delayed propagation path. We note that this indoor CIR is not particularly dispersive, however, at the 155 Mbps WATM rate, the dispersion corresponds to1 l sample periods, which would require a high-performance channel equaliserin a serial modem. The last CIR path arrives at a delay of 48.9 ns due to the reflection with an excess path length of about 15 m with respect to the line-of-sight path, which again corresponds to l I sample periods. The impulse response exhibits a Root Mean Squared (RMS) delay spread of 1.5276 . lop8 S, and is shown in Figure 13.3(a). The resulting frequency domain transfer function for this WATM impulse response is given in Figure 13.3(b), which exhibits an undulating behaviour across the 5 l 2 subcarriers. This suggests that the high-quality subcarrier may be able to use several bits per subcarrier, while others may have to be disabled. This issue will be further detailed during our later discourse. 13.3.3 Effects of Time-DispersiveChannels The effectsof the time-variant and time-dispersive channels on the data symbols transmitted in an OFDM symbol’s subcarriers are diverse. Firstly, if the impulse response of the channel is longer than the duration of the OFDM guard interval, then energy will spill over between consecutive OFDM symbols, leading tointer-OFDM-symbol interference. We will not elaborate on these effects here, since the length of the guard interval is generally chosen to be longer than the longest anticipated CIR.