Title of Invention	MINIMISATION OF PERTURBATION OF A RADIO SIGNAL BETWEEN A TRANSMITTER AND A RECEIVER BY CONDITIONING EQUALISER INPUT
Abstract	Apparatus (212) for conditioning a communications signal destined for equalisation by an equaliser, the apparatus (212) comprising a pre-filter (222) located before the equaliser, wherein the apparatus further comprises means (223) for modifying the filtering characteristic of the pre-filter by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter (222) due to multipath structure in said signal.

Title of Invention

MINIMISATION OF PERTURBATION OF A RADIO SIGNAL BETWEEN A TRANSMITTER AND A RECEIVER BY CONDITIONING EQUALISER INPUT

Abstract

Apparatus (212) for conditioning a communications signal destined for equalisation by an equaliser, the apparatus (212) comprising a pre-filter (222) located before the equaliser, wherein the apparatus further comprises means (223) for modifying the filtering characteristic of the pre-filter by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter (222) due to multipath structure in said signal.

Full Text	Field of Invention : The invention relates to apparatus for, and to methods of, conditioning communications signals and configuring filters. Background of Invention : In a radio system, a signal travelling between a transmitter and a receiver will be perturbed by the environment in which the transmitter and the receiver exist.' The performance of a receiver, i.e. its ability to accurately recover from a received signal information that was put into the signal by a transmitter, depends, to a significant extent, on the ability of the receiver to remove perturbations caused by the environment. Environmental perturbations of a transmitted signal are usually ascribed to three main sources. These are noise, interference and multipath propagation. Noise may have many sources, including thermal noise from the general environment and circuit noise from within the receiver in question. Interference arises from other signals that are being transmitted in the vicinity of the receiver in question. Multipath propagation arises when a signal reaches a receiver from a transmitter by more than one path. An example of a situation in which multipath propagation occurs is shown in Figure 1. Figure la shows a transmitter 100 attempting to send a radio signal to a receiver 110. Three objects 112, 114 and 116 are located in the vicinity of the transmitter and the receiver 110. The radio signal emitted by the transmitter 100 can reach the receiver 110 by various paths. First, the radio signal can reach the receiver 110 by the direct path 118 leading from the transmitter 100. The radio signal from the transmitter 100 can also reach the receiver 110 by reflection from objections 112 to 116. For example, the radio signal can travel along paths 120, 122 and 124 with reflection from objects 112, 114 and 116, respectively. It will be apparent that the paths 118 to 124 have different lengths with the result that the versions of the radio signal travelling along these paths will arrive at the receiver 110 with differing relative delays. It is common practice to characterise multipath propagation in terms of a graph showing how the power of a received signal is distributed with delay. Figure lb illustrates how such a plot might look for a radio signal travelling from transmitter 100 to receiver 110 in the presence of objects 112 to 116. Figure lb illustrates that the power in the received signal is spread over a range of delays and Figure 1c illustrates a sampled or digitised version of the power versus delay profile of Figure lb. Commonly, the information to be conveyed by a radio signal emitted by a transmitter exists in the form of a series of information symbols. Multipath propagation can cause a phenomenon known as intersymbol interference (ISI). ISI occurs when paths leading to a receiver from a transmitter differ in length in a way which causes one information symbol in the radio signal arriving along one path to arrive at the receiver at the same time as another information symbol in the radio signal arriving at the receiver along another path. ISI can also be introduced by other mechanisms. For example, in an Enhanced Data rates for GSM Evolution (EDGE) system, a filter is used to constrain the bandwidth of a modulated GMSK or 8PSK signal prior to its emission from a transmitter (see, for example, 3GPP TS 45.004 3rd Generation Partnership Project; Technical Specification Group GSM/EDGE; Radio Access Network). Moreover, a signal acquired by a receiver will usually undergo filtering of one form or another (for example, to reject interference in adjacent frequency channels that would otherwise reduce dynamic range). Such filtering mechanisms can be another source of ISI. It is usual to design a receiver to counteract, in the presence of interference and noise, ISI due to, for example, one or more of multipath propagation, filtering in a transmitter and filtering in a receiver and to estimate the information that was put into the signal by the transmitter that originated the signal. See, for example, "Digital Communications", John G. Proakis, McGraw-Hill International Series, 3rd Edition. There are many methods of counteracting ISI, which methods are known generically as "equalisation". Among these the following are worthy of note: 1. Linear equalisation (see, for example, "Digital Communications", John G. Proakis, McGraw-Hill International Series, 3rd Edition); 2. Maximum-Likelihood Sequence Estimation (MLSE) using the Viterbi algorithm (see, for example, ("Maximum-likelihood sequence estimation of digital sequences in the presence of intersymbol interference", Forney, G., Jr.; IEEE Transactions on Information Theory, Volume: 18, Issue: 3, May 1972 Pages: 363-378); 3. Maximum A-posteriori Probability (MAP) (see, for example, "Optimum and sub- optimum detection of coded data disturbed by time-varying intersymbol interference [applicable to digital mobile radio receivers]", Koch, W.; Baier, A.; Global Telecommunications Conference, 1990, and Exhibition. "Communications: Connecting the Future", GLOBECOM '90., IEEE, 2-5 Dec. 1990 Pagra: 1679-1684 vol. 3), and its lower complexity versions, the max-log-MAP; and 4. The soft-output Viterbi Algorithm (SOVA) ("A Viterbi algorithm with soft-decision outputs and its applications", Hagenauer, J,; Hoeher, P,; Global Telecommunications Conference, 1989, and Exhibition. "Communications Technology for the 1990s and Beyond". GLOBECOM '89., IEEE, 27-30 Nov. 1989 Pages: 1680-1686 vol. 3). A number of reduced complexity ISI suppression methods exist, in particular Reduced State Sequence Estimation (RSSE) (see, for example, "Reduced-state sequence estimation with set partitioning and decision feedback", Eyuboglu, M. V.; Qureshi, S.U.H.; IEEE Transactions on Communications, Volume: 36, Issue: 1, Jan. 1988 Pages: 13-20) which can be of use for higher order modulation schemes such as 8PSK. In mathematical terms, it is possible to regard a received signal as arriving at a receiver from a transmitter through a channel consisting of a filter describing the various sources of ISI that affect the received signal. In this document, such a filter shall be referred to as a "channel" and the impulse response of such a filter shall be referred to as a "channel response" and an estimate or measurement of such an impulse response shall be referred to as a "channel estimate". In order for an equaliser to work successfully on a received signal, it is desirable to endow the equaliser with an accurate channel estimate for the signal. This consideration advocates the use of a channel estimate which is at least as long as the true channel. However, the length L of a channel estimate used by an equaliser will directly affect the complexity of the equaliser. In the case of a linear equaliser, the complexity will vary linearly with L. In the case of a SOVA equaliser, the complexity will rise exponentially with L. Thus, the length allowed for a channel estimate in an equalisation scheme is normally limited. If an equaliser uses a channel estimate whose length is shorter than the channel response, then the performance of the equaliser will suffer. In general terms, the greater the proportion of the power of the received signal that lies in the portion of the channel response that lies beyond the length of the channel estimate, the worse the performance of the equaliser will be. For this reason, careful thought must be given to the design of filtering processes in transmitters and receivers to avoid unduly increasing the length of the channel response of signals presented for equalisation. It has been found that enhanced equalisation performance can be achieved by using an equalisation scheme comprised of a linear equaliser for conditioning a received signal prior to the processing of the signal using a more complicated equalisation scheme, such as a delayed decision feedback sequence estimation (DDFSE) scheme. See, for example, "Design of an interference-resistant equaliser for EDGE cellular radio systems" M. Barberis, S. Heinen, P. Guerra; Vehicular Technology Conference 2002. Proceedings. VTC 2052-Fall. 2002 IEEE 56th, Volume: 3, 24-28 September 2002, Pages: 1622-1626 vol. 3). In such a composite equalisation scheme, the linear equaliser is used to partially remove ISI and to condition noise and interference, thereby aiding the channel estimation and equalisation processes of the succeeding higher complexity equalisation Scheme. The inventors of the present patent application have realised that a linear equaliser used in such a capacity must be carefully controlled in order to avoid unduly lengthening the channel response of the signal as perceived by the higher complexity equalisation scheme following the linear equaliser relative to the length of the channel estimate that is used by the higher complexity equalisation scheme. Sunrary of Ihvrotion : According to one aspect, the invention provides apparatus for conditioning a communications signal destined for equalisation by an equaliser, the apparate comprising a pre-filter located before the equaliser and means for adjusting the effect of the pre-filter to inhibit lengthening of the channel response of the communications signal by the pre- filter due to multipath structure in said signal. The invention also consists in a method of conditioning a communications Signal destined for equalisation by an equaliser, the method comprising pre-filtering the communications signal prior to the equaliser and adjusting the effect of the pre-filtering step to inhibit lengthening of the channel response of the communications signal by the pre-filtering step due to multipath structure in said signal. In certain embodiments, the pre-filtering operation is adjusted by causing the signal that is to be subjected to the pre-filtering to contain energy beyond the Nyquist limit effective in the pre-filtering operation. In certain embodiments, the filtering characteristic that the pre-filtering operation is to apply is altered by convolution with a pulse-shaped impulse response. In certain embodiments, additional filtering is provided in-line with the pre-filter, the additional filtering having a pulse-shaped impulse response. According to another aspect, the invention provides a method of modifying the filtering characteristic of a pre-filter for an equaliser, wherein the pre-filter is arranged to apply a filtering characteristic to a communications signal which is destined for equalisation by the equaliser»and the method comprises providing a filtering characteristic which is intended to be deployed in the pre-filter and modifying the filtering characteristic by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter when the filtering characteristic is deployed in the pre-filter. The invention also consists in apparatus for modifying the filtering characteristic of a pre- filter for an equaliser, wherein the pre-filter is arranged to apply a filtering characteristic to a communications signal which is destined for equalisation by the equaliser and the apparatus comprises means for modifying the filtering characteristic by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter when the filtering characteristic is deployed in the pre-filter. The pre-filtering operation may comprise linear equalisation. Brief description of acccqpanyirg drawings : By way of example only, certain embodiments of the invention will now be described with reference to the accompanying drawings, in which: Figure 1 is a series of diagrams illustrating the principles of multipath propagation; Figure 2 is a block diagram of a communications system; Figure 3 is a channel response of a signal containing three multipath components; Figure 4 is a diagram showing the channel responses of the individual multipath components of Figure 3 superimposed on one another; Figure 5 is a diagram illustrating channel response impulses corresponding to the multipath structure of Figure 3; Figure 6 illustrates the distortion that can arise in the channel response of a multipath component after filtering; Figure 7 illustrates how the distortion shown in Figure 6 can be controlled; Figure 8 illustrates how the correction shown in Figure 7 can distort the chf-onel responses of other multipath components; Figure 9 illustrates in more detail the control unit associated with the pre-filter in Figure 2; Figure 10 illustrates a part of a variant of a system shown in Figure 2; Figure 11 illustrates a part of a variant of the system shown in Figure 2; Figure 12 illustrates another variant of the system shown in Figure 2; Figure 13 illustrates some normalised frequency responses (in each of which the horizontal axis values denote frequency in radians per sample when multiplied by n ) associated with the system shown in Figure 12; Figure 14'illustrates part of a variant of the system shown in Figure 2; and Figure 15 illustrates part of a variant of the system shown in Figure 2. Dataikd description of preferred enfcodiroants : Figure 2 shows an EDGE radio system 200 in which a transmitter 210 attempts to convey an information signal u to a receiver 212. The use of bold text denotes that the information signal is a vector consisting of a stream of samples un where n=0,1,2,3 ... and the same convention is used below in relation to other signals and some impulse responses. In the transmitter 210, the information signal u undergoes error protection coding in unit 214 to produce an error-coded signal d. An example of error protection coding is the convolutional coding used in an EDGE radio system (see 3GPP TS 45.003 3rd Generation Partnership Project; Technical Specification Group GSM/EDGE; Radio Access Network). The error-coded signal d is then supplied to unit 216 where the signal undergoes modulation and becomes modulated signal x. The signal x then passes through a filter 217. The role of the filter 217 is to constrain the bandwidth of the signal x. , The impulse response sof the filter 217 is a. After filtering by filter 217, the signal x is emitted from the transmitter 210 and travels through a channel 218 and arrives at the receiver 212 as a signal s. The channel describes the effect of the environment on the signal x and has a channel response c. In the receiver 212, the signal s is initially processed by a front-end section 220. Within the front-end section 220, the signal s is filtered by a filter 221. The role of the filter 221 is to reject signals in adjacent frequency channels that might interfere with the desired signal, s. The signal s emerges from the front-end section 220 as a signal r which is then conditioned by a pre-filter 222 and becomes signal y which is then supplied to a demodulation unit 224. The demodulation unit 224 uses a SOVA scheme to equalise the signal r as conditioned by the pre-filter 222. The pre-filter 222 is an MSE linear equaliser and its purpose is to partially remove ISI present in signal r and to condition the noise and interference present in symbol r, thereby aiding the channel estimation and equalisation processes carried out in the demodulation unit 224. The operation of the pre-filter 222 is controlled by control unit 223, whose function will be described later. Together, the pre- filter 222 and the SOVA scheme used in the demodulation unit 224 constitute a composite equalisation scheme. A demodulated version e of the signal r emerges from the demodulation unit 224 and proceeds to an error decoding unit 226. The error decoding unit 226 utilises knowledge about the error protection coding applied by unii 214 in the transmitter 200 in order to produce a signal i that is an estimate of the information signal u. It will be apparent that most of the signal processing shown in Figure 2 occurs in the digital domain with digital to analogue and analogue to digital conversion occurring before filter 217 and after filter 221, respectively. The signal r that enters the composite equalisation scheme can be thought of as a version of the signal x that has arrived through a channel having a channel response h where h has the form h = a®c®b, where denotes the convolution operation. An example form for the signal channel response h is shown in Figure 3. In the example shown, the channel response r contains three contains three multipath components n, r2 and r-3. These multipatti components manifest themselves as peaks 300, 310 and 312. The channel response h is a summation of the channel responses of the individual mulitpath components such that h = hi + h2 + h3 where hi, h2 and h3 are the channel responses of the multipath components n, r2 and r3i respectively. The channel responses hi, h2 and h3 are shown superimposed on one another in Figure 4 as curves 400, 410 and 412, respectively. The pre-filter 222 has an impulse response p. If one notionally decomposes r into its multipath components ri, r2 and r3 and considers the pre-filter 222 as acting on the multipatil components n, r2 and r3 individually, then, at the output of the pre-filter, the channel response of ri becomes p®h1 = g!_ the channel response of r2 becomes p®h2 = gz and the channel response of r3 becomes p®h3 = g3. Ideally, the pre-filter 222 operates to produce residual channel responses gi, g2 and g3 in the form of single impulses denoting the positions of the multipath components ri, r2 and r3 in the channel respond h. In order for the pre-filter 222 to produce such a result, however, it is required that r is such that hi = j®gi> hz,= j®g2 and h3 = j®g3 where gi, g2 and g3 are single impulses and that p can be set to a vector w where w is such that wj®x = x where x is an arbitrary signal. For the sake of clarity in the following description of the function of control unit 223, assume now that r conforms to the requirements stated in the previous sentence. The three impulses gi, g2 and g3 are shown superposed in Figure 5, as impulses 500, 510 and 512, respectively. Since the pre-filter 222 performs filtering in the digital domain with a finite sample rate, the pre-filter can only produce an approximation w' of the impulse response w required to convert hi, h2 and h3 into the single impulses gi, g2 and g3. Assume now that w' is designed to operate on ri to convert hi into a residual channel response gi' that is a single impulse, i.e. gi = gi. The channel response gi' is shown in Figure 5, in which the single impulse is labelled 500. The pre-filter 222 with its impulse response w' will also operate on multipath components tz and r3 to convert h2 and h3 into residual channel responses gz and g3', respectively. Whether or not the pre-filter 222 produces g2' in the form of a single impulse that maps on to g2 will depend on the separation A1.2 between gi and g2, shown in Figure 5, and enumerated in terms of a number of sample periods of the pre-filter 222. If A1-2 is an integer number, then g2' will be a single impulse that maps on to g2. This result is shown in Figure 6 as impulse 600. However, if A1-2 is an non-integer number, then g2 will be broadened from an impulse into the curve shown in Figure 6 as a broken lfae and having a peak 610 and slowly decaying tails 612 and 614. The situation as regards g3 and its conformation to g3 is dependent in an analogous manner on the separation A1-3, again shown in Figure 5, between gi and g3. It will be apparent that, largely due to the slowly decaying tails 612 and 614, many more taps are required to describe the residual channel response represented by the broken line in Figure 6 than the single tap that is required to characterize impulse 600. Consequently, due to the finite sampling rate of the pre-filter 222, the channel response of signal r at the output of the pre-filter is likely to contain peaks encumbered with slowly decaying tails in place of impulses 510 and 512. Thus, the application of pre-filter 222 can result in a channel response that is longer than the length of the estimate used in demodulation 224; this worsens the performance of the receiver. In order restore efficacy to the pre-filter 222, the pre-filter is configured by the control unit 223 so that the pre-filter does not have impulse response w' but rather has an impulse response w" = w'q where q is an impulse response in the form of a Gaussian pulse having a shape proportional to: where t is the sample count and T is the sampling interval of the pre-filter 222. Accordingly, the pre-filter 222 now performs the operations -w"®hi, w"®h2 and w"®h3 on the multipath components ri, r2 and r3 respectively to produce residual channel responses &"» g2" and ga" respectively. The residual channel response g2" is shown in Figure 7 alongside g2". The former is shown as a solid line and the latter as a broken line. It will be apparent that the tails are greatly suppressed in g2 when compared to g2'. Therefore, fewer taps are required to characterize the residual channel response g2" than are required to characterize the response g2'. The effect of w" on h\ is shown in Figure 8, in which gV is shown as a solid line and gi" is shown as a broken line. It will be apparent that the single impulse of g/ has in gi broadened and lowered into a peak of appreciable thickness. (The appearance of &" will, of course, depend on the separation A,_3 in the same way that g2" depends on g2. However, the residual channel response gs" is not shown or described here for the sake of brevity.) The width — of the Gaussian pulse in q controls the degree of suppression of the tails in a g2" and the broadening of the peak in g/. The smaller a becomes^ the less severe the suppression of the tails in g^ will become and the narrower the peak in gi will become. Conversely, the greater a becomes, the greater the suppression of the tails will become in g2" and the broader the peak in gi" will become. It is usually possible to choose a value of a that provides a useful compromise in terms of the tail suppression and the peak broadening through trial and error in order to ensure that the pre-filter 222 succeeds in reducingsthe length of the channel response z of signal y. The shaping of the Gaussian pulse within q is dictated by control unit 223. The control unit 223 is shown in more detail in Figure 9. In addition to application to pre-filter 222, the signal r is also applied to an estimation unit 900 within the control unit 223. Estimation unit 900 calculates some metrics of the signal r and supplies them to filter configuration unit 910. Filter configuration unit 910 periodically recalculates V, a, q and w" and reconfigures the pre-filter 222 to operate with the new version of w" thus obtained. In simpler variants, w" or q is fixed at a form derived by trial and error research. Some alternatives to the pre-filter 222 and its control unit 223 will now be described with reference to Figures 10, 11, 14 and 15. In Figure 10, the pre-filter 222 and its control unit 223 are replaced by a group of pre-filters 222a to c and a decision unit 1000. The pre-filters 222a to c have differing impulse responses* pa, pb and pc, respectively, and they operate in parallel on signal r to produce signals sa, Sb and sc, respectively. The decision unit 1000 calculates a quality metric \i for each of signals sa, Sb and sc and supplies the one of these signals with the best Value of \i to the demodulation unit 224 for further processing. The quality metric \x can be, for example, a noise power value calculated by first estimating the channel of a signal from a section of the signal containing a training sequence. This channel estimate is then convolved with the known training sequence and the result is subtracted from the section of the signal that was used to produce the channel estimate. The power in the residual signal section thus produced yields a value for u. At least one of the impulse responses pa, Pb and pc is derived in the same manner as w" in the system of Figure 2, i.e. involving convolution with a Gaussian pulse in order to suppress in the manner shown in Figure 7 tails, of the kind described with reference to Figure 6. Since the impulse responses pa, Pb an- pc are fixed, the arrangement shown in Figure 10 avoids the data processing overheads associated with the periodic recomputation of w" by control unit 223. In Figure 11, the control unit 223 is omitted and the adjustable pre-filter 222 is replaced with pre-rfilter 222d that uses a fixed, predetermined impulse response pa having the form wigiqa where w[a®b] = 1 and q suppress in the manner shown in Figure 7 tails of the kind shown in Figure 6. It will be recalled that a and b are the known impulse responses of filters 217 and 221, respectively. In Figure 14, instead of arranging that the pre-filter impulse response is modified by convolution with a Gaussian pulse, the pre-filter 222e is preceded by an inhibiting filter 1400 that has an impulse response in the shape of a Gaussian pulse. Together, pre-filter 222e and inhibiting filter 1400 have the same effect as pre-filter 222 in Figure 1. For example, given the scenario used in conjunction with Figure 2 wherein the pre-filter is given an impulse response w'®q, the pre-filter 222e in Figure 14 would be given impulse response w' and the inhibiting filter 1400 would be given impulse response q. It will be apparent to the skilled person that the inhibiting filter 1400 can be implemented with an impulse response that is, for example, periodically recalculated in an analogous manner to the recalculation of q as described in Figure 9, selected from amongst a number of available impulse responses in an analogous manner to the selection process performed by the decision unit 1000 in Figure 10, or fixed in analogy with impulse response qd in Figure 11. It will also be apparent to the skilled person that the inhibiting filter could be relocated to a position following the pre-filter 222, as shown in Figure 15 (in which the inhibiting filter is relabelled 1500). In the systems described with reference to Figures 2,10 and 11, a Gaussian pulse is used in a convolution step to control tail growth in the channel response z of signal y and, in the systems described with reference to Figures 14 and 15, an inhibiting filter with an impulse response \|n the shape of a Gaussian pulse is likewise used to control tail growth in channel response z. It will, however, be apparent to the skilled person that other kinds of pulse shape can be used in place of the Gaussian pulse and yet achieve the same result in controlling the length of the channel response z. For example, the pulse could be square, triangular or raised cosine in shape, the efficacy of the pulse depending on its precise shape. in another variant of the scheme described with reference to Figure 2, w does not attempt to eliminate the effect of a and b but only a such that w®a = 1. By arranging the pre-filter 222^0 counteract a, the co-channel interference affecting r is whitened and this can benefit the operation of the SOVA equaliser operating on signal y within demodulation unit 224. Another variant of system 200 is shown in Figure 12. Reference symbols in Figure 12 that appear in Figure 2 continue to denote the same features as in Figure 2. For the purpose of simplifying the following discussion of the EDGE system 1200 shown r* Figure 12, assume that signal r has the structure described with reference to Figure 3 and that the three multipara components ri, r2 and r3 again have channel responses jigtgi, j®g2 and jg3, respectively, where, as before, gl5 g2 and g3 are single impulses and Aw denotes the separation between impulses gi and g2 and A,_3 denotes the separate between impulses gi andg3. It will be recalled that in the system 200 the configuration of the pre-filter 222 with an impulse response w' having the property w'®j®gi = gi is rejected in favour of a configuration w" = w'q, where q has a Gaussian pulse shape, in order to inhibit the lengthening of the channel response z of signal y. In system 1200, however, the control unit 223 configures the pre-filter 222 to have an impulse response w' having the property w'®j®gi = gi and undesirable growth in the length of z is instead inhibited through the agency of filters 217 and 221, as will now be described. * The filters 217 and 221 with their impulse responses a and b respectively can, notionally, be replaced with a single filter F whose impulse response k is a®b. Conventionally, the filter 221 includes anti-aliasing filtering so that the frequency response of notional filter F is quiescent beyond the Nyquist limit of the pre-filter 222. However, in system 1200, the filters 217 and 221 are arranged such that notional filter F has a non-zero response in the part of its frequency response that lies beyond the Nyquist limit of the pre-filter 222. This extension of the frequency response of notional filter F causes aliasing in the* operation of the pre-filter 222 which, in turn, leads to suppression of the spreading effect in the channel response z of signal y that will arise if either or both of the alignments A,_2 and A,_3have non-integer values. A An example of the frequency response of notional filter F in system 1200 is shown in Figure 13 a. The frequency response shown there is sampled at a rate eight times higher than the sampling rate of the pre-filter 222 and features a non-zero response beyond the Nyquist limit of pre-filter 222, indicated by line A-A. Figure 13b shows two curves 1312 and 1314 which illustrate how the response of the notional filter F would appear to a signal at the sample rate used by the pre-filter 222. Curve 1312 applies when the samples of the signal are aligned with the taps of the notional filter F and curve 1314 applies when the samples of the signal experiencing filter F are offset by a fraction of a sample relative to the taps of filter F. It will be apparent that curve 1314 drops away from curve 1312 at higher frequencies and the effect of this difference on the operation of the pve-filter 222 will now be described. For the sake of simplicity, assume now that j = k such that the pre-filter 222 performs the convolutions w'®(k®gi) to produce gi', w'®(k®g2) to produce & and w'®(k®g3) to produce &' for the three multipath components. Curve 1300 in Figure 13c is the frequency domain representation of impulse response w'®k®gi sampled at the sampling rate of the pre-filter 222. As is apparent from Figure 13c, curve 1300 is fiat and this characteristic corresponds to a single impulse in the time domain, which is indeed the result rf w'®k®gi since it will be recalled that w' is designed to convert hi into the pure impulse gi. If A,_2 is integer, then w'®h2 will map to pure impulse g2 and the frequency response of w'®h2 (i.e. of w'®k®g2) will match curve 1300. However, if A^2 is non-integer, then w'®h2 will behave like curve 1310 and, after initially conforming to curve 1300, drop away due to the aliasing within the pre-filter 222 of the non-zero frequency response of k beyond the Nyquist limit of the pre-filter 222. In the time domain, this drop translates into a suppression, as in Figure 7, of the tails that form around the peak in the residual channel response w'®k®g2. The frequency response of w'®k®g3 likewise depends orf the value of A,_3. The system 1200 includes additional control units 1212 and 1210 for adjusting filters 217 and 221, respectively, for the purpose of adjusting k to manipulate the drop in the curve 1310 so as to produce a desired degree of tail suppression in residual channel responses responses g2' and g3'. In practice, trial and error would be used to find settings for filters 217 and 221 that are optimum in terms of a balance between the degree of tail suppression in g2* and & and other operating criteria. In certain embodiments, only h is rendered adjustable for the purpose of varying k. Several embodiments of the invention have now been described in which tail growth is suppressed in the channel response of the output of a pre-filter to an equalisation scheme. It will be apparent to the skilled person that many other variants of the invention exist. For example", the described embodiments assume that the information transmitted between the transmitter and receiver is formatted in blocks of bits rather than as a continuous stream. It will, however, be apparent that the invention is applicable to systems in which the information is transmitted as a continuous stream. Furthermore, it will be appreciated that the various units that operate on digital signals, for example, units 214 and 224, can be implemented as hardware structures or as software running on a general data processor or used in conjunction with customised data processing hardware. We Claim : 1. Apparatus (212) for conditioning a communications signal destined for equalisation by an equaliser, the apparatus (212) comprising a pre-filter (222) located before the equaliser, wherein the apparatus further comprises means (223) for modifying the filtering characteristic of the pre-filter by convolution with a pulse- shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre- filter (222) due to multipath structure in said signal. 2. Apparatus as claimed in claim 1 further comprising filtering means which is located before the pre-filter and which is arranged to cause the communications signal, at the pre-filter input, to contain energy in a frequency region beyond the Nyquist limit of that signal to cause aliasing in the pre-filter that inhibits said lengthening. 3. Apparatus as claimed in claim 1 or 2, wherein the modifying means comprises means in line with the pre- filter for applying to the communications signal filtering having a pulse-shaped impulse response to inhibit said lengthening by said pre-filter. 4. Apparatus as claimed in any one of claims 1 to 3, wherein said pulse is Gaussian. 5. Apparatus as claimed in any one of claims 1 to 4, wherein the pre-filter is a linear equaliser. 6. A method of conditioning a communications signal destined for equalisation by an equaliser, the method comprising pre-filtering the communications signal prior to the equaliser wherein the method further comprises modifying the filtering characteristic of the pre-filtering step by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filtering step due to multipath structure in said signal. 7. A method as claimed in claim 6, further comprising filtering the communications signal prior to the pre- filtering step such that the communications signal, as supplied to the pre-filtering step, contains energy in a frequency region beyond the Nyquist limit of the signal to cause aliasing in the pre-filtering step that inhibits said lengthening. 8. A method as claimed in claim 6 or 7, wherein the modifying step comprises a step of applying to the communications signal in line with the pre-filter filtering having a pulse-shaped impulse response to inhibit said lengthening by said pre-filter. 9. A method as claimed in claim 7 or 8, wherein said pulse is Gaussian. 10.A method as claimed in any one of claims 6 to 9, wherein the pre-filtering step performs linear equalisation of the communications signal. 11.A method of modifying the filtering characteristic of a pre-filter (222) for an equaliser, wherein the pre-filter (222) is arranged to apply a filtering characteristic to a communications signal destined for equalisation by the equaliser, wherein the method comprises providing a filtering characteristic which is intended to be deployed in the pre- filter (222) and modifying the filtering characteristic by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter when the filtering characteristic is deployed in the pre-filter. 12.A method as claimed in claim 11, wherein said pulse is Gaussian. 13. A method as claimed in claim 11 or 12, wherein said pre- filter is a linear equaliser. 14. Apparatus for modifying the filtering characteristic of a pre-filter (222) for an equaliser, wherein the pre-filter (222) is arranged to apply a filtering characteristic to a communications signal destined for equalisation by the equaliser, wherein the apparatus comprises means (223) for modifying the filtering characteristic by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre- filter (222) when the filtering characteristic is deployed in the pre-filter (222). 15.Apparatus as claimed in claim 14, wherein said pulse is Gaussian. 16.Apparatus as claimed in claim 14 or 15, wherein said pre- filter is a linear equaliser. ABSTRACT "Minimisation of perturbation of a radio signal between a transmitter and a receiver by conditioning equaliser input" Apparatus (212) for conditioning a communications signal destined for equalisation by an equaliser, the apparatus (212) comprising a pre-filter (222) located before the equaliser, wherein the apparatus further comprises means (223) for modifying the filtering characteristic of the pre-filter by convolution with a pulse-shaped impulse response to inhibit lengthening of the channel response of the communications signal by the pre-filter (222) due to multipath structure in said signal.

Full Text

Field of Invention :
The invention relates to apparatus for, and to methods of, conditioning communications
signals and configuring filters.
Background of Invention :
In a radio system, a signal travelling between a transmitter and a receiver will be perturbed
by the environment in which the transmitter and the receiver exist.' The performance of a
receiver, i.e. its ability to accurately recover from a received signal information that was
put into the signal by a transmitter, depends, to a significant extent, on the ability of the
receiver to remove perturbations caused by the environment.
Environmental perturbations of a transmitted signal are usually ascribed to three main
sources. These are noise, interference and multipath propagation. Noise may have many
sources, including thermal noise from the general environment and circuit noise from
within the receiver in question. Interference arises from other signals that are being
transmitted in the vicinity of the receiver in question. Multipath propagation arises when a
signal reaches a receiver from a transmitter by more than one path. An example of a
situation in which multipath propagation occurs is shown in Figure 1.
Figure la shows a transmitter 100 attempting to send a radio signal to a receiver 110.
Three objects 112, 114 and 116 are located in the vicinity of the transmitter and the
receiver 110. The radio signal emitted by the transmitter 100 can reach the receiver 110 by
various paths. First, the radio signal can reach the receiver 110 by the direct path 118
leading from the transmitter 100. The radio signal from the transmitter 100 can also reach
the receiver 110 by reflection from objections 112 to 116. For example, the radio signal
can travel along paths 120, 122 and 124 with reflection from objects 112, 114 and 116,
respectively. It will be apparent that the paths 118 to 124 have different lengths with the
result that the versions of the radio signal travelling along these paths will arrive at the
receiver 110 with differing relative delays.
It is common practice to characterise multipath propagation in terms of a graph showing
how the power of a received signal is distributed with delay. Figure lb illustrates how

such a plot might look for a radio signal travelling from transmitter 100 to receiver 110 in
the presence of objects 112 to 116. Figure lb illustrates that the power in the received
signal is spread over a range of delays and Figure 1c illustrates a sampled or digitised
version of the power versus delay profile of Figure lb.
Commonly, the information to be conveyed by a radio signal emitted by a transmitter
exists in the form of a series of information symbols. Multipath propagation can cause a
phenomenon known as intersymbol interference (ISI). ISI occurs when paths leading to a
receiver from a transmitter differ in length in a way which causes one information symbol
in the radio signal arriving along one path to arrive at the receiver at the same time as
another information symbol in the radio signal arriving at the receiver along another path.
ISI can also be introduced by other mechanisms. For example, in an Enhanced Data rates
for GSM Evolution (EDGE) system, a filter is used to constrain the bandwidth of a
modulated GMSK or 8PSK signal prior to its emission from a transmitter (see, for
example, 3GPP TS 45.004 3rd Generation Partnership Project; Technical Specification
Group GSM/EDGE; Radio Access Network). Moreover, a signal acquired by a receiver
will usually undergo filtering of one form or another (for example, to reject interference in
adjacent frequency channels that would otherwise reduce dynamic range). Such filtering
mechanisms can be another source of ISI.
It is usual to design a receiver to counteract, in the presence of interference and noise, ISI
due to, for example, one or more of multipath propagation, filtering in a transmitter and
filtering in a receiver and to estimate the information that was put into the signal by the
transmitter that originated the signal. See, for example, "Digital Communications", John
G. Proakis, McGraw-Hill International Series, 3rd Edition.
There are many methods of counteracting ISI, which methods are known generically as
"equalisation". Among these the following are worthy of note:
1. Linear equalisation (see, for example, "Digital Communications", John G. Proakis,
McGraw-Hill International Series, 3rd Edition);

2. Maximum-Likelihood Sequence Estimation (MLSE) using the Viterbi algorithm
(see, for example, ("Maximum-likelihood sequence estimation of digital sequences
in the presence of intersymbol interference", Forney, G., Jr.; IEEE Transactions on
Information Theory, Volume: 18, Issue: 3, May 1972 Pages: 363-378);
3. Maximum A-posteriori Probability (MAP) (see, for example, "Optimum and sub-
optimum detection of coded data disturbed by time-varying intersymbol
interference [applicable to digital mobile radio receivers]", Koch, W.; Baier, A.;
Global Telecommunications Conference, 1990, and Exhibition. "Communications:
Connecting the Future", GLOBECOM '90., IEEE, 2-5 Dec. 1990 Pagra: 1679-1684
vol. 3), and its lower complexity versions, the max-log-MAP; and
4. The soft-output Viterbi Algorithm (SOVA) ("A Viterbi algorithm with soft-decision
outputs and its applications", Hagenauer, J,; Hoeher, P,; Global
Telecommunications Conference, 1989, and Exhibition. "Communications
Technology for the 1990s and Beyond". GLOBECOM '89., IEEE, 27-30 Nov.
1989 Pages: 1680-1686 vol. 3).
A number of reduced complexity ISI suppression methods exist, in particular Reduced
State Sequence Estimation (RSSE) (see, for example, "Reduced-state sequence estimation
with set partitioning and decision feedback", Eyuboglu, M. V.; Qureshi, S.U.H.; IEEE
Transactions on Communications, Volume: 36, Issue: 1, Jan. 1988 Pages: 13-20) which
can be of use for higher order modulation schemes such as 8PSK.
In mathematical terms, it is possible to regard a received signal as arriving at a receiver
from a transmitter through a channel consisting of a filter describing the various sources of
ISI that affect the received signal. In this document, such a filter shall be referred to as a
"channel" and the impulse response of such a filter shall be referred to as a "channel
response" and an estimate or measurement of such an impulse response shall be referred to
as a "channel estimate".
In order for an equaliser to work successfully on a received signal, it is desirable to endow
the equaliser with an accurate channel estimate for the signal. This consideration

advocates the use of a channel estimate which is at least as long as the true channel.
However, the length L of a channel estimate used by an equaliser will directly affect the
complexity of the equaliser. In the case of a linear equaliser, the complexity will vary
linearly with L. In the case of a SOVA equaliser, the complexity will rise exponentially
with L. Thus, the length allowed for a channel estimate in an equalisation scheme is
normally limited. If an equaliser uses a channel estimate whose length is shorter than the
channel response, then the performance of the equaliser will suffer. In general terms, the
greater the proportion of the power of the received signal that lies in the portion of the
channel response that lies beyond the length of the channel estimate, the worse the
performance of the equaliser will be. For this reason, careful thought must be given to the
design of filtering processes in transmitters and receivers to avoid unduly increasing the
length of the channel response of signals presented for equalisation.
It has been found that enhanced equalisation performance can be achieved by using an
equalisation scheme comprised of a linear equaliser for conditioning a received signal prior
to the processing of the signal using a more complicated equalisation scheme, such as a
delayed decision feedback sequence estimation (DDFSE) scheme. See, for example,
"Design of an interference-resistant equaliser for EDGE cellular radio systems" M.
Barberis, S. Heinen, P. Guerra; Vehicular Technology Conference 2002. Proceedings.
VTC 2052-Fall. 2002 IEEE 56th, Volume: 3, 24-28 September 2002, Pages: 1622-1626
vol. 3). In such a composite equalisation scheme, the linear equaliser is used to partially
remove ISI and to condition noise and interference, thereby aiding the channel estimation
and equalisation processes of the succeeding higher complexity equalisation Scheme. The
inventors of the present patent application have realised that a linear equaliser used in such
a capacity must be carefully controlled in order to avoid unduly lengthening the channel
response of the signal as perceived by the higher complexity equalisation scheme following
the linear equaliser relative to the length of the channel estimate that is used by the higher
complexity equalisation scheme.
Sunrary of Ihvrotion :
According to one aspect, the invention provides apparatus for conditioning a
communications signal destined for equalisation by an equaliser, the apparate comprising
a pre-filter located before the equaliser and means for adjusting the effect of the pre-filter

to inhibit lengthening of the channel response of the communications signal by the pre-
filter due to multipath structure in said signal.
The invention also consists in a method of conditioning a communications Signal destined
for equalisation by an equaliser, the method comprising pre-filtering the communications
signal prior to the equaliser and adjusting the effect of the pre-filtering step to inhibit
lengthening of the channel response of the communications signal by the pre-filtering step
due to multipath structure in said signal.
In certain embodiments, the pre-filtering operation is adjusted by causing the signal that is
to be subjected to the pre-filtering to contain energy beyond the Nyquist limit effective in
the pre-filtering operation.
In certain embodiments, the filtering characteristic that the pre-filtering operation is to
apply is altered by convolution with a pulse-shaped impulse response.
In certain embodiments, additional filtering is provided in-line with the pre-filter, the
additional filtering having a pulse-shaped impulse response.
According to another aspect, the invention provides a method of modifying the filtering
characteristic of a pre-filter for an equaliser, wherein the pre-filter is arranged to apply a
filtering characteristic to a communications signal which is destined for equalisation by the
equaliser»and the method comprises providing a filtering characteristic which is intended to
be deployed in the pre-filter and modifying the filtering characteristic by convolution with
a pulse-shaped impulse response to inhibit lengthening of the channel response of the
communications signal by the pre-filter when the filtering characteristic is deployed in the
pre-filter.
The invention also consists in apparatus for modifying the filtering characteristic of a pre-
filter for an equaliser, wherein the pre-filter is arranged to apply a filtering characteristic to
a communications signal which is destined for equalisation by the equaliser and the
apparatus comprises means for modifying the filtering characteristic by convolution with a
pulse-shaped impulse response to inhibit lengthening of the channel response of the

communications signal by the pre-filter when the filtering characteristic is deployed in the
pre-filter.
The pre-filtering operation may comprise linear equalisation.
Brief description of acccqpanyirg drawings :
By way of example only, certain embodiments of the invention will now be described with
reference to the accompanying drawings, in which:
Figure 1 is a series of diagrams illustrating the principles of multipath propagation;
Figure 2 is a block diagram of a communications system;
Figure 3 is a channel response of a signal containing three multipath components;
Figure 4 is a diagram showing the channel responses of the individual multipath
components of Figure 3 superimposed on one another;
Figure 5 is a diagram illustrating channel response impulses corresponding to the multipath
structure of Figure 3;
Figure 6 illustrates the distortion that can arise in the channel response of a multipath
component after filtering;
Figure 7 illustrates how the distortion shown in Figure 6 can be controlled;
Figure 8 illustrates how the correction shown in Figure 7 can distort the chf-onel responses
of other multipath components;
Figure 9 illustrates in more detail the control unit associated with the pre-filter in Figure 2;
Figure 10 illustrates a part of a variant of a system shown in Figure 2;
Figure 11 illustrates a part of a variant of the system shown in Figure 2;

Figure 12 illustrates another variant of the system shown in Figure 2;
Figure 13 illustrates some normalised frequency responses (in each of which the horizontal
axis values denote frequency in radians per sample when multiplied by n ) associated with
the system shown in Figure 12;
Figure 14'illustrates part of a variant of the system shown in Figure 2; and
Figure 15 illustrates part of a variant of the system shown in Figure 2.
Dataikd description of preferred enfcodiroants :
Figure 2 shows an EDGE radio system 200 in which a transmitter 210 attempts to convey
an information signal u to a receiver 212. The use of bold text denotes that the information
signal is a vector consisting of a stream of samples un where n=0,1,2,3 ... and the same
convention is used below in relation to other signals and some impulse responses.
In the transmitter 210, the information signal u undergoes error protection coding in unit
214 to produce an error-coded signal d. An example of error protection coding is the
convolutional coding used in an EDGE radio system (see 3GPP TS 45.003 3rd Generation
Partnership Project; Technical Specification Group GSM/EDGE; Radio Access Network).
The error-coded signal d is then supplied to unit 216 where the signal undergoes
modulation and becomes modulated signal x. The signal x then passes through a filter 217.
The role of the filter 217 is to constrain the bandwidth of the signal x. , The impulse
response sof the filter 217 is a. After filtering by filter 217, the signal x is emitted from the
transmitter 210 and travels through a channel 218 and arrives at the receiver 212 as a
signal s. The channel describes the effect of the environment on the signal x and has a
channel response c.
In the receiver 212, the signal s is initially processed by a front-end section 220. Within
the front-end section 220, the signal s is filtered by a filter 221. The role of the filter 221 is
to reject signals in adjacent frequency channels that might interfere with the desired signal,
s. The signal s emerges from the front-end section 220 as a signal r which is then
conditioned by a pre-filter 222 and becomes signal y which is then supplied to a
demodulation unit 224. The demodulation unit 224 uses a SOVA scheme to equalise the

signal r as conditioned by the pre-filter 222. The pre-filter 222 is an MSE linear equaliser
and its purpose is to partially remove ISI present in signal r and to condition the noise and
interference present in symbol r, thereby aiding the channel estimation and equalisation
processes carried out in the demodulation unit 224. The operation of the pre-filter 222 is
controlled by control unit 223, whose function will be described later. Together, the pre-
filter 222 and the SOVA scheme used in the demodulation unit 224 constitute a composite
equalisation scheme. A demodulated version e of the signal r emerges from the
demodulation unit 224 and proceeds to an error decoding unit 226. The error decoding unit
226 utilises knowledge about the error protection coding applied by unii 214 in the
transmitter 200 in order to produce a signal i that is an estimate of the information signal u.
It will be apparent that most of the signal processing shown in Figure 2 occurs in the
digital domain with digital to analogue and analogue to digital conversion occurring before
filter 217 and after filter 221, respectively.
The signal r that enters the composite equalisation scheme can be thought of as a version
of the signal x that has arrived through a channel having a channel response h where h has
the form h = a®c®b, where denotes the convolution operation. An example form for
the signal channel response h is shown in Figure 3. In the example shown, the channel
response r contains three contains three multipath components n, r2 and r-3. These
multipatti components manifest themselves as peaks 300, 310 and 312. The channel
response h is a summation of the channel responses of the individual mulitpath
components such that h = hi + h2 + h3 where hi, h2 and h3 are the channel responses of the
multipath components n, r2 and r3i respectively. The channel responses hi, h2 and h3 are
shown superimposed on one another in Figure 4 as curves 400, 410 and 412, respectively.
The pre-filter 222 has an impulse response p. If one notionally decomposes r into its
multipath components ri, r2 and r3 and considers the pre-filter 222 as acting on the
multipatil components n, r2 and r3 individually, then, at the output of the pre-filter, the
channel response of ri becomes p®h1 = g!_ the channel response of r2 becomes p®h2 = gz
and the channel response of r3 becomes p®h3 = g3. Ideally, the pre-filter 222 operates to
produce residual channel responses gi, g2 and g3 in the form of single impulses denoting
the positions of the multipath components ri, r2 and r3 in the channel respond h. In order

for the pre-filter 222 to produce such a result, however, it is required that r is such that hi =
j®gi> hz,= j®g2 and h3 = j®g3 where gi, g2 and g3 are single impulses and that p can be set
to a vector w where w is such that wj®x = x where x is an arbitrary signal. For the sake
of clarity in the following description of the function of control unit 223, assume now that
r conforms to the requirements stated in the previous sentence. The three impulses gi, g2
and g3 are shown superposed in Figure 5, as impulses 500, 510 and 512, respectively.
Since the pre-filter 222 performs filtering in the digital domain with a finite sample rate,
the pre-filter can only produce an approximation w' of the impulse response w required to
convert hi, h2 and h3 into the single impulses gi, g2 and g3. Assume now that w' is
designed to operate on ri to convert hi into a residual channel response gi' that is a single
impulse, i.e. gi = gi. The channel response gi' is shown in Figure 5, in which the single
impulse is labelled 500. The pre-filter 222 with its impulse response w' will also operate
on multipath components tz and r3 to convert h2 and h3 into residual channel responses gz
and g3', respectively. Whether or not the pre-filter 222 produces g2' in the form of a single
impulse that maps on to g2 will depend on the separation A1.2 between gi and g2, shown in
Figure 5, and enumerated in terms of a number of sample periods of the pre-filter 222. If
A1-2 is an integer number, then g2' will be a single impulse that maps on to g2. This result is
shown in Figure 6 as impulse 600. However, if A1-2 is an non-integer number, then g2 will
be broadened from an impulse into the curve shown in Figure 6 as a broken lfae and having
a peak 610 and slowly decaying tails 612 and 614. The situation as regards g3 and its
conformation to g3 is dependent in an analogous manner on the separation A1-3, again
shown in Figure 5, between gi and g3.
It will be apparent that, largely due to the slowly decaying tails 612 and 614, many more
taps are required to describe the residual channel response represented by the broken line in
Figure 6 than the single tap that is required to characterize impulse 600. Consequently, due
to the finite sampling rate of the pre-filter 222, the channel response of signal r at the
output of the pre-filter is likely to contain peaks encumbered with slowly decaying tails in
place of impulses 510 and 512. Thus, the application of pre-filter 222 can result in a
channel response that is longer than the length of the estimate used in demodulation 224;
this worsens the performance of the receiver.

In order restore efficacy to the pre-filter 222, the pre-filter is configured by the control unit
223 so that the pre-filter does not have impulse response w' but rather has an impulse
response w" = w'q where q is an impulse response in the form of a Gaussian pulse
having a shape proportional to:

where t is the sample count and T is the sampling interval of the pre-filter 222.
Accordingly, the pre-filter 222 now performs the operations -w"®hi, w"®h2 and w"®h3 on
the multipath components ri, r2 and r3 respectively to produce residual channel responses
&"» g2" and ga" respectively.
The residual channel response g2" is shown in Figure 7 alongside g2". The former is shown
as a solid line and the latter as a broken line. It will be apparent that the tails are greatly
suppressed in g2 when compared to g2'. Therefore, fewer taps are required to characterize
the residual channel response g2" than are required to characterize the response g2'. The
effect of w" on h\ is shown in Figure 8, in which gV is shown as a solid line and gi" is
shown as a broken line. It will be apparent that the single impulse of g/ has in gi
broadened and lowered into a peak of appreciable thickness. (The appearance of &" will,
of course, depend on the separation A,_3 in the same way that g2" depends on g2.
However, the residual channel response gs" is not shown or described here for the sake of
brevity.)
The width — of the Gaussian pulse in q controls the degree of suppression of the tails in
a
g2" and the broadening of the peak in g/. The smaller a becomes^ the less severe the
suppression of the tails in g^ will become and the narrower the peak in gi will become.
Conversely, the greater a becomes, the greater the suppression of the tails will become in
g2" and the broader the peak in gi" will become. It is usually possible to choose a value of
a that provides a useful compromise in terms of the tail suppression and the peak
broadening through trial and error in order to ensure that the pre-filter 222 succeeds in
reducingsthe length of the channel response z of signal y.

The shaping of the Gaussian pulse within q is dictated by control unit 223. The control
unit 223 is shown in more detail in Figure 9. In addition to application to pre-filter 222, the
signal r is also applied to an estimation unit 900 within the control unit 223. Estimation
unit 900 calculates some metrics of the signal r and supplies them to filter configuration
unit 910. Filter configuration unit 910 periodically recalculates V, a, q and w" and
reconfigures the pre-filter 222 to operate with the new version of w" thus obtained. In
simpler variants, w" or q is fixed at a form derived by trial and error research.
Some alternatives to the pre-filter 222 and its control unit 223 will now be described with
reference to Figures 10, 11, 14 and 15.
In Figure 10, the pre-filter 222 and its control unit 223 are replaced by a group of pre-filters
222a to c and a decision unit 1000. The pre-filters 222a to c have differing impulse
responses* pa, pb and pc, respectively, and they operate in parallel on signal r to produce
signals sa, Sb and sc, respectively. The decision unit 1000 calculates a quality metric \i for
each of signals sa, Sb and sc and supplies the one of these signals with the best Value of \i to
the demodulation unit 224 for further processing. The quality metric \x can be, for example,
a noise power value calculated by first estimating the channel of a signal from a section of
the signal containing a training sequence. This channel estimate is then convolved with the
known training sequence and the result is subtracted from the section of the signal that was
used to produce the channel estimate. The power in the residual signal section thus
produced yields a value for u. At least one of the impulse responses pa, Pb and pc is
derived in the same manner as w" in the system of Figure 2, i.e. involving convolution with
a Gaussian pulse in order to suppress in the manner shown in Figure 7 tails, of the kind
described with reference to Figure 6. Since the impulse responses pa, Pb an- pc are fixed,
the arrangement shown in Figure 10 avoids the data processing overheads associated with
the periodic recomputation of w" by control unit 223.
In Figure 11, the control unit 223 is omitted and the adjustable pre-filter 222 is replaced
with pre-rfilter 222d that uses a fixed, predetermined impulse response pa having the form
wigiqa where w[a®b] = 1 and q suppress in the manner shown in Figure 7 tails of the kind shown in Figure 6. It will be
recalled that a and b are the known impulse responses of filters 217 and 221, respectively.

In Figure 14, instead of arranging that the pre-filter impulse response is modified by
convolution with a Gaussian pulse, the pre-filter 222e is preceded by an inhibiting filter
1400 that has an impulse response in the shape of a Gaussian pulse. Together, pre-filter
222e and inhibiting filter 1400 have the same effect as pre-filter 222 in Figure 1. For
example, given the scenario used in conjunction with Figure 2 wherein the pre-filter is
given an impulse response w'®q, the pre-filter 222e in Figure 14 would be given impulse
response w' and the inhibiting filter 1400 would be given impulse response q. It will be
apparent to the skilled person that the inhibiting filter 1400 can be implemented with an
impulse response that is, for example, periodically recalculated in an analogous manner to
the recalculation of q as described in Figure 9, selected from amongst a number of
available impulse responses in an analogous manner to the selection process performed by
the decision unit 1000 in Figure 10, or fixed in analogy with impulse response qd in Figure
11. It will also be apparent to the skilled person that the inhibiting filter could be relocated
to a position following the pre-filter 222, as shown in Figure 15 (in which the inhibiting
filter is relabelled 1500).
In the systems described with reference to Figures 2,10 and 11, a Gaussian pulse is used in
a convolution step to control tail growth in the channel response z of signal y and, in the
systems described with reference to Figures 14 and 15, an inhibiting filter with an impulse
response |n the shape of a Gaussian pulse is likewise used to control tail growth in channel
response z. It will, however, be apparent to the skilled person that other kinds of pulse
shape can be used in place of the Gaussian pulse and yet achieve the same result in
controlling the length of the channel response z. For example, the pulse could be square,
triangular or raised cosine in shape, the efficacy of the pulse depending on its precise
shape.
in another variant of the scheme described with reference to Figure 2, w does not attempt
to eliminate the effect of a and b but only a such that w®a = 1. By arranging the pre-filter
222^0 counteract a, the co-channel interference affecting r is whitened and this can benefit
the operation of the SOVA equaliser operating on signal y within demodulation unit 224.
Another variant of system 200 is shown in Figure 12. Reference symbols in Figure 12 that
appear in Figure 2 continue to denote the same features as in Figure 2. For the purpose of
simplifying the following discussion of the EDGE system 1200 shown r* Figure 12,

assume that signal r has the structure described with reference to Figure 3 and that the
three multipara components ri, r2 and r3 again have channel responses jigtgi, j®g2 and
jg3, respectively, where, as before, gl5 g2 and g3 are single impulses and Aw denotes the
separation between impulses gi and g2 and A,_3 denotes the separate between impulses gi
andg3.
It will be recalled that in the system 200 the configuration of the pre-filter 222 with an
impulse response w' having the property w'®j®gi = gi is rejected in favour of a
configuration w" = w'q, where q has a Gaussian pulse shape, in order to inhibit the
lengthening of the channel response z of signal y. In system 1200, however, the control
unit 223 configures the pre-filter 222 to have an impulse response w' having the property
w'®j®gi = gi and undesirable growth in the length of z is instead inhibited through the
agency of filters 217 and 221, as will now be described. *
The filters 217 and 221 with their impulse responses a and b respectively can, notionally,
be replaced with a single filter F whose impulse response k is a®b. Conventionally, the
filter 221 includes anti-aliasing filtering so that the frequency response of notional filter F
is quiescent beyond the Nyquist limit of the pre-filter 222. However, in system 1200, the
filters 217 and 221 are arranged such that notional filter F has a non-zero response in the
part of its frequency response that lies beyond the Nyquist limit of the pre-filter 222. This
extension of the frequency response of notional filter F causes aliasing in the* operation of
the pre-filter 222 which, in turn, leads to suppression of the spreading effect in the channel
response z of signal y that will arise if either or both of the alignments A,_2 and A,_3have
non-integer values.
A
An example of the frequency response of notional filter F in system 1200 is shown in
Figure 13 a. The frequency response shown there is sampled at a rate eight times higher
than the sampling rate of the pre-filter 222 and features a non-zero response beyond the
Nyquist limit of pre-filter 222, indicated by line A-A. Figure 13b shows two curves 1312
and 1314 which illustrate how the response of the notional filter F would appear to a signal
at the sample rate used by the pre-filter 222. Curve 1312 applies when the samples of the
signal are aligned with the taps of the notional filter F and curve 1314 applies when the
samples of the signal experiencing filter F are offset by a fraction of a sample relative to

the taps of filter F. It will be apparent that curve 1314 drops away from curve 1312 at
higher frequencies and the effect of this difference on the operation of the pve-filter 222
will now be described.
For the sake of simplicity, assume now that j = k such that the pre-filter 222 performs the
convolutions w'®(k®gi) to produce gi', w'®(k®g2) to produce & and w'®(k®g3) to
produce &' for the three multipath components. Curve 1300 in Figure 13c is the frequency
domain representation of impulse response w'®k®gi sampled at the sampling rate of the
pre-filter 222. As is apparent from Figure 13c, curve 1300 is fiat and this characteristic
corresponds to a single impulse in the time domain, which is indeed the result rf w'®k®gi
since it will be recalled that w' is designed to convert hi into the pure impulse gi. If A,_2 is
integer, then w'®h2 will map to pure impulse g2 and the frequency response of w'®h2 (i.e.
of w'®k®g2) will match curve 1300. However, if A^2 is non-integer, then w'®h2 will
behave like curve 1310 and, after initially conforming to curve 1300, drop away due to the
aliasing within the pre-filter 222 of the non-zero frequency response of k beyond the
Nyquist limit of the pre-filter 222. In the time domain, this drop translates into a
suppression, as in Figure 7, of the tails that form around the peak in the residual channel
response w'®k®g2. The frequency response of w'®k®g3 likewise depends orf the value of
A,_3.
The system 1200 includes additional control units 1212 and 1210 for adjusting filters 217
and 221, respectively, for the purpose of adjusting k to manipulate the drop in the curve
1310 so as to produce a desired degree of tail suppression in residual channel responses
responses g2' and g3'. In practice, trial and error would be used to find settings for filters
217 and 221 that are optimum in terms of a balance between the degree of tail suppression
in g2* and & and other operating criteria. In certain embodiments, only h is rendered
adjustable for the purpose of varying k.
Several embodiments of the invention have now been described in which tail growth is
suppressed in the channel response of the output of a pre-filter to an equalisation scheme.
It will be apparent to the skilled person that many other variants of the invention exist. For
example", the described embodiments assume that the information transmitted between the
transmitter and receiver is formatted in blocks of bits rather than as a continuous stream. It

will, however, be apparent that the invention is applicable to systems in which the
information is transmitted as a continuous stream. Furthermore, it will be appreciated that
the various units that operate on digital signals, for example, units 214 and 224, can be
implemented as hardware structures or as software running on a general data processor or
used in conjunction with customised data processing hardware.

We Claim :
1. Apparatus (212) for conditioning a communications
signal destined for equalisation by an equaliser, the
apparatus (212) comprising a pre-filter (222) located
before the equaliser, wherein the apparatus further
comprises means (223) for modifying the filtering
characteristic of the pre-filter by convolution with a pulse-
shaped impulse response to inhibit lengthening of the
channel response of the communications signal by the pre-
filter (222) due to multipath structure in said signal.
2. Apparatus as claimed in claim 1 further comprising
filtering means which is located before the pre-filter and
which is arranged to cause the communications signal, at
the pre-filter input, to contain energy in a frequency region
beyond the Nyquist limit of that signal to cause aliasing in
the pre-filter that inhibits said lengthening.
3. Apparatus as claimed in claim 1 or 2, wherein the
modifying means comprises means in line with the pre-
filter for applying to the communications signal filtering
having a pulse-shaped impulse response to inhibit said
lengthening by said pre-filter.
4. Apparatus as claimed in any one of claims 1 to 3, wherein
said pulse is Gaussian.
5. Apparatus as claimed in any one of claims 1 to 4, wherein
the pre-filter is a linear equaliser.
6. A method of conditioning a communications signal
destined for equalisation by an equaliser, the method
comprising pre-filtering the communications signal prior
to the equaliser wherein the method further comprises
modifying the filtering characteristic of the pre-filtering
step by convolution with a pulse-shaped impulse response

to inhibit lengthening of the channel response of the
communications signal by the pre-filtering step due to
multipath structure in said signal.
7. A method as claimed in claim 6, further comprising
filtering the communications signal prior to the pre-
filtering step such that the communications signal, as
supplied to the pre-filtering step, contains energy in a
frequency region beyond the Nyquist limit of the signal to
cause aliasing in the pre-filtering step that inhibits said
lengthening.
8. A method as claimed in claim 6 or 7, wherein the
modifying step comprises a step of applying to the
communications signal in line with the pre-filter filtering
having a pulse-shaped impulse response to inhibit said
lengthening by said pre-filter.
9. A method as claimed in claim 7 or 8, wherein said pulse is
Gaussian.
10.A method as claimed in any one of claims 6 to 9, wherein
the pre-filtering step performs linear equalisation of the
communications signal.
11.A method of modifying the filtering characteristic of a
pre-filter (222) for an equaliser, wherein the pre-filter (222) is
arranged to apply a filtering characteristic to a
communications signal destined for equalisation by the
equaliser, wherein the method comprises providing a filtering
characteristic which is intended to be deployed in the pre-
filter (222) and modifying the filtering characteristic by
convolution with a pulse-shaped impulse response to inhibit
lengthening of the channel response of the communications
signal by the pre-filter when the filtering characteristic is
deployed in the pre-filter.

12.A method as claimed in claim 11, wherein said pulse is
Gaussian.
13. A method as claimed in claim 11 or 12, wherein said pre-
filter is a linear equaliser.
14. Apparatus for modifying the filtering characteristic of a
pre-filter (222) for an equaliser, wherein the pre-filter (222) is
arranged to apply a filtering characteristic to a
communications signal destined for equalisation by the
equaliser, wherein the apparatus comprises means (223) for
modifying the filtering characteristic by convolution with a
pulse-shaped impulse response to inhibit lengthening of the
channel response of the communications signal by the pre-
filter (222) when the filtering characteristic is deployed in the
pre-filter (222).
15.Apparatus as claimed in claim 14, wherein said pulse is
Gaussian.
16.Apparatus as claimed in claim 14 or 15, wherein said pre-
filter is a linear equaliser.

ABSTRACT

"Minimisation of perturbation of a radio signal between a
transmitter and a receiver by conditioning equaliser input"
Apparatus (212) for conditioning a communications signal
destined for equalisation by an equaliser, the apparatus (212)
comprising a pre-filter (222) located before the equaliser,
wherein the apparatus further comprises means (223) for
modifying the filtering characteristic of the pre-filter by
convolution with a pulse-shaped impulse response to inhibit
lengthening of the channel response of the communications
signal by the pre-filter (222) due to multipath structure in said
signal.

MINIMISATION OF PERTURBATION OF A RADIO SIGNAL BETWEEN A TRANSMITTER AND A RECEIVER BY CONDITIONING EQUALISER INPUT

Documents:

Inventors:

PCT Conventions: