GB2271489A

GB2271489A - Processing of digital communication signals

Info

Publication number: GB2271489A
Application number: GB9226484A
Authority: GB
Inventors: David John Hoyle
Original assignee: Roke Manor Research Ltd
Current assignee: Roke Manor Research Ltd
Priority date: 1992-10-07
Filing date: 1992-12-18
Publication date: 1994-04-13
Anticipated expiration: 2012-12-18
Also published as: GB9226484D0; GB2271489B

Abstract

The decision feedback sequence estimator is for use in a digital communication system which suffers from inter-symbol interference. The estimator 8 includes a trellis of signal paths between trellis states and the decision feedback sequence estimator is adapted to allow more than one sequence of the signal symbols to pass each trellis state. This allows the signal symbol sequences to intersect in the trellis without necessarily causing the elimination of one of the intersecting signal symbol sequences. Application is to mobile radios. <IMAGE>

Description

PROCESSING OF DIGITAL COMMUNICATION SIGNALS The present invention relates to digital communication systems and more specifically to such systems which, in service, suffer from inter-symbol interference. Inter-symbol interference arises in pulsemodulation systems whenever the effects of one transmitted pulse are not allowed to die away before the transmission of the next pulse. However, the invention also can be used to recover some of the performance lost from any reduced complexity near maximum likelihood sequence estimator in any channel where either the modulation alphabet size, or the channel memory length (caused by inter-symbol interference) are of a size such as to force a reduced state operating technique to be used.

In addition to inter-symbol interference, a mobile radio channel is corrupted by a number of effects, one of the most prominent being multi-path fading caused by the existence of a multiplicity of possible signal paths between the transmitter and the receiver of a signal. At the receiver the different delays caused by the different signal path lengths may add destructively, causing the received signal to fade. Also, if the receiver is in motion, then in addition, the fading of the received signal will vary in time. The range of delays that the different signal path lengths introduce to the transmitted signal is referred to as the multi-path delay spread of the signal channel concerned. If the multipath delay spread is large enough, more than one transmitted symbol affects the received signal at a given instant.Frequency selective fading, which affects the frequency spectrum of the received signal, will occur.

This is equivalent to time varying inter-symbol interference where the channel memory is defined by the multi-path spread.

It is known that an effective receiver for a signal which is impaired by inter-symbol interference consists of a whitened matched filter followed by a symbol rate sampler and a maximum likelihood sequence estimator.

Such a receiver is described in a paper entitled maximum Likelihood Sequence Estimation of Digital Sequences in the presence of Inter-symbol Interference" by G.D. Forney IEEE Transactions on Information Theory Vol IT-18 No. 3 May 1972. However, because of the arduous operating conditions of mobile radio communication systems, the complexity which would be required of the maximum likelihood sequence estimator makes such a receiver impracticable for use in mobile radio communication systems.

As a way of increasing the capacity utilization of a digital communications system, coded modulation can be employed, in which coding gain can be achieved without an expansion in the data bandwidth which is required for separate coding and modulation. A particularly suitable form of coded modulation for fading channels is trellis coded modulation, because of the ease with which soft decision information can be used in the decoding process.

However because of the delay that would be caused if interleaving were employed, the transceiver architecture in which this would be employed, must be used without interleaving. Furthermore, to reduce the complexity that would otherwise be required, the receiver will include a combined equalizer/decoder unit.

The principle of trellis coded modulation is that nk information bits are encoded into (m + 1)k code-bits, which are mapped in an optimum way by means of a mapping function onto a Q = 2 (m + 1)k - point signalling constellation. The mapping of the (m + 1)bit codewords onto the signal constellation points is optimised for the modulation scheme being used and the channel over which the information is being transmitted.Ungerboek in a paper entitled "Channel Coding with Multi-level/Phase Signals" published in IEEE Trans., Information Theory Vol IT-28 No. 1, Jun '82 disclosed a set of codes which were optimum for a signal channel which included additive white Gaussian noise insofar as they employed a process known as set partitioning which caused successive codewords to select signal constellation points which are separated by the maximum Euclidean distance in the signal vector space. The case when k = 1 corresponds to what is known as conventional trellis coded modulation, whereas the case when k > 1 corresponds to what is known as multi-dimensional trellis coded modulation, where the encoded symbols correspond to points in a 2k-dimensional Euclidean vector space. Such codes have been shown to provide an increase in coding gain when compared with conventional trellis coded modulation.The present invention is concerned primarily with conventional trellis coded modulation but it can be used with multidimensional trellis coded modulation systems and uncoded modulation systems.

The sampled signal at the output of a receiver filter, z, at time n, corrupted by inter-symbol interference and additive white Gaussian noise, is given by the equation

where hi = 0, 1, ...L are the coefficients of the channel impulse response of length L symbols and wn is a sample of additive white Gaussian noise, assuming that as part of a receiver filter the symbol sequence has been passed through a whitened matched filter. As mentioned above, the use of a full state maximum likelihood sequence estimator is impracticable for mobile communication systems.However, by effectively truncating the length of the signal channel memory from a value L to a lesser value M, a family of reduced complexity, although suboptimum, equalizers which operate on a reduced number of combined inter-symbol interference and code states can be created. Such equalizers are known as decision feedback sequence estimators. In operation such equalizers only cancel M symbols affected by inter-symbol interference.

In order to lessen the resulting reduction in performance, the L-M affected symbols which are not cancelled specifically are estimated and subtracted by a metric computation using information on past symbols taken from the history of paths taken by those symbols within a trellis which forms part of the decoder. In order to optimise the performance of any maximum likelihood sequence estimator, a delay must be allowed before determining the equalized symbols to ensure that there is a high probability that symbol sequence paths in the trellis within the estimator merge, at which point the survivor sequence is the maximum likelihood sequence.

In a decision feedback sequence estimator tentative decisions from the survivor paths in the trellis are used in the metric calculations for subsequent decisions on received symbols. As these decisions are made with a delay of between 1 and L symbols only, the metric calculations are made on decisions which have a high probability of being wrong. Since at each step in the trellis only one survivor symbol sequence is maintained, and as the decision as to whether or not to maintain a given symbol sequence depends on metric calculations which are highly likely to be based on an incorrect symbol sequence, there is a similar likelihood that the subsequent step will be wrong and error propagation will occur.

It is an object of the present invention in one aspect to provide an improved decision feedback sequence estimator and in another aspect to provide a radio communication system utilising such a decision feedback sequence estimator.

According to the present invention there is provided a decision feedback sequence estimator for use in digital communication systems which suffer from inter-symbol interference, wherein the decision feedback sequence estimator includes a trellis of signal paths between trellis states and the decision feedback sequence estimator is adapted to allow more than one sequence of signal symbols to pass each trellis state thereby to allow signal symbol sequences to intersect in the trellis without necessarily causing the elimination of one of the intersecting signal symbol sequences.

Preferably the signal symbol sequences are trellis coded modulated signals.

According to the invention in another aspect, there is provided a radio communication system including means for producing symbol sequences of bits representative of information, a transmission filter, a multi-path transmission channel linking the transmitter to a receiver consisting of a receiver filter, a sampler arranged to sample the output from the receiver filter and an equalizer/decoder for estimating transmitted symbol sequences in the presence of inter-symbol interference in the multi-path channel between the transmitter and the receiver, wherein the equalizer/decoder includes a decision feedback sequence estimator including a trellis of signal paths between trellis states and the decision feedback sequence estimator is adapted to allow more than one sequence of received signal symbols to pass each trellis state thereby allowing signal symbol sequences to intersect in the trellis without necessarily causing the elimination of one of the intersecting signal symbol sequences.

Preferably the means for producing the symbol sequences of bits representative of information includes means for applying trellis coded modulation to the information bits.

The invention will now be described and explained, by way of example with reference to the accompanying drawings in which Figure 1 is a block circuit diagram of a digital radio communication system embodying the invention, Figure 2 shows a model representing the combination of an encoder and multi-path signal channel included in the system of Figure 1, Figure 3(a) illustrates a trellis of states and signal paths between trellis states within a decision feedback sequence estimator embodying the invention and included in the system of Figure 1, Figure 3(b) shows a mapping of signal symbol sequences onto branches of the trellis shown in Figure 3 (a), Figures 4 and 5 show bit error curves for a decision feedback sequence estimator embodying the invention when used in conjunction with trellis coded modulated transmitted symbol sequences and Figure 6 shows bit error curves for a decision feedback sequence estimator embodying the invention when used in conjunction with uncoded transmitted symbol sequences.

Referring to the drawings, a digital radio communication system including an equalizer/decoder embodying the invention consists of a pulsed mode transmitter, shown generally by the numeral 1, a multipath signal channel 2 and a receiver shown generally by the numeral 3. The signal channel 2 is assumed to cause signals received by the receiver 3 to be subject to inter-symbol interference and additive white Gaussian noise.

The transmitter 1 includes an encoder 4 for (X,) information bits per signalling interval and a transmission filter 5. Within the encoder 4 are a convolution encoder 4 and a signal constellation mapper 4 . The receiver 3 includes a receiver filter 6, a sampler 7 for signal symbol sequences emerging from the filter 6 and a decision feedback sequence estimator equalizer/decoder 8.

In use, m information bits per signalling interval (Xn) are applied to the encoder 4 which produces m + 1 encoded bits Yn which are assigned a symbol an taken from a 2m+l - ary signal constellation in accordance with the coding rule used by the encoder 4 - in this case a trellis modulation code. The trellis coded modulated signal sequences are then filtered and transmitted as signal symbol sequences s(t).The received signal symbol sequences rc(t) having passed through the multi-path signal channel 2, and being degraded thereby by the effects of inter-symbol interference and additive white Gaussian noise are passed through the receiver filter 6 and then sampled at the symbol rate such that the sampled receiver output per symbol Zn is fed to the decision feedback sequence estimator equalizer/decoder 8 which produces an estimate (X,) of the transmitted signal symbol sequence.

For the estimation of the transmitted signal symbol sequence, a discrete time model may be constructed, which includes the encoder 4 and the filter 45. The filter 45 has a memory L where LT 2 Tn) which is the length of the signal channel 2 in symbols. This discrete time model is shown in Figure 2 in the form of a finite state machine.

The states of the discrete time model shown are given by the equation: n=(an-L,an-L+1,...an-1,#n) . . (2) where the symbol sequence (an-LI an-L+1...an-1} corresponds to a path which takes the encoder 4 from a previous state Gn-L to a present state Gn in compliance with the encoding rule of the convolution encoder 4 . The finite state model of the encoder 4 and channel 2 leads to a combined inter-symbol interference and code trellis.

Associated with each trellis code state are inter-symbol interference states. Hence, when the encoder 4 is an S-state trellis encoder in conjunction with a signal constellation having Q = 2m+l points, the combined code trellis has 8(Q/2)L states with Q/2 transitions emerging from each state.

As explained above, the decision feedback sequence estimator 3 effectively truncates the length of the channel 2 from the value L to a lesser value M. The states of the combined code trellis are now given by the equation: H and the decision feedback sequence estimator 8 estimates the L-M terms not represented in equation (3) and subtracts them by means of a computation using information on past signals taken from the path history of symbol sequences through the trellis of the decoder 3.

The decoder 3 computes the survivor path metric recursively according to the equation:

Where the 8,~1(11M n) signifies that the ISI terms due to symbols (a,~i), M+1 < i < L, not represented by the truncated trellis state Mn, are estimated using decisions taken from the path history associated with predecessor state RMn. As usual with the Viterbi algorithm, the minimisation of the metrics is taken over all trellis branches originating from states (C1M,) and leading to the successor state The least complicated form of equalizer/decoder would be that for which M = 0, and for convenience this case is considered in describing the present invention, although the invention has application for all values of M < L.In the case when M = 0, equation (4) reduces to:

Figure 3 illustrates the application of this equation to the decision feedback sequence estimator 8 embodying the invention in the receiver 3 of Figure 1 In Figure 3, the term p~ij denotes the metric associated with state i and path j.In the diagram, the operation p ; * S where S s {A,B,C,D}, the signal set, denotes the operation on the path metric of:

are are the trellis code modulated symbols, corresponding to the tentative decisions on the information symbols #Xn = {Xn-1, Xn-2, Xn-3 ... xn-M} taken from the store of past decoded symbols, and z, is the sample of the received symbol, and the hi is the channel estimate.

Referring to Figure 3 of the drawings1 consider a trellis of a states and m connections to other states.

Initially, each trellis state is arranged to have ss path stores for data and symbols. On each iteration along the trellis, mss symbol sequences will merge at each state of the trellis. The best ss metrics from the list so far computed and the symbol sequences associated with these metrics will be stored, in no particular order, at that state in its ss path stores. In Figure 3(a) a 4-state trellis code transmitting 4-phase shift keyed symbols is shown. Figure 3(b) shows the mapping of the 4-phase shift keyed symbols onto the branches of the 4-statetrellis code. In the interests of clarity, in Figure 3(a) only the branches of the trellis which are highlighted in bold characters are detailed. Also in Figure 3(a) the operation identified by an asterisk denotes the metric calculation of equation (5).In the trellis of Figure 3, the number of hypothetical signal paths ss is 2. These are identified by the sub-scripts a and b at each trellis state, together with the number of the relevant trellis state. The number of trellis states, a is 4 and the number of connections, m, is 2.

Each signal path through the trellis has its own metric and storage of decoded symbols and updates its metric with an appropriate new symbol for the next branch of the trellis. The number of metrics joining at each state of the trellis is mp, that is, 4. The two lowest metrics, highlighted in the Figure 3(a) in bold, and the signal paths they represent are then copied to the new paths on the next trellis state. The sort routine has to be called for each state. It must sort out the ss different metrics together with their associated path stores.They have to be different because the data associated with the same metrics, even though they are on different signal paths within the trellis, are the same, therefore to gain the diversity of signal paths required and to prevent duplication, the metrics have to be different. In the embodiment of the invention a bubble sort algorithm is used to sort the metrics.

A description of the algorithm is: 1. Initialise path metrics to different small numbers p~j~i=i.ele = 10-2.

2. flag = reset 3. index = 1 4. if p~i~j-p~(j-l < le-3 then p~i,j = pij+a2 5. if p-i-j < p-i-(j-1) then swap p-i-j & p-i-(j 1), set flag 6. index = index+l 7. goto 4 until index=mn 8. if flag = set goto 2 9. Process the lowest n path metrics p~i~j 10. end s2 is the noise power estimate.

Line 4 tests for the event of copies of paths in separate stores. The metrics are regarded as Euclidean distances and paths with too similar metric are diversified by incrementing them by the noise power.

This doesn't increase the likelihood of one path over the other significantly but gives enough metric diversity for the sort algorithm.

In the embodiment of the invention described the decision feedback sequence estimator 8 is adapted for the combined equalisation and decoding of trellis coded modulated signals but the invention can be used if the transmitted signals are modulated in an uncoded way. The argument given above applies equally, except that there is no relationship between one received signal and the next.In this case, given any one estimated symbol ân which can take on any of the Q = 2M symbols from the signal constellation the next symbol ân+lt can also take on any symbol from the constellation. (Note that in this case there are only Q = 2M points in the signal constellation instead of 2M+1). Hence the states of the finite machine model shown in Figure 2 would be given by the equation: (6) As each symbol can take on any value in the constellation, each subsequent state in the trellis can be reached from Q predecssor states. Apart from this, the arguments which lead to the concept of the decision feedback sequence estimator still are applicable and equations (1), (4) and (5) are applicable also.

In a trial study for a communication system according to Figure 1, the transmitter and receiver filters 5 and 6 respectively, were of root raised cosine type with a roll-off factor of P = 1.0. Analysis of the trellis-coded modulation tranceiver structure was carried out for the COST 207 typical urban channel, at a carrier rate of 1M symbols/sec; the power-delay profile of the channel taps at 0.0, 0.2, 0.5, 1.2, 2.3 and 5.0 s were 3, 0, -2, -6, -8 and -10 dB, respectively. A mobile speed of 50 km/h was used together with perfect channel estimation and frequency hopping, with a constant channel estimate across each transmitted burst.

Figures 4 and 5 show results for the case of a 1/2 rate code mapped onto a 4-phase shift keyed modulation scheme for 4- and 8-state codes respectively. For both Figures 4 and 5, three-bit error rate curves are presented for the case of 4-, 2- and 1- hypothesised path tracked per state. The results presented in Figure 6 are for the case of uncoded 4-phase shift keyed modulation, showing the effect of tracking 1-, 4- and 16-paths per state.

As can be seen from Figures 4, 5 and 6, the performance of the system improves steadily as the number of signal paths tracked is increased.

The most significant result shown is that of Figure 5 where, with an 8-state modulation code and 4hypothesized signal paths per trellis state, the intersymbol interference is removed showing that the bit error rate reduces at high signal to noise ratios.

The present invention can be used to enhance the performance of any reduced state equalizer of the decision feedback sequence estimator type where the states of the equalizer encompass only some truncation of the signal channel length M and the remaining (L-M) inter-symbol interference terms are cancelled by symbols from the signal paths of each of the states (j) where j E (M...L) are the symbols from the survivor stores at each state. Hence the invention is not restricted to the application of trellis coded modulation, but can be used in any case where inter-symbol interference has occurred.

Claims

1. A decision feedback sequence estimator for use in digital communication systems which suffer from intersymbol interference, wherein the decision feedback sequence estimator includes a trellis of signal paths between trellis states and the decision feedback sequence estimator is adapted to allow more than one sequence of signal symbols to pass each trellis state thereby to allow signal symbol sequences to intersect in the trellis without necessarily causing the elimination of one of the intersecting signal symbol sequences.

2. A decision feedback sequence estimator according to Claim 1 adapted to operate in conjunction with a signal channel the memory of which is reduced from a length of L symbols to a length of M symbols where M < L and residual inter-symbol interference is cancelled from L-M sequence decisions made within the decision feedback sequence estimator.

3. A decision feedback sequence estimator according to Claim 1 or Claim 2 adapted to operate upon trellis code modulated signal symbol sequences.

4. A decision feedback sequence estimator according to Claim 2 or Claim 3 adapted to represent only those states which exist in the trellis modulation code applied to the signal symbol sequences.

5. A radio communication system including means for producing symbol sequences of bits representative of information, a transmission filter, a multi-path transmission channel linking the transmitter to a receiver consisting of a receiver filter, a sampler arranged to sample the output from the receiver filter and an equalizer/decoder for estimating transmitted symbol sequences in the presence of inter-symbol interference in the multi-path channel between the transmitter and the receiver, wherein the equalizer/decoder includes a decision feedback sequence estimator including a trellis of signal paths between trellis states and the decision feedback sequence estimator is adapted to allow more than one sequence of received signal symbols to pass each trellis state thereby allowing signal symbol sequences to intersect in the trellis without necessarily causing the elimination of one of the intersecting signal symbol sequences.

6. A radio communication system according to Claim 5 wherein the transmission signal channel memory is reduced from a length of L symbols to a length of M symbols where M < L and residual inter-symbol interference is cancelled from L-M sequence decisions made within the decision feedback sequence estimator.

7. A radio communication system according to Claim 5 or Claim 6 wherein the transmitter includes an encoder adapted to apply trellis coded modulation to the information bits.

8. A radio communication system according to Claim 7 wherein the encoder comprises an S-state trellis encoder in conjunction with a signal constellation having Q = 2m+l signal points where m is the number of information bits per signalling interval of the transmitter.

9. A radio communication system according to Claim 7 or Claim 8 wherein the decision feedback sequence estimator equaliser/decoder is adapted to represent only those states which exist in-the trellis modulation code applied to the information bits.

10. A radio communication system according to any of Claims 7 to 9 wherein the modulation scheme is a phase shift keyed modulation scheme.

11. A radio communication system according to Claim 10 in which the modulation scheme is a 4-phase shift keyed modulation scheme.

12. A decision feedback sequence estimator substantially as hereinbefore described and with reference to Figure 3 of the accompanying drawings.

13. A radio communication system substantially as hereinbefore described and with reference to the accompanying drawings.