GB2407007A

GB2407007A - Adaptive Space Time Decoding using soft decisions

Info

Publication number: GB2407007A
Application number: GB0323669A
Authority: GB
Inventors: Magnus Sandell
Original assignee: Toshiba Research Europe Ltd
Current assignee: Toshiba Europe Ltd
Priority date: 2003-10-09
Filing date: 2003-10-09
Publication date: 2005-04-13
Anticipated expiration: 2023-10-09
Also published as: GB2407007B; GB0323669D0

Abstract

A symbol estimator 29 for a wireless receiver feeds soft values bextr back into an equaliser 19 via a statistical computing block 21 in order to output further improved soft decisions. The received signal r may be block coded via several transmitters, thus requiring iterative Space Time Block Code (STBC) decoding, or from a single transmitter via multiple paths. The equaliser uses Minimum Mean Squared Error (MMSE) correction, and the statistical block calculates the mean and covariance from the a priori information, comprising log likelihood values, supplied by the soft demodulator 23 and optionally also the Soft Input Soft Output (SISO) channel decoder 11. After 2 or 3 iterations, or once a certain reliability Ree is achieved, the soft bit estimates are output as final hard decision values via a deinterleaver (25).

Description

ESTIMATOR FOR MIMO RECF,IVER

Field of the Invention

This invention relates to apparatus, methods and computer program code for equalisation in communication systems in which one or more receivers receive signals from one or more transmit antennas.

Background of the Invention

In practical data communication systems, multipath within a channel results in intersymbol interference (ISI), which is often corrected with a combination of equalisation and forward error correction coding. For example a linear equaliser effectively convolves the received data with an inverse of the channel impulse response to produce data estimates with ISI substantially removed. An optimal equaliser may employ maximum likelihood (ML) sequence estimation or maximum a priori estimation (MAP), for example using a Viterbi algorithm. Where data has been protected with a convolutional code a soft input Viterbi decoder may be employed, usually together with data interleaving to reduce the effects of burst errors. Such approaches provide optimal equalisation but become impractical as the symbol alphabet size and sequence length (or equivalently channel impulse response length) increases.

Turbo equalisation achieves results which are close to optimal but with substantially reduced complexity compared to non-iterative joint channel equalisation and decoding.

Broadly speaking turbo equalisation refers to an iterative confidence building process in which "soft" information is exchanged between an equaliser and a decoder until a consensus is reached. "Soft" information is information relating to the likelihood of a particular bit, symbol or signal value as opposed to, for example, hard information resulting from a hard bit decision defining a bit as a logic one or zero. The effect of the channel response on the data symbols is treated similarly to an error correction code and typically a soft output Viterbi algorithm (SOYA) is used for both. Again, however, such techniques are impractically complex for large delay spreads and symbol alphabets, particularly as several processing iterations may be needed to achieve convergence for a single data block. These difficulties are significantly exacerbated where signals from more than one transmit antenna must be disentangled and equalised, with a different channel response for each transmit antenna or transmit-receive antenna pair.

A paper by Tuchler et al. (Minimum Mean Squared Error Equalization Using A-priori Information, Michael Tuchler, Andrew Singer, Ralf Koetter, IEEE Transactions on Signal Processing, vol. 50, pp. 673 - 683, March 2002) describes a simplified approach to turbo-equalisation where a single transmit antenna is employed. In this paper the conventional MAP equaliser is replaced by a linear equaliser (that is by a linear or transversal filter) with filter coefficients which are updated using a minimum mean square error (MMSE) criterion evaluated over both the distribution of noise and the distribution of symbols. A linear estimate En of a transmitted symbol an is determined using an observation Zn via the equation in= anH Zn + En where superscript H denotes the Herrnitian operator and an and bn are the coefficients of the estimator (strictly the estimate should be termed affine rather than linear because of the constant ten). The coefficients are chosen to minimise the MSE cost E(lxn - En l2) where E( ) denotes a mean or expectation value. As information is fed back to the equaliser from the error correction decoder the filter coefficients change with time and are thus recomputed for each data symbol to be estimated.

A related technique is described in WO 02/15459, which is implemented in a data communications system, and provides an estimator having a softinput-soft-output (SISO) MMSE equaliser that is modified to accept a priori information from a SISO decoder coupled to its output, and thus implement an iterative decoder. The additional information provided by the channel decoder (which knows the forward error coding FEC utilised) is used by the estimator to improve its symbol estimates. However iterative decoding is complex to implement and processor intensive.

In multiple transmit antenna arrangements a MIMO (multiple-input-multipleoutput), multiple receiver antennas, or a MISO (multiple-input-singleoutput), single receive antenna, channel is provided. In such arrangements different symbols are transmitted from the transmit antennas at given times, resulting in the receive antenna(s) receiving a combination of these transmitted symbols which interfere with each other - known as multi-stream interference (MSI). In a manner analogous to an equaliser for an ISI channel, an equaliser for a MIMO or MISO channel effectively convolves the received data with an inverse of the channel impulse response to produce data estimates with MSI substantially removed. In the case where no ISI is assumed, such equalisers are typically known as space-time decoders. Where ISI is assumed, the equaliser is known as a MIMO equaliser and effectively has a more complex inverse channel response to convolve with the received data symbols.

Summary of the Invention I

In general terms in one aspect there is provided an estimator comprising and equaliser and which uses internal feedback to improve its estimate of a received symbol or bit (s).

A soft value(s) for the estimated received bits of the received symbol is used as a priori information for the equaliser in a subsequent iteration of outputting received soft bit estimates. After a predetermined number of iterations the soft bit estimates are outputted, and may be converted directly into hard bit estimates for use in the rest of the receiver, or used as an input for a soft-input channel decoder.

This internal feedback arrangement improves performance while reducing the complexity of the decoder architecture. It allows the usual iterative structure of a turbo decoder or similar to remove the SISO decoder feedback, without reducing the performance of the equaliser, but whilst significantly reducing its complexity. In addition, this approach provides an arrangement which doesn't rely on a channel decoder, for example in an un-coded system.

The term equaliser is used in the specification to refer to a device or process which receives data or symbols and multiplies or convolves these with an equaliser waveform or response, for example this might be the inverse of the estimated channel through which those data or symbols have travelled in some equalisers.

The term estimator is used in the specification to refer to a device or process which comprises an equaliser, and in addition provides the equaliser with the necessary information to perform its convolving operation on the incoming data. For example for a MMSE equaliser, the estimator provides mean and co-variance information on the received symbols. Where necessary, the estimator also provides a demodulation function to provide estimated for the received data bits. Estimators are also known in the art as decoders, for example space-time decoders in the case of a MIMO channel without ISI. Various other terminology may be used in the art, however the term estimator is used in this specification to encompass any device or process having the same function or operation.

In particular according to a first aspect, the present invention provides an estimator for a wireless communications receiver; comprising: means for equalising the received symbols in order to produce soft bit estimates from the symbols; wherein the equalising means is arranged to accept a priori information in order to generate statistical data for said equalisation; wherein said a priori information is provided from the output of the equahsmg means.

Preferably the equalising means is a soft-input-soft-output equaliser for providing soft symbol estimates and coupled to a soft demodulator for providing soft bit estimates from the soft symbol estimates.

Preferably the equalising means is a minimum mean squared error (MMSE) equalising means.

Preferably the estimator further comprises a statistics computation block for computing the mean and co-variance of the soft bit estimates which correspond to said a priori information.

Preferably the a priori information are log-likelihood values of the bit estimates.

There is also provided a receiver for a wireless communications system comprising an estimator according to the definitions above.

Preferably the receiver further comprises a channel decoder coupled to the output of the estimator.

The receiver may be for a MIMO, MISO, or SISO channel. The receiver may also be used in a channel with out without ISI, an example of the latter being in an orthogonal frequency division multiplex (OFDM) communications system.

There is also provided a communications system comprising a receiver according to the above definitions.

In particular in another aspect there is provided a method of estimating received signals in a wireless communications receiver; comprising: equalising received symbols in order to produce soft bit estimates from the symbols; wherein the equalising comprises generating statistical data for said equalisation from a priori information; wherein said a priori information is provided by the soft bit estimates.

Preferably the equalisation is a minimum mean squared error (MMSE) equalisation.

Preferably the equalization further comprises computing the mean and covariance of the soft bit estimates.

There is also provided a method of receiving signals comprising estimating the signals according to the estimation methods defined above and channel decoding the output of the estimating method.

There is also provided processor control code for implementing a method according to the above definitions. This is preferably provided on a carrier medium such as a CD ROM or a transient carrier such as an Internet download signal. The code may also be provided on a DSP or similar platform which forms part of a receiver for implementing the defined methods.

In particular in another aspect there is provided an estimator for a wireless communications receiver; comprising: a soft-in-soft-out equaliser for estimating received symbols, and coupled to a soft-in-soft- out demodulator for estimating bits from the symbols; wherein the equaliser arranged to accept a priori information to assist said symbol estimation; the estimator further comprising an internal feedback path from the output of the demodulator to the a priori input of the equaliser such that said a priori information is generated from said estimated bits.

Preferably there is a statistics computation block for computing the mean and co- variance of the estimated demodulated bits which correspond to said a priori information.

A receiver comprising this estimator may also comprise a hard output discriminator coupled to the output of the estimator for providing hard output bits.

Alternatively a channel decoder coupled to the output of the estimator. The channel decoder is preferably a soft-in-soft-out decoder, and the output of said decoder is fed back and added to the soft demodulator feed back output to generate the a priori information for said equaliser.

Brief Description of the Drawings

Embodiments of the invention will be further described with reference to the accompanying drawings, by way of example only and without intending to be limiting, in which: Figure I is a block diagram of a known art communications system; Figure 2 is a block diagram of a known MMSE estimator block; Figure 3 is a block diagram of an MMSE estimator block according to a first embodiment; Figure 4 is a block diagram of a communications system according to an embodiment; Figure 5 is a block diagram of a communications system according to another embodiment; Figure 6 is a graph showing the performance of an embodiment.

Detailed Description

The embodiments are described with respect to a MIMO channel without ISI, for as an OFDM (orthogonal frequency division multiplexed) based MIMO system. Estimators for such applications are also known as STBC (space time block code) decoders, and so these terms will be used interchangeably in this description. Ilowever, the described estimators can also be used in single antenna systems in multi-path propagation environments and so suffering from ISI. Similarly, the described estimators can be used in a MIMO system with ISI, such as a non-OFDM based system with multiple antennas at the transmitter (and receiver). Finally, the estimator may also be implemented as a multi-user detector in which the interference affecting the channel matrix is from other users. In each case, an equaliser response or waveform (for example the inverse channel response) is still calculated and applied to incoming symbols in order to provide bit estimates. These calculations may be more complex where multiple interference sources are effecting the channel matrix H. An optimal STBC decoder or estimator uses the A Posteriori Probability (APP) algorithm. However its complexity grows exponentially with the data rate, so it becomes impractical with a large number of antennas. Several suboptimal techniques have been proposed based on the APP and recently constellation-independent algorithms, such as the MMSE, have been investigated as described above.

Iterative space-time block code (STBC) and channel decoding, (also known as turbo decoding) can provide excellent performance. Since the optimal soft-in, soft-out STBC decoding complexity grows exponentially with the data rate, numerous sob-optimal techniques have been proposed. One of them is the minimum mean-squared error (MMSE) estimator that can be modified to accept a priori information from a channel decoder, and thus be possible to use in iterative decoding.

Normally a priori information is supplied by a channel decoder, thus providing new information for the STBC decoder.

The baseband equivalent signal model is r = Hx + v where r is the received signal, H is the channel matrix, x the transmitted signal and v the noise. As noted above, for the described embodiments, H represents the channel matrix affecting the transmitted symbols here the symbols interfere with each other in space. However in further embodiments for example in a non-OFDM single-input-single-output antenna system in a practical multi-path environment, the symbols interfere with each other in time (ISI).

Further embodiments may implement non-OFDM based MIMO or MISO systems in which the symbols interfere with each other in both space and time (MSI and ISI). In this case the same model holds, however the channel may well be more complex to estimate and may therefore require extra processing resources.

There follows a mathematical description of an MMSE estimator with a priori information. With a priori information available, the vector x is no longer zero-mean.

The MMSE estimate is XMMSE =I1+(H H+NoR J (y-H) where = E{x}, R = El(x Fax - Il)H 11 and E{vvH}= No I. However, in iterative decoding it is important to use the extrinsic output. This is the estimate of symbol m when using no a priori information for that symbol. Hence to estimate symbol m, we must use the mean and covariance matrices Him) = - Them it(m) = R - (E)Xm - /Um | i- Ex imemH, where emis the m:th unit vector (all zeros except a one in position m) end Exis the variance of xm without any a priori information.

The extrinsic estimate for symbol m now becomes X(mX7 =(m)+(HHH+NoR(m/) HH(y-H,(m)). Note that the covariance matrix is dependent on m. This means that for every symbol we wish to estimate, a new inverse will have to be calculated since the covariance matrix changes. By using the matrix inversion lemma, the estimate can be written as x(rm)r = 'I(m) + Ym(HHH + NoR) HH (y-H'l(m)) = Il,(m) +ymW-'HH (y - Hp(m)) where Ym = E R. -N (E -R. OFT-' Since this estimate is only valid for symbol m, we x m,m O x m,m m, m get xm = em x( ) = Ymem W H (y-H,( )), as em p( ) = 0. Note that W is independent of m and only has to be computed once. In fact this is more or less the "standard" MMSE estimator so very few changes are needed for it to accept a priori information.

The error variance for symbol m can be expressed as E|Xexr m -Xm| i= NOWmimYm This variance is needed when demodulating the symbol estimates to bit estimates.

Performance can be sensitive to this parameter so it is important to estimate it as accurately as possible.

An MMSE estimator accepting a priori information is shown in Figure 1. A communications systems 1 comprises an encoder block 13, a channel model block 15 and an iterative decoder block 17. A data signal a is encoded by a FEC encoder 3 as a bit stream b which is interleaved by an interleaver 5 and transmitted as a modulated signal x. For example, a binary, quadrature or other phase shift keying or other modulation scheme could be employed. The signal x is transmitted over a channel 7, which distorts the signal by introducing ISI and which adds noise. The signal is received at a receiver as r = Hx + v, where H is the channel matrix, and where v is the AWGN component, and fed into the MMSE estimator 9.

The structure of the MMSE estimator 9 using a priori information is shown in Figure 2.

The received signal r is one of the inputs into an MMSE equaliser 19, which also receives some additional information about the symbol sequence in the form of soft information from a compute statistics block 21. The compute statistics block 21 computes the mean and the covariance R of the transmitted symbols by analysing the a priori information La produced by the SISO decoder 11. With this information, the MMSE equaliser 19 delivers soft estimates xexr of the transmitted symbols and the estimation error covariance Ree (the reliability of the symbol estimate). These two values are passed to the soft demodulator block 23 which converts the symbol estimate (e.g. QPSK) to a soft bit estimate (e.g. 00), boxer.

Returning to Figure 1, the soft bit estimate is deinterleaved by the deinterleaver 25, whose characteristics are the inverse of the interleaver 5, and presented to the SISO decoder 11. The SISO decoder l l produces a soft output which is an estimate of the coded bits be. This output is then interleaved by the second interleaver 27 and fed back to the SISO MMSE estimator 9 as a priori information, La. The iterative process continues until the SISO blocks 9, l l agree or converge sufficiently, or until a fixed number of iterations are performed. The combination and convergence test block 29 determines whether the predetermined iteration termination criterion is satisfied, and if the criterion is satisfied the soft input to the SISO decoder l l is combined with the soft output of the SISO decoder ll. If necessary a hard decision is generated for the combined soft value.

On the first pass through the iterative decoder block 17, when no a priori information is available from the SISO decoder l l, initialization data is used, until the SISO decoder l l produces a soft output.

Figure 3 is a block diagram of the MMSE estimator 29 according to an embodiment.

As with the estimator 9 of figure 2, the estimator 29 of figure 3 comprises an MMSE equaliser 19 which receives signals r from the channel 7, and have estimated symbol xenon and reliability outputs Ree coupled to a soft demodulator 23, which in turn outputs soft bit estimates bear. As described above, the compute statistics block 21 computes the mean and the covariance R of the transmitted symbols. However the a priori information used to calculate this is derived from the feedback path 30 from the output of the soft demodulator to the statistics computing block 21. The soft bit estimate extrinsic output flexor is fed back into the compute statistics block 21, instead of the soft output from a SISO decoder such as La in figure 2. The operation of the equaliser l9 and compute statistics block is as described above with respect to figure I and 2.

It has been shown that this internal feedback improves the performance of the estimator 29. This is despite the lack of additional information from the channel decoder 11.

Although, in general, internal feedback does not provide any additional information to improve an earlier estimate, as the MMSE equaliser 19 is a sub-optimal equaliser, the estimate generated can be improved by using internal feedback as an a priori probability of the transmitted bit.

The performance is sufficiently improved that no iterative feedback from a channel decoder (11) is necessary which considerably simplifies the implementation of the decoder block, as is illustrated in figure 4.

Figure 4 is a block diagram of a communications system l incorporating the MMSE estimator block 29 of figure 3. The feedback loop of figure I from the SISO decoder 11 has been removed and all iterations are performed within the MMSE estimator block 29. Note however that in certain cases there may be an advantage to maintain the feedback from the SISO decoder 11, as indicated by the dashed-line input La to the compare statistics block 21 in Figure 3. For example if there is a soft output from a channel decoder available, then this can be fed back to provide a further improvement in the estimator's performance.

The data bits to be transmitted are encoded and interleaved as they are in the prior art system I of Figure 1 in order to provide a robust communications system. The iterative MMSE estimator 29 is most likely to be deployed when transmitting over poor channels where a robust communications system is required, hence it is expected that block coding and interleaving will additionally be deployed; although this is not essential.

The estimator 29 receives incoming signals from the receive antenna which have been effected by the channel as described above. A number of feedback iterations are performed on each received signal sample. The equaliser 19 generates a first estimate xe'r' which is then demodulated by the soft demodulator to obtain the bit estimates bear for the estimated symbol xexr. The statistics (mean 11 and covariance R) fed to the a priori information input for the equaliser 19 are based on initialization values for the first iteration, and are then calculated by the compute statistics block 21 from the soft bit estimates bear on subsequent iterations.

After a predetermined number of iterations, or after a predetermined reliability value Ree for the symbols estimates is reached (or both), the soft bit estimates bear for one of the received signal samples r is provided at the output of the estimator 29. Preferably 2 or 3 iterations are performed for each sample r.

As with the arrangement of figure 1, the soft bit estimate output of the estimator 29 is fed to a deinterleaver 25 which deinterleaves a sequences of estimated bits in an inverse fashion to the interleaving performed by interleaver 5 on the transmission side (13) of the system. The deinterleaved bit estimates are then fed to a SISO decoder 11 which generates decoded soft bit estimate outputs, which can then be fed to a hard output discriminator for conversion to hard bits for use by the rest of the receiver.

Alternatively, the soft bit estimate output of the SISO decoder 11 may be fed back as a priori information to the estimator 29, to be added to the fed back demodulated bit estimates bear as shown in figure 3. In a further alternative, the decoder may be a soft- input hard-output type.

Whilst a priori information from a decoder can be added to the fed back information to further improve performance, a further advantage of the estimator 29 is that it can be used in a communications system in which there is no channel decoder. Whilst a coded system is preferred for robust communication over wireless channels, a simpler embodiment can also be implemented using no coding (nor interleaving), as shown in Figure 5. The data bits are modulated as a BPSK signal which is transmitted over the ISI channel 7 which also introduces AWGN. The MMSE estimator 29 outputs a soft estimate boxy of the transmitted bits which is then converted directly to a hard estimate of the data bits by a hard output discriminator block 31.

Figure 6 shows simulation results for a 4 Tx, 4 Rx antenna system with QPSK modulation and a rate 'Liz convolutional code (polynomials 5 and 7) on a slow fading channel. The results show bit-error-rate verses signalto-noise ratio, and as can be seen the performance improvement is significant.

The improvement is particularly significant for weaker or less robust channel codes such as convolutional codes with a short constraint length or a high rate Turbo code.

The gains can be in the order of l OdB, for example simulations show a 1 2dB improvement for 4/5 rate Turbo code.

In the case of BPSK the soft demodulator may be replaced with a function block to take the real part of the soft symbol estimate xenon and scale it (for example according to the SNR), in order to provide a soft bit estimate bear. The compute statistics block 21 then still has the loglikelihood values it needs to calculate the mean u and covariance R values needed for the MMSE equaliser.

In further alternative arrangements, other sub-optimal equalisers may be deployed instead of the MMSE equaliser, to achieve improved performance. In this case a modified statistics compute block will be required if statistics other than the mean u and covariance R of the transmitted symbols are required.

As already mentioned, in further embodiments, the modified estimator with internal feedback may also be applied in ISI channel applications with or without multiple transmit antennas.

Whilst the internal feedback arrangement has been described with respect to an estimator, it can alternatively or additionally be applied to other parts of the receiver, for example the channel decoder block.

The skilled person will recognise that the above-described apparatus and methods may be embodied as processor control code, for example on a carrier medium such as a disk, CD- or DVD-ROM, programmed memory such as read only memory (Firmware), or on a data carrier such as an optical or electrical signal carrier. For many applications embodiments of the invention will be implemented on a DSP (Digital Signal Processor), ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array). Thus the code may comprise conventional programme code or microcode or, for example code for setting up or controlling an ASIC or FPGA. The code may also comprise code for dynamically configuring re-configurable apparatus such as re- programmable logic gate arrays. Similarly the code may comprise code for a hardware description language such as Verilog _ or VHDL (Very high speed integrated circuit Hardware Description Language). As the skilled person will appreciate, the code may be distributed between a plurality of coupled components in communication with one another. Where appropriate, the embodiments may also be implemented using code running on a field-(re)programmable analog array or similar device in order to configure analog hardware.

The skilled person will also appreciate that the various embodiments and specific features described with respect to them could be freely combined with the other embodiments or their specifically described features in general accordance with the above teaching. The skilled person will also recognise that various alterations and modifications can be made to specific examples described without departing from the scope of the appended claims.

Claims

CLAIMS: 1. An estimator for a wireless communications receiver;

comprising: means for equalising the received symbols in order to produce soft bit estimates from the symbols; wherein the equalising means is arranged to accept a priori information in order to generate statistical data for said equalisation; wherein said a priori information is provided from the output of the equalising means.
2. An estimator according to claim 1 wherein the equalising means is a soft-input- soft-output equaliser for providing soft symbol estimates and coupled to a soft demodulator for providing soft bit estimates from the soft symbol estimates.
3. An estimator according to claim 1 or 2 wherein the equaliser is a minimum mean squared error (MMSE) equaliser.
4. An estimator according to claim 3 further comprising a statistics computation block for computing the mean and co-variance of the soft bit estimates which correspond to said a priori information.
5. An estimator according to any one preceding claim wherein the a priori information are log-likelihood values of the bit estimates.
6. A receiver for a wireless communications system comprising an estimator according to any one preceding claim.
7. A receiver according to claim 6 further comprising a channel decoder coupled to the output of the estimator.
8. A receiver according to claim 6 or 7 wherein the estimator is for a multiple transmission antenna communications system.
9. An orthogonal frequency division multiplex (OFDM) receiver according to claim 6, 7 or 8.
10. A method of estimating received signals in a wireless communications receiver; comprising: equalising received symbols in order to produce soft bit estimates from the symbols; wherein the cqualising comprises generating statistical data for said equalisation from a priori information; wherein said a priori information is provided by the soft bit estimates.
11. A method according to claim 10 wherein the equalisation is a minimum mean squared error (MMSE) equalisation.
12. A method according to claim 10 or 1 1 wherein the equalisation further comprises computing the mean and co-variance of the soft bit estimates.
13. A method according to any one of claims 10 to 12 wherein the a priori information are log-likelihood values of the bit estimates.
14. A method of receiving a signal comprising estimating said signals according to the method of any one of claims 10 to 13 and channel decoding the output of the estimating method.
15. A method according to any one of claims 10 to 14 wherein the signals are orthogonal frequency division multiplex (OFDM) signals.
16. A method of receiving signals in a multiple transmission antenna communications system, the method comprising a method according to any one of claims 1 0 to 15.
17. Processor control code for implementing a method according to any one of claims 10 to 16.
18. A carrier medium carrying processor control code according to claim 17.