NO316488B1 - Method and apparatus for receiving digital communication signals - Google Patents

Description

Field of the invention

The present invention relates to the reception of digital communication signals sent over an air interface, using iterative channel estimation, and iterative smoothing and decoding, also known as turbo smoothing. It is particularly suitable for baseband processing in a receiver in a high-frequency (HF) modem The following description will focus on such use. However, the inventive solutions can also be used in other systems based on wireless transmission links, for example in mobile communication systems.

Technical background

Communication in the HF band (3-30 MHz) offers one of the few alternatives for communication beyond visual range without the use of existing infrastructure or satellites However, there are physical and regulatory limitations on the data transfer rates that can be achieved in HF systems When it comes to the last factor is bandwidth - it is limited to 3KHz due to frequency allocations Most HF systems are based on signals reflected from the ionosphere, which vary over the day and season and with the sunspot cycle To a certain extent these variations can be predicted, but the ionosphere degrades also the signals, which makes it difficult to achieve high-speed digital communication This is particularly the case with HF communication at high latitudes due to auroral effects Rapid fluctuations in the ionosphere are the cause of Doppler spread (fading), multiple transmission paths lead to delay spread , also known as multipath, which introduces intersymbol interference, and the aurora borealis b-sorption reduces the received signal to noise ratio

(SNR for signal-to-noise ratio)

In this invention we primarily deal with intersymbol interference (ISI) In a signal that experiences such interference, neighboring symbols will interfere with each other, which causes difficulties for receivers trying to detect the individual symbols and read the information content.

The relevant communication modes are phase shift keyed modulation (PSK) or quadrature amplitude modulation (QAM). Particular reference is made to the standards STANAG 4285, STANAG 4539 and MIL-STD-188-110B

A number of techniques have been introduced to improve the reliability of data transmission. These include error-correcting coding, interleaving, and multiplexing training sequences into the transmitted data stream.

On the receiving side, smoothing is used to combat the defects in the transmission channel. Currently, standard receivers use decision feedback equalizers (DFE for decision feedback equalization) with non-linear processing for this task. A channel estimation process is used to estimate the channel impulse response, which is used to calculate the coefficients of the equalizer. The channel estimation process uses known symbols sent over the channel, i.e. training sequences, to estimate the channel impulse response Between training sequences, tentative decisions from the equalizer can be used to follow the time-varying channel Alternatively, the channel impulse response between training sequences can be estimated by interpolation

In recent years, iterative smoothing and decoding, or turbo smoothing, has been proposed as a powerful method for receiving ISI-disturbed data In a receiver that uses turbo smoothing, the received signal is first passed through an equalizer to reduce ISI A de-interleaver and a channel decoder follows the equalizer The decoder emits soft information about the code bits, which is fed back through an interleaver to the equalizer Then a further round of equalization is performed on the same received signal, using soft bits returned from the decoder as a priori information. When soft information is repeated between the equalizer and the decoder several times, the error rate of the data bits from the decoder will generally decrease

In a receiver with turbo equalization, the equalizer and decoder are Soft-in Soft-out (SISO) modules. The optimal SISO equalizer is a trellis-based "maximum a priori probability" (MAP) equalizer which is too complex for HF modems when the channel impulse response is long and/or the signal constellation is larger than 2-PSK. Suboptimal SISO equalizers based on linear filters and soft ISI cancellation are a good alternative when MAP equalization is too complex (M. Tuchler et al "Minimum Mean Squared Error Equalization Usmg A priori Information", IEEE Trans.Sig.Proe , vol 50, no 3, pp. 673-683, Mar 2002).

Known technique

The use of turbo equalization in a receiver for digital communication signals was first described in C. Douillard et al. "Iterative Correction of Intersymbol Interference Turbo-Equalization", European Trans. Telecommun, vol 6, no 5, pp. 507-511, Sept -Oet. 1995. This method uses a trellis-based equalizer, which is too complex to be used in the applications considered here. A reference discussing the relative performance of different trellis-based equalizers can be found in G Bauch & V Franz. "A comparison of soft-m/soft-out algorithms for Turbo-detection", Proe. 5th Int Conf. on Telecommunications, pp 259-263, June 1998

Later solutions have chosen a suboptimal method using SISO equalizers based on linear filters, for example as described in A Glavieux et al.: "Turbo equalization over a frequency selective channel," Proe Int Symp on Turbo codes, Brest, France, pp. 96-102, Sept 1997 The articles mentioned so far concern turbo equalization, that is, the problem of channel estimation to find the optimal coefficients for the equalizer is not considered

INN Nefedov & M. Pukkila: "Turbo Equalization and Iterative (Turbo) Estimation Techniques for Packet Data Transmission," 2nd Int Symp. on Turbo Codes & Related To-pics, Brest, France, pp 423-426, 2000 a method is described which involves both a turbo smoothing loop and a separate loop for estimating the channel impulse response The last loop includes a channel estimator which receives hard decisions from the decoder Initially the parameters are estimated from a training sequence, but later the estimate is updated from mformation symbols However, this is a step-by-step process, where the parameters are updated and locked to the current values during a complete frame

From P. Strauch et al.: "Iterative Channel Estimation for EGPRS," IEEE 52nd Vehicular Tech Conf , Boston, USA, vol 5, pp. 2271-2277, Sept 2000, a conventional (non-Turbo) receiver is known where the channel estimation is formed based on hard feedback from either the detector or the decoder between training sequences. The authors point to the possibility of using soft feedback to extend the training sequences

A number of methods are known from the patent literature which describe the use of a basal turbo feedback loop for the detection of communication signals in cellular communication systems, for example GB application 2 354 676, EP application 959 580 and EP application 948 140 However none of these documents consider the question of how the smoothing parameters should be estimated

Concise summary of the invention

There is a purpose to it. present invention to provide a method for an arrangement in a receiver with turbo equalization for receiving digital communication signals, with an improved ability to handle intersymbol interference

This is achieved in an apparatus according to the attached claim 1 and a method according to claim 8. The invention is applicable in a receiver that uses a SISO equalizer based on soft ISI cancellation and a linear filter. According to the mventive method, the coefficients of The SISO equalizer updated in each symbol interval using a time-varying channel estimate and a priori information on the code bits

The scope of the invention will be apparent from the attached patent claims.

Brief description of the drawings

The invention will now be described with reference to the attached drawings, where

Figure 1 shows a schematic presentation of the coding process used for PSK data transmissions, Figure 2 is a block diagram showing the principle of Turbo equalization, Figure 3 shows a corresponding receiver that uses turbo equalization according to the present invention. Figure 4 is a graph showing the improvement in signal-to-noise ratio that can be achieved with the present invention.

Detailed description of the invention

Figure 1 is a schematic representation showing the code steps in a transmitter that encodes an incoming data stream into a PSK signal

The incoming data bits are encoded using an error-correcting code, in an encoder 1 The code has the ability to correct errors that occur in individual symbols that are randomly scattered However, interference will often appear as bursts of noise (eg static energy) , which will affect several adjacent symbols Fading will also cause several adjacent symbols to have a low received signal level Scan error patterns are not easy to handle for the error correction code To avoid this problem the signal is interleaved in the interleaver 2

In the next step, the data bits are modulated into data symbols in the symbol mapper 3 (symbol mapper) Next, training sequences containing known symbols are multiplexed into the data stream at 4 The training sequences are repeated in each block of data frames

The stream of modulated data symbols is further handled by the transmitter circuit, delivered to an antenna and broadcast on the air In the transmission channel, noise is added and the symbols distorted due to multipath fading The fading is a frequency-selective time-varying phenomenon, or expressed in another way, will the channel's impulse response constantly changes

The following description assumes that the received baseband signal is sampled once per symbol interval. The invention can also be modified to use several samples per symbol, so-called oversampling

Figure 2 shows the general principle of a receiver with turbo equalization Symbols from a radio receiver (not shown) are fed to a soft-in soft-out (SISO) equalizer 200 Soft information at the input and output of a SISO module takes the form of logarithmic probability ratios ( LLRs for Lo-ganthmic Likelihood Ratios) on Code Bit Ten, L( ct

Further information on "Logarithmic Likelihood Ratios" is available in J. Hagenauer et al "Iterative decoding of binary block and convolutional codes," IEEE Trans on Information Theory, pp 429-445, March 1996

The LLRs out of the equalizer are de-interleaved at 201, and decoded in the SlSO decoder 202 (for the error correction code) The LLR values from the decoder are obtained as intrinsic information, i.e. the information content of the input soft information, and extrinsic information, which is information generated from the redundancy in the code The extrinsic information is obtained by subtracting the LLRs at the input of the decoder from the corresponding LLR values at the output of the decoder in the subtractor 203. The resulting LLRs are fed back to the equalizer 200 through an interleaver 204 The interleaver 204 is included to produce signals that can be compared with the interleaved signals in the equalizer 200 (since the equalizer 200 is upstream of the de-mterleaver 201) The feedback information is used as a priori information in a new attempt at equalization and decoding on the same set of data - this is the first iteration After a few iterations the error rate will improve significantly In this way smoothing becomes and decoding performed repeatedly for a fixed number of iterations, or until a termination criterion stops the iterative (Turbo) process

The turbo process described so far depends on the channel impulse response being known to the equalizer A separate channel estimation process (not shown in figure 2) is therefore necessary to provide an estimate of the channel impulse response

Figure 3 shows a receiver with turbo smoothing according to the present invention. The turbo loop comprises a SISO equalizer 300, a demterleaver 301, a SISO decoder 302, a subtractor 303 and an interleaver 304

The SISO equalizer used here is based on soft ISI cancellation followed by a linear filter. The LLRs fed back from the decoder are converted to an a priori expected value (mean value) of the transmitted symbols, from which an expected value of the intersymbol interference can be calculated. Soft ISI cancellation subtracts this expected value of the intersymbol interference from the received signal. The coefficients of the subsequent linear equalizer filter are calculated to minimize the expectation of the squared error (minimum mean square error criterion, MMSE for minimum mean square error) of each symbol emitted from the equalizer, given the a priori information and the estimated channel impulse response The input of the SISO equalizer is described in detail in Tuchler 2002, but the procedure has been modified to incorporate a time-varying channel estimate The new procedure is described in the appendix.

Extrinsic LLRs for each code bit are calculated from the output symbols from the equalizer using the formula in Tuchler 2002

Other SISO equalizers may be used within the scope of this invention

In addition to the turbo feedback loop, a further channel estimator loop has been introduced. This outer loop includes a further interleaver 306 connected directly to the SISO decoder 302. This will produce the total LLR output from the decoder, including the intrinsic information These LLRs is converted to soft symbols (the expected value of each symbol given the LLRs) using the formula in Tuchler 2002 In addition, a hard decision assigner 308 is introduced, which provides a hard output signal from the denominator. Furthermore, the signals are routed using two switches Sl and S2.

The channel estimator uses the transmitted and received symbols to provide a time-varying estimate of the channel impulse response. The transmitted symbols are known during training sequences. Outside the training sequences, estimates of the unknown information symbols must be used. For initial equalization, hard symbols directly from the equalizer output are used as estimates {in this case the channel estimate must be delayed) In subsequent iterations, soft symbols calculated from the decoder output are used Switch S2 is used to switch between training symbols and estimated symbols, and switch Sl is used to switch between hard symbols from the equalizer and soft symbols from the decoder The turbo loop is still used for detection of the information symbols In this way the channel estimate will also improve for each iteration

Thus, the invention involves a two-stage channel estimation and equalization process comprising the subsequent individual steps

1 Initial smoothing Sl is placed in position 1 (as shown in Figure 3) S2 is switched between a first position (when training sequences are received), where known symbols are fed to the channel estimator and the equalizer, and a second position where channel estimates are performed using hard decisions received from the hard decision mapper 308 In this case, a delay (not shown) equal to the delay in the equalizer must be added to the channel estimate. 2. Iterative channel estimation, smoothing and decoding Then move the switch Sl to position 2 This will close both loops and establish normal operation of the receiver The upper loop will form a conventional iterative turbo-smoothing loop The lower loop will now continuously update the channel estimate based on extrinsic and mtrinsic information received from the SISO decoder 303 S2 will shift between position 1 when receiving training sequences and position 2 when receiving information symbols In this case the channel estimate does not need to be delayed The channel estimator can use any algorithm for adaptive filtering or adaptive system identification, for example the stochastic gradient least mean squares algorithm (SG-LMS) or recursive least squares algorithm (RLS), as described in S. Haykin, Adaptive Filter Theory, 3rd edition, Prentice Hall, 1996 The channel estimator provides the equalizer with a time-varying estimate of the channel impulse response and of the error variance The error variance is defined as v the variance of the error (noise) signal one obtains when the actual channel impulse response has been replaced by the estimated channel impulse response

With notation as in the appendix

Figure 4 shows the simulation result with bit error rate versus SNR in a receiver that uses a conventional (non-turbo) receiver using a decision feedback smoother and a receiver according to the present invention. The simulation is performed using the 2400 bps HF waveform defined in MIL-STD-188-110B on a fading channel defined as mid-latitude disturbed conditions in ITU-R

F 1387

The graphs are shown for four different configurations of the turbo equaliser. Without iterations, the receiver will give poorer performance compared to a conventional receiver. This is as expected, because the SISO equalizer is a linear filter, unlike the decision feedback equalizer in the conventional receiver. However, performance has been markedly improved already after one iteration Two iterations further improve performance, while the improvement from two to three iterations is small. Compared to a conventional receiver using a decision feedback equalizer, an improvement of about 2-3 dB can be achieved in the required SNR to achieve a certain bit error rate.

Appendix: SISO equalizer for time-varying channels

The transmitted symbols are denoted xn, where n is the bd index. The received symbols Zn from a channel with lntei symbol interference are given by

where hn = [ hoiTl hi_ iin] T is the time-varying estimate of the channel impulse response, L is the length of the channelb fed and e„ is an error signal containing received noise and excess noise caused by uncertainty in the channel estimator An estimate of the variance <j\ of is provided from the channel estimatorii The equalizer file operates on N = N\ + A/j +1 received symbols at a time, collected in the vector % n=[.«n+Ni Zn- thf Dcanc vector can be written Zn H„x„ + e„, wherexn = [rEn +j/1 *»-./c,_L+i]r, Herden decreasing N x ( N + L — 1) channel convolution matrix

and e„ = [ en+ Nl e„_ ^ F

The LLRs of the code bits, blbacke<q>led from the decoder from the previous iteration, are connected to an a priori mean value x„ = E{ x„} = [x«+Ni £n-JVi-L+i] and covariance matrix V „ = E{ xnx%}— xnxnI of the vector Vn is a diagonal matrix with elements «n-Nb-L+i on the diagonal When known symbols have been sent, we have xn=Xn and vn=Q

The equalizer outputs an estimate of each symbol sent

Here sn is the ( Ni + l)th column of H„, and f„ contains the coefficients in the equalizer filter, of length N The equalizer filter is calculated as where £n is a diagonal matn se with elements <^ l+^ 1 °?i- n3 The matrix U„ is calculated recursively from Un_i using the following algorithm where Un_t and U„ have been divided as follows

Claims (16)

1 Apparatus for detecting an encoded digital signal received over a communication channel, including a Soft In Soft Out (SISO) equalizer (300) and a SISO decoder (302) forming an iterative turbo equalization loop, characterized in that the apparatus further includes a channel estimator ( 309) which supplies channel-impulse response estimates to the SISO equalizer (300), a source of known symbols (tn), first switching means (S2) arranged to supply either soft symbols calculated from extrinsic information from the SISO decoder (302) to the channel estimator ( 309) when symbols containing information are received, or supply known symbols (tn) to the channel estimator (309) and the SISO equalizer (300) when training sequences are received 2 Apparatus according to claim 1, characterized in that said first switching device (S2) is arranged to deliver soft symbols calculated from intrinsic information in addition to external soft information from the SISO decoder (302) in one of its positions 3 Apparatus according to claim 1 or 2, characterized in that the apparatus further includes a hard decision allocator (308) which receives soft symbols from the SISO equalizer (300) and second switching means (S2) arranged to deliver hard decision symbols to the channel estimator (309) from said assigns (308) 4 Apparatus according to claim 3, characterized in that said SISO equalizer (300) includes a linear equalizer adapted to minimize the mean square error (MMSE), when a priori information and a time-varying channel estimate are available 5. Apparatus according to claim 4, characterized in that said channel estimator (309) is arranged to estimate the impulse response of the transmission channel using a recursive least squares (RLS) algorithm 6 Apparatus according to claim 4, characterized in that said channel estimator (309) is designed to estimate the impulse response of the transmission channel using a least mean squares (LMS) algorithm 7 Apparatus according to one of the preceding claims, characterized in that said SISO equalizer (300) is arranged to sample the received signal several times per symbol interval 8 Method for detection of a coded digital signal received over a communication channel, where the signal is equalized in a SISO linear equalizer, and decoded in a decoder, said equalizer and decoder forming an iterative turboout equalization loop, characterized in that the impulse response of the communication channel is estimated in a) an initial step where estimates are formed from known training symbols when training sequences multiplexed into the signal stream are received and from hard decisions received from the equalizer when information symbols are received, b) a subsequent step where estimates are formed from known training symbols when training sequences and soft symbols are received from the decoder when information symbols are received 9 Method according to claim 8, characterized by ati step b) said soft symbols include external soft information from the decoder 10. Method according to claim 8, characterized by step b) said soft symbols include intrinsic and extrinsic soft information from the decoder 11 Method according to claim 9 or 10, characterized in that received symbols are sent through an equalizer that minimizes mean square errors (MMSE), when a priori information and a time-varying channel estimate are available 12 Method according to claim 11, characterized in that the equalizer estimates each transmitted symbol as where s„ is the (2^+1) th column of H„, H„ is the time-varying N x (JT+Ii-l) channel convolution matrix, a band diagonal matrix whose rows contain the vectors hTn+ Ni and zeros, h„ is the time-varying estimate of the channel impulse response, and fn contains the coefficients of the equalizer filter of length N=N1+ N3+ 1 at time period n 13. Method according to claim 12, characterized in that the equalizer filter is calculated as: where En is a diagonal matrix with elements o~ l+ Ni , the time-varying estimate of the error variance, and the matrix Un is calculated recursively from using the following algorithm where Un.! and Un has been divided as 14 Method according to claim 11 or 13, characterized in that said channel impulse response is estimated using a recursive least squares (RLS) algorithm. Method according to claim 11 or 13, characterized in that said channel impulse response is estimated using a least mean squares (LMS) algorithm 16 Method according to one of claims 8-15, characterized in that the received signal is sampled several times per symbol interval