EP1103124A1

EP1103124A1 - Receiver for digital signals propagated in channels with multiple paths

Info

Publication number: EP1103124A1
Application number: EP99940115A
Authority: EP
Inventors: Roberto Cusani; Jari Mattila
Original assignee: Telit Mobile Terminals SpA
Current assignee: Telit Mobile Terminals SpA
Priority date: 1998-08-05
Filing date: 1999-07-30
Publication date: 2001-05-30
Also published as: IT1301907B1; ITMI981839A1; WO2000008816A1; CA2339499A1; CN1369162A; AU5418099A; ITMI981839A0

Abstract

A method and a receiver for equalization of received digital signals r(n) propagating in a scattered communication channel with multiple time-unvarying and also time-varying paths which give rise to time and frequency distortions with channel pulsed response (CIR) having total length equal to L symbol intervals. In accordance with the method the scattered channel state is divided in three components and to wit the observable, close hidden and distant hidden states. The channel is then estimated on the basis of a vector p'(n) of the a posteriori probabilities (APP) of the observable state b'(n) and on the basis of the received sequence r(n) and the received symbol decision is made on the basis of a maximum a posteriori probability (MAP) based on the above mentioned APP vector p'(n) of the channel observable state. The CIR can be estimated in relation to a probe sequence contained in the received signal and the result of the estimate is used during reception of the remaining part of the signal and if necessary updated if the channel is time-varying.

Description

"RECEIVER FOR DIGITAL SIGNALS PROPAGATED IN CHANNELS WITH MULTIPLE PATHS"

The present invention relates to a treatment method for equalization in particular of signals and to a receiver applying this method for digital connections through channels affected by multiple paths varying rapidly in time as well as noisy. The intersymbolic interference (ISI) from multiple paths constitutes one of the main factors limiting the performance of terrestrial mobile radio systems. The temporal spread ts induced by the transmission channel can reach 15-20 μsec regardless of the carrier frequency (900/1800Mhz for GSM transmissions, 2GHz for the UMTS).

In the present GSM transmission where modulation is binary at 270kbps this corresponds to an equivalent time-discrete Channel Pulsed Response (CIR) of length L up to L=6 symbol intervals (samples) so that a Viterbi equalizer (VE) with 2⁶=64 states constitutes a feasible solution as implemented in present GSM receivers. With reference to systems more advanced than GSM greater symbol speeds (up to 2Mbps) give rise to greater values of L and the conventional VE is excessively complex. A similar increase in complexity is achieved by passing to higher order constellations (e.g. QPSK).

In addition a reduction in complexity as compared with the 64- state VE can be useful even in the present GSM in order to be able to activate additional functions such as e.g. channel estimation along the entire received timeslot and not only in relation to the so- called midamble as takes place in present GSM receivers.

In such situations it is necessary to use the VE in favor of less complex techniques such as a cross filter or a DFE which however give much poorer performance than the VE.

The equalizer proposed belongs to the family of so-called symbol-to-symbol with maximum a posteriori probability (SBS-MAP) equalizers. The equalizer in accordance with the present invention has a recursive structure and operates on the basis of the step-by- step channel knowledge and sequence observation (noisy) received in relation to the channel output.

The general purpose of the present invention is to overcome the above shortcomings by supplying a signal treatment method and a receiver in accordance with this method permitting less implementation complexity than conventional receivers in which complexity depends on the total impulse response length.

In view of this purpose it is sought to provide in accordance with the present invention a method for the equalization of received digital signals r(n) propagated in a communication channel with multiple paths of the time-unvarying type or even time-varying giving rise to distortions in time and frequency and with pulsed response of total length L and comprising the phases of:

- preliminary estimation of the channel pulsed response termed CIR,

- taking the NNZ coefficients of the estimated CIR termed 'nonzero' coefficients which have much greater amplitude than the remaining L-NNZ coefficients called 'zero' coefficients to divide the channel state in: a) 'visible state' defined by a vector b'(n)≡[b-(n-i'1)...b(n-i'NNZ)] consisting of the NNZ information data allocated in relation to the non-zero CIR coefficients, b) 'close hidden state' defined by a vector b"(n)≡[b(n-i"1)...b(n- i"NZV)] consisting of the NZV data having a non-zero CIR coefficient in relation to the following CIR position, c) 'distant hidden state' defined by a vector b"'(n)==[b(n- i'"1)...b(n-i'"NZL)] consisting of the remaining NZL=L-NNZ-NZV data,

- calculation of the vectors p_'(n), p_"(n) and p_'"(n) of the a posteriori probabilities (APP) of the channel states b'(n), b"(n) and b'"(n) respectively,

- deciding the received symbol on the basis of a maximum a posteriori probability (MAP) criterion based on the above mentioned vector APP p_' (n) of the visible state of the channel, and - updating step-by-step the initial estimate of the channel on the basis of a vector APP p_'(n) of the visible state b'(n) in addition to the received sequence r(n).

According to the method it is sought to provide a receiver of received digital signals r(n) propagated in a communication channel with multiple paths of the time-unvarying type or even time-varying giving rise to distortions in time and frequency and with pulsed response of total length L comprising:

- an estimator performing estimation of the channel pulsed response

(CIR), - a detector of non-zero coefficients of the estimated CIR to identify the following three channel states: a) 'visible state' defined by a vector b'(n)≡[b(n-i'1)...b(n- i'NNZ)] consisting of the NNZ information data allocated in relation to the non-zero CIR coefficients, b) 'near hidden state' defined by a vector_b"(n)≡[b(n-i"1)...b(n- i"NZV)] consisting of the NZV data having a non-zero CIR coefficient opposite the following position in the CIR, and c) 'distant hidden status' defined by a vector b'"(n)s[b(n- i'"1)...b(n-i'"NZL)] consisting of the remaining NZL=L-NNZ-NZV data, - a calculator of the vectors p_'(n), p_"(n) and p_"'(n) of the a posteriori probability (APP) of the channels states b'(n), b"(n) and b'"(n) respectively,

- a maximum a posteriori probability (MAP) decider based on this vector APP p_'(n) of the visible state of the channel,

- a step-by-step channel estimator based on the vector APP p'(n) of the visible state b'(n) as well as on the received sequence r(n).

To clarify the explanation of the innovative principles of the present invention and its advantages compared with the prior art there is described below with the aid of the annexed drawings a possible embodiment thereof by way of non-limiting example applying said principles. In the drawings -

FIG 1 shows the block diagram of an SBS-MAP receiver of known type, and FIG 2 shows the block diagram of an SBS-MAP receiver applying the method in accordance with the present invention.

The method in accordance with the present invention provides a basic version for time-unvarying channel (Tl) where it is assumed that the CIR is known or estimated preliminarily on the basis of a known data sequence and a version for time-variant channel (TV) in which the channel estimate is updated step-by-step in relation to and on the basis of the received sequence.

It is provided that the digital signals to be processed can be advantageously made up of information data and probe sequences (consisting of known data) variously interleaved. For example data sequences and probe sequences can be regularly spaced or the signal can be made up of timeslots with a probe sequence in the center (midamble) and two information data sequences at the sides or the timeslot can consist of a probe sequence at the beginning (preamble) followed by a known data sequence etc. Digital signals of these types are well know to one skilled in the art who will select the probe and information data composition he deems preferable for the specific application. With the classical equivalent model in time-varying base band

(TV) (i.e. the channel varies rapidly in time or the CIR is different from symbol to symbol even in the same timeslot) sampled at symbol time Ts, the received sequence can be expressed as:

L-1 r(n)= ∑ c_k(n)b(n-k)+w(n) (1) k=0

where b(n) is the information data sequence belonging to a generally complex constellation and with dimension S (e.g. S=2 and b(n)=±1 in the BPSK, S=4 and b(n)=(±1 ±j/v^/2 in the QPSK), w(n) is a typically Gaussian white additive noise sequence with average nul and bilateral power spectral density of N₀/2, c_k(n), k=0 L-1 are the coefficients of the CIR which vary with the temporal index n. In the particular case if unvarying time channel (Tl) equation (1) becomes: L-1 r(n)= Σ c_kb(n-k)+w(n) k=0

with the coefficients of the CIR c₀,...c_L-ι which become unvarying with n and are assumed known or estimated on the basis of an appropriate probe sequence present together with the information sequence.

With reference to FIG 1 in both cases (TV or Tl channel) the SBS-MAP receiver of known type comprises an SBS-MAP type equalizer which performs equalization of the received signal with the steps:

1) Define the channel state vector b≡[b(n)...b(n-L+1)] which can assume N=S^L different configurations mι,...m_N.

2) Define the probabilities of the possible states of the channel in step n conditioned to the observation of step 1 to step n of the entire received sequence rι,_n≡[r(1)...r(n)]. These are grouped in the a posteriori probability (APP) vector: p(n)=[p(b(n)=m_l|r₁,_n)...P(b(n)=mN|rι_ιn)]^τ where ^τ denotes the transposition of the vector y

3) Define the matrix F of the transition probability (PT) of the channel state with elements:

Fij i,j=1 ,...N which is used to calculate as Fp_(n-1 ) the prediction at a step of p_(n) based on the observation of r_1>n-1.

4) In step n is calculated recursively p_(n) on the basis of its prediction Fp_(n-1) and the observation of r(n) as well as knowledge of the CIR. 5) From β(n) are calculated the probabilities of the constellation symbols in step n-D (or with decision delay D equal to the channel memory i.e. D=L-1) adding the elements of p_(n)corresponding to the channel states displaying that symbol datum in position n-D. 6) Lastly a MAP decider on the basis of the known MAP rule decides for the constellation symbol having the highest probability. The decided data b^Λ(n-D) have delay D.

In the case of channel TV in addition to the APP calculator block (APPC) it is also necessary to have a channel estimator (CE) resupplying step-by-step the SBS-MAP equalizer with the updated estimate of the CIR found on the basis of the received sequence r(n) and usually of the data decisions (or of the known data in relation to any probe sequence). C_k(n) are the estimated CIR coefficients.

In this case the SBS-MAP equalizer is particularly efficient because the decisions output display low delay (of D) as compared with VE which operates with a decision delay typically of 5L or 6L This allows the CE to supply updated CIR estimates to improve the performance of the equalizer. A conventional solution for the CE consists of a Kalman filter guided by the decisions. A possible alternative proposed recently consists of a nonlinear adaptive Kalman filter guided by the vector p_(n) of the channel APPs in addition to the received sequence r(n).

Up to this point a virtually known method is described.

For both Tl and TV it is readily verifiable that implementation complexity depends on the total length L of the pulsed response. In the presence of multiple paths the pulsed response there can even have very long length L making the above mentioned method unproposable if satisfactory accuracy is desired.

But in the presence of multiple paths the pulsed channel response (CIR) generally consists of some coefficients which are non-null and null or in any case of negligible amplitude for equalization purposes. Indeed, it is observed that although in the presence of high temporal spread ts the number of solvable multiple paths or with not small differences between arrival times compared with the symbol interval is generally limited. Thus only some of the CIR coefficients are non-zero or in any case do not have negligible amplitude and the channel is defined as scattered.

In this case a possible known solution consists of the introduction of an adaptive cross filter to reduce the ISI caused by the most distant multiple path echoes so that the following VE can operate with better efficiency. But the performance of such a system could be unsatisfactory because of the only partial cancellation of the distant echos and intensification of the noise introduced by the cross filter.

Another solution consists of an appropriately pruned version of the VE where the presence of CIR coefficients of zero is allowed for in calculation of the metrics associated with the different paths in the VE tree. But the VE must be designed ad hoc for each of the possible zero and non-zero sample configurations in the CI R. Since this design is quite complicated and no general solution for it is available today the pruned VE does not at present constitute a solution with practical interest. FIG 2 shows a solution in accordance with the present invention. In accordance with the present invention it is sought innovatively to divide the scattered channel state in three components, to wit the observable or visible state, the hidden nearby sate and the hidden distant state. The prior art APP calculator is replaced by three calculation blocks.

Let c_i ..., Civ..., Cj'_MMz be the NNZ non-zero CIR coefficients whose positions are identified by the whole indices i'l<i'k..<i'NNZ<i'NNZ. These CIR coefficients are known or estimated with good accuracy and their positions are assumed. To limit the complexity of the receiver an upper NNZ limit can be assumed and then the NNZ CIR coefficients with higher amplitude selected as non-zero while ignoring all the others. For the sake of simplicity null coefficients are discussd here even if it is intended that this term indicate coefficients with amplitude less than a preset limit selected at will. Obviously the lower this limit the higher the complexity of the receiver but the better the multiple path response.

The received sequence expression can be written as follows: NNZ r(n)= Σ Cπ_<b(n-i'k)+w(n) k=1

and a vector b'(n)≡[b(n-i'1)...b(n-i'NNZ)] of the state of the channel observable in step n consisting of NNZ information data allocated in relation to non-zero CIR coefficients can be defined. b'(n) can assume N'=S^NNZ different configurations mY--. m'_N., and the APP vector can be defined as follows for it: β^«(n)_s[p(b'(n)=m^,ι|rι,n)...p(b^,(n)=m^lN'|rι,n)]^τ (2) The information data allocated in relation to the remaining L-

NNZ null coefficients of the CIR constitute the hidden state of the channel in step n. Among these let there be grouped in the vector b"(nHb(n-i"1)...b(n-i"NZV)] the NZV data displaying a non-zero CIR coefficient in relation to the following position in the CIR which are observable in the following step n+1 and which are therefore a close hidden state. Instead let there be grouped in the vector b"'(n)≡[b(n- i'"1)...b(n-i'"NZL)] the remaining NZL=L-NNZ-NZV data which will remain hidden even in the following step n+1 and thus are a distant hidden state. The vectors b"(n) and b'"(n) can assume respectively the

N"=S^NZV configurations m"1 ,...,m"_N and the N'"=S^NZL configurations Jϋ,'"ι,— .D-Tri and with these are associated in step n the APP vectors p."(n) and p_"'(n), defined in the same manner as for (2) for p_'(n) with obvious substitutions. The three calculation blocks in which the APP calculator (APPC) is divided each calculate the APP of the corresponding channel state. The APP calculator comprises a detector (NNZD=NNZ Detector) of non-zero coefficients of the estimated CIR to identify the three states of the channel and allow the three blocks to calculate the APPs of their associated state.

It is observed that the configuration taken from b'(n) in the following step n+1 depends not only on b'(n) itself but also on b"(n).

Introducing the matrices F'_k, k=1 N" of the PTs conditioned in relation to b"(n) with elements: where i,j=1 ,...,N' and k=1 ,...,N" the prediction at one step of p_'(n) can be expressed as follows. N" Σ {[F'_kβ'(n)]P(b"(n)=m"_k)}= k=1

N"

{Σ [F'_kP(b"(n)=rn"_k)]}_g'(n)=F(n)fi'(n) (3) k=1

where the matrix F(n) is the average of the F'_k weighted for their probability in step n and its elements are the non-conditioned PTs of

F_ij(n)=P(_β'(n)=m^li|fi'(n-1)=m^,j), i,j=1 ,...,N'. It is observed that the step-by-step calculation of F(n) does not appear in the non-scattered channel equalizer SBS-MAP.

Concerning the calculation of the APP vectors p_'(n), p_"(n) and β"'(n), they are initialized with equal values (1/N¹, 1/N", 1/N'" respectively) assuming therewith equally probable all the possible states. If there is a probe sequence the above mentioned vectors are initialized on the basis of the known data.

As may be seen in FIG 2 in step n the vectors p_"(n) and ρ_'"(n) are calculated on the basis of the vectors rj'(n) calculated and memorized in the previous steps. It is observed that the received sample r(n) is not used in calculating p_"(n) and p_"'(n) which have been hidden.

On the other hand p_'(n) is calculated on the basis of its prediction at one step estimated as in (3) and of the present received symbol r(n).

If the scattered channel is time-variant (TV-S) it is assumed that the positions of the NNZ non-zero CIR coefficients are fixed in time, i.e. the index sets (i'1 ,..,i'NNZ), (i"1 ,..,i"NZV), (i'"1 ,..,i'"NZL). The channel model then becomes:

NNZ r(n)= Σ Ci._k{n)b(n-i'k)+w(n) k=1

and the CE which supplies the estimated CIR coefficients must update the trajectories only of the non-zero NNZ coefficients Cπ(n),..., C _k(n),...,c_iΗNZ(n).

For this purpose it is possible to use a conventional Kalman (vectorial) filter based on r(n) and on the decisions concerning b(n- i'1),..,b(i-i'NNZ) or on any known data in relation to a probe sequence.

As an alternative in the present invention there can be used a Kalman non-linear adaptive filter based on r(n) and on the APP vector β'(n) of the visible state b'(n), instead of on the APP vector p_(n) of the whole state of the channel b(n) as is done in the similar Kalman non-linear adaptive filter described above with reference to the non-scattered channel.

In the cases of both scattered and non-scattered time-variant channels in relation to the training sequence the CE (the Kalman type filter) is modified to allow for the fact that the state of the channel is completely known (in the training sequence) or is partly known (in the passage from the training sequence to the known data). Again with reference to FIG 2 the SBS-MAP equalization algorithm in accordance with the present invention can be summarized as follows.

1) Preliminarily the CIR is estimated with an appropriate estimator on the basis of the probe sequence. This can be done with a simple algorithm, e.g. calculating the cross correlation between the probe sequence transmitted and the one received as in the GSM.

This way there are identified the non-zero CIR coefficients and hence the index sets (i'1 ,..,i'NNZ), (i"1 ,..,i"NZV), (i^,"1 ,...,i'"NZL). To limit the complexity of the receiver an upper limit for NNZ can be assumed and then the NNZ CIR coefficients of greatest amplitude selected as non-zero while ignoring all the others.

2) If the channel proves scattered i.e. if NNZ>L the matrices F'k' _k=1 ,...,N" are calculated on the basis of the channel index sets. 3) If necessary the CIR can be recalculated with higher accuracy than in step 1 by using the Kalman non-linear adaptive filter for TV channel and known canal state described above with initial values of CIR equal to those found in step 1. 4) If the channel is TV-S the procedure described above for a time-variant scattered channel is followed, i.e. the APP vectors are initialized on the basis of the training sequence and then the information data sequence is moved along where in step n -

4.1) the APP vectors are updated on the basis of the received sample r(n) and the channel prediction calculated in the previous step,

4.2) on the basis of p_'(n) this channel prediction is then updated and made available for the following step n+1 ,

4.3) from p_^*(n) the probabilities of the symbols in step n-D with which the MAP decider supplies the MAP decision are calculated. The decided data b^A(n-D) have delay D.

5) If the channel is Tl-S the procedure described above for the time-unvarying scattered channel is followed i.e. one proceeds as set forth in above steps 1 to 3 but with the channel estimator kept inactive along the data sequence and with the SBS-MAP equalizer using the channel estimate found in step 1 and if necessary finished as in step 3.

6. If in step 1 the channel is not scattered, the SBS-MAP solutions already known (described above) for TI-NS or those described for TV-NS are applied as required.

As regards the computing complexity of the equalizer described, for channel NS it is the same as the similar SBS-MAP solutions proposed by the prior art.

The solution proposed in the present invention for the scattered channel, whether Tl or TV, displays added complexity as compared with an NS channel with the same number of non-null coefficients because the hidden states of the channel must also be allowed for. In particular for each timeslot received it is necessary to select the NNZ CIR coefficients of greatest amplitude. The complexity of this operation depends on the algorithm used for preliminary estimation of the channel.

Then there are calculated the N" matrices of the F'_k PTs to be used as shown in (2) and each of which has dimension (N')² for a total of N"x (N¹)² elements to be calculated.

It must be considered however that the effective number of elements to be calculated is in reality much smaller if one allows for the symmetries and properties of the PT matrices. Assuming that the CIR configuration remains virtually unchanged for the entire timeslot the PT matrix calculation shown above is performed only once for each timeslot received and the computing cost therefor can be considered negligible as compared with that for the timeslot information data step-by-step equalization.

The design complexity of the equalizer for the present configuration of the CIR coefficients is negligible as compared with that for information data equalization.

In particular it is necessary to calculate step-by-step for each sample received not only the APP vector p_'(n) of the observable state but also the APP vectors p_"(n) and p_'"(n) of the hidden states. While calculation of ρ_'(n) is equivalent to calculation of p_(n) for an NS channel with the same number of non-null coefficients, to calculate p_"(n) and p_'"(n) some additional operations are necessary. In particular it has been estimated that for each symbol received there are needed N"x (NNZ-1) products to calculate p_"(n) from the previous vectors p_'(n) appropriately memorized. N" x N'² products and (N"-1)xN'² sums to calculate the matrix F(n) of the weighed PTs. (NNZ-1 )xSxN' products and (NNZ-1 )xSx(N'-1) sums to update the estimates of the probabilities corresponding to the positions of the L CIR coefficients. It must also be considered that the complexity of the solution in accordance with the present invention is in any case much less than that which there would be in the scattered channel solution in which allowance is made for the whole length L of the pulsed response. It has been estimated that while the complexity of the conventional SBS-MAP solution increases exponentially with L the complexity of the solution in accordance with the present invention is not more than twice that of the SPS-MAP equalizer with the same number of non-null coefficients. It is now clear that the preset purposes have been achieved.

The complexity of a receiver in accordance with the present invention is linked only to the number of non-null coefficients of the CIR and not to the total number of the pulsed response coefficients also including the null coefficients as occurs in the prior art receivers and in particular those based on the Viterbi algorithm.

Complexity reduction allows making cases in which length L of the pulsed response is such as to not be otherwise confrontable except with known alternative techniques of much poorer performance computationally acceptable. Complexity reduction also allows insertion of additional functions in those cases where the prior art allows only implementation of the basic equalization functions.

Naturally the above description of an embodiment applying the innovative principles of the present invention is given by way of non- limiting example of said principles within the scope of the exclusive right claimed here. For example the channel coding of the received sequence could be the one considered most appropriate even if the Trellis channel coding (TMC - Trellis Code Modulation) of a convolutional or block type were found preferable. Again advantageously in transmission the sequence could be subject to differential or interlacing coding.

Claims

1. Method for the equalization of received digital signals r(n) propagated in a communication channel with multiple paths of the time-unvarying or even time-varying type giving rise to time and frequency distortions and with pulsed response having total length L and comprising the phases of:

- preliminary estimation of the channel pulsed response termed CIR, - taking the NNZ coefficients of the estimated CIR termed 'nonzero' coefficients which have much greater amplitude than the remaining L-NNZ coefficients called 'zero' coefficients to divide the channel state in: a) 'visible state' defined by a vector b'(n)Γëí[b(n-i'1)...b(n-i'NNZ)] consisting of the NNZ information data allocated in relation to the non-zero CIR coefficients, b) 'near hidden state' defined by a vector b"(n)Γëí[b(n-i"1)...b(n- i"NZV)] consisting of the NZV data having a non-zero CIR coefficient in relation to the following CIR position, c) 'distant hidden state' defined by a vector b'"(n)Γëí[b(n- i'"1)...b(n-i'"NZL)] consisting of the remaining NZL=L-NNZ-NZV data,

- calculation of the vectors D_'(n), p_"(n) and ╬▓'"(n) of the a posteriori probabilities (APP) of the channel states b'(n), b"(n) and b'"(n) respectively, - deciding the received symbol on the basis of a maximum a posteriori probability (MAP) criterion based on the above mentioned vector APP p_'(n) of the visible state of the channel, and

- updating step-by-step the initial estimate of the channel on the basis of a vector APP p_'(n) of the visible state b'(n) in addition to the received sequence r(n).

2. Method in accordance with claim 1 characterized in that the CIR is estimated in relation to a probe sequence contained in the received signal and the result of the estimate is used during reception of the remaining part of the signal.

3. Method in accordance with claim 1 characterized in that the transmitted sequence is subject to interlacing.

4. Method in accordance with claim 1 characterized in that the transmitted sequence is subject to modulation and/or differential coding.

5. Method in accordance with claim 1 characterized in that the transmitted sequence is subject to channel coding selected preferably from among the block, convolutional or Trellis codings.

6. Receiver of received digital signals r(n) propagated in a communication channel with multiple paths of the time-unvarying or even time-varying type giving rise to time and frequency distortions and with pulsed response of total length L comprising:

- an estimator (CE) performing preliminary estimation of the channel pulsed response (CIR), - a detector (NNZD) of non-zero coefficients of the estimated CIR to identify the following three channel states: a) 'visible state' defined by a vector b'(n)Γëí[b(n-i'1)...b(n- i'NNZ)] consisting of the NNZ information data allocated in relation to the non-zero CIR coefficients, b) 'near hidden state¹ defined by a vector_b"(n)Γëí[b(n-i"1)...b(n- i"NZV)] consisting of the NZV data having a non-zero CIR coefficient in relation to the following position in the CIR, and c) 'distant hidden status'defined by a vector b'"(n)Γëí[b(n- i'"1)...b(n-i'"NZL)] consisting of the remaining NZL=L-NNZ-NZV data, - a calculator (APPC) of the vectors p_'(n), p_"(n) and ╬▓"'(n) of the a posteriori probabilities (APP) of the channels states b'(n), b"(n) and b'"(n) respectively,

- a maximum a posteriori probability (MAP) decider based on this vector APP ╬▓'(n) of the visible state of the channel,

- a step-by-step channel estimator (CE) for updating the preliminary estimate based on the vector APP ╬▓'(n) of the visible state b'(n) and on the received sequence r(n).

7. Receiver in accordance with claim 6 characterized in that CIR is estimated in relation to a probe sequence contained in the received signal and the result of the estimate is used during reception of the remaining part of the signal.

8. Receiver in accordance with claim 6 characterized in that the transmitted sequence is subject to interlacing.

9. Receiver in accordance with claim 6 characterized in that the transmitted sequence is subject to modulation and/or differential coding.

10. Receiver in accordance with claim 6 characterized in that the transmitted sequence is subject to channel coding selected preferably from among the block, convolutional or Trellis codings.