Method of channel estimation and corresponding receiver

The field of the invention is that of the transmission of digital data, in particular in transmission channels having, or may exhibit, high Doppler fading and low intersymbol interference (IES). More specifically, the invention relates to the estimation of the transmission channel and the demodulation in receivers of signals transmitted through such channels.

A preferred, although not exclusive, field of application of the invention is that of digital communications between satellites and mobiles. It can be considered in particular within the framework of the ICO and Iridium projects. More generally, the invention is advantageously applicable in all communication systems where the channel has a strong Doppler and a low IES.

In conventional digital communications systems, synchronization symbols are frequently used. They allow the receiver not only to synchronize, but also to properly estimate the channel to ensure the smooth running of the demodulation phase.

In the case of these conventional communication systems, the synchronization symbols are consecutive and form a synchronization sequence, usually placed at the beginning of the data streams. The synchronization at the receiver is performed by detecting a threshold crossing of the correlation of the received samples with those of the sequence used for the synchronization. This synchronization makes it possible, on the one hand, to detect the beginning of a data stream and, on the other hand, to precisely determine the maximum opening instants of the eye diagram.

In the framework of the studies carried out for the GSM system, attempts have been made to construct synchronization sequences adapted to low Doppler fading channels and high IES channels. These studies have shown the interest, for this kind of channels, of placing these synchronization sequences in the middle of each data stream.

The patterns of synchronization sequences proposed lead to synchronization algorithms and channel estimation that are intuitive and simple to implement. As a result, they have been adopted in several digital communications systems.

They are also envisaged in satellite telephony projects. However, they are much less efficient for this type of channel, and lead to significant limitations in performance when dealing with high Doppler and low IES channels, such as a satellite channel.

The basic principle of time synchronization and channel estimation by the receiver is to transmit a sequence of known symbols thereof in the data streams sent by the transmitter. By using a few predefined algorithms, these symbols are used to guarantee not only a good synchronization of the receiver, but also a reliable estimate of the channel, thus allowing a good progress of the demodulation phase.

In known systems, a synchronization pattern consisting of a sequence of grouped symbols having good correlation properties is used. These properties are mainly used to synchronize well at the receiver. In the context of radio-mobile communication systems, such as GSM, the channel is quasi-static during a data stream but nevertheless has very severe IES. These correlation properties then prove to be well adapted and even necessary for a simple and direct estimation of the impulse response of the channel.

Channel estimation systems of this type are described, for example, in the documents <patcit id="pcit0001" dnum="WO9428661A"><text>WO-94/28661</text></patcit> and "on pilot symbol assisted detection of MSK and GTFM in fast fading channels".

These synchronization sequences can be used in communication systems between satellites and mobiles. It should be noted, however, that if the channels encountered in this type of application have negligible IES relative to the symbol rate, their variations in the data streams received are very important. In other words, the transmission channel can not be considered as quasi-static during the duration of a data stream (in terms of time-frequency representation, it can be said that these channels are the dual ones of those presented here. -above).

As a result, the use of a conventional synchronization sequence in this type of system is purely arbitrary, and solves only the single synchronization problem at the receiver.

More precisely, if the choice of group synchronization in the context of quasi-static channels with high IES (classical case of GSM in particular) is well founded, it is not necessarily the same for high Doppler channels and low IES. The latter show considerable variations in the data streams received. The conventional sync symbol sequence can ensure good receiver timing. On the other hand, the quality of the channel estimation is severely compromised because it is not representative of the state of the channel during the reception of the useful data of a data stream.

The conventional synchronization is therefore not suitable for systems communicating through a high Doppler channel, for the channel estimation and demodulation function.

The aim of the invention is in particular to overcome these drawbacks of the state of the art in digital transmission systems organized in data streams and which can be confronted, at least in certain situations and / or at certain times, with high Doppler channels, can not be considered quasi-static over the duration of a data stream.

More precisely, one of the objectives of the invention is to provide structures that make it possible in particular to make reliable channel estimation and demodulation in all circumstances.

Another object of the invention is to simultaneously optimize, in the receivers, the synchronization, the quality of the channel estimation and, where appropriate, the efficiency of the interleaving of the coded data. In other words, it must make a compromise between good synchronization, good channel estimation at the receiver and good interleaving of the encoded data.

Another object of the invention is also to limit the losses in useful rate, and therefore to limit the number of reference data required.

Another object of the invention is to provide channel estimation methods, and corresponding receivers, which can manipulate arbitrary patterns of synchronization.

A complementary objective of the invention is to provide a channel estimator that can significantly improve its performance by taking into account all or part of the coded structure of the data transmitted.

These objectives, as well as others which will appear later, are achieved according to the invention by means of an estimation method according to the main claim.

Advantageous features of this method are the subject of the secondary claims.

The invention uses a signal of structure quite different from that of conventional signals, in which the reference symbols, used to estimate the channel, are systematically grouped into a single synchronization block (generally at the beginning of the data stream, or possibly in the middle of it).

The solution of the invention, which goes against this conventional technique, provides an effective response to the problems associated with high Doppler channels.

A second method of estimating a transmission channel of such a digital signal comprises a step of writing said estimate in the form of a combination of predetermined basic functions, of a bandwidth greater than or equal to that of the Doppler power spectrum (SPD) of said signal.

Preferably, said basic functions are restrictions of discrete flattened spheroidal sequences (SSADs, known in the English literature as "Discrete Prolate Spheroidal Sequences") [8].

This type of function makes it possible to obtain a very accurate estimate of the channel, from a reduced number of basic functions (for example 3 to 8).

According to an advantageous embodiment of the invention, this method comprises a preliminary step of adapting the characteristics of said basic functions, in particular according to said Doppler power spectrum.

This gives a base adapted to the best channel to estimate.

The invention also relates, of course, channel estimation devices and receivers implementing the methods described above, and the transmitters and receivers of signals according to the invention.

Other characteristics and advantages of the invention will appear on reading the following description of a preferred embodiment of the invention, given by way of a simple illustrative and nonlimiting example, and the appended drawings, among which:<ul id="ul0001" list-style="dash" compact="compact"><li>Figure 1 is a block diagram of a transmitter of a signal according to the invention;</li><li>FIG. 2 is a block diagram of a receiver according to the invention, able to receive the signals of the invention;</li><li>Figure 3 schematically illustrates the principles of the estimation of the transmission channel according to the invention;</li><li>FIG. 4 shows the operation of a receiver implementing a method according to the invention;</li><li>FIG. 5 illustrates the rapid decay of the eigenvalues ​​of the correlation matrix of the discrete channel at the level of transmitted data streams;</li><li>FIGS. 6A to 6G illustrate the eigenvectors corresponding to the most significant eigenvalues ​​of the correlation matrix of the discrete channel at the level of transmitted data streams;</li><li>FIG. 7 illustrates the weightings of the eigenvectors, of the correlation matrix of the discrete channel at the level of transmitted data streams, used in the estimation of the channel by the receiver.</li></ul>

1. Overview

The invention therefore applies in particular to stationary channels with low IES with SPDs of any width and shape. In particular, these channels may be of Rice or Rayleigh type with SPD horn or flat. In particular, they may include the case of a static frequency offset ("offset") resulting from a partial correction of the frequency error of the local oscillator at the receiver. This correction is generally accomplished through a conventional phase locked loop.

The invention proposes a channel estimation technique that uses a new signal structure, in which a plurality of reference blocks are distributed in each data stream.

Thus, an object of the invention is to provide for a Doppler channel and a given characteristics communication system the most suitable synchronization pattern. This reason must be sufficiently distributed over a more or less extensive part of the data streams to characterize and estimate the most significant variations of the channel and to distribute the coded data as well as possible. It should therefore not give priority to the receiver's performance in terms of synchronization at the cost of sacrificing the quality of the channel estimate and the interleaving of the coded data.

The corresponding channel estimation method proposed according to the invention is based in particular on a suitable modeling of the real channel and a new and simplified representation thereof.

This representation should take into account as much as possible of all known information on the channel. This information must include at least the maximum width of the Doppler band, which must be known by the receiver.

In the case where the shape of the SPD of the channel is not known, the algorithm adopts a flat modeling of this one. This modeling is best suited from the point of view of information theory because it leads to maximum channel entropy [7]. In this case, the discrete channel can be written as a combination of discrete flattened spheroidal sequence (SSAD) restrictions to the medium of the data stream to be demodulated. These SSADs have a narrow band of width equal to or greater than that of the SPD.

There are essentially two criteria that can be used to estimate one of the unknown parameters: the maximum likelihood (MV) criterion and the posterior maximum (MAP) criterion [1]. In the case of the MV criterion, the parameters are assumed deterministic but unknown. In the case of the MAP criterion, these parameters are assumed to be random and characterized by a known probability density (PD).

The criterion MV is best suited to problems of estimation of the energy per emitted symbol or of the noise variance at the output of a given filter, when these are not known by the receiver. The MAP criterion is best suited to the estimation problem of the discrete channel at the transmitted data streams.

With the simplified representation of the channel developed by the inventors and subsequently presented, and (not necessarily) the reference symbols sent by the transmitter, the receiver implements the EM algorithm to perform an iterative estimation of a part or of all these parameters according to the corresponding criteria.

In other words, this algorithm makes it possible to find the most likely channel realization conditionally to the signal observed at the receiver. It also makes it possible to best estimate the energy per transmitted symbol and / or the noise variance at the output of a given filter when these are not fully known.

One of the advantages of the EM algorithm is that it can optimally use synchronization symbols (explicit reference) as well as all or some of the characteristics of the information symbols (implicit reference) constituting a train of data to complete the channel estimate. More precisely, this algorithm makes it possible to advantageously use the coding experienced by the information data sent to significantly improve the quality of the channel estimation. It also allows to take into account the memory possibly due to the modulation used.

This algorithm avoids the systematic use of conventional cluster synchronization patterns, which owe their success to their immediate exploitation by intuitive and simple algorithms.

The choice of the location of the reference symbols in the data streams is indeed important for the performance of the channel estimator. It is also of paramount importance in the good determination of the conditions i paramount nitrate of the EM algorithm. Indeed, the latter generally tends to converge towards local maxima of the conditional probability (defined as the posterior probability of an embodiment of the channel conditionally to the received signal), if the location of the synchronization symbols is badly chosen.

With the aid of the estimation of the channel, the receiver performs a standard demodulation and / or decoding of the information data received according to the criterion of the MV.

2. Principle of a transmitter

Figure 1 shows a simplified block diagram of a transmitter of a data train signal according to the invention. The emission process is deduced directly.

An arbitrary bit rate information source 11 is considered that generates binary or non-binary data corresponding to source signals of any type (sounds, images, data, etc.). This data is optionally subjected to a source coding 12, followed by an error correcting coding 13 adapted to the Rice type channels without lES.

The coded data generated from these codes (useful symbols) are then organized into data streams and modulated (14). They are thus appropriately distributed and interleaved on several data streams in order to provide the necessary diversity and to decorrelate the fading affecting the symbols transmitted. Reference elements are also introduced into each data stream, according to the distribution principles specified below. Finally, the data is modulated in phase.

The baseband signal generated by the modulator 14 is obtained by formatting filtering of the symbols composing this train. In the case of MDP2 and MDP4 modulations, the shaping filter is generally a Nyquist root filter to avoid any IES at the output of the receiver's adapted filter. The baseband signal thus generated is then transposed into frequency, amplified and transmitted (15) through the channel.

3. Example: the ICO system

By way of example, the transmission characteristics of the satellite radiocommunication (ICO) system (Intermediate Circular Orbit) are presented here, to which the invention advantageously applies.

This system is based on a Time-Division Multiple Access (TDMA) technique. It is based on data frames, each composed of 6 time slots (known in English as "time slots"). These frames are transmitted in the uplink (mobile-satellite) and downstream (satellite-mobile) directions for the channels dedicated to the voice as well as those dedicated to the signaling. Each time interval corresponds to a transmitted data stream composed of<i>NOT</i> = 120 or 240 symbols, depending on the case.

The set of logical channels used by the ICO system is very close to that of GSM. It includes, in particular, a uplink and uplink Traffic CHannel (TCH) channel, a Down Channel Broadcast Common Channel (BCCH) channel, and a Random Access Channel (RACH). at the uplink level.

The modulation adopted for the downlink is MDP4 for TCH and MDP2 for BCCH. In these two cases, the number of symbols per data stream is 120 (1 bit per symbol in MDP2, 2 bits per symbol in MDP4). The continuous phase GMSK modulation, and therefore with memory, is retained at the uplink, for reasons of non-linearity at the mobile transmitters. The number of transmitted symbols is 240 for the TCH, and 120 for the RACH. In both cases, an emitted symbol corresponds to an emitted bit.

Each logical channel has its own channel coding and interleaving techniques. However, this coding and interleaving are organized so as to allow, as much as possible, to unify the structure of the decoder. Each logical channel can implement the following sequence of operations (each of these operations being optional):<ul id="ul0002" list-style="dash" compact="compact"><li>the bits of information are coded with a cyclic and systematic block external code;</li><li>the binary elements resulting from this coding are then coded by the internal code, either of the extended Golay type or of the convolutional type;</li><li>the coded bits are then coded using a repetition code;</li><li>the coded bits obtained are finally interleaved with an interleaving function.</li></ul>

The RACH channel carries, among other things, the identity of a mobile terminal that wants to access the services of the ICO system. It requires a higher link margin than the traffic channels. For this reason, it implements a Golay code followed by a repetition code (3.1 3).

According to the invention, the reference symbols are distributed in several blocks (of at least one symbol) in each data stream.

The number of symbols used for synchronization depends on the characteristics of the actual transmission channel.

The transmission channel underlying the logical channel TCH is a Rice type channel (that is to say composed of a direct path and a multipath portion with a relative zero delay relative to the latter. ) presenting a report <i>K</i> between the power of the direct path and that of the favorable multipaths, with values ​​of the order of 7 to 12 dB. The TCH channel may use as references only 10% of the symbols of the data stream, ie 12 reference symbols per transmitted data stream for the downlink, and 24 reference symbols per transmitted data stream for the uplink.

These symbols may be uniformly distributed, one by one, two by two or four by four over a whole data stream (reference blocks comprising 1, 2, or 4 reference symbols).

For a better calculation of the initial conditions for the estimation of the discrete channel, it is advisable to implement one of the first two forms of synchronization for the downlink.

Table 1 illustrates an implementation mode in which the reference blocks comprise 2 reference symbols. This implementation is used for the TCH downlink channel.<tables id="tabl0001" num="0001"><table frame="all"><title>Table 1</title><tgroup cols="3"><colspec colnum="1" colname="col1" colwidth="24mm" /><colspec colnum="2" colname="col2" colwidth="32mm" /><colspec colnum="3" colname="col3" colwidth="49mm" /><thead><row><entry align="center" valign="top">symbol number</entry><entry align="center" valign="top">length of the field</entry><entry align="center" valign="top">field content</entry></row></thead><tbody><row><entry align="center">0-1</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row><row><entry align="center">2-8</entry><entry align="center">7</entry><entry align="center">useful symbols</entry></row><row><entry align="center">9-10</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">11-28</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">29-30</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">31-48</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">49-50</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">51-68</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">69-70</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">71-88</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">89-90</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">91-108</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">109-110</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">111-117</entry><entry align="center">7</entry><entry align="center">useful symbols</entry></row><row><entry align="center">118-119</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row></tbody></tgroup></table></tables>

For the uplink, the third form (table 2) is the most suitable because it allows to get rid of the memory of the GMSK modulation at lower cost in the choice of initial conditions. In this particular case, only a portion of the waveform corresponding to the blocks of four reference symbols is used for the calculation of these initial conditions.<tables id="tabl0002" num="0002"><table frame="all"><title>Table 2</title><tgroup cols="3"><colspec colnum="1" colname="col1" colwidth="24mm" /><colspec colnum="2" colname="col2" colwidth="32mm" /><colspec colnum="3" colname="col3" colwidth="49mm" /><thead><row><entry align="center" valign="top">symbol number</entry><entry align="center" valign="top">length of the field</entry><entry align="center" valign="top">field content</entry></row></thead><tbody><row><entry align="center">0-3</entry><entry align="center">4</entry><entry align="center">guard symbols</entry></row><row><entry align="center">4-17</entry><entry align="center">14</entry><entry align="center">useful symbols</entry></row><row><entry align="center">18-21</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">22-57</entry><entry align="center">36</entry><entry align="center">useful symbols</entry></row><row><entry align="center">58-61</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">62-97</entry><entry align="center">36</entry><entry align="center">useful symbols</entry></row><row><entry align="center">98-101</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">102-137</entry><entry align="center">36</entry><entry align="center">useful symbols</entry></row><row><entry align="center">138-141</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">142-177</entry><entry align="center">36</entry><entry align="center">useful symbols</entry></row><row><entry align="center">178-181</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">182-217</entry><entry align="center">36</entry><entry align="center">useful symbols</entry></row><row><entry align="center">218-221</entry><entry align="center">4</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">222-235</entry><entry align="center">14</entry><entry align="center">useful symbols</entry></row><row><entry align="center">236-239</entry><entry align="center">4</entry><entry align="center">guard symbols</entry></row></tbody></tgroup></table></tables>

In particular, the BCCH logical channel is used by the mobile terminals to synchronize at the level of the data streams. The underlying real transmission channel is Rice with a value of<i>K</i> of the order of 0 dB. For these two reasons, 32 explicit reference symbols are provided.

Table 3 illustrates the corresponding structure of the data stream.<tables id="tabl0003" num="0003"><table frame="all"><title>Table 3</title><tgroup cols="3"><colspec colnum="1" colname="col1" colwidth="24mm" /><colspec colnum="2" colname="col2" colwidth="32mm" /><colspec colnum="3" colname="col3" colwidth="83mm" /><thead><row><entry align="center" valign="top">symbol number</entry><entry align="center" valign="top">length of the field</entry><entry align="center" valign="top">field content</entry></row></thead><tbody><row><entry align="center">0-1</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row><row><entry align="center">2-8</entry><entry align="center">7</entry><entry align="center">useful symbols</entry></row><row><entry align="center">9-10</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">11-28</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">29-30</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">31-46</entry><entry align="center">16</entry><entry align="center">useful symbols</entry></row><row><entry align="center">47-72</entry><entry align="center">26</entry><entry align="center">explicit reference symbols (synchronization word)</entry></row><row><entry align="center">73-88</entry><entry align="center">16</entry><entry align="center">useful symbols</entry></row><row><entry align="center">89-90</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">91-108</entry><entry align="center">18</entry><entry align="center">useful symbols</entry></row><row><entry align="center">109-110</entry><entry align="center">2</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">111-117</entry><entry align="center">7</entry><entry align="center">useful symbols</entry></row><row><entry align="center">118-119</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row></tbody></tgroup></table></tables>

Part of these symbols (synchronization word) can then be grouped in the middle of the data stream to maintain acceptable performance in terms of synchronization and determination of the maximum aperture of the eye diagram at the receiver. The other part is then distributed over the rest of the data stream, as is the case for the logical channel TCH.

The RACH logical channel is used in the uplink by the mobile terminals to request access to the services of the ICO system. The underlying real transmission channel is generally of the same type as that of the BCCH. Unlike the BCCH channel, which is repetitive, the RACH is transmitted in isolation and must be detected in a single pass. Therefore, the number of reference symbols is 44, (instead of 32 for the BCCH), as specified in Table 4.<tables id="tabl0004" num="0004"><table frame="all"><title>Table 4</title><tgroup cols="3"><colspec colnum="1" colname="col1" colwidth="24mm" /><colspec colnum="2" colname="col2" colwidth="32mm" /><colspec colnum="3" colname="col3" colwidth="49mm" /><thead><row><entry align="center" valign="top">symbol number</entry><entry align="center" valign="top">length of the field</entry><entry align="center" valign="top">field content</entry></row></thead><tbody><row><entry align="center">0-1</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row><row><entry align="center">2-7</entry><entry align="center">6</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">8-31</entry><entry align="center">24</entry><entry align="center">useful symbols</entry></row><row><entry align="center">32-47</entry><entry align="center">16</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">48-71</entry><entry align="center">24</entry><entry align="center">useful symbols (repetition)</entry></row><row><entry align="center">72-87</entry><entry align="center">16</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">88-111</entry><entry align="center">24</entry><entry align="center">useful symbols (repetition)</entry></row><row><entry align="center">112-117</entry><entry align="center">6</entry><entry align="center">explicit reference symbols</entry></row><row><entry align="center">118-119</entry><entry align="center">2</entry><entry align="center">guard symbols</entry></row></tbody></tgroup></table></tables>

As in the case of the BCCH logical channel, part of these reference symbols can be grouped to form synchronization references (2 blocks of 16 symbols). However, since the GMSK modulation used is a memory modulation, the remaining reference symbols must also be grouped before being distributed over the rest of the train. This grouping (two blocks of 6 reference symbols) naturally allows the receiver to temporarily free itself from the memory of the modulation to calculate the initial conditions of the iterative channel estimation algorithm.

In this situation, there are therefore four reference blocks consisting of 6 or 16 explicit reference symbols, and three blocks consisting of 24 implicit reference symbols, corresponding to the repetition of the same useful symbols.

An advantageous compromise is thus achieved between:<ul id="ul0003" list-style="dash" compact="compact"><li>the presence of relatively long reference blocks (6 or 16 symbols), facilitating synchronization;</li><li>the interleaving of the repetition code (3 blocks distributed on the data stream);</li><li>the number of divided blocks (4 + 3) for the estimation of the channel.</li></ul>

As already indicated, the presence of explicit reference symbols is not mandatory, although it simplifies the processing. Implicit reference symbols alone, obtained by repetition, can also be used. Moreover, it is recalled that the use of a repetition code is only an advantageous embodiment, by its simplicity, for obtaining implicit reference symbols. Indeed, these can be obtained whatever the code used, by analyzing the links generated between the symbols useful by this code [6, 9].

When both types of reference symbols are present, the implicit reference symbols are advantageously used to refine the estimation of the channel.

4. Principle of a receiver

Figure 2 illustrates an example of a receiver of a signal according to the invention as well as the corresponding reception method.

The signal corresponding to a received data stream is preamplified 31, then converted to intermediate frequency 32 in order to perform the channel 33 adapted filtering. The intermediate frequency signal is then converted to baseband 34 over two channels in quadrature, or on only one low intermediate frequency channel, then sampled (37).

The sampled signal is then demodulated (39). Moreover, it feeds a synchronization module 36 and an estimation module of the transmission channel 38.

The synchronization module 36 uses the synchronization sequence to estimate the symbol rate as well as the instants corresponding to the maximum opening of the digraph of the eye of the baseband signal received. The signal is sampled (37) at these precise times.

The samples corresponding to a data stream are also used by the receiver to determine an estimate 38 according to the MAP criterion of the realization of the discrete channel at these levels. This estimate 38 is carried out by means of the simplified representation of the discrete channel at the level of the received data stream, the use of the EM algorithm and possibly of an algorithm such as that of Bahl [9].

This iterative algorithm, described later, starts from arbitrary initial conditions that can be obtained advantageously through the synchronization symbols known to the receiver.

The synchronization distributed over the entire data stream according to the invention proves to be of great utility because it makes it possible not only to avoid the convergence of the algorithm towards local maxima of the conditional probability, but also of accelerate this convergence at a lower cost.

The receiver then uses the estimate 38 at the MAP of the discrete channel at a received data stream to demodulate 39 according to a given criterion the information-bearing symbols of this train. The demodulator 39 may in particular provide weighted outputs to improve the performance of the decoder 310.

These outputs, weighted or not, are first deinterleaved 311 and grouped in accordance with the interleaving and distribution performed at the transmitter, then decoded by the channel decoder 310. The demodulated data are then decoded by the decoder source 312, to provide an estimate of the transmitted source signal.

The channel estimate is also used by the receiver to control the progress of preamplification 31, by using an automatic gain correction (AGC) function 35.

5. Example: the ICO system

By way of example, and with reference to FIG. 3, some of the characteristics of the receiver of the ICO system whose transmitter is presented in the previous section are presented below. At the level of the mobile terminals, the synchronization 41 of the data streams is performed in a rough manner 411, in a first step, thanks to the detection of the power profile on the logical channel BCCH.

This synchronization is then refined 412 by means of the synchronization symbols of the data streams of the BCCH channel.

At the space segment level (formed, in the case of the ICO system, by all the satellites and earth stations, each satellite only acting as a repeater), the coarse synchronizations 411 and fine 412 are performed in the same way as by the mobile terminals using the RACH logical channel.

At the level of the mobile terminals and the spatial segments, the estimate 421, 422 of the discrete channel at the level of a received data stream is carried out according to the MAP criterion. This estimate is made thanks to the algorithm described later. The receiver begins by first using the timing symbols to establish suitable initial conditions for the proper operation of the iterative algorithm.

At this stage, he has no idea, even partial, of the values ​​probably taken by the coded information symbols. He then assigns to these information symbols uniform conditional probabilities with maximum entropy 431 (according to the method described more precisely below).

At the downlink level, the symmetry of the MDP2 constellations for the BCCH and the MDP4 for the TCH then leads to a zero contribution of the information symbols in the calculation of the initial conditions of the iterative algorithm 44 for estimating channel. These initial conditions are then used by this algorithm to improve the estimation of the channel 422, this time taking into account the additional contribution 432 of the coded information symbols.

At the uplink level, the channel estimate at the RACH and TCH logical channels follows roughly the same procedures as for mobile terminals. The only difference here is that the GMSK modulation used in the uplink is a memory modulation.

Since the synchronization symbols transmitted by the mobiles are grouped on a basis of at least four by four symbols (in the case of the TCH), the IES due to the memory of the modulation and the adjacent unknown symbols do not affect the symbols in the middle of each grouping. These symbols of the middle of each group can then be used to calculate the initial conditions of the iterative algorithm. For the following iterations, the iterative algorithm then uses the modulation grid of the GMSK to also take into account the remaining synchronization symbols and the coded information symbols (for example using the Bahl algorithm).

In the particular case of the RACH channel, the reception of the data streams by the space segment is almost asynchronous. In addition, the implicit references corresponding to the 3 blocks of 24 symbols obtained by the repetition of a Golay code are used. This characteristic of the invention makes it possible to greatly improve the quality of the estimation of the channel, while also maintaining a good quality of synchronization, thanks to the grouped explicit references.

The position chosen for the implicit reference symbols makes it possible to obtain a good interleaving of the words of the repetition code (3,1,3). Since taking into account the repetition code (3,1,3) can be integrated directly into the estimation algorithm (in the case where the modulation memory can be neglected in the first order), it generates an imperceptible increase of the complexity of it.

6. Modeling and representation of the discrete channel according to the invention

One of the aspects of the invention is based on a modeling and a new simplified representation of the discrete channel seen at the output of the receiver-adapted filter at the level of the data streams sent.

6.1 Modeling the transmission channel

The transmission channel is assumed to be of the Rice type. It consists of a direct path and a multipath portion with a relative delay of almost zero compared to the latter. The direct path is specified by a constant complex attenuation factor. The multipath part resulting from field reflections is characterized as a stationary Gaussian random process of zero mean [1, 16].

This channel is mainly characterized by two parameters. The first is the report,<i>K</i>, between the power of the direct path and that of the multipaths. The second parameter is the channel SPD function<i>S_c(F).</i>

This function is a power spectrum that gives the intensity of the transmission channel as a function of the Doppler frequency <i>f.</i> It is equal to the Fourier transform of the autocorrelation function φ<i>_C</i>(τ) of the channel. It has a bounded breadth of support<i>B</i>_D called the Doppler stretch of the channel. In the context of the invention, this range is assumed to be small compared to the symbol rate 1 /<i>T</i> .

6.2 Modeling and characteristics of transmitted and received signals

The invention can be applied to many types of modulation. In the context of satellite communications, it is particularly concerned with receiving phase-modulated data streams. This includes conventional modulations (MDP2, MDP4, MDP8, MDP16, ...) as well as modulations with a memory such as continuous phase modulations (MSK, GMSK, ...).

The baseband signal corresponding to a data stream transmitted from <i>NOT</i> In the case of a CDM modulation, the symbols are written as (easily adaptable to other types of modulation [1]): <maths id="math0001" num=""><math display="block"><mi>s</mi><mfenced><mi>t</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munderover><mo>Σ</mo><mrow><mi>k</mi><mo>=</mo><mn>0</mn></mrow><mrow><mi>NOT</mi><mo>-</mo><mn>1</mn></mrow></munderover></mstyle><msub><mi>at</mi><mi>k</mi></msub><mo>⁢</mo><mi>x</mi><mo>⁢</mo><mfenced separators=""><mi>t</mi><mo>-</mo><mi mathvariant="italic">kT</mi></mfenced><mo>,</mo></math><img file="EP0802656B1_D0001.tif" /></maths>or <i>T</i> is the symbol period, <i>x</i>(<i>t</i>) is a standard unit Nyquist root shaping filter and the <i>at_k</i> are complex symbols belonging to an arbitrary alphabet <i>AT</i>. The modulus of these symbols is equal to the square root of the baseband energy per transmitted symbol<i>2E_s.</i>

The signal at the input of the receiver corresponds to the baseband signal emitted distorted by the channel and corrupted by a Gaussian white complex additive noise of spectral power in baseband 2<i>NOT</i>₀.

6.3 Modeling the sampled signal at the output of the adapted filter

Doppler spreading <i>B</i>_D is assumed to be weak compared to the symbol / <i>T</i>. Therefore, the samples obtained by matched filtering of the received signal<i>r</i>(<i>t</i>) by the adapted filter <i>x</i>* (-<i>t</i>) and sampling of the resulting signal at the symbol rate at the maximum opening instants of the eye diagram can be approximated, irrespective of the type of modulation [1], by <maths id="math0002" num=""><math display="block"><msub><mi>r</mi><mi>k</mi></msub><mo>=</mo><msub><mi>c</mi><mi>k</mi></msub><mo>⁢</mo><msub><mi>at</mi><mi>k</mi></msub><mo>+</mo><msub><mi>not</mi><mi>k</mi></msub><mo>,</mo><mi>k</mi><mo>=</mo><mn>0</mn><mo>,</mo><mn>1</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>NOT</mi><mo>-</mo><mn>1</mn></math><img file="EP0802656B1_D0002.tif" /></maths>

In this expression, the <i>c_k, k =</i> 0,1, ...,<i>NOT -</i> 1 represent an embodiment of the discrete channel at the transmitted data stream. In the case of the CDM,<i>not_k</i>as for them, represent a realization of a discrete Gaussian additive complex noise. They are independent and have<i>2N</i>₀ as variance. The autocorrelation function,<i>φ_c</i>(<i>l</i>), the discrete channel is deduced directly from that of the transmission channel, <i>φ_C</i>(τ), by sampling it at the symbol rate.

6.4 Simplified representation of the discrete channel at the data stream level

One of the objectives of the invention is to propose a completely new approach to the simplified representation of the outputs of the discrete channel at the level of transmitted data streams. For each data stream, the receiver uses the samples<maths id="math0003" num=""><math display="block"><msub><mi>r</mi><mi>k</mi></msub><mo>=</mo><msub><mi>c</mi><mi>k</mi></msub><mo>⁢</mo><msub><mi>at</mi><mi>k</mi></msub><mo>+</mo><msub><mi>not</mi><mi>k</mi></msub><mo>,</mo><mi>k</mi><mo>=</mo><mn>0</mn><mo>,</mo><mn>1</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>NOT</mi><mo>-</mo><mn>1</mn></math><img file="EP0802656B1_D0003.tif" /></maths> at the output of the adapted filter to detect the corresponding transmitted symbols.

Let (-) 'the transpose operator, <maths id="math0004" num=""><math display="inline"><mi mathvariant="normal">c</mi><mo>⁢</mo><mover><mo>=</mo><mi mathvariant="normal">Δ</mi></mover><mrow><mo>(</mo><msub><mi>c</mi><mn>0</mn></msub><mo>,</mo><msub><mi>c</mi><mn>1</mn></msub><mo>,</mo><mo>...</mo><msub><mi>c</mi><mrow><mi>NOT</mi><mo>-</mo><mn>1</mn></mrow></msub><mo>⁢</mo><msup><mrow><mo>)</mo></mrow><mi>t</mi></msup></mrow></math><img file="EP0802656B1_D0004.tif" /></maths> the vector representing the realization of the discrete channel at the transmitted data stream and <b>The</b> its covariance matrix. The vector<b>c</b> can then be broken down into the form <maths id="math0005" num=""><math display="block"><mi mathvariant="normal">c</mi><mo>=</mo><msqrt><msub><mi>φ</mi><mi>c</mi></msub><mfenced><mn>0</mn></mfenced></msqrt><mstyle displaystyle="true"><munderover><mo>Σ</mo><mrow><mi>k</mi><mo>=</mo><mn>0</mn></mrow><mrow><mi>NOT</mi><mo>-</mo><mn>1</mn></mrow></munderover></mstyle><msub><mi>e</mi><mi>k</mi></msub><mo>⁢</mo><msub><mi mathvariant="normal">b</mi><mi>k</mi></msub><mo>,</mo></math><img file="EP0802656B1_D0005.tif" /></maths>where the <b>b</b><i>_k</i> are the eigenvectors of the covariance matrix <b>The</b> of the vector <b>c</b> and <i>e_k</i> are independent complex Gaussian random variables whose variances are equal to eigenvalues <i>λ_k</i> of the matrix <b>The</b> associated with vectors <b>b</b><i>_k</i> divided by <i>φ_c</i>(0).

The vectors <i>b_k</i> form an orthonormal basis of the complex canonical space at <i>NOT</i> dimensions. The corresponding eigenvalues ​​are assumed to be arranged in descending order. DP of the vector^e = <i>(e₀, e₁, ... e_N-1) ',</i> denoted <i>p</i>(e), is equal to the product of the Gaussian DPs of these components.

The covariance matrix <b>The</b> is poorly conditioned because Doppler spread <i>B_D</i> is small in front of the clock rate 1 /<i>T</i>. Therefore, eigenvalues<i>λ_k</i> present a very steep decay and vanish quickly. One of the contributions of the invention is to take advantage of these qualities of independence of the random variables<i>e_k</i> and rapid decay of eigenvalues ​​λ<i>_k</i> to substantially reduce the complexity of the discrete channel estimator.

In the general case, the receiver has an incomplete knowledge of the characteristics of its local oscillator and those of the SPD <i>S_c(F)</i> the transmission channel. In this case, the most unpredictable SPD from the point of view of information theory [7] is a bounded support flat spectrum of width equal to Doppler spreading.<i>B_D.</i>

The eigenvectors b_k are equal in this case to the SSAD data stream support restrictions [8] that are well defined. The first eigenvalues ​​λ_k corresponding, normalized by the multiplicative factor <i>B_DT</i> / <i>φ_c(0)</i> are shown in Figure 5 for <i>NOT</i> = 120 and <i>B_DT =</i> 1/45. The eigenvectors b_k corresponding are also shown in Figures 6A-6G.

7. Estimation of the discrete channel at the level of a data stream according to the invention

The invention proposes a simple iterative algorithm (FIG. 3) allowing the joint estimation of all unknown parameters of the receiver that it requires to carry out the tasks incumbent upon it.

Among these parameters, one can cite, in a non-exhaustive way, the realization <b>c</b> of the discrete channel, the variance of the noise 2<i>NOT</i>_o [6] and the average energy per symbol received <maths id="math0006" num=""><math display="inline"><mn>2</mn><mo>⁢</mo><msub><mover><mi>E</mi><mo>~</mo></mover><mi>s</mi></msub><mo>⁢</mo><mover><mo>=</mo><mi mathvariant="normal">Δ</mi></mover><mo>⁢</mo><msub><mi>φ</mi><mi>c</mi></msub><mfenced><mn>0</mn></mfenced><mn>.2</mn><msub><mi>E</mi><mi>s</mi></msub><mn>.</mn></math><img file="EP0802656B1_D0006.tif" /></maths> If the DP _at priori of an unknown parameter of the receiver is known, then it is recommended to estimate it according to the criterion MAP. In the opposite case, it is advisable to use the MV criterion for the proper fulfillment of this estimate.

RFP of an achievement <b>c</b> of the discrete channel at a transmitted data stream is determined indirectly by the DP, <i>p</i>(<b>e</b>), <b>e</b> which is of course well known to the receiver. The vector<b>c</b> can then be estimated according to the MAP criterion. DPs of 2<i>NOT</i>₀ and 2<i><o ostyle="single">E_s</o></i> are generally unknown to the receiver. Their estimation is then generally performed according to criterion MV.

By way of example, and without any limitation, FIG. 4 presents the case of the estimate according to the criterion of the MAP of the discrete channel. The estimation of additional parameters such as energy per transmitted symbol can be performed together with that of the channel without much change to the final algorithm [6].

Let's note respectively by <b>r</b> = <i>(<b>r</b>₀, <b>r</b>₁...., <b>r</b>_No.1) '</i><b>at</b> = <i>(at₀, at₁,..., at_N-1)</i>', and n = <i>(not₀, not₁,..., not_N-1)^t</i> received sample vectors, transmitted symbols and noise. For reasons of error correction, synchronization and estimation of the discrete channel at the transmitted data streams, some transmitted symbols are coded or fixed. The vector<b>at</b> of transmitted symbols is then characterized by the discrete DP <i>a priori P</i>(<b>at</b>). MAP estimate<b>ĉ</b> of <b>c</b>, or in an equivalent way, <b>ê</b> of <b>e</b>, is the value <maths id="math0007" num=""><math display="block"><mover><mi mathvariant="normal">e</mi><mo>^</mo></mover><mo>=</mo><mi>arg</mi><mspace width="1em" /><munder><mrow><mi>max</mi><mspace width="1em" /></mrow><mi>e</mi></munder><mo>⁢</mo><mi>p</mi><mfenced separators=""><mi mathvariant="normal">e</mi><mrow><mo mathvariant="normal">|</mo><mi mathvariant="normal">r</mi></mrow></mfenced></math><img file="EP0802656B1_D0007.tif" /></maths> which maximizes the conditional DP <i>at</i> posteriori <i>p</i>(<b>e</b>|<b>r</b>).

The direct resolution of this equation is a problem that is difficult to solve. One of the contributions of the invention is to propose a simple solution to this problem by using the iterative algorithm EM. This algorithm inductively re-estimates the vector<b>e</b> in such a way that a monotonic growth of the conditional PD a posteriori p (<b>e</b>|<b>r</b>) be guaranteed. This reestimation is achieved by maximizing an auxiliary function<i>Q</i>(<b>e, e '</b>) based on Kullback-Leibler information measurement and vector function <b>e</b> and the new vector <b>e</b>.

Given the vector received <b>r</b>, the EM algorithm starts with an arbitrary initial value 51 <b>e</b>⁽⁰⁾ of the vector <b>e</b>. The evolution of the estimate<b>e</b>^(<i>i</i>) to the reestimation <b>e</b>^(<i>i</i>+1) is performed via the auxiliary function <i>Q</i>(<b>e, e '</b>) by implicitly performing the following estimation and maximization steps:<ul id="ul0004" list-style="dash" compact="compact"><li>Estimation step 52: Calculate Q (<b>e</b>^(I),<b>e</b>'),</li><li>Maximization step 53: Finding the estimate <b>e</b>^{(I + 1)} who maximizes <i>Q</i>(<b>e</b>^(<i>i</i>),<b>e '</b>) in terms of <b>e</b>.</li></ul>

Without any limitation, the iterative estimate (54) of <b>e</b> can be done a finite number <b><i>I</i></b> of times. This number is chosen in such a way that the estimate<b>e</b>⁽¹⁾ reached is close enough on average to the optimal estimate <b>ê</b> to ensure an imperceptible degradation of the receiver performance.

For a data vector issued <b>at</b> and a discrete channel realization <b>c</b> given, taking into account the independence of noise vector components <b>not</b> allows to write the conditional RFP <i>p</i>(<b>r</b>|<b>a, e</b>) of the received sample vector <b>r</b> as the product of conditional Gaussian DPs of the components of this vector.

Taking explicit PD expressions into account <i>p</i>(<b>e</b>) and <i>p</i>(<b>r</b>|<b>a, e</b>) and the previous formulation of the EM algorithm, gives <maths id="math0008" num=""><math display="block"><msubsup><mi>e</mi><mi>m</mi><mfenced separators=""><mi>i</mi><mo>+</mo><mn>1</mn></mfenced></msubsup><mo>=</mo><msub><mi>w</mi><mi>m</mi></msub><mstyle displaystyle="true"><munderover><mo>Σ</mo><mrow><mi>k</mi><mo>=</mo><mn>0</mn></mrow><mrow><mi>NOT</mi><mo>-</mo><mn>1</mn></mrow></munderover></mstyle><mfenced><mstyle displaystyle="false"><mstyle displaystyle="true"><munder><mo>Σ</mo><mrow><mi>at</mi><mo>∈</mo><mi>AT</mi></mrow></munder></mstyle><mfrac><msup><mi>at</mi><mo>*</mo></msup><msqrt><mn>2</mn><mo>⁢</mo><msub><mi>E</mi><mi>s</mi></msub></msqrt></mfrac><mo>⁢</mo><mi>p</mi><mfenced separators=""><msub><mi>at</mi><mi>k</mi></msub><mo>=</mo><mi>at</mi><mrow><mo>|</mo><mi mathvariant="normal">r</mi><mo>,</mo><msup><mi mathvariant="normal">e</mi><mfenced><mi>i</mi></mfenced></msup></mrow></mfenced></mstyle></mfenced><mo>⁢</mo><mfrac><msub><mi>r</mi><mi>k</mi></msub><msqrt><mn>2</mn><mo>⁢</mo><msub><mover><mi>E</mi><mo>~</mo></mover><mi>s</mi></msub></msqrt></mfrac><mrow><mo>(</mo><msubsup><mi>b</mi><mi>k</mi><mfenced><mi>m</mi></mfenced></msubsup><mo>⁢</mo><msup><mrow><mo>)</mo></mrow><mo>*</mo></msup></mrow></math><img file="EP0802656B1_D0008.tif" /></maths> as an explicit expression of the <i>m</i>-th component of the estimate <b>e</b>^{(I + 1)}.

In this expression, <maths id="math0009" num=""><math display="inline"><msubsup><mi>b</mi><mi>k</mi><mfenced><mi>m</mi></mfenced></msubsup></math><img file="EP0802656B1_D0009.tif" /></maths> represents the <i>k</i>-th component of the basic vector <b>b</b><i>_m</i> (55) as defined above. In order to further improve the method, it can be provided that the vector base 55 is adapted to better correspond to the channel to be estimated, according to various criteria and in particular the width of the Doppler spread 58. The parameter ω<i>_m</i> is a weighting factor 57 that depends on the average signal-to-noise ratio <o ostyle="single">E_s</o>, <i>l N</i>₀ 59 at the receiver input and eigenvalue <i>λ_m</i> divided by the power <i>φ_c</i>(0) of the discrete channel.

Weighting factors ω<i>_m</i> are also obtained in the case of the estimate of the discrete channel according to the least-squares (MMC) method with perfect knowledge of the transmitted data. They measure the quality of the input of a basic vector<b>b</b><i>_m</i> in the representation of the MAP estimate of the discrete channel.

When the power λ<i>_m</i> / φ<i>_c</i>(0) of the <i>m</i>-th component <i>e_m</i> of <b>e</b> is more important than the normalized noise variance <i>NOT₀</i> / <i><o ostyle="single">E_s</o></i>, then the contribution of the basic vector <b>b</b><i>_m</i> in the representation of the channel estimate is very accurate. The corresponding weighting<i>w_m</i> is then very close to 1.

When this power is smaller than the normalized variance of noise, taking into account the basic vector <b>b</b><i>_m</i> in the estimation of the channel provides more noise than useful information. The weighting<i>w_m</i> is then very close to 0. In sense, the weights <i>w_m</i> play the same role and are based on the same principle as adapted filtering.

As shown in Figure 7 for <i>NOT</i> = 120 and <i>B_DT =</i> 1/45, the weighting coefficients <i>ω_m</i> fade quickly depending on the index <i>m</i> for not too high values ​​of the average signal-to-noise ratio <o ostyle="single">E<i>_s</i></o> / <i>NOT</i>₀. This is of course due to the very rapid decrease in eigenvalues ​​λ<i>_m</i>. One of the advantages of the invention is then to be able to keep in the channel estimation process, without noticeable degradation of the performance of the estimator, only the first few coefficients of the<b>e</b>^(I).

One of the other advantages of the invention is to be able to improve the performance of the estimate of the discrete channel by taking into account by the receiver the memory due to continuous phase modulation (MSK, GMSK, etc.) , synchronization symbols and some or all of the coded structure of the other symbols of the received data stream.

The consideration of synchronization symbols in the channel estimation process is immediate. Let<i>S</i> the set of indices of synchronization symbols in a data stream and <i>α_k</i>, <i>k∈S,</i> the values ​​taken by these symbols. Since these values ​​are known to the receiver, then the conditional DP<i>p</i>(<i>at_k</i> = <i>at</i>|<b>re</b>^(I)<i>), k∈S,</i> can be systematically replaced by 1 for <i>at</i> = <i>at_k</i> and by 0 if not.

Taking into account the memory of the phase modulation used or the coded structure of the remaining symbols is generally less systematic. It can for example be carried out indirectly by determining, via the Bahl algorithm [9] or the SOVA algorithm [10], conditional PDs.<i>p</i>(<i>at_k = a</i>|<b>re</b>^(I)) from the modulation trellis and codes used at the transmitter [13-15].

In the particular case where the modulation is without memory and the symbols are uncoded or are with simple codes such as the repetition codes, the coding structure can be integrated directly into the preceding formula.

This is also valid in the case where, for reasons of complexity, part or all of the coding is not taken into account by the estimator. A typical example is concatenated serial codes where the internal code is often the only one to be considered in the channel estimate. Another typical example is that of coded memory modulations where only the memory of the modulation is taken into account in the estimation process.

The value of the estimate of the channel to which the EM algorithm converges is largely conditioned by the initial conditions used by this algorithm. If these initial conditions are poorly chosen, the EM algorithm can converge to estimates,<b>ê</b> of <b>e</b>, corresponding to local maxima of the conditional DP <i>a posteriori p</i>(<b>e</b>|<b>r</b>)<i>.</i>

The best way to obtain these initial conditions is of course to use the synchronization symbols which are perfectly known to the receiver. Other techniques for overcoming ambiguity are also possible.

The choice of the location of these synchronization symbols in the transmitted data streams is not only decisive for the quality of the channel estimation but also for the good convergence of the EM algorithm towards the optimal estimation which maximizes <i>p</i>(<b>e</b>|<b>r</b>). The uniform distribution of the synchronization symbols at the level of transmitted data streams is beneficial not only for a better estimation of the channel but also for a better stabilization and even an acceleration of the convergence of the EM algorithm.

For a smooth synchronization of the data streams at the receiver, it is sometimes imperative to group the synchronization symbols or distribute them over a small area of ​​data streams. In this case, the EM algorithm has instabilities that can severely compromise the quality of the channel estimate.

One of the contributions of the invention is to be able to stabilize this algorithm by taking into account part or all of the coded structure of the information symbols of the received data stream. Simple codes such as repetitive codes whose codewords are well interlaced at the data stream level can be advantageously used to accomplish this stabilization task at a lower cost.

8. Demodulation and decoding of data according to the invention

Note by <b>ê</b> the <i>I</i> -thest reestimation, <b>e</b>^(<i>l</i>), from <b>e</b>. Based on this estimate and on the vector<b>r</b> of samples at the output of the matched filter, the detector or decoder 510 may be designed to minimize a given criterion. In particular, it can minimize the probability of error of the information data or coded modulated symbols.

By way of example, the optimal detector or decoder 510 which minimizes the probability of symbol error must be found for an information symbol <i>at_k, k∉S,</i> given the symbol <maths id="math0010" num=""><math display="block"><msub><mover><mi mathvariant="italic">at</mi><mo>^</mo></mover><mi>k</mi></msub><mo>=</mo><mi>arg</mi><mspace width="1em" /><munder><mrow><mi>max</mi><mspace width="1em" /></mrow><mrow><mi>at</mi><mo>∈</mo><mi>AT</mi></mrow></munder><mo>⁢</mo><mi>p</mi><mfenced separators=""><msub><mi>at</mi><mi>k</mi></msub><mo>=</mo><mi mathvariant="normal">at</mi><mrow><mo mathvariant="normal">|</mo><mi mathvariant="normal">r</mi></mrow><mo>,</mo><mover><mi mathvariant="normal">e</mi><mo>^</mo></mover></mfenced></math><img file="EP0802656B1_D0010.tif" /></maths> which maximizes the conditional DP <i>p</i>(<i>at_k</i> = <i>at</i>|<b>re</b>).

This detector can take into account the memory of the phase modulation used or the coded structure of the transmitted symbols. This consideration can be realized in particular via the Viterbi algorithm [11, 12], the Bahl algorithm [9] or the SOVA algorithm [10]. These last two algorithms can be used advantageously in the case where the receiver requires confidence values ​​of the decoded data. This is particularly the case for communication systems using serial concatenated coding and where the external decoder needs weighted outputs of the internal code to improve its performance.

In the particular case where the modulation is without memory and the symbols are uncoded or are with simple codes such as the repetition codes, the coding structure can be integrated directly into the preceding formula.

This is also valid in the case where, for reasons of complexity, part or all of the coding is not taken into account by the decoder. A typical example is concatenated serial codes where the internal code is often decoded separately from the external code. In this particular case, the structure of the internal code makes it possible to systematically generate weighted outputs for the external decoder at a lower cost.

APPENDIX: REFERENCES

<ul id="ul0005" list-style="none" compact="compact"><li>[1] <nplcit id="ncit0001" npl-type="b"><text>JG Proakis, "Digital Communications," McGraw-Hill, New York, 1989</text></nplcit>.</li><li>[2] <nplcit id="ncit0002" npl-type="s"><text>AP Dempster, NM Laird and DB Rubin, "Maximum Likelihood from Incomplete Data via the EM Algorithm", J. Roy. Stat. Soc., Ser. 39, 1977</text></nplcit>.</li><li>[3] <nplcit id="ncit0003" npl-type="s"><text>Baum, T. Petrie, G. Soules, and N. Weiss, "A Maximization Technique Occurring in the Statistical Analysis of Probabilistic Functions of Markov Chains," Ann. Math. Stat., Vol. 41, 1970</text></nplcit>.</li><li>[4] <nplcit id="ncit0004" npl-type="s"><text>Liporace, "Maximum Likelihood Estimate for Multivariate Observations of Markov Sources", IEEE Trans. Inform. Theory, IT-28, September 1982</text></nplcit>.</li><li>[5] <nplcit id="ncit0005" npl-type="s"><text>BH Juang, "Maximum Likelihood Estimation for Multivariate Mixture Stochastic Observations of Markov Chains," AT &amp; T Technical Journal, vol. 64, no. 6, July-August 1985</text></nplcit>.</li><li>[6] <nplcit id="ncit0006" npl-type="s"><text>GK Kaleh, "Joint Carrier Phase Estimation and Symbols Decoding of Trellis Codes," International Symposium on Information Theory, San Diego, CA, January 1990</text></nplcit>.</li><li>[7] <nplcit id="ncit0007" npl-type="b"><text>Louis L. Scharf, "Statistical Signal Processing: Detection, Estimation, and Time Series Analysis," Addison-Wesley Publishing Company, New York, 1991</text></nplcit>.</li><li>[8] <nplcit id="ncit0008" npl-type="s"><text>D. Slepian, "Prolate Spheroidal Wave Functions, Fourier Analysis and Uncertainty --- V: The Discrete Case," BSTJ, May-June 1978</text></nplcit>.</li><li>[9] <nplcit id="ncit0009" npl-type="s"><text>LR Bahl, J. Cocke, Jelinek F. and J. Raviv, "Optimal Decoding of Linear Codes for Minimizing Symbol Error Rate", IEEE Transactions on Information Theory, vol. IT-20, March 1974</text></nplcit>.</li><li>[10] <nplcit id="ncit0010" npl-type="s"><text>J. Hagenauer and P. Hoeher, "A Viterbi Algorithm with Soft-Decision Outputs and its Applications," GLOBECOM'89, Dallas, Texas, November 1989</text></nplcit>.</li><li>[11] <nplcit id="ncit0011" npl-type="s"><text>GD Forney, Jr., "The Viterbi Algorithm", Proc. IEEE, 61, March 1973</text></nplcit>.</li><li>[12] <nplcit id="ncit0012" npl-type="s"><text>GD Forney, Jr., "Maximum Likelihood Sequence Estimation of Digital Sequences in the Presence of Intersymbol Interference", IEEE Trans. Inf. Theory, IT-18, May 1972</text></nplcit>.</li><li>[13] <nplcit id="ncit0013" npl-type="s"><text>JK Wolf, "Efficient Maximum Likelihood Decoding of Linear Block Codes Using a Trellis," IEEE Transactions on Information Theory, vol. IT-24, no. January 1978</text></nplcit>.</li><li>[14] <nplcit id="ncit0014" npl-type="s"><text>GD Forney, Jr., "The Viterbi Algorithm", Proc. of the IEEE, vol. 61, no. 3, March 1973</text></nplcit>.</li><li>[15] <nplcit id="ncit0015" npl-type="s"><text>GD Forney, Jr., "Coset Codes --- Part II: Binary Lattices and Related Codes," IEEE Transactions on Information Theory, vol. 34, no. 5, September 1988</text></nplcit>.</li><li>[16] <nplcit id="ncit0016" npl-type="b"><text>RS Kennedy, "Fading Dispersive Communication Channels", John Wiley &amp; Sons, 1969</text></nplcit>.</li></ul>