Method of combatting interference by spatial filtering or spatio-temporal filtering in a multi-channel receiver

Abstract

A method for receiving a signal and for rejecting interference in a multichannel receiver, comprises the steps of: reception, transposition and discretization of the signal received on each of the channels of the receiver, so as to obtain a discretized multichannel signal, synchronization of the discretized multichannel signal, computation, on the basis of the discretized and synchronized multichannel signal, of a matrix {circumflex over (R)} of correlation of the total noise, computation, on the basis of the matrix {circumflex over (R)} of correlation of the total noise, of a vector w comprising amplitude phase weighting coefficients of a multichannel filter, and application, to the discretized and synchronized multichannel signal, of a multichannel filtering processing on the basis of the vector w, and then of a single-channel equalization processing to the filtered signal.

Claims

1. A method for receiving a signal and for rejecting interference in a multichannel receiver, the said method comprising the steps of: reception, transposition and discretization of the said signal received on each of the channels of a receiver, so as to obtain a discretized multichannel signal, synchronization of the discretized multichannel signal, and further comprising the steps of: computation, on the basis of the discretized and synchronized multichannel signal, of a matrix {circumflex over (R)} of correlation of the total noise, computation of a matrix {circumflex over (R)}.sub.XD of intercorrelation between a training sequence and the discretized and synchronized multichannel signal received, computation, on the basis of the matrix {circumflex over (R)}.sub.XD, of an estimate of a director vector associated to one path, computation, on the basis of the said matrix {circumflex over (R)} of correlation of the total noise and of the estimate of the director vector associated to one path, of a vector w comprising amplitude phase weighting coefficients of a multichannel filter, and application, to the discretized and synchronized multichannel signal, of a multichannel filtering processing on the basis of the said vector w, and then of a single-channel equalization processing to the filtered signal.

2. The method for receiving a signal and for rejecting interference according to claim 1, wherein the computation of the matrix {circumflex over (R)} of correlation of the total noise received on the set of channels of the receiver is carried out on the basis: of a matrix {circumflex over (R)}.sub.XX of autocorrelation of the discretized and synchronized multichannel signal received, of a matrix {circumflex over (R)}.sub.DD of autocorrelation of the said training sequence, according to the formula:
{circumflex over (R)}={circumflex over (R)}.sub.XX{circumflex over (R)}.sub.XD.sup.H{circumflex over (R)}.sub.DD.sup.1{circumflex over (R)}.sub.DX.

3. The method for receiving a signal and for rejecting interference according to claim 2, wherein the inverse of the matrix {circumflex over (R)}.sub.DD is pre-computed and stored in a memory of the multichannel receiver.

4. The method for receiving a signal and for rejecting interference according to claim 1, wherein the computation of the vector w comprising amplitude phase weighting coefficients to be applied to each of the channels in the multichannel filtering part (201) comprises: the computation of r.sub.xd(i.sub.0), r.sub.xd(i.sub.0) being the mathematical expectation of the correlation between a training sequence and the multichannel signal received affected with a delay i.sub.0, the computation of according to the formula ={circumflex over (R)}.sup.1r.sub.xd(i.sub.0).

5. The method for receiving a signal and for rejecting interference according to claim 4, wherein the delay i.sub.0 is determined with respect to the position of the path of strongest power.

6. The method for receiving a signal and for rejecting interference according to claim 4, wherein the delay i.sub.0 is determined with respect to the position which maximizes the signal-to-noise plus interference ratio at the output of the multichannel filtering.

7. The method for receiving a signal and for rejecting interference according to claim 6, wherein the position which maximizes the signal-to-noise plus interference ratio is obtained through the formula:
i.sub.0=ArgMax.sub.i{SNIR(i)={circumflex over (r)}.sub.xd(i).sup.H{circumflex over (R)}.sup.1{circumflex over (r)}(i)}.

8. The method for receiving a signal and for rejecting interference according to claim 1, wherein the said single-channel equalization is carried out on the basis of an estimation of the single-channel propagation pathway g given by the formula g=w.sup.H where is a multisensor pathway matrix estimated according to the formula ={circumflex over (R)}.sub.XDR.sub.DD.sup.1.

9. The method for receiving a signal and for rejecting interference according to claim 1, wherein the said multichannel filtering is a spatial filtering.

10. The method for receiving a signal and for rejecting interference according to claim 1, wherein the said multichannel filtering is a spatio-temporal filtering, the method furthermore comprising a step of determining at least one delay .sub.j associated with a multiple path of the signal received, and a step of forming a spatio-temporal multichannel signal vector on the basis of the said discretized and synchronized signal and of the said discretized and synchronized signal delayed by the said delay or delays .sub.j, steps of computing a matrix {circumflex over (R)} of correlation of the total noise, and of applying a multichannel filtering processing taking the said spatio-temporal multichannel signal as synchronized and discretized multichannel signal.

11. A multichannel receiver configured to receive a signal on a plurality of channels and to reject interference, comprising: at least two radio chains, for the reception, transposition and discretization of the said signal received on each of the channels of a receiver, so as to obtain a discretized multichannel signal, and computation circuits configured to synchronize the said discretized multichannel signal, wherein the computation circuits are further configured to: compute, on the basis of the discretized and synchronized multichannel signal, a matrix {circumflex over (R)} of correlation of the total noise, compute a matrix {circumflex over (R)}.sub.XD of intercorrelation between a training sequence and the discretized and synchronized multichannel signal received, compute, on the basis of the matrix {circumflex over (R)}.sub.XD, of an estimate of a director vector associated to one path, compute, on the basis of the said matrix {circumflex over (R)} of correlation of the total noise and of the estimate of the director vector associated to one path, a vector w comprising amplitude phase weighting coefficients of a multichannel filter, and apply, to the discretized and synchronized multichannel signal, a multichannel filtering processing on the basis of the said vector a and of a single-channel equalization processing to the filtered signal.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) The invention will be better understood and other characteristics and advantages will become better apparent on reading the nonlimiting description which follows, and by virtue of the appended figures among which:

(2) FIG. 1a presents a frame structure making it possible to implement the invention;

(3) FIG. 1b presents a conventional functional diagram of multichannel receiver in which a multichannel filtering method according to one embodiment of the invention is implemented;

(4) FIG. 2 presents a functional diagram of multichannel antijammed equalizer allowing the implementation of the invention;

(5) FIG. 3 represents an exemplary result obtained by correlation between a received signal and a training sequence;

(6) FIG. 4 presents a functional diagram of multichannel antijammed equalizer according to another embodiment of the invention, in which the spatial filtering part comprises an additional temporal dimension with respect to the prior art;

(7) FIG. 5 is a representation in the form of a state chart of the steps of an embodiment of the method according to the invention;

(8) FIG. 6 is a representation in the form of a state chart of the steps of an embodiment of the method according to the invention.

DETAILED DESCRIPTION

(9) FIG. 1a presents the frame structure 100 of a signal making it possible to implement the invention. It applies to any type of mono-carrier signal comprising training sequences 101, consisting of reference symbols known by the sender and the receiver, transmitted in alternation with information sequences 102, consisting of useful symbols, or information symbols. This frame structure is common to most current single-carrier and multicarrier communication standards. The GSM standard, the 3G (3.sup.rd generation) standard, the LTE standard (the acronym standing for Long Term Evolution, or 4G), the various DVB standards (the acronym standing for Digital Video Broadcasting), and many others will be cited by way of example.

(10) The purpose of the training sequences is to allow the temporal and frequency synchronization of the receiver, as well as the estimation of the propagation pathway with a view to its compensation by way of an equalizer or of any other suitable technique. Their positions in the frame are known to the receiver.

(11) The length of the training sequences, and their distribution within the signal transmitted, depends on the constraints of implementation of the waveform (such as for example the length of the frames, the maximum displacement speed, the type of propagation pathway, . . . ). Hereinafter in the document, it will be considered that the training sequences are of identical sizes equal to P symbols, and that the data sequences are of identical sizes equal to N symbols. The invention applies in an identical manner when the sizes of the training sequences and of the data sequences vary, or when the training sequences are not all identical.

(12) FIG. 1b presents a conventional functional diagram of multichannel receiver in which a multichannel equalization method according to one embodiment of the invention is implemented.

(13) The receiver 110 comprises at least two antennas (or sensors) 111 and 112, intended to receive a signal. Each of the antennas is the input point of one of the channels of the receiver. It also comprises a radio chain 121 and 122 per channel, the object of which is to transform the analogue signal received on carrier frequency into a discretized baseband signal. Accordingly, the radio chain implements means for converting the signal carrier frequency. A conventional embodiment consists in transposing the signal firstly into intermediate frequency, and then in transforming it into equivalent discrete baseband signals by way of a local oscillator, of a phase-shifter and of mixers creating respectively a first channel I in phase and a second channel Q in phase quadrature with the input signal. The signals delivered by the mixers are discretized by analogue-digital converters, or ADCs. Means for filtering the signal in diverse steps of the signal transposition can also be implemented so as to avoid the phenomena of spectral upturns related to the frequency transpositions and to the reductions in the sampling rate. The signal at the radio chain output is a baseband digital signal sampled at a rate greater than the symbol rate of the modulation (generally an integer multiple of the symbol rate).

(14) Downstream of the radio chain, the receiver comprises computation circuits 130 in charge of the execution of all the signal processing algorithms required to ensure good reception of the signal. These computation circuits can take, for example, the form of a reprogrammable computation machine, such as a processor, a DSP (the acronym standing for Digital Signal Processor), or a microcontroller, the form of a dedicated computation machine, such as a set of logic gates like an FPGA (the acronym standing for Field-Programmable Gate Array) or an ASIC (the acronym standing for Application Specific Integrated Circuit), or the form of any other hardware module allowing the execution of computations. They have, or are linked to, a storage space making it possible to store data, such as for example training sequences or matrices during the diverse processings that they carry out.

(15) The computation circuits are in charge of the synchronization 140 of the receiver. The object of the synchronization is to select the best sampling instant, and to measure and compensate the temporal and frequency shifts due to the propagation of the signal. The synchronization is generally carried out by considering the training sequences. The temporal and frequency position of synchronization determined is identical for the whole set of channels. It can be determined by synchronizing each of the channels independently and then by choosing the position offering the best performance, or by a more specific method of multichannel synchronization, such as that described in European patent EP 0665665 B1, the latter solution offering better performance in terms of robustness when faced with noise and with interference. The resulting signal then takes the form of a synchronized multichannel signal, generally sampled at the symbol rate.

(16) The computation circuits are also in charge of the filtering and anti-jammed multichannel equalization 150 of the signal, which constitutes the subject of the invention. This multichannel equalization comprises an anti-jammed multichannel filtering, followed by an equalization of the single-channel signal resulting from the multichannel filtering. The role of the anti-jammed multichannel filtering consists in optimizing the reception of the set of multipaths associated with the useful signal and in removing the undesired interfering signals by generating voids in the radiation pattern of the antennas in the direction of the interferences and multi-paths, the spreading of whose delays is outside of the window of the single-channel equalizer placed at output. The length of the window of the single-channel equalizer is an implementational choice, determined as a function of a compromise between the probabilities of occurrence of a propagation pathway whose spreading is greater than this size, and the complexity required for its implementation. The single-channel equalization can be carried out either in the time domain or in the frequency domain.

(17) Finally, the computation circuits are in charge of the set of signal processing algorithms 160 required for reconstructing the useful bits transmitted on the basis of the estimate of the string of emitted symbols. These algorithms are for example algorithms for pathway decoding, for deinterleaving, for decryption, or the like, the choice and the implementation parameters of which depend on the communication standard used. According to the type of algorithm used, the estimated symbols transmitted to the computation circuits may be decided or undecided symbols. They may also be associated with an estimation of the SNR so as to allow or otherwise a Log Likelihood Ratio (LLR) computation.

(18) Hereinafter, the following notation will be used: A (upper case) designates a matrix, a (bold) designates a vector, ^ (hat) designates an estimated quantity, .sup.H designates the transpose-conjugate operator, E[a] designates the mathematical expectation of a. P is the length of the training sequences, in number of symbols, N is the length of the information sequences, in number of symbols, K designates the number of channels of the receiver, L is the length of the window of the equalizer, in number of symbols. This window corresponds to the maximum spreading of the paths that are able to be equalized. When the propagation pathway comprises multi-paths whose spreading is greater than this distance L, these paths are seen as interfering signals. x.sub.j(n) designates the signal received on channel j for symbol n, x(n) designates the multisensor signal vector received for symbol n, of dimension K, d(n) designates the signal emitted on the training sequences, s(n) designates the signal emitted on the information sequences, (n) designates the signal estimated on the information sequences, b(n) is a vector of dimension K designating the total reception noise (background noise plus possible interference, plus possible useful multipaths situated outside of the equalization window, that is to say arriving before the first path included in the equalization window or whose delay is greater than L symbols with respect to this first path), h(n), for n=0, . . . , L1, is the vector, of dimension K, containing the impulse response of the multisensor useful propagation pathway associated with the paths entering the equalization window, R designates the autocorrelation matrix of the total reception noise, R.sub.DD designates the autocorrelation matrix of the training signal, R.sub.DX designates the intercorrelation matrix for the cross-correlation between the training signal and the corresponding received signal, and R.sub.XX designates the autocorrelation matrix of the received signal.

(19) We will denote in particular: d(n)=[d(n), d(n1), . . . , d(nL+1)].sup.T the vector of size L formed on the basis of the training symbols, and H=[h(0), h(1), . . . , h(L1)] the multisensor pathway matrix formed on the basis of the vectors h(i) of size (K, L),
as well as the matrices:

(20) $\begin{array}{cc}{\hat{R}}_{\mathrm{DD}}=\frac{1}{P-L+1}\underset{n=L-1}{\overset{P-1}{.Math.}}d\left(n\right){d\left(n\right)}^H,& \left(1\right)\\ {\hat{R}}_{\mathrm{XD}}=\frac{1}{P-L+1}\underset{n=L-1}{\overset{P-1}{.Math.}}x\left(n\right){d\left(n\right)}^H,& \left(2\right)\\ {\hat{R}}_{\mathrm{XX}}=\frac{1}{P-L+1}\underset{n=L-1}{\overset{P-1}{.Math.}}x\left(n\right){x\left(n\right)}^H.& \left(3\right)\end{array}$

(21) The invention therefore applies to a multichannel signal, and follows upon the synchronization step 140.

(22) It relates to the computation of the coefficients which will be used for the multichannel filtering and the equalization of the received signal, and therefore rests in particular on the training sequences.

(23) FIG. 2 presents a functional diagram of multichannel antijammed equalizer allowing the implementation of the invention.

(24) After synchronization, the signal vector received on the array of sensors may be written:

(25) $\begin{array}{cc}x\left(n\right)=\underset{i=0}{\overset{L-1}{.Math.}}e\left(n-i\right)h\left(i\right)+b\left(n\right)& \left(4\right)\end{array}$
with e(n) the signal emitted, equal to d(n) for the training sequences and s(n) for the information sequences.

(26) Denoting by e(n)=[e(n), e(n1), . . . , e(nL+1)].sup.T the vector of size L formed on the basis of the symbols emitted by the useful source, then the above relation can be written x(n)=He(n)+b(n).

(27) The multichannel antijammed equalizer according to the invention comprises a multichannel filtering part 201 and a part 202 carrying out a single-channel equalization. The signal x(n) is received on each of the K channels (in the example two channels 211 and 212 receive respectively x.sub.1(n) and x.sub.2(n)). The multichannel filtering part implements a filter with a single complex coefficient per channel (221, 222), which makes it possible to weight the multichannel input signal by an amplitude phase weighting vector w. The signals arising from the weighting with each of the complex coefficients are summed by a summator 231. The single-channel equalization part 202 implements an equalizer charged with recombining the signals arising from the various useful multi-paths, on the basis of an estimate g of the propagation pathway associated with the recombined signal. This estimate g is of finite length L.

(28) If the structure of the antijammed multisensor equalizer is known, the novelty of the invention resides in the computation of the coefficients w.

(29) In patent EP 0867079 B1, the computation of the spatial filter w is done jointly with the computation of a temporal filter v, so as to minimize a mean square error (MSE) criterion on the training sequences:
EQM=E(w.sup.H.Math.x(n)v.sup.Hd(n).sup.2).(5)

(30) Two schemes making it possible to compute these coefficients, each one exhibiting advantages and drawbacks, are described in the patent.

(31) The first scheme seeks to optimize the global energy of all the paths received, by computing the spatial filter according to the formula:
={circumflex over (R)}.sub.XX.sup.1{circumflex over (R)}.sub.DX.sup.Hv.(6)

(32) The vector v is obtained by computing the eigenvector associated with the minimum eigenvalue of the matrix B, with
B={circumflex over (R)}.sub.DD{circumflex over (R)}.sub.DX{circumflex over (R)}.sub.XX.sup.1{circumflex over (R)}.sub.DX.sup.H.(7)

(33) The computation of v then requires the decomposition into eigenvalues of the matrix B, which is a matrix of size (LL). Therefore, the more significant the length of the pathway to be equalized, the more expensive is this decomposition, this rapidly becoming a major limitation for the use of this processing in current communication standards.

(34) The second scheme is aimed at optimizing the energy of the maximum power path. In this scheme, the secondary paths are not seen, that is to say that the multichannel filtering seeks neither to optimize the gain in the direction of the secondary paths, as in the first scheme, nor to reject them, as in the SMFR.

(35) Accordingly, patent EP 0867079 B1 proposes to compute the spatial filter:
=A.sup.1.Math.(r.sub.xd(i.sub.0){circumflex over (R)}.sub.DX.sup.H{circumflex over (R)}.sub.DDr.sub.Dd(i.sub.0))(8)
with
A={circumflex over (R)}.sub.XX{circumflex over (R)}.sub.DX.sup.H{circumflex over (R)}.sub.DD.sup.1{circumflex over (R)}.sub.DX(9)
i.sub.0 being the position associated with the main synchronization spike, {circumflex over (R)}.sub.DX and {circumflex over (R)}.sub.DD being the matrices {circumflex over (R)}.sub.DX and {circumflex over (R)}.sub.DD obtained by using the vector d(n) instead of d(n), d(n) being the vector d(n) stripped of the component i.sub.0 and r.sub.Dd(i.sub.0) being the correlation product of d(n) and d(ni.sub.0).

(36) Two inversions are required in order to solve the above equations: the inversion of the matrix A, which is of size (KK), K being the number of sensors of the receiver, and the inversion of the matrix {circumflex over (R)}.sub.DD, of size ((L1)(L1)).

(37) The inversion of the matrix A does not a priori pose any problem, the size of this matrix not being related to the length of the equalized pathway but to the number of sensors, this number hardly being able to attain values such that the inversion of this matrix becomes too constraining.

(38) Concerning the matrix {circumflex over (R)}.sub.DD, the latter is computed on the basis of the training sequences. These sequences being known, the inverse of this matrix can be precomputed upstream and stored in memory. This matrix depending on the value of i.sub.0, it is necessary to precompute an inverted matrix {circumflex over (R)}.sub.DD for each of the L possible values of i.sub.0. The memory space required for the storage of these matrices {circumflex over (R)}.sub.DD therefore equals L(L1)(L1).

(39) Certain waveforms, in particular military waveforms, seek to guarantee a low probability of interception and of jamming. Accordingly, they take care to reduce to the maximum any sort of radio signature, signature related to the employing of repetitive patterns, which could allow an exterior receiver to characterize the type of communication employed, and which are so many weak points in terms of robustness against jammers, the jamming of the training sequences alone making it impossible to receive the whole of the transmission.

(40) Within the framework of these waveforms, it is therefore customary to provide a large number of different training sequences, known at one and the same time to the sender and to the receiver, and then to select the sequences used in the transmission from among these sequences, making this choice depend on a known parameter of the whole set of players in the transmission, such as for example the time of transmission and an encryption key.

(41) The implementation of the second scheme of patent EP 0867079 B1 then requires the memory storage of M inverted matrices {circumflex over (R)}.sub.DD, the size of these matrices then becoming a dimensioning criterion for the receiver (M*L*(L1)(L1)).

(42) The subject of the invention consists in proposing a novel mode of computation of the spatial filter w, making it possible to reduce the memory space occupied, while taking care in respect of the implementational complexity related to the inversion of the matrices, and in improving the performance of the processing.

(43) Accordingly, the invention proposes to consider the column vectors of the matrix {circumflex over (R)}.sub.XD, which are denoted hereinafter r.sub.xd(i):
{circumflex over (R)}.sub.XDr.sub.xd(0), . . . , r.sub.xd(L1)].(10)

(44) These vectors constitute estimates of the direction vectors associated with each useful path. Denoting by i.sub.0 the position of the path of strongest power, determined during the synchronization phase as the position corresponding to the maximum of the synchronization criterion, r.sub.xd(i.sub.0) constitutes an estimate of the direction vector associated with the main path.

(45) In contradistinction to the antijammed schemes according to the prior art implementing the Spatial Matched Filter with Reference (SMFR) defined by the amplitude phase weighting vector of the channels:
={circumflex over (R)}.sub.XX.sup.1r.sub.xd(i.sub.0),(11)
the invention implements the spatial matched filter with noise reference only (SMF-NRO) defined by the following vector w:
={circumflex over (R)}.sup.1r.sub.xd(i.sub.0),(12)
with {circumflex over (R)} the total-noise correlation matrix, that is to say of the background noise plus possible interferences, and of the noise related to the possible useful multipaths situated outside of the equalization window.

(46) Indeed, the SMFR exhibits the major drawback of processing the useful multipaths which are decorrelated from the path i.sub.0 as entirely separate interference (they appear in the correlation matrix {circumflex over (R)}.sub.XX), whilst they could be processed effectively by the single-channel equalizer placed at the output of the spatial part. The consequence is that the SMFR may even lead to lower performance than that of a single-channel equalization processing.

(47) In the method according to the invention of spatial matched filtering with noise reference only, the useful multipaths decorrelated from the path i.sub.0 do not appear in the correlation matrix {circumflex over (R)}: consequently, the SMF-NRO does not seek to reject them, but seeks to reject solely the interferences which cannot be exploited by the single-channel equalizer.

(48) The implementation of the spatial matched filtering according to the invention (SMF-NRO) requires the estimation of the total-noise correlation matrix R, which can be obtained directly on the basis of the estimated matrices {circumflex over (R)}.sub.XX, {circumflex over (R)}.sub.XD and {circumflex over (R)}.sub.DD through the formula:
{circumflex over (R)}={circumflex over (R)}.sub.XX{circumflex over (R)}.sub.XD{circumflex over (R)}.sub.DD.sup.1{circumflex over (R)}.sub.XD.sup.H.(13)

(49) These computations require only the inversion of the total-noise correlation matrix R, which is of size (KK), and that of the matrix R.sub.DD, which is of size (LL) and which can be computed upstream and stored in memory.

(50) Concerning {circumflex over (R)}, its size is not dependent on the length of the spread propagation pathway, but on the number of sensors. It is therefore very improbable that this number is high, and its inversion ought not to pose any particular problems.

(51) Concerning {circumflex over (R)}.sub.DD, just like {circumflex over (R)}.sub.DD previously, it can be precomputed on the basis of the training sequences. Since it is a complete matrix, it is necessary to store only a single version of this inverted matrix. Thus, the memory space required is of size (LL) for the inversion of the matrix {circumflex over (R)}.sub.DD, versus L(L1)(L1) for the inversion of the matrix {circumflex over (R)}.sub.DD of the prior art. The memory space required for the storage of this matrix therefore decreases by an order of magnitude close to L, this being particularly advantageous when a large number of these matrices have to be stored.

(52) The above equations make it possible to compute the vector , which represents the complex weighting factor to be applied to each of the channels of the part corresponding to the spatial filtering 201. The signal y(n) resulting from the spatial filtering of the multichannel signals is therefore a single-channel signal equal to y(n)=.sup.Hx(n). This spatial filtering operation makes it possible to remove the jammers and the useful multi-paths situated outside of the window of the equalizer (of size L), and to optimize the gain in the direction of the main path while preserving the various useful paths situated in the window of the equalizer.

(53) When the noise is white temporally, the spatial filtering thus carried out is adapted, that is to say that it makes it possible to maximize the signal-to-noise ratio on the symbol s(ni.sub.0). However, this operation does not make it possible to remove the inter-symbol interference related to the propagation pathway on the symbols (s(n), s(n1), . . . , s(nL+1). This is the reason why this filtering operation must be followed by a step of single-channel equalization of the signal y(n) to obtain the string of estimated symbols (n).

(54) The invention stands out from the schemes described in patent EP 0867079 B1 through the following points:

(55) the first mode of computation of the spatial matched filter described in EP 0867079 B1 is suitable when the spatial filtering is followed by a step of single-channel equalization based on an MLSE equalizer. This type of equalizer is efficacious when the length L of the equalization window is small, and when the constellation used has few states. Specifically, this type of equalizer is not suited to current communications standards for which it leads to worse performance and to higher computational power. Moreover, the invention requires the inversion of a matrix of size (KK), versus the eigenvalue decomposition of a matrix (LL) for this mode of computation of the EP patent application;

(56) the second mode of computation of the spatial matched filter described in EP 0867079 B1 is suitable when the spatial filtering is followed by a step of single-channel equalization based on an MMSE equalizer. The equalization method according to the invention exhibits performance equivalent to this mode of computation, while being more stable numerically and requiring less memory storage space (of the order of LL versus L(L1)(L1) for the EP patent application), thus exhibiting a non-negligible interest, in particular when a large number of training sequences must be stored.

(57) In the method for receiving a multichannel signal according to the invention, the coupling between the part 201 carrying out the spatial filtering of the signals and the part 202 carrying out the single-channel equalization can be carried out in various ways, depending on whether one wishes to optimize the computational power or minimize the interaction between the two parts.

(58) The sequencing of the processings carried out in a single-channel equalizer 202 is the following: Temporal synchronization of the received signal: the input signal is sampled at the sample rate. This step makes it possible to select the right sampling instant, and to under-sample the signal so as to bring it down to the symbol rate. Advantageously, this step can also comprise a frequency synchronization aimed at removing a frequency shift related to the Doppler effect, to clock drifts, etc. . . . Estimation of the single-channel propagation pathway, Equalization properly speaking by any type of single-channel equalizer, such as for example of an MLSE equalizer that can be implemented by way of a suitably matched filtering followed by a Viterbi algorithm, or of an MMSE equalizer, implemented by a transverse equalizer, a decision feedback equalizer (DFE), linear or block-based (BDFE), by an equalization in the frequency domain (FDE), etc. . . .

(59) According to a first embodiment, the part 201 carrying out the multichannel filtering of the signals and the part 202 carrying out the single-channel equalization of the receiver are totally independent. On output from the spatial filtering, the single-channel signal is sampled at the same rate as at input and the single-channel equalizer works in the same manner as if it had to process a single-channel signal directly received on the antenna. This procedure makes it possible to minimize the revamping of code to be performed on the single-channel equalizer when adding a multichannel processing to a single-channel receiver, but is not optimal in terms of computational power required.

(60) According to a second embodiment, the block 140 of the receiver, which carries out the initial multichannel synchronization, is configured to transmit to the multichannel filtering and single-channel equalization block 150 information relating to the optimal synchronization position and the estimated frequency shift. The block 150 then carries out the spatial filtering of the multichannel signal, and is configured to decimate the single-channel signal at the symbol frequency, and optionally to compensate the frequency shift. Thus, the signal exiting the spatial filtering part 201 is synchronized, sampled at the symbol rate, and possibly corrected of an estimated frequency shift. This signal is provided to the single-channel equalizer which does not then have to carry out the step of synchronization and decimation of the signal.

(61) According to a third embodiment, the spatial filtering part decimates the signal at the symbol frequency, and compensates for frequency shifts, in an identical manner to what is done in the second embodiment. On output from the spatial filtering part, the signal is synchronized, sampled at the symbol rate and possibly corrected of an estimated frequency shift. This signal is provided to the single-channel equalizer, as well as an estimate of the single-channel propagation pathway. The single-channel equalizer then performs nothing more than the equalization step properly speaking.

(62) In this third embodiment, the estimate of the propagation pathway is provided to the single-channel equalizer, which uses it to remove the intersymbols interference. This estimate is the propagation pathway vector after spatial filtering, denoted g, of size (1, L), obtained on the basis of an estimation of the multisensor propagation pathway matrix , of size (K, L) through the following formula:
g=w.sup.H(14)
where is the multisensor pathway matrix estimated directly on the basis of the previously computed matrices:
={circumflex over (R)}.sub.XDR.sub.DD.sup.1.(15)

(63) This embodiment, in which the spatial filtering part and the single-channel equalization part of the multichannel receiver are coupled, is particularly advantageous in terms of implementational complexity. Indeed, the spatial matched filtering with noise reference only then exhibits reduced complexity with respect to the prior art, but also makes it possible to reduce the complexity of the processings operated downstream during the single-channel equalization of the signal.

(64) The invention also proposes a second avenue of improvement of the prior art, which relates to the determination of the value of the position i.sub.0, used for the computation of the vector used by the spatial matched filter. A way of determining this value is given in patent EP 0867079 B1, and consists in choosing i.sub.0 as being the position of the path of strongest power, this position being determined during the synchronization phase preceding the computation of the coefficients of the spatial filter.

(65) In the presence of multi-paths, this position is not necessarily optimal. Indeed, the synchronization does not take into account the existence of an equalization algorithm positioned further on in the processing chain. Thus, the synchronization is aimed at selecting the strongest path, by considering the set of other paths to be jammers, for the purpose of pointing the spatial filter in the direction of this path.

(66) The invention proposes to choose i.sub.0 during the step of computing the vector w. The computation can then be carried out on the basis of the total-noise autocorrelation matrix {circumflex over (R)}, which comprises solely the Gaussian white noise and the set of paths that are situated outside of the window of the equalizer. The position chosen by using this matrix will therefore be that which will maximize the rejection of these jammers, in contradistinction to the prior art, in which the set of multiple paths, be they situated in or outside the window of the equalizer, are considered to be jammers.

(67) The position i.sub.0 therefore corresponds to the position which maximizes the SNIR at the output of the spatial filter, and is determined according to the formula:
i.sub.0=ArgMax.sub.i{SNIR(i)={circumflex over (r)}.sub.xd(i).sup.H{circumflex over (R)}.sup.1{circumflex over (r)}.sub.xd(i)}.(16)

(68) Finally, the invention proposes a third avenue of improvement with respect to the prior art, which consists in extending the proposed spatial filtering structure by affording it an additional temporal component.

(69) FIG. 3 represents an exemplary result obtained by correlation between a received signal and a training sequence. The abscissa axis is a time axis graduated in symbol times, while the ordinate axis gives the modulus of the correlation.

(70) In this example, three multi-paths 301, 302 and 303 are observed. The first path 301 is the strongest path. The second path 302 is a path of weaker power, its disparity with respect to the first path allows it to be taken into account by an equalizer of length L positioned on the first path. The third path 303 is a distance away, greater than the length of the window of the equalizer. By positioning the synchronization on the first path, the third will be outside of the horizon of the equalizer, and will be considered to be a jammer.

(71) The addition of an additional temporal dimension to the antenna filtering makes it possible to recombine this signal with the useful signals, and therefore to take it into account, thus improving the signal-to-noise ratio, reducing the inter-symbol interference, and therefore thus improving the performance of the receiver.

(72) FIG. 4 presents a functional diagram of multichannel antijammed equalizer according to another embodiment of the invention, in which the multichannel filtering part comprises an additional temporal dimension with respect to the prior art. One then speaks of spatio-temporal filtering.

(73) In this example, the architecture of the multichannel antijammed equalizer is such that it is able to process only a single additional path. The person skilled in the art wishing to extend this structure to more than one path could very easily achieve same on the basis of the teachings given hereinafter.

(74) The multichannel equalizer such as defined by the invention is based on the multichannel filtering structure of FIG. 2, whose part 401 dedicated to the spatial filtering is modified by the addition, for each of the input channels 211 and 212, of an additional branch comprising a delay line 431 and 432, and a filter with a complex coefficient 421 and 422. Additional paths can be taken into account by adding an additional branch on each of the input channels per additional path.

(75) The delay lines 431 and 432 introduce a delay corresponding to the disparity between the synchronization position and the path that must be taken into account. In the example presented in FIG. 3, this delay corresponds to the disparity between the path 301 and the path 303. The additional path taken into account could also be the path 302, so as to reset it in phase with the main path upstream of the equalizer. In practice, the paths on which the adaptation is carried out are chosen during the initial multichannel synchronization step.

(76) The spatio-temporal filtering presented preserves the spatial-filtering interference rejection capacity presented in FIG. 2 and its optimality for non-spread pathways (propagation in free space and flat fading), and improves it when the useful propagation pathway is frequency selective, by resetting in phase the secondary paths selected between the various channels. This phase resetting can also be applied to the multiple paths exiting the horizon of the equalizer, this exhibiting an interest in the case of propagation pathways whose temporal spreading may be greater than the horizon of the equalizer. An interesting case of application relates to aircraft, for which the use of an equalizer is desirable in order to take account of the multi-paths related to the reflections on the cabin or the neighbouring buildings when the aircraft is on the ground (delays typically less than 5 or indeed 10 s), but may turn out to be very expensive when it entails taking account of delays related to ground reflection when the aeroplane is at altitude (delays of perhaps greater than 20 s for two aeroplanes at high altitudes). The spatio-temporal filtering proposed by the invention affords a solution making it possible to take account of these off-horizon paths of the equalizer in receivers whose single-channel equalizer is not dimensioned to process such paths.

(77) The spatio-temporal filtering structure described is compatible with the various types of waveform comprising reference sequences.

(78) The coefficients w of the spatio-frequency part of this embodiment can be computed in an identical manner to those presented previously, on replacing the vectors x(n) and w by their spatio-temporal equivalents:

(79) $\begin{array}{cc}\tilde{x}\left(n\right)=\left[\begin{array}{c}x\left(n\right)\\ x\left(n+\right)\end{array}\right]={\left[x_1\left(n\right),x_2\left(n\right),x_1\left(n+\right),x_2\left(n+\right)\right]}^T\mathrm{and}& \left(17\right)\\ \tilde{w}=\left[\begin{array}{c}{w}_1\\ w_2\end{array}\right],& \left(18\right)\end{array}$
with w.sub.1 the coefficients determined on the basis of the undelayed path, and w.sub.2 the coefficients determined on the basis of the additional path delayed by .

(80) The output signal of the spatio-temporal filter, after summation, may be written:
y(n)=.sup.H{tilde over (x)}(n)=w.sub.1.sup.Hx(n)+w.sub.2.sup.Hx(n+r)(19)
and is provided to the single-channel equalizer suitably matched to the waveform considered.

(81) Another advantage afforded by the spatio-temporal filtering structure such as described in the invention resides in the capacity to correct the defects of mispairing between the reception channels.

(82) Indeed, the receiver's reception chains are generally imperfect, that is to say that not all the reception channels are exactly identical. These imperfections are due mainly to the variations specific to the analogue hardware components. The antenna filtering processings make the assumption that the signals received on each of the sensors are processed in an identical manner by the radio chains. The imperfections of the hardware components then create variations in the band of the signal which degrade the performance of these processings.

(83) When the antenna filtering exhibits solely a spatial component (see FIG. 2), its frequency response is flat in the band of the signal. The spatial filtering cannot therefore compensate for the mispairing-related frequency variations of the signal. Conversely, the introduction of an additional temporal notion to this antenna filtering processing (FIG. 4) allows the filter to afford a frequency response, and therefore to compensate for the variations related to mispairings.

(84) The invention therefore pertains to a method of multichannel equalization in a receiver, the equalization comprising a spatial or spatio-temporal multichannel filtering part coupled to a single-channel equalization part, as well as in the receiver configured to implement this method.

(85) FIG. 5 is a representation in the form of a state chart of the steps of an embodiment of the method according to the invention, when the envisaged multichannel filtering structure is that represented in FIG. 2.

(86) The first step 501 of the method consists of the reception of the signal, the signal comprising training sequences and data sequences, on a plurality of sensors, or antennas. It also comprises the set of processings culminating in the digitization of the signal, namely in particular the transposition of the signal to baseband and the associated filterings and the analogue-digital conversion.

(87) The second step 502 of the method consists of the temporal and frequency synchronization of the multichannel signal received, this step being able advantageously to be carried out by means of a multichannel synchronization algorithm. The object of the synchronization is to find the sampling instant making it possible to optimize the reception of the signal, and to determine the position in the received signal of the training sequences and of the data sequences. The synchronization can also comprise a frequency synchronization of the signals, aimed at compensating for a shift between the clocks of the transmit and receive equipment.

(88) The third step 503 of the method consists in computing, on the basis of the training sequences of the signal received on the set of channels of the receiver, a matrix {circumflex over (R)} of correlation of the total noise received, the computation of this matrix being able advantageously to be done on the basis of the formula:
{circumflex over (R)}={circumflex over (R)}.sub.XX{circumflex over (R)}.sub.XD{circumflex over (R)}.sub.DD.sup.1{circumflex over (R)}.sub.XD.sup.H.(20)

(89) In particular, this computation requires the inversion of a matrix R.sub.DD of autocorrelation of the training sequence on which the estimate of the coefficients of the spatio-temporal equalizer is made. In order to limit the computational burden, this inverse matrix can advantageously be stored in a memory of the multichannel receiver according to the invention.

(90) The fourth step 504 of the method consists in computing, on the basis of the matrix {circumflex over (R)} and of a value r.sub.xd(i.sub.0), the vector w containing the complex coefficients that must be applied by the receiver during the multichannel filtering step. In this case, the multichannel filtering is a spatial filtering. The value r.sub.xd(i.sub.0) is the mathematical expectation of the correlation between a training sequence and the multichannel signal received affected with a delay i.sub.0. This delay i.sub.0 can be selected as being the delay between the synchronization position and the position of the signal of strongest power, or else the position which maximizes the SINR on output from the spatial processing.

(91) The fifth step 505 of the method consists in applying the multichannel spatial filtering computed to the received signal, to obtain a single-channel signal, and then in equalizing this signal so as to remove the residual inter-symbol interference.

(92) On completion of the spatial filtering, the signal is a single-channel signal that may be written y(n)=(n).sup.Hx(n), and whose interferences and multiple paths situated outside of the window of the single-channel equalizer have been attenuated. The object of the single-channel equalization which follows the filtering is to exploit the set of residual multiple paths contained in the equalization window to remove the interferences between symbols.

(93) The method according to the invention therefore makes it possible to compute the coefficients of the vectors that must be used by the spatial filtering part and by the single-channel equalization part of a spatio-temporal equalizer so as to combat the interference.

(94) Steps 501 to 504 of the method are carried out on the basis of the training sequences received, these sequences then being known to the receiver. The computations can be carried out for each training sequence. The processing is then suitable for following the fast variations of the propagation pathway. They can also be carried out for each training sequence, and combined together, so as to smooth the noise-related computation errors.

(95) Step 505 is applied to the useful data sequences. The equalized and spatially filtered useful symbols which result therefrom can thereafter be processed by the signal processing algorithms as a function of the communication standard used, in such a way as to reconstruct the useful bits transmitted.

(96) According to one embodiment, the multichannel filtering block according to the invention furthermore comprises a temporal dimension on each of the reception channels: the multichannel filtering of the input signals is carried out by a spatio-temporal structure.

(97) FIG. 6 is a representation in the form of a state chart of the steps of an embodiment of the method according to the invention, when the structure of the envisaged multichannel filtering block is that represented in FIG. 4, for which the filtering of the input signals is carried out by a spatio-temporal structure.

(98) Thus, the method comprises, in addition to steps 501 and 502 of reception, transposition and discretization of the signal and of multichannel synchronization, a step 601 of determining at least one delay (.sub.1, .sub.2, . . . , .sub.N) between the synchronization position and one or more paths of the received signal. This or these delays are advantageously estimated jointly with the step 502 of synchronization on the basis of the function for the correlation between the received signal and the training sequences.

(99) A step 602 consists in delaying each of the channels of the receiver by the computed delay or delays and in forming a multichannel signal vector on the basis of the discretized and synchronized multichannel signal received x(n), and of the delayed versions of this same signal:

(100) $\begin{array}{cc}\tilde{x}\left(n\right)=\left[\begin{array}{c}x\left(n\right)\\ x\left(n+{}_1\right)\\ \mathrm{.Math.}\\ x\left(n+{}_N\right)\end{array}\right].& \left(21\right)\end{array}$

(101) The following steps of the method are identical to those of the method of FIG. 5, except that they use the signal {tilde over (x)}(n) instead of x(n).

(102) Thus, step 503 of computing the total-noise correlation matrix {circumflex over (R)} is done on the basis of the signal {tilde over (x)}(n). This step comprises the computation of the matrices {circumflex over (R)}.sub.XX and {circumflex over (R)}.sub.XD formed on the basis of the spatio-temporal signal {tilde over (x)}(n).

(103) The computation 504 of the vector w is carried out on the basis of {circumflex over (R)}, w constituting the spatio-temporal filter that must be applied to the multichannel signal {tilde over (x)}(n).

(104) Finally, the method comprises step 505 of multichannel filtering on the basis of the vector w, but this step is applied to the signal {tilde over (x)}(n). It then entails a spatio-temporal filtering, the resulting single-channel signal having thereafter to be processed by a single-channel equalizer.