WO2007090995A2 - Method for receiving a signal using an improved propagation channel estimate, corresponding receiver device and computer programme product - Google Patents

Abstract

The invention concerns a method for receiving a signal formed by a temporal succession of symbols comprising, at spectral level, at least one unmodulated guard element and one modulated useful data element, said useful data elements including reference data elements, and informative data elements. According to the invention, such a method uses a step of estimating at least one propagation channel between at least one transmitter and said at least one receiver, including: a sub-step of determining (22) a first estimate of said propagation channel; a sub-step of determining an improved estimate (23) of said propagation channel by multiplying said first estimate vector by a predetermined decorrelation matrix based on at least one structural characteristic of said signal.

Description

A method of receiving a signal implementing an improved estimate of a corresponding propagation channel, receiving device and computer program product.

FIELD OF THE INVENTION The field of the invention is that of fixed or mobile communications, over the air (radiocommunications) or by wire.

More specifically, the invention relates to the reception and estimation of propagation channels, between at least one transmitter and at least one receiver, in a system of MIMO type ("Multiple Input Multiple Output") or SISO ("Single Input Single"). Output "), from the transmission of signals comprising reference data elements called pilots, known from at least one receiver.

The invention applies in particular to any system implementing an equalization in the frequency domain in reception, for example the muti-carrier type systems.

Thus, the invention finds inter alia applications in systems implementing OFDM type techniques (Orthogonal Frequency

Multiplex Division "), OFDMA (Orthogonal Frequency Division Multiple

Access "), Multi-Carrier Coded Division Multiple Access (MC-CDMA), IFDMA (Interleaved Frequency Division Multiple Access), LP-OFDM

("Linear Precoded Orthogonal Frequency Division Multiplex"), or in single carrier systems implementing equalization in the frequency domain.

In addition, the invention applies to uplink communications (from a terminal to a base station) as well as to downlink communications (from a base station to a terminal).

2. Prior Art

Receiving propagation channel estimation in digital communications has been the subject of much research. Indeed, this estimate directly affects the performance of a communication system. More precisely, it is recalled that it is strongly recommended to estimate the propagation channel efficiently at a receiver, in order to be able to equalize the signal received via a coherent demodulation, and to detect the transmitted data, in particular in the absence of using differential modulations. Indeed, in a high-speed communication system, the noise level is doubled. Differential modulations are thus not used, or little, in such a system, since these modulations only apply to phase modulations (for example of the QPSK type for "Quadrature Phase Shift Keying"), and not to amplitude modulations (for example of QAM type for "Quadrature Amplitude Modulation"), the latter making it possible to achieve high data rates.

It is therefore desirable to be able to make a correct estimate of the propagation channel, approaching as far as possible the real transmission channel, especially in high-speed systems.

There are mainly two classes of channel estimation techniques: estimation techniques in the frequency domain, and estimation techniques in the time domain.

2.1 Estimate in the frequency domain

Conventionally, the estimation of the channel in the frequency domain is carried out from the insertion in the useful data stream to transmit, before transmission, reference data elements called pilots, to known locations of the receiver.

In reception, the values taken by these reference sequences are read, and the complex gain of the channel at these locations is determined.

A major disadvantage of these estimation techniques in the frequency domain is that they do not allow to estimate directly, that is to say without interpolation in the frequency domain, that as many coefficients of the propagation channel as pilots introduced.

Thus, this class of estimation technique leads to a moderate channel estimation quality, the SNR (Signal to Noise Ratio) level. Indeed, compared to a perfect knowledge of propagation channel, the latter usually results in a degradation of 2dB in terms of bit error rate BER (or BER for "Binary Error Rate" in English).

2.2 Estimation in the time domain The following is the second class of channel estimation techniques: the estimation in the time domain. A. Ratings

For this purpose, for example, the transmission of an OFDM symbol comprising only reference data elements (ie known drivers of the receiver) in a multi-carrier system is considered.

More precisely, this system is defined by the parameters N, corresponding to the size of the Fourier Transform (FFT - "Fast Fourier

Transfrom "), N _moi , corresponding to the number of modulated carriers, and Δ, corresponding to the size of the cyclic prefix (CP -" Cyclic Prefix "), also called guard interval.

In reception, the signal received y (n) after demodulation is of the form: y (n) = U (ω) b _p (n) + w (n) with: b _p (n) a vector of size N, formed drivers of the transmitted OFDM symbol;

H (ω) a diagonal matrix of size (N, N), including the frequency response of the channel; and w (n) a vector of size N corresponding to Gaussian white additive noise (BABG). In particular, we denote (W (Cu) the vector defined by the main diagonal of

H (ω): tf (ω) = diag {U (ω)} => /} vFh with: h the impulse response of the propagation channel, in the time domain; and F a size matrix (JV, Δ + 1), comprising the first Δ + 1 columns of the square Fourier matrix F (JV, JV) (again respecting the Matlab writing - trademark: F = F (:, l: Δ + 1)); where the Fourier matrix F is of the following form:

It can notably be noted that the vector h is a vector of size Δ + 1, where Δ corresponds to the size (length) of the cyclic prefix.

Finally, in the case where null subcarriers are emitted at the edge of the spectrum, for example to respect an emission mask, a partial Fourier matrix F 'of size (JV _mod , Δ + 1) is defined, which includes

N _mod lines corresponding to the modulated subcarriers of the Fourier matrix

F and the first Δ + 1 columns of the Fourier matrix F.

These different matrices F, F and F 'are schematically illustrated in relation to FIGS. 1A, 1B and 1C, respectively. B. Channel estimation in the time domain

As presented in the doctoral dissertation of JF Hélard "Coded Modulations in Treilles Associated with a Multiplex of Orthogonal Carriers, in Presence of Affected Channels of Multiple Pathways" (University of Rennes 1, May 1992), to estimate the impulse response of the H-channel in the time domain, an inverse Fourier transform, referred to as IFFT (Inverse Fast Fourier Transform), is applied to the frequency estimation of the channel, retaining a value equal to 10 vvaakleurs, using the matrix F (where corresponds to the Hermitian operator):

où la division entre les deux vecteurs y(n) et b_p(n) dénote une division élément par élément. On peut notamment remarquer que la multiplication par l'inverse de la racine carrée de N permet de normaliser l'estimation de la réponse impulsionnelle du canal par la taille de la matrice de Fourier F (de taille (N, N)).

where the division between the two vectors y (n) and b _p (n) denotes a division element by element. It can notably be noted that the multiplication by the inverse of the square root of N makes it possible to normalize the estimation of the impulse response of the channel by the size of the Fourier matrix F (of size (N, N)).

We then return to the frequent domain! el via an FFT: SW V "iv ^FF * - b _p (»)

It can notably be noted that this time domain channel estimation technique gives good results, since the entire spectrum of the signal is used.

However, a major disadvantage occurs in the case where carriers or elements of zero value (said null or guard) are inserted at the edge of the spectrum so as to respect a certain emission template (allowing the signal issued especially not to have a spectral occupancy interfering with the neighboring bands, and thus to ensure a spectral regulation of the systems).

Indeed, in this case, the propagation channels estimated according to this technique have edge effects, which leads to an error floor for high SNR (of the order of 18 dB), significantly limiting performance.

To remedy this problem, it was then proposed a direct adaptation of this channel estimation technique, in the case where null subcarriers exist. To do this, we consider vectors y (n) and b _p (n) no longer of length N, as previously, but of length N _mod , and the Fourier transforms are calculated using the matrix F ', of size (iV _mod , Δ + 1).

This gives the following estimate:

Cependant, il s'avère que cette technique appliquée sur des vecteurs de longueur N_moά est moins performante que Ia technique présentée précédemment appliquée à des vecteurs de longueur N,

However, it turns out that this technique applied to vectors of length N _moά is less efficient than the previously presented technique applied to vectors of length N,

Indeed, this technique applied on vectors of length -V _mod is again characterized by the appearance of an error floor for the highs SNR significantly limiting performance, including edge effects that this estimate of the channel presents.

3. Objectives of the invention

The invention particularly aims to overcome these disadvantages of the prior art.

More specifically, an object of the invention is to provide a reception propagation channel estimation technique, which performs better than the known estimation techniques, when elements or carriers that are not modulated by information to be transmitted are inserted at the edge of the transmission spectrum of the data signal.

In particular, it is an object of the invention to provide such a technique with improved performance compared with conventional frequency domain estimation techniques, for example for OFDM-type multi-carrier systems, and with respect to the known techniques of estimation in the time domain.

Thus, an object of the invention is to provide such a technique delivering an estimated channel having no or little effect of edges, and not leading to an error floor.

Another objective of the invention is to implement a technique for estimating the impulse response of a channel that is not very complex to implement.

It is another object of the invention to provide such a technique which is suitable for SISO or MIMO type systems, for single-carrier or multicarrier modulations, possibly combined with a multiple access technique (for example of the CDMA type).

4. Presentation of the invention

These objectives, as well as others which will appear later, are achieved by means of a method of receiving a signal formed of a temporal succession of symbols comprising, at the spectral level, at least one guard element. not modulated by at least one piece of information to be transmitted and a piece of useful data modulated by at least one piece of information to be transmitted, the useful data elements comprising, on the one hand, reference data items called pilots, the value at the sending of the pilots being known to at least one receiver intended to perform a reception of said signal, and on the other hand informative data elements.

According to the invention, such a method implements a step of estimating at least one propagation channel between at least one transmitter and at least one receiver, comprising:

a sub-step of determining a first estimate of the propagation channel from the reference data elements, delivering a first channel estimation vector;

a substep of determining an improved estimate of the multiplication propagation channel of the first estimation vector by a predetermined decorrelation matrix as a function of at least one structural characteristic of the signal, enabling at least part of the decorrelation to be decorrelated; improved estimate for the guard elements of the improved estimate for the useful data items.

Thus, the invention is based on a completely new and inventive approach to the estimation of at least one propagation channel, in reception, making it possible to improve this estimate, especially when the transmitted signal includes guard elements, c that is, elements or carriers that are not modulated by information to be transmitted, in the spectral domain.

Indeed, such guard elements are conventionally added to the ends of the spectrum during the transmission of the signal, so that the emitted signal does not have a spectral occupation interfering with the neighboring bands, and respects a certain emission mask. Thus, these guard elements correspond, for example, to carriers of zero value, in a multicarrier system, or elements of zero value, in an IFDMA system, making it possible to regulate the spectral occupation of the transmitted signal. However, these guard elements disturb the estimation of the propagation channel.

Thus, the invention proposes to determine an improved estimation of the propagation channel, by multiplying a first estimation of the channel by a decorrelation matrix, making it possible to decorrelate the estimate of the propagation channel for the unmodulated guard elements by information to transmitting, from the estimation of the propagation channel for the useful data elements modulated by information to be transmitted.

For example, in the context of receiving a multicarrier signal, the decorrelation matrix does not take into account, according to the invention, values of the propagation channel in the part where there are null carrier regions.

This decorrelation matrix is in particular predetermined according to at least one structural characteristic of the signal, such as the size of a cyclic prefix inserted between two symbols before transmission, the number of modulated useful data elements (respectively carriers or valuable elements non-zero in a multi-carrier or single-carrier system), ....

Advantageously, the sub-step of determining an improved estimate implements the following steps:

transformation of the first estimation vector from the frequency domain to the time domain;

multiplying the first estimation vector in the time domain by the decorrelation matrix, delivering a corrected estimate of the impulse response of the propagation channel;

transformation of the corrected estimation of the time domain into the frequency domain, delivering the improved estimation of the propagation channel.

Thus, according to the invention, a degraded estimate of the propagation channel in the frequency domain (first estimate in the frequency domain), which is expressed in the time domain (first estimate in the time domain), is first determined. This vector is then multiplied first estimation by the decorrelation matrix, delivering a corrected estimate of the propagation channel in the time domain. This time domain corrected estimate is finally transformed into the frequency domain, delivering the improved estimate of the propagation channel. Preferably, the decorrelation matrix is a pseudo-inverted matrix of the hermitian matrix product of a partial Fourier matrix by the corresponding partial Fourier matrix: p = (F ' ^H F0 ⁺ with: P the decorrelation matrix; F 'the partial Fourier matrix denotes the Hermitian operator ^* denotes the pseudo-inverse operator, the partial Fourier matrix F' being an extract of a Fourier matrix F defined by:

avec : N un entier ;

with: N an integer;

2n

W _N : = e ^JN .

Recall that classically called by a pseudo-inverse matrix a matrix A ^ obtained by inverting only the invertible part of the original matrix A.

More precisely, the partial Fourier matrix F 'comprises the coefficients of the Fourier matrix F corresponding to the time / frequency locations of the reference data elements.

Advantageously, the step of transforming the first estimation vector implements a multiplication of the first estimation vector in the frequent domain by the Hermitian of the size Fourier matrix F.

(N, N), delivering the first estimate vector in the time domain, where JV is the total number of unmodulated guard elements and user data elements modulated in the spectrum of the transmitted signal.

The multiplication step implements a selection of JV coefficients of the first estimation vector in the time domain, among the JV coefficients obtained, and a multiplication of the JV 'coefficients by the decorrelation matrix P of size (JV', JV ' ), delivering the corrected estimate in the time domain.

The transformation of the corrected estimation implements a multiplication of the time domain corrected estimation by the partial Fourier matrix F ', of size (JV _mod , JV'), delivering the improved estimation of the propagation channel in I frequency domain, where JV _mod is the number of modulated useful data elements. We therefore, classically, JV _mod less than or equal to JV. More specifically, when the transmitted signal is formed of a succession of symbols comprising, at the spectral level, at least one guard element and a useful data element, JV _{mod is} strictly less than JV. By taking again the notations thus defined, the improved estimate thus determined is of the form:

où : représente, aux emplacements temps/fréquence des éléments de

données de référence, une division du signal reçu par l'élément de données de référence situé à l'emplacement temps/fréquence, élément de données de référence par élément de données de référence.

where: represents, at the time / frequency locations of the elements of

reference data, a division of the signal received by the reference data element at the time / frequency location, reference data element per reference data element.

In particular, the signal y _p (n) takes the values of the signal received after demodulation y (n) at the time / frequency locations where the reference data elements are positioned. According to an advantageous embodiment of the invention, a cyclic prefix of length Δ being inserted between the symbols in the time domain, the number JV 'is equal to the length of the cyclic prefix plus one:

JV '≈ Δ + 1. The number N 'thus corresponds to the maximum spread of the temporal response of the propagation channel.

The method according to the invention is also remarkable in that it implements, when the reference elements are distributed in the space time / frequency ("scattered pi lots"):

a step of determining at least two subsets of carriers each comprising at least one reference data element;

a step of determining a partial improved estimate of the channel applied to each of the subsets; a step of two-dimensional interpolation from the partial improved estimates, in order to determine an estimate of the improved channel for each carrier of the time / frequency space. Thus, the invention applies whether the pilots are distributed within the data signal to be transmitted ("scattered pilots"), or continuous within the same symbol ("full pilots").

Advantageously, the method according to the invention furthermore implements a step of equalizing the signal received in the frequency domain, based on the improved estimation of the propagation channel.

This equalization of the signal received in the frequency domain can notably be implemented by dividing the signal by the estimate of the propagation channel.

The invention also relates to a device for receiving a signal formed of a temporal succession of symbols comprising, at the spectral level, at least one guard element that is not modulated by at least one piece of information to be transmitted, and a modulated useful data item. by at least one piece of information to be transmitted, the useful data elements comprising, on the one hand, reference data elements called pilots, and on the other hand informative data elements.

According to the invention, such a reception device comprises means for estimating at least one propagation channel between at least one transmitter and one receiver, comprising:

means for determining a first estimate of the propagation channel from the reference data elements, delivering a first channel estimation vector; means for determining an improved estimation of the multiplication propagation channel of the first estimation vector by a predetermined decorrelation matrix as a function of at least one structural characteristic of the signal, enabling the improved estimate to be decorrelated at least in part; for the guard elements of the improved estimate for the useful data items.

Such a device can in particular implement the reception method as described above. It is therefore suitable for determining an improved estimate of at least one propagation channel between at least one transmitter and a receiver of the following form: tf (ω) = * F '(F' ^H F ') ^f ¥' ^H y _P (n> b _D (n) by taking again the notations defined previously.

The invention finally relates to a computer program product downloadable from a communication network and / or stored on a computer readable medium and / or executable by a microprocessor, comprising program code instructions for the implementation of the method of reception previously described.

5. List of figures

Other characteristics and advantages of the invention will appear more clearly on reading the following description of a preferred embodiment, given as a simple illustrative and nonlimiting example, and the appended drawings, among which: FIGS. , IB and IC respectively illustrate, schematically, the matrices F, F and F '; Figure 2 shows the general principle of the reception method according to the invention; FIGS. 3A and 3B illustrate the construction of a partial Fourier matrix F ', implemented in the reception method of FIG. 2, as a function of the distribution of the reference data elements; FIGS. 4A, 4B and 4C show the comparative performances of the estimation techniques according to the prior art and according to the invention; Figure 5 is a simplified representation of the hardware structure of a receiving device according to the invention. 6. Description of an embodiment of the invention

The general principle of the invention is based on the weighting, in reception, of a vector of first channel estimation in the time domain, delivering a corrected estimate of the impulse response of the propagation channel, and on the transformation of this corrected estimate. from the time domain to the frequency domain, delivering an improved estimate of the propagation channel.

To do this, the first estimation vector is multiplied by a decorrelation matrix P, making it possible to decorrelate the channel estimate for the guard elements of the channel estimate for the useful data elements of the received signal. Thus, for the carriers corresponding to the modulated useful data elements, an improved estimate is obtained close to the actual propagation channel.

In relation to FIG. 2, the general principle of receiving a signal according to the invention is presented. More particularly, a preferred embodiment of the invention is described according to which the received signal is formed of a temporal succession of OFDM symbols.

The skilled person will easily extend this teaching to the emission of a single carrier signal, emitted for example as part of an IFDMA transmission. In particular, each OFDM symbol includes JV carriers, including N _modulators modulated by information to be transmitted, also called useful data elements, and (N-N _moά ) carriers unmodulated by information to be transmitted, also called guard elements. . A cyclic prefix of size Δ is also inserted between each OFDM symbol.

It is recalled in particular that in the time domain, the channel is assumed to have a size which does not exceed that of the cyclic prefix plus one, that is to say Δ + 1, which thus makes it possible to avoid interference between symbols. .

Therefore, using exactly the same amount of information via the reference data elements (pilots), we need to estimate much less propagation channel coefficients, as soon as Δ </ V _mod (and in the in most cases, we check that Δ "N _n ^).

In relation with FIG. 2 and taking again the notations presented previously in relation with the prior art, the transmission of an OFDM symbol comprising only reference data items (pilots) is first considered. Recall that the receiver knows this reference sequence.

During a first step 21, the reception method implements a reception of the signal y (n), ie: y (n) = H (ii) b _/ , ()) + w ()) with: b _p (n) a vector of size N, formed by the drivers of the transmitted OFDM symbol;

H (G)) a diagonal matrix of size (N, N), including the frequency response of the channel; and w (n) a vector of size N corresponding to the Gaussian white additive noise.

Then, the reception method according to the invention implements a step 22 of determining a first estimate of the propagation channel, in the frequency domain, from the reference data elements. This first estimate, delivering a vector of first channel estimation in the frequency domain, is for example implemented by dividing the vector y (n) corresponding to the signal received by the vector b _p (n) carrying Ia.

emplacements temps/fréquence des éléments de données de référence, une division du signal reçu par ï' élément de données de référence à l'emplacement temps/fréquence correspondant, élément de données de référence par élément de données de référence.precisely, represents, to

time / frequency locations of the reference data elements, a division of the signal received by the reference data element at the corresponding time / frequency location, reference data element per reference data element.

The reception method according to the invention then implements a step 23 of determining an improved estimate of the propagation channel, from the first estimation vector.

To do this, during a step 231, an inverse Fourier transform is implemented on the first estimation vector, making it possible to express this first-time estimation vector in the time domain. This transformation notably implements a multiplication of the first estimation vector, in the frequency domain, by the Hermitian of the Fourier matrix F of size (JV, N), corresponding to the sampled useful band of the signal, less than the band. sampled signal. It is recalled that the Fourier matrix F is of the following form:

During a next step of multiplication 232, JV 'coefficients among the JV coefficients of the first estimate vector in the time domain are retained. Note that N 'is an integer less than JV, defined from the properties of the propagation channel. Thus, advantageously, N 'is greater than or equal to the maximum spread of the temporal response of the channel. For example, N 'is chosen to be equal to the maximum size of the channel in the time domain, that is to say: iV' = Δ + 1. It is recalled that the channel is assumed to have a size that does not exceed Δ + 1 in the time domain, which avoids interference between symbols. The vector of N 'coefficients thus obtained is then multiplied by a decorrelation matrix P, corresponding to a pseudo-inverted matrix of the Phermitian matrix product of a partial Fourier matrix by the corresponding partial Fourier matrix:

P = (F ^ F) ^{1 "} with: P the decorrelation matrix;

F 'the partial Fourier matrix; ^H denotes the Hermitian operator; 'denotes the pseudo-inverse operator.

This partial Fourier matrix F 'is a matrix extracted from the Fourier matrix F. More precisely, this partial Fourier matrix F', of size

(N _mod , N '), includes the coefficients of the Fourier matrix corresponding to the time / frequency locations of the reference data elements. It can notably be noted that this matrix is not of full rank.

During the multiplication step 232, the obtained vector is also normalized by the size of the Fourier matrix F by multiplying the vector obtained by the inverse of the square root of N.

FIG. 3A thus illustrates the construction of the partial Fourier matrix F ', according to this exemplary embodiment in which an OFDM symbol comprising only pilots (denoted' x ') is transmitted. In this FIG. 3A, the guard elements are denoted by O ', and the useful data elements are denoted by' D '. Thus, considering for example:

- N = S, where N is the number of unmodulated guard elements and useful data elements modulated per symbol;

- N _mod = 6, with N _mod the number of user data elements modulated per symbol; - Δ = 1 and iV '= Δ + l =: 2, with Δ the length of the guard interval; and whereas (N - N _moά ) = (8 - 6) = 2 _unmodulated guard elements are located at both ends of the spectrum, that is, a guard element not modulated by information to be transmitted from each side of the spectrum, then the vector b _p (n) formed of the drivers of the transmitted symbol, is of size (6, 1), and the Fourier matrix F, of size (8, 8) is of the following form:

p _.

; où i dénote l'unité imaginaire ( i = -1 ).

; where i denotes the imaginary unit (i = -1).

We deduce the partial Fourier matrix F ', of size (6, 2):

) soit encore :

) again:

This gives a corrected estimate of the impulse response of the channel h in the time domain:

In a next step 233, the corrected estimate is transformed from the time domain into the frequency domain by means of a Fourier transform, delivering the necessary feedback control.

Then, during a step 234, at least one interpolation between the different estimated symbols is carried out in the time / frequency space.

Specifically, once the channel for two OFDM symbols including reference data elements has been estimated, an interpolation, which can be linear, in the time domain (ID interpolation) is applied to obtain the coefficients of the channel corresponding to the informative data elements. Interpolation of a higher order is also possible.

Finally, the reception method according to the invention implements an equalization 24 of the signal received in the frequency domain, from this estimated response of the channel, for example by dividing the signal received by the improved estimation of the propagation channel, delivering an estimate of the transmitted signal.

Thus, considering the problem of minimization min h φ (tt)

cherchant à égaliser à zéro la dérivée partielle de Φ(n) par rapport à h — L_ = o , les inventeurs de la présente demande de brevet ont proposé uneleading to and

seeking to equalize to zero the partial derivative of Φ (n) with respect to h - L_ = o, the inventors of the present patent application have proposed a

et dans le domaine fréquentiel :technique for improving the estimation of a propagation channel, leading, in the time domain, to a corrected estimation of the impulse response of the channel of the form:

and in the frequency domain:

It can thus be seen that by multiplying the estimate of the impulse response of the channel in the time domain by a decorrelation matrix P, with P = (F ' ^H F ^, where * denotes the pseudo-inverse operator, we do not hold the value of the channel in the regions in which the guard elements are located: the channel estimate for the guard elements of the channel estimate is decorrelated for the useful data elements. that the Fourier transform of a channel is a continuous curve, which really means that the neighboring pilots are strongly correlated Thus, by using a decorrelation matrix according to the invention, the values of the channel are no longer taken into account. part where there are carriers or elements of null value (or in any case, it is considered that such an estimate is impossible as soon as an element is not modulated by information to be transmitted).

In the context of the reception of a multinar signal, the main effect of the invention is thus based on the decorrelation of the two regions of modulated carriers and null carriers. The case in which the OFDM symbols do not present continuous reference data elements, but pilots distributed in the time / frequency plane (in English "scattered pilots"), according to one of FIG. regular pattern (in English "pattera").

Consider for example a pattern according to which the reference data elements, denoted V in relation to FIG. 3B, occupy at a first instant T ₁ the modulated carriers with an even index, and at a second instant T ₂ the modulated carriers with an odd index. The guard elements are denoted ^* 0 ', and the useful data items are denoted by' D '. For example, Z ₁ and I ₂ respectively denote the sets of indices of the odd and even pilots, that is to say:

- i ₁ ≈ {l, 3, ..., iV _mod - l};

-; ₂ = { ² > ⁴ '"-># mod} Moreover, ω _iχ and ω _{i2 denote} the Fourier frequency sets corresponding respectively to the indices / _t and I ₁ .

For each set of frequencies ω _lχ and (O ₁ , we determine a partial improved estimate of the channel, so for the set 6J _j1 we have:

avec :

with:

F {comprising (O _j rows of the matrix F 'and N' columns of the matrix F '

(advantageously N '= A + 1), or respecting the writing Matlab - registered trademark: F ₁ ' = F '^, :), and - yp (n) = y (n) at the time / frequency locations where are positioned the reference data elements; and for the set ω _i2 , we have:

avec F£ comprenant (ύ_t lignes de la matrice F' , et N' colonnes de Ia matrice F' (avantageusement N' - A + 1), soit en respectant l'écriture Matlab - marque déposée : Fi = Ψ'(ω_k ,:).

with F £ comprising (ύ _t rows of the matrix F ', and N' columns of the matrix F '(advantageously N' - A + 1), respecting the Matlab writing - trademark: Fi = Ψ '(ω _k , :).

These two improved partial estimates thus make it possible to measure a subsampled version of the propagation channel.

Thus, considering for example: - N = S, with N the number of unmodulated guard elements and useful data elements modulated by symbol; - iV _mod = 6, with N _moά the number of user data elements modulated per symbol; - Δ = 1 and N '≈ Δ + 1 = 2, with Δ the length of the guard interval; and whereas (N - iV _mod ) = (8 - 6) = 2 unmodulated guard elements are located at both ends of the spectrum, that is, a guard element not modulated by information to be transmitted from each side of the spectrum, then we obtain two vectors b _p (ή) of size (3, 1) spaced in time, noted b _p ι (n) and b _p 2 (ri), the first vector b _p ι (n) corresponding to the pilots of even index, and the second vector b ^ ") corresponding to the pilots of odd index, and the Fourier matrix F, of size (8, 8) is of the following form:

où i dénote l'unité imaginaire ( i² = -1 ).

where i denotes the imaginary unit (i ² = -1).

We deduce the partial Fourier matrix F ', of size (6, 2):

r -0.707 1 0, 707 i

J soit encore :

J again:

et les matrices F₁' et F2 , de taille (3, 2) :

and the matrices F ₁ 'and F2, of size (3, 2):

A two-dimensional interpolation is then performed to determine the value of the channel at all locations in the time / frequency space.

The invention thus proposes a new algorithm for estimating the impulse response of a propagation channel, in particular for systems seen as multicarrier systems in reception.

This new approach notably makes it possible to correct the estimation of the channel, especially in the case where guard elements (carriers or elements that are not modulated by information to be transmitted, and therefore carrying a value of zero or almost zero) are introduced at the ends. of the spectrum representative of the transmitted signal, by multiplying a first estimation of the channel in the time domain by a decorrelation matrix. It is particularly considered that this matrix is of small size, of the order of the cyclic prefix Δ. In the example presented above, we consider more precisely a size decorrelation matrix (Δ +1, Δ +1).

Moreover, this decorrelation matrix is constant, and independent of the propagation channel. This decorrelation matrix can thus be pre-calculated as a function of at least one structural characteristic of the signal, depending on the parameters of the transmission system, such as the size of the prefix cyclic, the number of modulated carriers iV _mod , or the position of the guard elements,

It should be noted, however, that the estimated vector h of the time domain propagation channel, also referred to as the time domain corrected estimate, is not an estimate of the real channel.

On the other hand, the frequency response ¹ Jf of the estimated channel is close to the real channel for the received modulated carriers. Indeed, thanks to the use of the decorrelation matrix, we are interested only in the modulated useful data elements for the channel estimation, since it is these modulated elements that carry the information to be transmitted (transmitted data).

It is thus found that the algorithm according to the invention requires O (Δ ² ) more operations than the time domain estimation technique of the prior art, applied to vectors of length N, which requires O (Nlog ₂ N) operations. However, considering that, conventionally, Δ "JV, it is found by comparing the techniques of the prior art and the invention that:

N - if Δ = -, then there is no increase of the complexity a principal until N = 512, then the same order of complexity for

N = 1024;

N - if A = -, then there is no increase in complexity

16 up to N = 2048, and then the same order of complexity for

N = 4096.

N N where Δ = - and Δ = - correspond to common cyclic prefix sizes.

8 16 ^F

FIGS. 4A, 4B and 4C show the simulation results obtained when applying the reception method according to the invention to a system of the MC-CDMA type, combining an allocation of the spectral resources in CDMA codes. to a multicarrier type modulation

OFDM. The results presented in FIGS. 4A to 4C were obtained by considering a typical context of 3GPP (in English "3rd Generation Partnership Project"), in downlink transmission.

The processing gain of the considered system (that is to say the length of the Walsh-Hadamard signatures corresponding to the spreading codes isolating the users of the same cell) is equal to 16.

OFDM pilots are also considered, with a pilot all twelve pieces of informative MC-CDMA data, that is to say 7.7% of pilots.

For example, the following values of the main parameters of the MC-CDMA system are chosen: JV-1024, N _moά = 704, and Δ-64.

More specifically, FIGS. 4A and 4B illustrate the bit error rate (BER, or BER) as a function of the signal-to-noise ratio (SNR) respectively for a four-phase phase modulation MDP4 (in English QPSK for "Quadrature Phase Shift Keying "), and for quadrature amplitude modulation 16-QAM, in the context of a single application in an ITU Vehicular A channel at 30km / h, with a spreading factor SF equal to 16, 15 users , and a yield R equal to 2/3.

More specifically, in relation with FIG. 4A (respectively FIG.

4B), the curve 4l _A (respectively _41B) illustrates the BER versus SNR in reception, with perfectly known channel (ideal case); curve 42 _A

(respectively 42 _δ ) illustrates the BER according to the SNR with an estimated channel according to the improved estimation technique according to the invention, taking into account a decorrelation matrix; curve 43 _A (respectively 43 _B ) illustrates the BER according to the SNR with a channel estimated according to the time domain estimation technique according to the prior art; and the curve 44 _A (respectively 44g) illustrates the BER according to the SNR with a channel estimated according to the frequency domain estimation technique according to the prior art.

These two figures show the efficiency of the technique according to the invention compared to the techniques of the prior art, whatever the level of the signal-to-noise ratio. More specifically, by estimating the channel in the time domain as proposed by the invention (curves 42 _A and 42 _B), BER degradation in reception there is only 0.2dB compared to a perfect knowledge of the propagation channel ( curves 41 _{and 41B)} at any point. It is also noted that the estimation in the frequency domain according to the prior art (curves 44 _A and 44 g) gives a BER at 2 dB of the ideal curve (curves 41 _A and 41 _B ), which constitutes a degradation of a ratio ten compared to the approach according to the invention.

Finally, it can be seen that the estimate in the time domain according to the prior art (curves 43 _A and 43 _B ) leads to an error floor for the high SNRs, and in particular the SNRs greater than 18 dB.

Finally, FIG. 4C illustrates the variations of the frequency domain propagation channel (carrier frequency on the ordinate, the real part of the values of the abscissa propagation channel), for a signal-to-noise ratio of 20 dB, for: a real channel ( curve 41 _C );

an estimated propagation channel according to the invention, taking into account a decorating matrix (curve 42 _C ); and

a propagation channel estimated according to the estimation techniques in the time domain of the prior art (curve 43).

This figure 4C clearly shows the edge effects of the approach according to the prior art (curve 43 _C ), corresponding to the unmodulated guard elements.

It is recalled in fact that in the example illustrated with reference to FIGS. 4A, 4B and 4C, OR has JV = 1024, iV _mod = 704, and Δ = 64. Thus, we have (N - N _mod ) = 320 elements of for example, distributed at the borders of the signal spectrum, or on the first 160 carriers, and the last 160 carriers.

According to the invention, on the other hand, there is an improvement in the estimation of the propagation channel throughout the spectrum of modulated subcarriers iV _mod . Finally, in connection with FIG. 5, a simplified representation of the hardware structure of a receiving device according to the invention is presented.

Such a reception device comprises a memory M 50, a processing unit P 51, equipped for example with a microprocessor μP, and controlled by the computer program Pg 52. At initialization, the code instructions of the program of the computer 52 are for example loaded into a RAM memory before being executed by the processor of the processing unit 51. The processing unit 51 receives as input a signal y (") 53, formed of a temporal succession symbol system comprising at least one unmodulated guard element, and a modulated useful data element.

The microprocessor μP of the processing unit 51 implements the steps of the reception method described above in relation to FIG. 2, according to the instructions of the program Pg 52. In particular, the processing unit makes it possible to determine an improved estimation of the propagation channel, used for the equalization of the signal received in the frequency domain.

The processing unit 51 outputs an estimate 54 of the transmitted signal.

The invention thus proposes a reception method having better performance than conventional reception methods, for systems implementing an equalization of the signal received in the frequency domain, thanks to a better estimation of the propagation channel.

Indeed, the determination of an improved estimate of the propagation channel notably makes it possible to correct the edge effects, and to deliver an estimated signal that does not have an error floor.

This technique according to the invention can thus be integrated into B3G ("Beyond 3G") systems for fixed or mobile, wired or wireless communication.

It may also be noted that this improved propagation channel estimation technique can be combined with the technique presented in the French patent application No. 0511082 filed in the name of the same Applicant on October 28, 2005 and not yet published, concerning a method of reception at within a given geographical unit of a signal disturbed by the interference due to the transmission of signals in neighboring geographical cells.

Claims

A method for receiving a signal formed of a temporal succession of symbols comprising, at the spectral level, at least one guard element that is not modulated by at least one piece of information to be transmitted and a piece of useful data modulated by at least one piece of information to transmit, said useful data elements comprising on the one hand reference data elements called pilots, the value on transmission of said drivers being known from at least one receiver intended to perform a reception of said signal, and other part of the informative data elements, characterized in that it implements a step of estimating at least one propagation channel between at least one transmitter and said at least one receiver, comprising: a substep of determining (22) a first estimate of said propagation channel from said reference data elements, delivering a first channel estimation vector; a sub-step of determining an improved estimate (23) of said propagation channel by multiplying said first estimation vector by a predetermined decorating matrix as a function of at least one structural characteristic of said signal, making it possible to decorrelate at least part said improved estimate for said guard elements, of said improved estimate for said payload data items. 2. Reception method according to claim 1, characterized in that said substep of determining an improved estimate implements the following steps: transforming (231) said first domain estimation vector from the frequency domain to the time domain; multiplying (232) said time domain first estimation vector with said decorrelation matrix, delivering a corrected estimate of the impulse response of said propagation channel; transformation (233) of said corrected estimation of the time domain towards the frequency domain, delivering said improved estimate of said propagation channel. 3. Reception method according to any one of claims 1 and 2, characterized in that said decorrelation matrix is a pseudo-inverted matrix of the matrix product of the hermitian of a partial Fourier matrix by said partial Fourier matrix. corresponding: P = (F ^ FO ¹ with: P said decorrelation matrix; F 'said partial Fourier matrix; denotes the Hermitian operator; denotes the pseudo-inverse operator; said partial Fourier matrix F' being an extract of a Fourier matrix F defined by:

with: N an integer; W _N = e ^J *. 4. Reception method according to claim 3, characterized in that said partial Fourier matrix F 'comprises the coefficients of said Fourier matrix F corresponding to the time / frequency locations of said reference data elements. 5. Reception method according to claims 2 and 3, characterized in that said step of transforming said first estimation vector implements a multiplication of said vector of first estimation in the frequency domain by the Hermitian of said Fourier matrix F of size (N, N); in that said multiplying step implements a selection of N 'coefficients of said time domain first estimate vector, among N coefficients, and multiplying said N 'coefficients by said decorrelation matrix P of size (W', N '), delivering said corrected estimate, and in that said transformation of said corrected estimate implements a multiplication of said corrected estimate in the time domain by said partial Fourier matrix F ', of size (N _mod , N'), delivering said improved estimate of said propagation channel, where iV _mod is the number of modulated payload elements. 6. Reception method according to claim 5, characterized in that said improved estimate is of the form:

where: represents, at the time / frequency locations of said elements

reference data, a division of said signal received by said reference data element at said time / frequency location, reference data element by reference data element. 7. Reception method according to any one of claims 5 and 6, characterized in that a cyclic prefix of length Δ being inserted between said symbols, the number N 'of coefficients is equal to the length of said cyclic prefix plus one: N '= A + 1. 8. Reception method according to any one of claims 1 to 7, characterized in that said reference data elements being distributed in the time / frequency space, said method implements the following steps: determination of at least two subsets of carriers each including at least one reference data element; determining a partial improved estimate of said channel applied to each of said subsets; bidimensional interpolation from said partial improved estimates to determine an improved channel estimate for each carrier of the space time / frequency. 9. Reception method according to any one of claims 1 to 8, characterized in that it also implements a step of equalizing (24) said received signal in the frequency domain, from said improved estimate of propagation channel, 10. Device for receiving a signal formed of a temporal succession of symbols comprising, at the spectral level, at least one guard element that is not modulated by at least one piece of information to be transmitted, and a useful data element modulated by at least one information to be transmitted, said useful data elements comprising, on the one hand, reference data items called pilots, the value at transmission of said pilots being known to at least one receiver intended to perform a reception of said signal, and on the other hand informative data elements, characterized in that it comprises means for estimating at least one propagation channel between at least one transmitter and said at least one receiver, comprising: means for determining a first estimate of said propagation channel from said reference data elements, delivering a first channel estimation vector; means for determining an improved estimate of said propagation channel by multiplying said first estimation vector by a predetermined decorrelation matrix as a function of at least one structural characteristic of said signal, enabling at least a part of said improved estimate to be said guard elements, said improved estimate for said payload data items. 11. Computer program product downloadable from a communication network and / or stored on a computer readable medium and / or executable by a microprocessor, characterized in that it comprises program code instructions for the implementation of at least one of claims 1 to 9.