EP1325600A1

EP1325600A1 - Method for optimal estimation of a propagation channel relying solely on pilot symbols and corresponding estimator

Info

Publication number: EP1325600A1
Application number: EP01974376A
Authority: EP
Inventors: Mohamed Siala; Emmanuel Jaffrot
Original assignee: France Telecom SA
Current assignee: Orange SA
Priority date: 2000-09-14
Filing date: 2001-09-13
Publication date: 2003-07-09
Also published as: US6947373B2; KR20030060892A; JP4777598B2; CN100431317C; FR2814011A1; WO2002023841A1; FR2814011B1; US20030016645A1; KR100832456B1; CN1475066A; JP2004509520A

Abstract

The invention concerns a method for optimal estimation of a propagation channel relying solely on pilot symbols. It consists in: constituting a base of eigenvectors from the correlation matrix of the observed channel only at the locations of the pilot symbols; breaking down the representation of the channel in said base; extending the base vectors to the given symbols, thereby producing an optimal estimation of the channel to the data symbols. The invention is applicable to OFDM, TDMA, CDMA systems.

Description

OPTIMAL ESTIMATION OF A CHANNEL

PROPAGATION BASED ONLY ON

PILOT SYMBOLS AND CORRESPONDING ESTIMATOR

DESCRIPTION

Technical area

The subject of the present invention is a method of optimal estimation of a propagation channel based solely on the pilot symbols and a corresponding estimator. It finds its application in radiocommunications, in particular for multiple access and modulation techniques of the OFDM type [1], [2], [3] (orthogonal frequency multiplexing or "Orthogonal Frequency Division Multiplexing"), or TDMA (access time division multiple) or CDMA (code division multiple access). It can be applied, among others, to HIPERLAN II, DAB [2] and DVB-T [3] systems for OFDM, IRIDIUM (Registered trademark) and ICO for TDMA, UMTS and CDMA-2000 for CDMA .

State of the art

Current digital communications systems can be classified into three main categories: - OFDM multicarrier systems with, among others, the case of HIPERLAN II, DAB and

VDB-T;

- multiple access systems by time division (TDMA) with _t among others, the GSM and DECT systems for the terrestrial radio and systems IRIDIUM (registered trademark) and ICO for satellite radio communications;

- conventional CDMA systems with, for example, the IS'95, UMTS and CDMA-2000 systems for terrestrial radiocommunications and the GLOBALSTAR (Trademark) system for satellite and time radiocommunications.

In all these systems, the resolution of reception problems linked to the knowledge of the characteristics of the propagation channel is facilitated by the introduction of pilot symbols (or reference symbols) or of common pilot channels, known from

1 advance by each of the receivers.

The following description emphasizes the OFDM multicarrier system, but the invention also applies to single-carrier systems, namely the systems:

- with conventional continuous transmission with non-selective or frequency-selective propagation channel, - with TDMA transmission without frequency hopping, with non-selective or frequency-selective propagation channel,

- with TDMA transmission with frequency hopping, with non-selective or frequency-selective propagation channel,

- CDMA transmission with frequency selective or non-frequency propagation channel, and

- OFDM transmission with propagation channel which can be very selective in frequency and in time. In all these cases, the modeling of the transmitted and received signal always corresponds to a particular case of the modeling which will be presented later in the context of the OFDM system.

a) OFDM systems

OFDM systems belonging to the first category, use multi-carrier modulation to distribute the users in the time-frequency plane. They allow high speed signals to be transmitted without the need for equalizers. They are widely used in broadcasting contexts such as DVB-T and DAB and in mobile radio contexts such as HIPERLAN II. The basic principle of OFDM is to produce a number of narrowband signals all orthogonal to each other. These orthogonality properties are then used by each receiver to find the corresponding transmitted data. A typical embodiment of an OFDM system implements an inverse discrete Fourier transform (TFDI) on transmission and a discrete Fourier transform (TFD) on reception.

Figure 1 attached illustrates a conventional OFDM transmission chain with a single sensor. This chain comprises a series-parallel conversion circuit 10 receiving symbols A, a reverse discrete Fourier transformation circuit 12, transmission means 14, reception means 20, a Fourier transformation circuit 22, a parallel-to-serial converter 24 and finally a decision means 26 which delivers the estimated symbols A.

A classic OFDM modulator processes the data flow in blocks. It manages this flow in sequences of N symbols and performs the inverse Fourier transform. This amounts to saying that this transform generates N subcarriers, each carrying one of the symbols of the starting sequence. This block, called the OFDM symbol, can contain both data symbols and pilot symbols used for synchronization and channel estimation purposes. Unlike conventional CDMA or TDMA signals, OFDM often requires a distribution of the pilot symbols over the entire time-frequency plane.

The radiomobile channel crossed during a communication between a transmitter and a receiver is generally of the multipath type with rapid Rayleigh fading. This phenomenon is generally due to the conjunction of the movement of the mobile and the propagation of the radio wave along several paths. It can also be produced artificially by a broadcasting pattern of "1" used for the digital broadcasting system DVB-T (which is a priori impossible to achieve in analog broadcasting).

The receiver can be fitted with L sensors providing L branches of diversity as shown in the attached FIG. 2. In this figure, the receiver comprises L sensors 30 ¹ , 30 ² , ..., 30 ^L , L Fourier transformation circuits, 32 ¹ , 32 ² , ..., 32 ^L , L parallel-series converters 34 ¹ , 34 ² , ..., 34 ^L , L channel estimation circuits 36 ¹ , 36 ² , ..., 36 ^L , L re-phase circuits 38 ¹ , 38 ² ,. .., 38 ^L and an adder 40 delivering the weighted outputs of the symbols to be estimated.

From the receiver's point of view, after demodulation, the channel affecting a time-frequency block can be represented in the form of a time-frequency matrix, or even a surface in time-frequency-amplitude space. The problem is therefore treated in a two-dimensional space, unlike TDMA where the problem is one-dimensional.

Channel estimation is based on the use of pilot symbols. These make it possible to directly obtain an estimate of the channel of each diversity branch at the locations of the pilots with a view to an interpolation or an extrapolation making it possible to estimate the channel affecting the data symbols.

b) TDMA systems

TDMA systems, belonging to the second category, are currently limited to the treatment of two extreme cases: channels that are very selective in frequency (very large delay spreading) but not selective in time (negligible Doppler spreading) and channels that are not selective in frequency but can possibly be very selective in time. The first case is often encountered in terrestrial radio systems such as GSM. The second is rather encountered in the systems of satellite communications such as ICO and IRIDIUM.

c) CDMA systems CDMA systems, belonging to the third category, introduce the concept of power control period (PCP). The signal strength sent by the transmitter remains constant during each PCP but can vary from one PCP to another to counteract slow fading (due to loss of space propagation and mask effects) and fading fast due to the effects of multiple paths (time selectivity).

The conventional receiver of a CDMA system first performs a suitable filtering of the received signal. The signal thus obtained is then despread and continued for each of the significant power paths retained by the receiver. Thereafter, each path is estimated at the level of each PCP by correlation of the pilot symbols with the despread samples associated with both this path and this PCP. This estimate is then used in the rest of the CFP to demodulate the data symbols for each of the paths and recombine them for decision-making. This is the principle of the so-called rake receiver ("RARE receiver" in English terminology).

For a slow movement of the terminal, and therefore a selectivity in low time, the estimation of the paths for a given PCP can be consolidated by weighting with those of a finite number of neighboring PCPs. Disadvantages of prior techniques

The channel seen by the receiver can vary significantly from one time-frequency block to another for OFDM, from one time interval to another for TDMA and from a power control period to 1 other for the AMRC. For the three multiple access categories considered, this variation is mainly due to changes in the propagation conditions between the transmitter and the receiver. However, for OFDM, this variation can be further accentuated in frequency with the artificial increase in the spread of the delays T _m generated by the use of a diffusion pattern at "1" (in particular as in the case of the DVB-T system).

In the case of OFDM, the variable character of the channel can be characterized by the product B _d T _{m /} where B _d represents the Doppler spread. The larger this product, the faster the channel varies in time and frequency. In the case of TDMA and CDMA, this variable character can be characterized by the product B _d T _s , where T _s represents the duration of a symbol. The larger this product, the faster the channel varies in time. Reception methods prior to the invention do not seek to optimize the estimation of the channel. They are content to carry out an estimation of this at the positions of the pilot symbols and then to extend, by linear interpolation, this estimation to the data symbols.

Five methods commonly encountered in the context of OFDM systems are described in the following. These methods have their counterpart in the TDMA and CDMA systems.

i) First method

In the first method, we take into account the three pilot symbols closest to the data symbol at which we want to estimate the channel. The plane passing through the three pilot symbols is calculated and an estimate of the channel at the point considered is deduced therefrom. Even respecting the Nyquist sampling theorem, from the point of view of the concentration and the placement of the pilot symbols in the time-frequency plane, this method always remains sensitive to strong channel variations and does not make it possible to effectively reduce the effects. related to thermal noise. ii) Second method

A second method uses a simple form of the MMSE ("Minimum Mean Square Error") criterion. It consists in searching for the constant plane by means of the values of the channel at the level of the pilot symbols and in deducing therefrom a constant estimate in time and in frequency of the channel affecting the transmitted data. This channel modeling is well suited to channels varying very little at the level of each block received and therefore having relatively small products B _d T _m . However, as soon as the channel becomes more selective, the plane modeling shows its limits because of the notable bias affecting the channel estimates at the level of the data symbols. iii) Third method

A third method also uses another form of the MMSE criterion with, this time, a non-constant sought channel estimation plan. This method therefore adapts better to slowly varying channels, but remains less suitable than the second for almost constant channels. iv) Fourth method

A fourth method is based on the use of a two-dimensional discrete Fourier transformation. Only the samples corresponding to the pilot symbols in the places they occupy in the block are kept from the received time-frequency block. All other positions, associated with the data symbols, are set to zero. A two-dimensional discrete Fourier transformation is then performed on this modified block. It is then followed by filtering isolating the channel space from the total information obtained in the Doppler-delay plane. Finally, it ends with an interpolation over the entire block carried out by an inverse two-dimensional discrete Fourier transformation. This technique has undesirable side effects and is therefore not suitable for small blocks. v) Fifth method

A fifth method consists in a representation of the channel by the eigenvectors of its correlation matrix. However, the projection of the signal received on the basis of the eigenvectors is not optimized since the restriction of the eigenvectors to the pilot symbols does not form an orthogonal basis. The first three methods are adapted to very specific propagation cases, but in no way to time and frequency selective channels. The last two methods can be used in the case of time and frequency selective channels but quickly show their limits for high values of the product B _d T _m .

The object of the present invention is precisely to remedy these drawbacks.

Statement of the invention

Unlike the previous methods, the method proposed by the invention is based on an optimal estimation of a channel, possibly very selective in time and / or in frequency, carried out from a modeling of its evolution in time and in frequency. This approach makes it possible to use the pilot symbols of neighboring blocks without any introduction of bias even for channels varying significantly from one block to another. The efficiency of this approach can therefore be put to good use by better fighting against channels that are more and more selective in time and frequency, by reducing the number of pilot symbols or even by reducing their power.

The object of the present invention is to improve the performance of existing or future OFDM, TDMA and CDMA systems. This improvement, obtained by optimizing the channel estimation, makes it possible to significantly increase capacity and coverage of these systems. It is generated by an optimization of the operation of the conventional receiver in the case of slow fading but also in the more difficult case of very rapid fading. It is further increased if the pilot symbols are judiciously distributed in time for TDMA and CDMA systems and in time-frequency for OFDM systems.

It is then possible to counteract the degradations in performance generated by a rapid variation in time (for the AMRT and the CDMA) or in time-frequency (for the OFDM) of the channel on each block received. It is also possible to advantageously use the pilot symbols of the neighboring blocks for OFDM, of neighboring time intervals for TDMA or of neighboring power control periods for CDMA, in the channel estimation without introduction of bias.

An object of the invention is to improve the quality of the information received by implementing an optimal method of channel estimation for OFDM, TDMA or CDMA systems. This method can be used regardless of how the pilot symbols are introduced or distributed in the flow of information transmitted. The invention makes it possible to reduce, at constant reception quality, the relative number and / or the power of the pilot symbols. This object is achieved by optimally taking into account an arbitrary number of pilot symbols of time-frequency blocks or consecutive time intervals in the channel estimation. The invention can be used independently. It can also advantageously be used to optimally initialize certain iterative semi-blind channel algorithms (using both the pilot symbols and the data symbols) often very sensitive to the initial conditions [4], [5], [6], [7].

This technique also allows, thanks to a theoretical performance formulation, to optimize by exhaustive research the position of the pilot symbols in the time-frequency blocks.

Specifically, the subject of the invention is a method for optimal estimation of a propagation channel in which:

A signal having passed through said channel is received, this signal comprising blocks of symbols either one-dimensional in time or in frequency, or two-dimensional in time and in frequency, each block comprising N digital symbols with N _P pilot symbols and N _D data symbols ,

- the signal received by a signal vector denoted R having N components is modeled,

- the propagation channel is modeled by a discrete multiplicative channel vector C with N components,

- an estimate C of the discrete channel vector C is deduced from the signal vector R, this method being characterized in that:

- for each transmitted block, a restriction vector R _P of this vector limited to the N _P pilot symbols is calculated from the signal vector R, - from the discrete multiplicative channel vector C, we define a restriction vector C _P limited to the pilot symbols,

- we define an orthonormal basis constituted by the N _P eigenvectors of the matrix H of covariance of the vector of the restriction C _P of the discrete channel, said matrix having N _P eigenvalues denoted T _e {£ = 0, 1, 2,

..., Np_ι), - the restriction vector R _P of the signal received in said orthonormal base is broken down

{B _P ^ _Q and 1 'we obtain N _P components G £,

- we extend the orthonormal base ^{to the N} _D data symbols to obtain an extended base {B ,,} ^,

- we obtain the optimal sought estimate C of the equivalent channel vector C by calculating the

sum in which the w _£ are N _P weighting factors defined by

1 / (l + N ₀ / r _f ), where No is the variance of the noise.

The subject of the invention is also an estimator which comprises the means capable of fulfilling the functions of the method thus defined. Brief description of the figures

- Figure 1, already described, shows a conventional OFDM transmission chain with a single sensor;

- Figure 2, already described, shows a conventional OFDM receiver with several sensors and L diversity branches;

FIG. 3 illustrates the calculation of the flexible outputs for the data symbols from the estimation of the equivalent channels of the L diversity branches;

- Figure 4 shows the following of the treatments carried out according to the invention for estimating the equivalent channel in - a branch of diversity; - Figure 5 shows the main normalized eigenvalues corresponding to the positions of the pilot symbols given in Figure 8B;

- Figures 6A, 6B, 6C, 6D show the modulus of the normalized main vectors of the extended base for B _d T _m = (1/32) ² , for pilot symbols distributed as in Figure 8B;

FIG. 7 shows the variation of the weighting factor for B _d T _m = (l / 32) ² for pilot symbols distributed as in FIG. 8B, and for three values of the ratio E / N ₀ ;

- Figures 8A and 8B show the optimal positions in time and frequency of the pilot symbols for B _d T _m = (l / 32) ² and E / N ₀ = 10 dB; for figure 8A the probability criterion errors, binary is average; for figure 8B, it is the worst; FIG. 9 shows the variations in the probability of gross binary errors for several values of BT _m , for a distribution of the pilot symbols corresponding to FIG. 8B.

Detailed description of particular modes of implementation The invention will be described only in the context of OFDM multicarrier modulation. This description easily extends to other single carrier modulations of the TDMA or CDMA type. It should be noted, however, that for CDMA systems, the signals output from the fingers of a rake receiver play an identical role to that of the signals output from multisensor systems with independent non-selective frequency channels. However, the proposed invention deals with the more general framework of multisensor systems with selective channels both in time and in frequency.

The principle of the invention, shown here for OFDM is to use ^'samples pilot symbols in the received signal to achieve an optimal estimate of the multipath channel correspondan.

The OFDM receiver obtained performs block-by-block processing each time a given number of OFDM symbols is available. It achieves an optimal estimate of the channel only from the pilot symbols. This method is optimal within the meaning of the criterion Due a posteriori (MAP). It can be reformulated simply by using an appropriate weighting of the projections of the samples received corresponding to the pilot symbols on an extended orthonormal basis. This basis is obtained by extending the Karhunen-Loève orthogonal decomposition from the pilot symbols to the remaining data symbols.

This method can be implemented as is to obtain an estimate of the channel based solely on the pilot symbols. It can also be used as an initialization phase for an iterative semi-blind channel estimator taking into account the samples of the data symbols. It can obviously be used for a multisensor system by placing such an estimator behind each of the sensors.

More specifically, the optimization of the operation of the OFDM receiver is obtained through the use of a multi-path channel estimator with Rayleigh fading exclusively based on the pilot symbols. The optimal structure of the estimator is based on the representation of the channel obtained thanks to the Karhunen-Loève orthogonal decomposition theorem. Unlike the other methods using this decomposition, that of the invention uses the eigenvectors of the correlation matrix of the channel observed only at the locations of the pilot symbols. This makes it possible to have a set of orthogonal vectors at the locations of these pilot symbols. These vectors are then suitably extended to the data symbols, which allows optimal interpolation of the channel estimation to the data symbols. This technique also allows, thanks to a theoretical performance formulation, to optimize by exhaustive research the position of the pilot symbols in the time-frequency blocks.

i) Representation of the OFDM signal transmitted In the example taken, it is assumed that the proposed receiver has L independent diversity branches with L sensors (or receiving antennas) sufficiently spaced. This receiver processes block by _. blocks the signals received in these branches. The size of each block processed does not necessarily depend on the number of carriers of the OFDM system and can take into account all or part of one or more OFDM symbols. The shape and size of the block processed on reception is free, so as to best adapt to the system. The channel estimation is performed block by block for each of the diversity branches taken separately. A block is made up of N symbols a ^ _. of energy E ^ and of two-dimensional position _mn = (mF, nT), where F and T are respectively the frequency and time spacings between two adjacent symbols (for TDMA and CDMA, each symbol is identified by a unique index n and has a one-dimensional position in time nT). Each block is composed of N _D data symbols indexed in the set S _D and N _P pilot symbols indexed in the set S _P. ii) Characteristics of the multi-path fading channel

The multi-path channel associated with each diversity branch seen by the transmitted OFDM signal exhibits temporal and frequency variations due to the Doppler effect and to the multiple paths. Each path is characterized by an average power and a Doppler power spectrum (SPD) depending on both the environment and the speed of the mobile.

The multi-path channel of each branch is generally characterized by its time-frequency autocorrelation function φ (Δ £, Δt), where Δf is the frequency spacing and Δt the time spacing. By way of example, the time-frequency autocorrelation function of a channel with a conventional Doppler spectrum and with an exponential intensity multipath intensity profile of average power φ (0.0) seen on a branch of diversity is given by:

J ₀ (πB _d Δt) φ (Δf, Δt) = φ (0.0) l + j27tT _m Δf where T _m and B _d represent respectively the spread of the delays and the Doppler spread and Jo _t. ) denotes the first-order Bessel function of order 0.

iii) Modeling of the received signal

The signal received from each diversity branch is first demodulated by a discrete Fourier transform (TFD). Assume that the .signal received j j ^th _oxanc _ιe ι corresponding to the _min symbol is written: "min ~ 4nιrmn"^{"^} mn _m where c ^l _n is the gain factor of the discrete spindle ^th channel _{seen by} _ a _m _th symbol and N ^ is a white Gaussian noise variance of N ₀ complex additive. The gain factors within the same branch of diversity are correlated with each other in time and frequency. However, the gain factors belonging to branches of different diversity are uncorrelated with each other.

The object of the invention is to estimate the gain factors (^ _min of the channels of all the diversity branches as well as to optimize the positions (mF, nT), of the pilot symbols.

To simplify the notations, we introduce the indexing function δ (k) between the one-dimensional set {k} _{k = 0} and the two-dimensional indexing set S = S _D US _P. We also introduce the indexing function δ _P (k) between the one-dimensional set the two-dimensional indexing set S _P corresponding to the only pilot symbols. Finally, we introduce the indexing function δ _D (k) between the one-dimensional set and the two-dimensional indexing set S _D corresponding to the only data symbols.

Let (.) ^{Τ be} the transposition operator. For each block transmitted, we introduce both the signal vector at the output of the adapted filter of the j ^th branch: and its restriction to pilot symbols ^R P ⁼ ( ^R δ _p (0) ' ^R δ _p (l)' ^{• • •} ' ^R δ _p ( _p -l) ^ ^•

To overcome the dependence of the amplitude of each symbol _min from its index (m, n), we introduce the vector of the normalized symbols of the transmitted block: A = (§ (0), Aδ (i),. .., δ ( _N -i)) ^'r with A5 ( _k ) = aδ ( _k ) / | a§ ( _k ) | - With these notations, it is possible to rewrite the components of the vector received in the form:

^RJ _{δ (k)} ⁼ C _{δ (k)} A _{δ (k)} + N ^J _{δ [k)} where C ₍ y is the k ^th component of the equivalent multiplicative discrete channel vector: ^c ^••• > | ^a δ (Ni) | ^c δ (Ni)) ^•

Finally, we introduce the restriction vector:

of the discrete multiplicative channel equivalent to the pilot symbols.

We now seek to perform, for each block and for each branch of diversity, an optimal estimate, C ^J , of the vector C ³ , conditionally on the restriction vector received Rp.

Let (.) ^* Be the complex conjugation operator. The receiver then calculates its decisions using complex flexible outputs:

for modulations with a number of states greater than 2

(MDP4, MDP8, ...) or their real parts for two-state MDP2 modulation. These operations are illustrated in FIG. 3 where we see L branches respectively receiving signals R °, ..., R ^{3 '} , • • -, R ^L_1 , L estimation circuits of the equivalent channel referenced 36 ¹ ,. .., 36 ³ , ..., 36 ^L_1 and delivering the estimates C ^ y, L 38 ° phase resetting circuits, ..., 38 ³ , ..., 38 ^L delivering the products:

^R δ _D (k) ( ^C δ _D (k)) 'an adder 40 delivering the sum of the products coming from the L rephasing circuits and finally a normalization circuit 42 dividing the sum obtained by N ₀ (noise variance) and delivering the quantity

iv) Calculation of the estimate of the equivalent channel

FIG. 4 illustrates more precisely the operations implemented for optimally estimating the equivalent channel for each branch of diversity. The letter R represents the signal vector received and applied to a circuit 50 for restricting this vector to the only pilot symbols. This circuit therefore delivers a vector denoted R _P. A circuit 52 projects this vector onto a base comprising N _P vectors and denoted {Bp ^} ^ ¹ . The projection shows the components: These N _P components are applied to a circuit 54 to reconstruct a channel estimation vector noted.

C, which is a weighted sum of the products G ^ B _f . The circuit 54 finally delivers the estimated channel vector C.

More precisely, we denote E [.] The operator mathematical expectation. Let H = E [C _P C _P ^T ] be the covariance matrix of the restriction C _P to the pilot symbols of the discrete channel vector C, with as (μ, v) ^th input:

^H μv = ^E δ _P (μ) ^E δ _p (v) Φ (Pδ _p (μ) - Pδ _p (v)) with μ, v = ⁽ U ..., N _P -l.

We denote {B _P1 } _{1 =} ^P ₀ ^~ the orthonormal base consisting of the N _P eigenvectors of the Hermitian matrix H. Let {ri} _{1 =} ^p ₀ ^~ the N _P eigenvalues, assumed to be classified in descending order, associated with these eigenvectors. These basic vectors are determined (to within an arbitrary phase) by the equation: FIG. 5 represents the distribution of the normalized eigenvalues T ₍ / K, (E being the average energy received for each symbol emitted), for the most important, for pilot symbols and data symbols of common emitted energy E _mn = E and for a time-frequency autocorrelation function channel with conventional Doppler spectrum and with exponential multi-path intensity profile.

We now introduce the extended orthonormal base {B _f } ^ ₌ ^P o to the data symbols of each block. This extensive database is completely determined from the orthonormal basis {B} ^ _Pf ₌ ^P o B ^y ^are the relationships:

Note that the restriction to the pilot symbols of this last base gives back the orthonormal base rR τ.Np-1 \ ^u P. S (= 0 ^•

Figures 6A, 6B, 6C and 6D show the modulus of the four main vectors of the extended base obtained for B _d T _m = (1/32) ² .

For each branch of diversity, the optimal estimate C of the associated equivalent channel vector C, conditionally on the restriction vector received R _P , is given by:

C = 2 _j W _e G _e B ê 'e = o or

G, —B p _j R _P , K. - U, J, ..., N _p —1, represents the decomposition of the restriction of the vector received, R _P , in the orthonormal base {Bp ^} ^ ¹ and the weighting factors w _f , £ -0, 1, ..., N _P -1, are given by:

1 w ”=,

I + N ₀ / Γ, where N ₀ is the variance of the noise. Figure 7 shows the evolution of the main weighting factors for B _d T _m = (l / 32) ² and for certain values of the signal to noise ratio E / N ₀

(OdB for curve 71, 5dB for curve 72 and 10dB for curve 73). For the propagation channels with Doppler spreads and spreads of weak delays, the eigenvalues T _e (£ = 0, 1, ..., N _P -1) decrease very fast and the weighting factors w _P are almost zero except for the first. The estimation algorithm can then be significantly simplified by not calculating the projections and by calculating the reconstruction of the C channel estimate only with the corresponding vectors of the extended base.

More precisely, let Q <N _{P be} the number of significant normalized eigenvalues associated with weighting factors close to unity.

The estimator calculates in a simple way and with very good precision the approximation C of C given by:

~ S " ¹ C = ∑W, G, B ,,

where only the coefficients are evaluated

The optimal estimation of the channel according to the invention is based solely on the pilot symbols of a received block. It can be used as is to demodulate and re-phase the contributions of all the diversity branches. It can also be used as an optimal initialization for several iterative semi-blind channel estimation algorithms based on all the symbols of a received block (pilot symbols and data symbols) [4], [5], [6], [7]. v) Optimization of the positions of the pilot symbols

It is possible to determine the theoretical performance of the estimator in terms of gross bit error rate for the MDP2 and MDP4 modulations. It is therefore possible to quickly determine, by exhaustive search, the positions of the pilot symbols corresponding to the optimal performance in terms of gross bit error rate. Among the optimization criteria, the average gross bit error rate can be used, the average being carried out on all the data symbols of a block. You can also use the worst raw bit error rate.

For each data symbol a _{§ (k)} , (k = 0, 1, ..., N _D -1), we introduce the vector:

^V k = - Pδ _D (k))> ^••• > V ^E δp (Np-l) ^E δ _D (k) Φ (Pδp (N _P -l) ^~ Pδ _D (k))) ^T

We also introduce the identity matrix I. For L branches of diversity, the raw bit error rate on the symbol a§ _(k) is given explicitly by:

or φ (0,0) E _{δD (k)} + N _{0 ι}

for MDP2 modulation and

for MDP4 modulation. By way of example, the optimal estimation method according to the invention can be applied to a system

OFDM with square blocks of N = 256 symbols. It is assumed that each block consists of N _P = 16 pilot symbols and N _D = 256-16 = 240 data symbols.

We assume that the propagation channel has a time-frequency spectrum autocorrelation function

Classic Doppler with an exponential multi-path intensity profile. We assume that the receiver has perfect knowledge of the characteristics of this channel.

FIGS. 8A and 8B represent the optimal position of the pilot symbols in each data block for B _d T _m = (l / 32) ² and E / N ₀ = 10dB. FIG. 8A presents the results obtained for the criterion of average gross bit error rate. Figure 8B presents the same results for the worst gross bit error rate criterion.

Figure 9 presents the probability of gross binary errors for three values of the product B _d T _m = (l / 16) ² for curve 81, (1/32) ² for curve 82, (1/64) ² for curve 83. For comparison, the same figure also shows the performances obtained with perfect knowledge of the channel (curve 84).

References

[1] J.A.C. Bingham, "Multicarrier Modulation for Data Transmission, An Idea Whose Time Has Corne", IEEE Communication Magazine, 28, 5, pp. 5-14, May 1990.

[2] European Telecommunications Standards Institute, "Digital Broadcast Systems; Digital Audio Broadcasting (DAB) to Mobile, Portable and Fixed Receivers", ETS 300 401 2nd Edition. [3] European Telecommunications Standards Institute, "Digital Video Broadcasting (DVD); Framing Structure, Channel Coding, and Modulation for Digital Terrestrial Télévision", ETS 300 744. [4] EPO 0 802 656

[5] FR-A-2 782 585

[6] FR-A-2 782 587

[7] EN 99 11415 of 13 September 1999

Claims

1. A method for optimally estimating a propagation channel in which: a signal having passed through said channel is received, this signal comprising blocks of symbols either one-dimensional in time or in frequency, or two-dimensional in time and in frequency, each block comprising N digital symbols with N _P pilot symbols and N _D data symbols,

- an estimate C of the discrete channel vector C is deduced from the signal vector R, this method being characterized in that: - for each block transmitted, a restriction vector R _P of this vector limited to the is calculated from the signal vector R N _P pilot symbols,

- from the discrete multiplicative channel vector C, we define a restriction vector C _P limited to the pilot symbols,

- we define an orthonormal basis constituted by the N _P eigenvectors of the matrix H of covariance of the vector of the restriction C _P of the discrete channel, said matrix having N _P eigenvalues noted T _e {£ = 0, 1, 2,

- the restriction vector R _P of the signal received in said orthonormal base is broken down

^and 1 ' ^we obtain N _P components G £,

- we extend the orthonormal base to N _D data symbols to obtain an extended base {B ^ _g , - we obtain the optimal sought estimate C of the equivalent channel vector C by calculating the

Np-l sum ^, _f G ^ B _f in which the w _e are N _P

weighting factors defined by WHERE N _O is the variance of the noise.

2. Method according to claim 1, in which:

• in the operation of decomposition of the restriction vector R _p in the orthonormal base

' ^we retain only a number Q of vectors B of the base where Q <N _P , for which the associated eigenvalues T _e are greater than a certain value,

• one calculates the estimate of the channel C only with the said Q vectors B ^ (^ = 0, 1, ..., Ql) retained, which gives an approximation C of the estimate C.

3. The method of claim 1, wherein the signal is transmitted by a multiple access technique by orthogonal frequency division multiplexing (OFDM).

4. The method of claim 1, wherein the signal is transmitted by a time division multiplex (TDMA) multiple access technique.

5. The method of claim 1, wherein the signal is transmitted by a code division multiplex (CDMA) multiple access technique.

6. Method according to claim 1, characterized in that it is implemented in the diversity branches of a “rake” type receiver and makes it possible to estimate the propagation channel of each of these branches.

7. Optimal propagation channel estimator for implementing the method according to claim 1, comprising:

Means for receiving a signal having passed through said channel, this signal comprising blocks of symbols either one-dimensional in time or in frequency, or two-dimensional in time and in frequency, each block comprising N numerical symbols with N _P pilot symbols and N _D data symbols,

Means for modeling the signal received by a signal vector denoted R having N components, means for modeling the propagation channel by a discrete multiplicative channel vector C with N components,

Means for deducing from the signal vector R an estimate C of the discrete channel vector C, this estimator being characterized in that it comprises:

- means for calculating, for each block transmitted, from the signal vector R, a restriction vector R _P of this vector limited to the N _P pilot symbols, - means for defining, from the discrete multiplicative channel vector C, a vector restriction C _P limited to pilot symbols,

- means to define an orthonormal basis

formed by the N _P eigenvectors of the covariance matrix H of the restriction vector C _P of the discrete channel, said matrix having N _P eigenvalues denoted T _ê (£ = 0, 1, 2,

Means for decomposing the restriction vector R _P of the signal received in said orthonormal base and we obtain N _P components G _e , (G ^ = Bp _f R _p with £ = 0, 1, ..., Np-i), - means to extend the orthonormal base

of data to obtain an extended base {B ^ J _Q -,

- means to calculate the sum in which the vr £ are N _P weighting factors defined by 1 / {1 + N ₀ / T £), which constitutes the estimate sought.

8. Estimator according to claim 7, in which:

• the means to decompose the restriction vector R _p in the orthonormal base

^{do not} retain a Q number of vectors B ,, the base where Q _<P N for which the eigenvalues _ê T are greater than a certain value,

• the estimation means of the channel C using only said Q vectors B _t (£ = 0, 1, ..., Ql) retained, and delivering an approximation C of the estimation C.