FR2738098A1 - RF Signal Multiplex/Demultiplexing system for SDMA Cellular Mobile Radio - Google Patents

Abstract

The system has a base station which transmits and receives information to several mobile units. The base station has a multiple channel transmit/receive unit (1), coupled to an antenna. Received signals are processed (2) using an estimator to determine the position of each mobile unit, especially where multiple trajectores change the mobile unit position. Spatial filtering (3) is then applied to optimise the returns from each mobile and reduce unwanted returns for each demultiplexed output.

Description

 The present invention relates to a multiplexageldemultiplexing method of digital radio signals organized in frames and transmitted in duplex on the same frequency and time channel, not requiring calibration, but exploiting a learning sequence, known a priori, included in the structure of the frame, and more particularly applies to cellular radiocommunications between at least one base station and several mobiles.

 With the ever-increasing demand for mobile communications and limiting the number of channels allocated to operators of cellular radio networks, saturation problems will become crucial in the coming years.

 Conventional techniques currently make it possible to multiplex frequency (FDMA), time (TDMA) or code (CDMA) communications.

 FDMA (Frequency Division Multiple Access) consists of allocating each pair of frequencies a pair of frequencies (one for the downstream channel and one for the upstream channel). By itself, this very simple multiplexing technique offers almost no prospect of improving the spectral efficiency.

 TDMA (Time Division Multiple Access) consists of allocating each communication a specific time interval. Only this technique, already optimized to improve the spectral efficiency in the context of speech transmission by exploiting the silence present on the speech signal, suggests little further improvement in this respect.

 CDMA (Code Division Multiple Access) consists in allocating each code a code, defining a frequency jump law over short time intervals. Although it allows a certain increase in spectral efficiency (gradual saturation by signal degradation and not brutal interruption of service as in the case of the two previous techniques), this method remains cumbersome and expensive to implement.

Several ways are currently being explored to improve the capabilities of cellular communication systems namely - the diversification of cells and waveforms with the implementation of cells of increasingly smaller sizes ("microcell" and "nanocell" of the ATDMA standardization project, abbreviation for "Advanced Time"
Multiple Access Division "for urban areas), and especially a new method of multiplexing that can be combined with existing techniques Spatial Distribution Multiple Access (AMRS or SDMA in English terminology).

 This technique is based on the idea that radio signals corresponding to several links between a base station and mobiles can be in many cases decorrelated spatially.

 The exploitation of this decorrelation makes it possible to separate the spatial channels (even if they all occupy the same frequency and time channel) by means of a multichannel transmission / reception system associated with an antenna base adapted to the frequency range used. .

Several methods implementing this concept have already been proposed. A first method is based on arrival direction estimation techniques such as the system described in the patent of Richard ROY (US Patent PCT / US92 / 10074, 12/1991, entitled "Spatial
Division Multiple Access Wireless Communication System ") A second method uses characteristics of cyclostationnarity of digital transmission signals to separate them by blind processing (US Patent No. 5260968, issued 9/11/93, entitled" Method and Apparatus for
Multiplexing Communications Signals through Blind Adaptive Spatial
Filtering "of William A. GARDNER and Stephan V. SCHELL).

 For the first method, estimating the arrival directions of different radio sources from the signals received on an antenna basis, requires perfect control of the wavefront which results in a calibration stress of the base of the antenna. antennas used. In addition to the calibration-induced overcost, the standard deviation of the indirect estimate of the direction vector from the estimated arrival direction and the calibration table of the antenna base is significantly larger than that obtained by direct estimation methods. The director vector corresponds to the value taken by the antenna base transfer function for the azimuth, the site and the carrier frequency of the incident electromagnetic wave emitted by a source. This information is then exploited by spatial multiplexing processing.

 On the other hand, second-order separation techniques using cyclostationarity impose on the signals to be separated either differences in symbol rates or carrier frequency shifts in order to function. These limitations make these techniques unsuitable for configurations of channels formed in radiocommunications between a base station and mobiles having uncorrelated multipaths which are not separable by the above methods and therefore not operational in most urban areas.

 Finally, the aforementioned antenna processing techniques implement to separate the different received signals a so-called purely spatial filtering structure constitutes a complex gain by reception channel. The anti-jamming capabilities of this type of structure in terms of the number of independent jamming jets (two uncorrelated paths coming from the same source constitute two independent jammers) are directly limited by the number of sensors.

 In the remainder of the description, the terms "sensors" and "antennas" will be used interchangeably as well as the terms "multi-sensor", "multi-sensor network" and "antenna base", on the other hand knowing that a base of antennas, or multisensor network, forms a number of transmission paths equal to the number of antennas, or sensors.

 The invention aims to overcome the imperfections and limitations of the above methods.

For this purpose, the subject of the invention is a method of multiplexagememultiplexing spatial digital radio signals arranged in frames comprising sequences known a priori, and transmitted in duplex between at least one base station and several mobiles communicating on the same frequency channel and time by means of a receiver and a multi-channel transmitter integrated in the base station and coupled to an antenna base, characterized in that it consists of:
estimating the spatial information relating to each mobile, from the signal received by the multi-channel receiver, for the reception and transmission frequencies, by means of source separation methods, by exploiting the known prior sequences, and from this information,
to isolate in the presence of multipaths in the channel, the respective paths to each mobile having a steering vector whose power is greater than a determined threshold, and to demodulate the signal received from each of the mobiles in connection with the base station, for enable spatial demultiplexing, and
- Simultaneously transmit in the direction of the most powerful paths among the isolated paths, the signals to the corresponding mobiles, protecting each mobile vis-à-vis the signals transmitted to others by spatial filtering, to allow spatial multiplexing.

 An advantage of the present invention is that it makes it possible, for carrying out the spatial multiplexing of radio transmissions between several mobiles and a base station, of adaptation techniques for spatial or spatio-temporal filters on replica (training sequence) up to now mainly used for anti-jamming in defense radiocommunication systems. Most digital cellular radio systems use, for particular synchronization and equalization purposes, learning sequences inserted into waveforms organized into frames.

The use of this information, known a priori, allows the implementation of more efficient processing and less restrictive at the level of the modulations used than so-called "transparent" techniques.

 Among the methods described below, those implementing spatio-temporal filtering structures (a FIR filter by channel, abbreviation Anglo-Saxon for "Finite Impulsional Response"), allow in the presence of multipath, to increase the capabilities of interfering signal rejection and error rate optimization after demodulation of a multichannel system versus narrowband filtering structures (one coefficient per channel); the control of the wavefront is not subject to constraints, which makes it possible to use an uncalibrated antenna base); and the compatible modulations of these methods encompass all modulations including Gaussian waveforms.

Other advantages and features of the present invention will appear more clearly on reading the description which follows and the appended figures which represent:
- Figure 1, the main steps of the method according to the invention,
FIG. 2, a block diagram of a device for implementing the method according to the invention,
FIG. 3, an exemplary structure of a "superframe" taking into account the constraints of the AMRS mode,
FIG. 4, a block diagram specifying the interface between the AMRS controller and the spatial multiplexing and multiplexingmultiplexing estimating means,
FIG. 5, a diagram of the decomposition of a frequency and time channel in spatial subchannels,
FIG. 6, an illustration of the notion of frequency and time channel,
FIG. 7, a spatial correlation table of the mobiles present in the cell (taken in pairs), and
FIG. 8, a block diagram of a switching and setting module of the device for implementing the method according to the invention,
FIG. 9, a multi-sensor spatial filter structure + single-sensor equalizer / demodulator,
FIG. 10, a general structure: multisensor spatio-temporal filter + single-sensor equalizer / demodulator,
FIG. 11, a first implementation corresponding to a transmission path of the general structure of FIG.
FAST-BBT-DFE, and
- Figure 12, a second implementation of the general structure of Figure Il called FAST-DFE.

 In the figures, the homologous elements are designated by the same reference.

 The method according to the invention, the main steps of which are illustrated in FIG. 1, consists, from a multipath transmission / reception 1, of estimating 2 the spatial information concerning each active mobile source present in the cell at the two frequencies transmission and reception alternately, then from this information to achieve a multiplexageldemultiplexing global 3 (temporal, frequency and spatial) optimal communications.

The general principle of the invention is as follows: the spatial information is estimated for each mobile source present in the cell at the two frequencies of the channel, transmission and reception media alternately, from a received signal. This spatial information is then used to adapt the coefficients of the spatial filtering structure in transmission. As regards the spatial demultiplexing and demodulation part, several variants are proposed by the present invention:
- a structure of type FAS-DFE (FAS, abbreviation for Filter
Adapted Spatial, and DFE, in English terminology for "Decision
Feedback Equalizer "), grouping a purely spatial filter (a complex coefficient per channel) and a decision organ;
- a structure of FAST-BBT-DFE type (FAST, abbreviation for "Adapt Spatio-temporal filter" and BBT abbreviation for white noise)
Temporarily, in English terminology "Decision Feedback
Equalizer "), consisting of a spatio-temporal filter associated with an independent decision-making organ;
a so-called FAST-DFE structure consisting of a spatio-temporal filter coupled to a decision element.

 Whereas the first structure (FAS-DFE) exploits to adapt the previously estimated information, or directly the learning sequence, the adaptation of the spatio-temporal filters implemented in the last two structures (FAST- BBT-DFE and FAST-DFE) directly use the known training sequences of the waveform.

 Spatial information characterizing a mobile is defined by the direction vectors of the multitrafts of the signal received on the basis of antennas.

 These direction vectors represent the response (amplitude and phase) of the antenna base to an incident signal according to an azimuth characteristic of the position of the source.

 The estimation 2 of the direction vectors of each source uses signals received from the source of interest in a single-emission configuration.

In other words, in the waveform, a periodic time slot, hereinafter called "burst", is allocated to each mobile for the updating of its spatial information. The method applied for this estimation 2 therefore exploits the non-Gaussian character of the source signals to be processed and functions perfectly in the absence of calibration of the antenna base. On the other hand, in view of the generally important difference between the uplink (mobile to base station) and downlink (base-to-mobile) frequencies, the estimation performed on the received signals and used for a spatial filtering in This is why, in the method according to the invention, a periodic permutation of the uplink and downlink frequency groups can be performed so as to enable learning and updating. spatial information also at transmission frequencies.

 Most base stations manage several frequency-time channels in duplex mode and in this context the method according to the invention can provide global management of these channels within which the spatial multiplexing generates AMRS subchannels in a to regroup preferentially on the same time / frequency channel the most spatially decorrelated mobiles.

 For each mobile, the estimated spatial information comprises one or more director vectors associated with the different detected paths as well as the relative power of each of the paths. From this data concerning all the mobiles whose transmissions are multiplexed together, a spatial filter is calculated in transmission for each link with the constraint of focusing the radiation pattern in the direction of the main path of the useful signal and on the contrary of drilling the radiation pattern towards the main path of each of the interfering signals (due to transmissions with the other mobiles). The signals are formed by channel summation through the outputs of the spatial multiplexing filters.

According to the variants implemented, the spatial / demodulation demultiplexing assembly enables:
- FAS-DFE: to take into account the main useful path and to reject the main interfering paths and the other useful paths decorrelated spatially with the main path within the limits of the capacities of the structure (a structure of N sensors, can reject Nl signals spatially decorrelated with the useful signal),
- FAST-BBT-DFE: to re-phase the energy of all the paths of the wanted signal and to optimally reject the interfering multipath signals. While this process brings a gain in terms of signal-to-noise ratio on the current symbol with respect to the FAS-DFE, it has the same capacities of rejection of the interfering signals except that the useful multipaths other than the main path are no longer assimilated to interfering signals, which frees up as many degrees of freedom for take into account other jammers,
- FAST-DFE: quasi-optimal processing of a scrambled multisensor signal by not only realizing the repetition of the multivitrajets of the useful signal but also that of the multipaths of the interfering signals before their rejection.This method makes it possible to eliminate a jamming source of n ' use only one degree of spatial freedom regardless of the number of multipaths that this source is received at the multisensor network.

 In addition, the method according to the invention updates the coefficients of the spatial filters in reception and in transmission, respectively performing spatial demultiplexing and spatial multiplexing, taking into account pairing faults, as a result of residual errors in the calibration of the channels. broadcast and in reception.

A device for carrying out the method according to the invention is shown diagrammatically in FIG. 2 and comprises the following subsets:
an antenna base 4 coupled to means 5 for switching frequency and tare,
a multi-channel receiver 6 comprising means for transposition radio amplification, and multi-channel analog / digital conversion means, not shown,
a spatial demultiplexer 7, comprising a determined number of spatial or space-time filters, coupled to a determined number of demodulators 8i equal to the maximum number of spatial channels fixed per frequency channel, for example three demodulators for a base comprising ten antennas,
a multi-channel transmitter 9 comprising means for transposition radio amplification and multi-channel digital-to-analog conversion, not shown,
a spatial multiplexer 10,
a determined number of modulators Ili equal to the number of demodulators 8i of the multipath receiver 6,
a digital processing module 12 delimited by a closed line in broken lines comprising a controller 13 and an estimator 14 specific to the AMRS processing, which is at the heart of the device for implementing the method according to the invention.

 The implementation of the AMRS concept by means of blind source separation methods, by the equipment it requires, the constraints of applied treatments and the new possibilities it offers, requires a system consideration. This translates into a number of control and management tasks necessary for optimal integration of spatial multiplexing into radio communications networks. These control and management tasks are performed in the device by the implementation of the method according to the invention.

 The separation, at the base station level, of the signals from or to the various "active" mobiles, that is to say in communication with the base station, in a given cell, involves the acquisition and the update of the specific spatial information of each mobile and requires for this a synchronization of the treatments.

 To achieve this function, the method according to the invention is based on an AMRT waveform with a temporal segmentation of the channel: the signal received by the base station, as shown in FIG. frames, frame # 1 to frame # K, each having a determined number M of "bursts". Each "burst", conventionally allocated to a user through spatial multiplexing is shared by several communications except for a "burst" per frame that remains allocated specifically to a mobile. These periodic "bursts" called "AMRS localization bursts" are therefore successively allocated to all the spatio-temporal channels potentially usable on the frequency channel (per frequency channel is a transmission / reception frequency pair).

 The structure of a "superframe", appearing in FIG. 3, corresponds to the period of reactualization of the spatial information of all space-time channels, ie a number K of frames equal to twice the number of these channels (for each space-time channel, the spatial information is acquired on two frames: one for each frequency channel frequency).

 Depending on the duration of the "superframe" with regard to stationarity of spatial information, this method must be adapted by allocating for example more than one "burst" per frame to the AMRS location to reduce the refresh period.

 It is obvious that the choice of the number of "bursts" of location per frame has a direct effect on the capabilities of the system.

The synchronization of the processing in this context essentially consists of directing the "bursts" received either towards the measurement of spatial information ("burst" of location of an active mobile), or towards the spatial demultiplexing ("burst" of traffic), or, nowhere ("burst" of unused location),
and thereafter distributing the spatial information to the transmit / receive multiplexing and transmitting filters according to the channel assignments and reassignments managed by the AMRS controller 13.

 FIG. 4 illustrates a block diagram of the interface between the AMRS controller 13, the AMRS estimator 14, and the transmit and receive spatial multiplexer / demultiplexer 10 and 7.

The orientation of the "bursts" is effected by means of a switch 15 receiving on a first input the signal received from the receiver 6 delimited by a discontinuous closed line, after transposition.
LF. The switch 15 also receives on a second input a control signal CTRL delivered by the AMRS controller 13. It comprises a first, a second and a third output respectively corresponding to the "burst" location of an active mobile, the "burst" traffic and the "burst" of unused location (mobile inactive).

The first output is coupled to the input of the estimator AMRS 14 estimating the director vector Sj of the source i corresponding to a determined active mobile i. The estimated steering vector S is injected into the controller
AMRS 13. The second output is inactive and the third output is coupled to the input of the spatial demultiplexer / demodulators 7 and 8i. The third output is also used for the refresh of spatio-temporal filters. The demultiplexer / demodulator 7 and 8i comprises N spatial or spatio-temporal filters (not shown), respectively receiving the same number N of spatial subchannels contained in the "burst" received traffic. The outputs of the spatial or spatio-temporal filters are respectively coupled at the input of the demodulators 8i, and deliver the data relating to the mobiles on the network of base stations shown in the figure by two parallel lines.

 A reactualization of the weights in reception of each coefficient of the spatial or spatio-temporal filters is carried out starting from the direction vector S estimated by the estimator AMRS 14 and after control by the controller AMRS 13.

 In transmission, modulators ili MOD # 1 to MOD # 3, delimited by a discontinuous closed line, respectively receive data received from the network of base stations, to be transmitted towards the mobile.

Each modulated information is injected respectively at the input of each spatial filter in transmission, not shown, of the spatial multiplexer 10 delimited by a discontinuous closed line.

 In the same way as in reception, a reactualization of the emission weight of the coefficients of the spatial filters is carried out from the same direction vector S estimated by the AMRS estimator 14 and controlled by the AMRS controller 13. The outgoing subchannels respectively of each spatial filter are summed and then transmitted after transposition in intermediate frequency F1 in the multipath transmitter 9 delimited by a discontinuous closed line.

 The spatial filter adapted to transmit to a mobile is calculated from its direction vector S estimated on the received signal. The need to distance sufficiently the transmission and reception frequencies, in particular to allow the use of a common antenna base for the two transmission / reception transmission directions, can make the source vectors of the sources estimated on the transmission station unusable. received signal.

 To solve this problem, a frequency permutation of the same frequency channel (transmission / reception) is carried out by the method according to the invention, resulting in the base station at a global permutation of the sub-bands allocated to the uplinks. and down. This permutation carried out at the frame rate makes it possible to estimate the spatial information on the frequency used in transmission with a shift frame with respect to the reception, which means that in the "superframe" two consecutive "bursts" of location are used to perform a complete estimate of the spatial information on each space-time channel. A frequency switching controller 16 is described in the following description with reference to FIG. 8.

 With spatial multiplexing, each channel, conventionally defined as the frequency and time support of a transmission between a base station SB and a mobile, is broken down into spatial subchannels.

FIG. 5 illustrates a schematization of this decomposition in which three X, Y and Z channels support the transmissions with respectively three mobiles MBx1, MBx2, MBx3, two mobiles MBy1, MBy2 and a mobile MBz1.

 Figure 6 gives an illustration of the notion of frequency and time channel for the X, Y, Z channels in the case of uplinks and in the case of downlinks.

 To perform a channel assignment, the method according to the invention is based on an inter-mobile spatial correlation table.

 This table, an example of which is shown in FIG. 7, contains the spatial correlation coefficients CS of each of the mobiles with all the others inside the fleet managed by the base station.

The spatial correlation coefficient CSij between two mobiles is established from their direction vectors ajb, ajb (estimated on the low frequency Fb of the channel) and aih, to \ h (estimated on the high frequency Fh of the channel) according to the formula:

 In the presence of multipaths, it is the vector of the most powerful path that represents the spatial information representative of the mobile with regard to the channel assignment.

 In FIG. 7, the mobiles are grouped in such a way as to reveal all the spatial correlations inside each channel respectively for the X channel, the Y channel and the Z channel (areas surrounded by dashed lines) whose values are more particularly taken into account in the channel assignment method.

 The hatched lines and columns correspond to unused spatial subchannels.

 With each new estimate of a manager vector relating to an active mobile (after the analysis of each AMRS location burst), the row and the associated column of the table are updated and channel assignment tests are applied. following a determined assignment algorithm. Its principle is as follows: an update of the distribution of mobiles on the different channels of the cell is performed every two frames after analysis of two "bursts" location allocated to each mobile successively on the two frequencies of the duplex channel.

If the last measured spatial information is associated with a mobile entering the cell, the AMRS controller applies to the new mobile p the following admission test:
for the mobile p to be admitted on the unsaturated channel Cn, it is necessary and sufficient that whatever the mobile x already present on one of the spatial subchannels of Cn:
CSxp threshold (2)
if this condition is not fulfilled on any unsaturated channel (with respect to these potential capacities, that is to say the maximum number of spatial subchannels) then the mobile can not be supported by the base station.

In the opposite case, the mobile is admitted on the channel Cn such that: Max ,, c, (CSpx) 2 Mini-r2NbCan [MaXy EC, (CSpy)] (3)
If the last measured spatial information is associated with a mobile p already present on the cell, then the AMRS controller first checks whether the mobile remains sufficiently spatially decorrelated from other mobiles present on the same channel by applying the following test:
CSxp (Threshold for all mobile x channel Cn (4)
if this condition is fulfilled, the assignment of the channels is not changed. In the opposite case, the AMRS control means tests the possibility of permuting the channel assignments between the mobile p and a mobile q on a different channel Cm. For this permutation to be possible, it is necessary and sufficient that the two following conditions are simultaneously realized:
CSyp threshold for all mobile y of the Cm channel (5)
CSqx threshold for all mobile x channel Cn (6)
If no permutation is possible, the mobile p can no longer be supported by the base station.

If, and this must be the most frequent case, several permutations are possible, the optimal choice criterion proposed is the following:
the mobile q retained to perform the permutation with the mobile p is the one that minimizes the expression:

 where Cm denotes the channel assigned to mobile q.

 As most spatial multichannel signal filtering techniques do not use a training sequence, the proposed method is applicable only if the transmit and receive chains do not introduce differential phase shift, gain, or distortions (amplification nonlinearities).

 It is very difficult and very expensive to produce a device satisfying these constraints, which is why a method is described below to correct by digital processing the drifts and distortions on the different transmission and reception channels.

 The correction of the differential phase-shifts and gain for each channel formed by each antenna of antennas is carried out by a calibration procedure decomposing in two steps. This procedure is implemented by a device whose functional diagram is illustrated in FIG.

 In a first step, the setting of the reception chain 1 7i of each channel i comprising means of transposition and of analog / digital conversion, is performed by injection upstream of the reception chain 1 7i of a calibration signal, then adapting the coefficients of an equalization filter 18i coupled to the output of the reception chain 17i (except the first socket as a reference) making it possible to compensate for the dispersion of the impulse responses of the different channels i, by a calibration analyzer common to all the channels formed by the antenna base 4.

 In a second step, the setting of the transmission channel 20i of each channel i comprising means for digital / analog conversion, filtering and frequency transposition is carried out analogously, by also injecting a calibration signal upstream. of the transmission channel 20i. Part of the transmitted signal is fed back to the reception channel 17i, previously tared, via a resistive coupler 21 i. Finally, an equalization filter 22i coupled at the input of the transmission channel 20i is adapted according to the same principle as that of the reception chain 17i.

 This figure also shows the controller 16 of the frequency swapping. A two-way switch ensures the transmission and reception respectively at the low frequency Fb, and the high frequency Fh of the received signal. These two frequencies Fb and Fh being separated by means of a duplexer 24i disposed between the resistive coupler 21 i and the two-way switch 23i. The frequency swap controller 16 Fb and Fh is coupled to the transmit and receive channels 20i and 17i of each channel i, the tare analyzer 19 and the two-way switch 23i.

 A first switch 25i arranged at the input of the reception channel 17i receives on a first input the reception calibration signal, on a second input the signal taken by the resistive coupler 21i and on a third input the reception signal to the low frequency Fb or high frequency Fh according to the position of the two-way switch 23i.

 A second one-way switch 26i is coupled to the output of the receive equalizer filter 1 8i and controls the receive tare.

 A third switch, a channel 27i, disposed at the input of the transmission channel before the equalization filter 22i receives on a first input the signal to be transmitted and on a second input the transmission setting signal.

 The calibration signal used is composed of N equuspaced sinusoidal lines in the useful band.

The principle of the adaptation of the p coefficients of the equalization filters 22i and 18i of the transmission and reception chains 20i and 17i is based on the minimization of the following criterion: m '- F. wiII2 (8)
where m; = 1 (Frequency response of channel i estimated at
Hi frequency of the calibration signal),
F. wj is the frequency response of the equalization filter of channel i,
with F matrix Nxp such that F (m, n) = e-2 j zfm n, and
one of the frequencies composing the setting signal.

The least squares solution for the equalization filter of channel i is therefore:

 where "+" in exponent corresponds to the function of conjugated transposition.

 The transmission channel formation as well as the independence of the sources means that the signals to be transmitted in AMRS mode do not have a constant envelope and that their dynamics can exceed the linear range of the amplifiers of the transmission chains 20i.

To avoid undesired distortions between channels, a pre-correction filter 20i is arranged at the input of the transmission channel 20i, the characteristics of which are as follows:
by modeling the amplitude and phase distortions of an amplifier as follows:

the pre-correction functions r and g 'established after learning the functions f and g are defined by:

<tb> If '<SEP> G.f109 <SEP> (11)
<Tb>
where G denotes a possible fixed gain of the pre-correction + amplifier filter assembly.

 In contrast to the acquisition of spatial information, the synchronization of the receiver on one of the transmission paths during the traffic phases (AMRS traffic bursts) is performed in a scrambled environment (all mobile sharing the same channel time-frequency transmitting simultaneously during traffic bursts).

 These reception conditions require the implementation of a more efficient synchronization method than conventional single-sensor synchronization (threshold detector at the filtering output adapted to the training sequence) that is inefficient in this context.

The synchronization method according to the invention was the subject of a French patent filed by the Applicant and published under No. 2,715,588 entitled: "Method and device enabling a modem to synchronize on a data transmitter in the presence of interferers "is implemented by an optimal multisensor detector under the following conditions:
- total noise (background noise + jammers) Gaussian,
- total circular noise (invariant by rotation),
- total white noise temporally (constant in the band, no multitrajet).

This detector aims at each sample the decision-making between two hypotheses:
HO: x (t) = b (t) (noise only)
H1: x (t) = d (t - To) ec (t) + b (t) (noise + signal) (12)
with c (t) designating the multichannel channel supporting the transmission, d (t) the transmitted signal and b (t) the total correlation matrix noise R.

The optimal detector in the sense of maximizing the probability of detection for a given false alarm probability corresponds to the likelihood ratio under H1 and HO over the time interval [noTe, (n0 + Ke) Te] (Te period of time). sampling)
A (x (n0 + k)) K p (x (n0 + k), 1 # k # Ke / H1) (13)
where p (x / HO), p (x / H1) respectively denote the probability density of the observation under HO, and HI.

The development of this analytic expression, under the assumptions defined above concerning the total noise, leads to the expression, function of x, c, b and R, as follows:

A statistic sufficient to solve the problem posed consists in submitting to a determined threshold n the following estimate:

 The spatial information characterizing the position of each mobile is an essential parameter for the adaptation of the channel formation filters in transmission. This is therefore defined for each of the mobiles by the source vector associated with the main energy arrival on the basis of antennas.

 This information is estimated on "bursts" allocated specifically and successively to each mobile and therefore in single-emission configuration (not scrambled).

For most cellular radio systems, standard channel models defining "typical" propagation conditions report a discrete impulse response corresponding to a number of paths. Therefore, for the estimation of spatial information, the modeling of the signal received by the antenna base is as follows:

where :: x (t) = [xl (t), x2 (t), ..., xN (t) iT, denotes the signal observed by the antenna base, 'T' by exponent defines the operation of transposition,
m (t), denotes the transmitted signal,
P, denotes the number of discrete and decorrelated temporally paths representing the impulse response of the channel,
garlic means the director vector associated with the path &num; i, more precisely aj integrates the product of two components ::
the response of the antenna base to an incident wavefront of azimuth O and of site zip (vector with N complex coefficients),
the response of the channel to the delay ti associated with the path &num; i (complex coefficient),
(the source vectors being defined to a close complex coefficient, in the continuation of the description the two components are most often confused in the notion of source vector),
b (t) = [b1 (t), b2 (t), ..., bN (t)] T, denotes the Gaussian white noise component on the N sensors. The principle of estimating the source vectors consists in extracting the outputs of a multi-sensor frame synchronization module. This module consists of N single-sensor correlators performing on each channel a filtering adapted to the training sequence.

In matrix notation this treatment is written:
y (t) = x (t) * d * (- t) (19)
for a sampled signal:

le vecteur constitué des N sorties des corrélateurs,
d = [d(o), d(i) d(L - 1), d(L - 2)] , la séquence d'apprentissage, et
x(t) = [x1(t), x2(t),..., 4j(t)], le signal observé sur les N capteurs.

the vector consisting of the N outputs of the correlators,
d = [d (o), d (i) d (L - 1), d (L - 2)], the training sequence, and
x (t) = [x1 (t), x2 (t), ..., 4j (t)], the signal observed on the N sensors.

The principal vector of the main path of the wanted signal is therefore estimated by the output of the N correlators at the frame synchronization instant (maximum of the multisensor detection function estimator), ie: j = y (tj) (21)
with tj = k1 path delay (composite or not) &num; i corresponding Fe to the main peak detected on the detection function implemented by the multi-sensor synchronization process.

There are other possible alternatives to this method of estimating spatial information that are compatible with the method that is the subject of the present invention, namely:
- estimation by separation of blind sources to higher orders with a limited scope of application to non-Gaussian signals,
- High resolution goniometry (with the additional constraint of having a calibrated antenna base).

 The step of demultiplexingemodulation in reception of the method according to the invention consists, by means of an antenna base coupled to a multi-channel receiver and a digital signal processing unit, to separate then demodulate the various components of the received signal. , associated with mobile stations transmitting simultaneously and on the same frequency to the base station.

 Several methods for demultiplexing followed by a reception demodulation according to the invention are described below, as well as a filter structure adapted to each method.

 These methods can be implemented from two types of filtering: spatial filtering or spatio-temporal filtering.

They all implement two identical consecutive digital processes, namely:
a multi-channel spatial or spatio-temporal filtering effecting the separation of the different components of the director vector and the optimization of the signal-noise ratio on the current symbol, and
a single-channel demodulation equalization that eliminates intersymbol interference and decision-making.

 They have in common the fact that they exploit the prior knowledge of learning sequences contained in the frames of the transmitted signal to adapt the coefficients of the filters implemented.

 Three structures allowing the implementation of these methods, already introduced previously, and which are recalled below.

A first FAS-DFE structure, illustrated in FIG. 9, applies to spatial filtering associated with a Viterbi-type equalization / demodulation module, encompassing two variants:
- FAS "estimated" -DFE, with adaptation of the spatial filter on the vector director of the source of interest, and
- FAS "replica" -DFE, with adaptation of the spatial filter directly on the reference sequence (replica).

a second and a third structure, a general structure of which is illustrated in FIG. 10, applies to spatio-temporal filtering associated with a DFE equalization / demodulation module and comprise respectively:
a FAST-BBT-DFE, illustrated in FIG. 11, having a decoupled structure, carrying out the anti-jamming and equalization / demodulation treatments, and whose implementation is optimal in the presence of total noise (background noise more interfering) ) white temporally, and
a FAST-DFE, illustrated in FIG. 12, jointly carrying out the spatial filtering processing and the equalization / demodulation function.

 These different methods and structures for their implementation are described in detail below.

 Concerning the FAS-DFE, the purely spatial filtering consists of optimizing for each mobile the antenna diagram according to its spatial information directly from the reference sequence of the useful signal, that is to say by forming lobes in the direction of the multipaths of the intended moving signal and the holes in the directions of the other moving signals. The base station in AMRS mode therefore applies as many spatial filters as there are active mobiles on the same channel.

The application of the spatial filter illustrated in Figure 10 is summarized as follows:
yi (t) wj + .x (t) (22)
where yj (t) represents the spatially filtered signal on which the contribution of the mobile has been amplified and that of the other mobile strongly attenuated.

 This filtering method makes it possible to reject N-1 jamming sources (uncorrelated multitrafts from the same scrambler are all sources of interference to be taken into account). In the presence of multitrafts of the useful signal, this method is not optimal because it does not make it possible to re-phase paths on the different channels before their summation.

In return, its implementation does not require a large processing capacity.

The two proposed adaptation variants of the reception spatial filter coefficients are known as "estimated" spatial filtering or "estimated" FAS, and adapted spatial filtering "replicates" consist of updating the spatial filter to isolate the mobile signal. i, according to the formula:
- For the "estimated" FAS: wi = (Rx + # 2j) -ai (23)
- For the FAS "replica":
Wi = (Rx + afl) eXd (24)
with::
Wi, the vector weight adapted to the separation of the source i,

Estimating the correlation matrix of the observations,

the composite directional vector of the mobile i constructed by summation of the direction vectors of the multipaths weighted by their relative powers,
a2, the fictitious noise term added to make the estimate more robust with respect to the risks of rejection of the useful signal, related to the errors made on âj (the addition of this noise term is not crucial for the separation method from source to blind, since the source vectors are directly and therefore precisely estimated from the received signals whereas in the case of direction finding each directional vector is deduced from an estimate of direction of arrival via calibration tables), and

Concerning the second structure relating to the implementation of the FAST-BBT-DFE method, the general problem of the adaptation on a replica of a spatio-temporal filtering structure for a multisensor signal is theoretically defined as follows:
the received signal modeled by
x (t) = G (t) * s (t) + b (t) (28)
where: x (t) = [x1 (t), x2 (t), ..., xN (t)] T denotes the signal received on the basis of antennas,
G (t) = [91 (t), g2 (t), ..., gN (t)], the modeling of the channel on the N sensors,
s (t) the transmitted signal, and
b (t) = [b1 (t), b2 (t) bN (t)] T, the noise + interfering component on the N sensors.

 The problem posed consists in estimating the coefficients of the filter W of the spatio-temporal structure realizing an optimal matched filtering with the multisensor signal.

The theoretical formulation of this filter is as follows:
W (t) = R-b1 (t) * G (t) (29)
with Rb (t) = E [b (t0), b + (t0 - t)] correlation matrix of the noise + interfering component.

The estimation of this expression is not simple in the general case also the first proposed method of adaptation (FAST-BBT) is placed in the restrictive hypothesis of a white noise temporally which is translated mainly concerning the interfering signals by the following assumptions:
- interfering signals are broadband (2 bands of the wanted signal),
- interfering signals are monotrajet.

 The problem to be solved is thus simplified by the form taken by the correlation matrix Rb.

Rb (t) = Rb S (t) with constant Rb. (6 (t) is the function of Dirac)
If the frequency domain is taken into account W is written:
W (f) = Rb1 G (f) (30)
Rb no longer depends on f and the problem is reduced to the separate estimates of G and Rb.

 The G channel estimate consists of N independent channel estimates on a single-sensor signal.

Given one of the channels to be estimated, the single-sensor modeling of the associated received signal x (t) at the frame synchronization time and sampled twice per symbol is as follows:
x (k) = gT s (k) + b (k) (31)
or:
s (k) = [s (kTe), s ((k - 1) Te) s (k - (L - 1) Te)] denotes the samples of the known or replicate learning sequence, and
b (k), the sample of the noise + scrambler component.

This equation, whose unknown is g, is generally solved by methods that minimize the mean squared error EQM defined by:

et dont la solution théorique est définie par: h = R rsx (33)

dont une estimation non biaisée est donnée par:

and whose theoretical solution is defined by: h = R rsx (33)

an unbiased estimate of which is given by:

and rsx = [s (t) x * (t)] whose unbiased estimate is given by:

être précalculée. In practice. Rss dependent exclusively. of the known prior learning sequence, the matrix Rss-1

to be precalculated.

 The estimation of the correlation matrix of the noise + interfering component Rb is based on the knowledge of the training sequence and the channel on the N sensors (matrix G).

From the previous modeling of the multisensor signal is deduced the expression of the estimate of the noise + interfering component on the following channel n:
bn (t) = xn (t) -gn (t) .s (t) (36)
hence the expression of the estimation of the correlation matrix Rb:

Several choices are possible to estimate W in particular to estimate the inverse of Rb. Among them is a method based on the eigenvariation of eigenvalues of Rb, namely:
Rb = UDU + (38)
with U, matrix of orthonormal eigenvectors and D diagonal matrix of eigenvalues Xj.

After this decomposition, the expression of W in the frequency domain becomes:
W (f) = U D-1U + G (f) (39)
This decomposition of W makes it possible to identify the different processing steps involved in the filtering structure shown in FIG. 11.

The expression of the filtered signal in the time domain thus becomes:
y (t) = W + (- t) x (t) (40)
with W + (- t) = G + (- t) U + D-1U (41)
The FAST-BBT-DFE method that has just been described therefore boils down to a filtering adapted on each channel not on the signal received directly but after a pre-treatment of the whitening of its component: background noise + scrambler; at the end of the processing chain, the channels thus constituted are summed after weighting each channel by a coefficient inversely proportional to the eigenvalue of the correlation matrix of the associated noise + jammer component.

The treatments break down sequentially as follows:
- channel estimation on scrambled signals on each channel,
- estimation of the correlation matrix of the noise + scrambler component Rb,
estimating the pretreatment matrix (matrix of the eigenvectors of the correlation matrix Rb),
preprocessing: projection of the received signal and impulse responses of the estimated channels on the basis of the eigenvectors,
filtering adapted to the channels projected on each channel for the re-phasing of the possible multipaths,
- combinations of the different channels, each being assigned a weight inversely proportional to the corresponding eigenvalue so that the components corresponding to the jammers (maximum eigenvalues) are taken into account in the combination with a minimum weight and vice versa,
- Suppression of intersymbol interference and decision-making by a DFE processing module adapting to a criterion of minimization of the mean squared error between "soft" decision and "hard" decision.

 The concatenation of the processing and the filtering constitute a spatio-temporal filtering structure differentiating from the purely spatial filtering by its ability to re-phase the decorrelated multipaths of the useful signal.

 Although this method is suboptimal, in the sense of the profitability of the antenna base, in the presence of multipath stray signals, it has the advantage of not requiring long learning sequences (adaptation of the space filters). temporally performed independently on each channel at the preprocessing output), the criterion for this parameter being the maximum time delay of the channel.

 Concerning the third structure relating to the implementation of the FAST-DFE adaptation method of a spatio-temporal filtering structure, no restrictive hypothesis concerning the statistical characteristics of noise have been taken.

 Its principle consists of the joint optimization of a spatio-temporal filter and the recursive part of a decision equalizer in the loop (decision feedback equalizer, DFE) on a global criterion of minimization of the mean squared error (EQM). ) between the "soft" decision at the equalizer output and either the learning sequence (at the beginning of the process) or the "hard" decision (later).

 The different elements of this structure and their arrangement are shown in FIG.

The signal at the output of the equalizer z (t) ("soft" decision) is therefore expressed as a function of the multi-sensor signal x (t) as follows:

where Wk is the temporal filter applied on the k-channel,
xk (n) xk (nT), xk (nT + T,), ..., xk (nT + (NHR-1) Te)] denotes the memory of the filter Wk on the channel k.

The MSE, minimization criterion adopted, is expressed according to the previous expression:

- pour un canal non stationnaire:

This criterion is estimated according to the type of channel in presence: - for a stationary channel:

- for a non-stationary channel:

 with x forgetting factor (0 (R <1) to be adjusted according to the degree of nonstationarity of the channel.

Numerous algorithms for estimating the set of filters (Wk with k = 1 to N, and HR) lead to each iteration to the minimization of the MSE, among which we can mention the recursive least squares (RLS) algorithm. , gradient algorithms (Gradient
Stochastic, Conjugated Gradients, etc.) and especially the algorithm "Recursive
Modified Gram Schmidt "(described in the Telecom thesis of JL FETY) probably the current algorithm best suited to make this estimate given its speed of convergence and its numerical stability.

 Simulations have confirmed the superiority of this method compared to the so-called FAST-BBT-DFE method in channel configurations with interferers with several paths (noise + noise component correlated temporally).

 However, this advantage has a cost in terms of the length of the reference sequence necessary to converge the overall estimate of all the coefficients of the filtering structure.

The table below presents the main advantages and disadvantages of the methods presented above:

<tb><SEP> Spatial <SEP> Filter <SEP> Spatial-Temper Spacer <SEP>
<tb><SEP><SEP> FASDFE <SEP> Method FAST-BBT-DFE <SEP> FAST-DFE
<tb><SEP> - <SEP> faiber <SEP> complexity <SEP> -remet <SEP> in <SEP> phase <SEP><SEP> -treatment <SEP> quasi
<tb><SEP> muftitrajets <SEP> useful <SEP> optimal, <SEP> which <SEP> that <SEP> is <SEP> the
<tb><SEP> - <SEP> sequence <SEP> channel
<tb><SEP> A <SEP> t <SEP> learning <SEP> short <SEP> - <SEP> sequence
<tb><SEP> van <SEP> ages <SEP> learning <SEP> short <SEP> - <SEP> sequence <SEP> short <SEP><SEP> requires <SEP> only one
<tb><SEP> degree <SEP> degree <SEP> of <SEP> freedom <SEP> spatial
<tb><SEP> by <SEP> scrambler <SEP> (what <SEP> that
<tb><SEP> be <SEP> the <SEP> number <SEP> of
<tb><SEP> trips)
<tb><SEP> - <SEP> sub-optimal <SEP> in <SEP> - <SEP> average complexity <SEP> average <SEP> - <SEP> complexity <SEP>
<tb><SEP> presence <SEP> of
<tb><SEP> multipath <SEP> useful <SEP> - <SEP> sub-optimal <SEP> in <SEP> - <SEP> sequence
<tb> Disadvantages <SEP> requires <SEP> a <SEP> degree <SEP> presence <SEP> of <SEP> multipath <SEP> learning <SEP> long
<tb><SEP> -need <SEP> a <SEP> degree <SEP> of <SEP> jammers
<tb><SEP> Freedom <SEP> Spatial <SEP> by <SEP> Trip
<tb><SEP> independent jammer <SEP>
<Tb>
Depending on the problem to be addressed: - configuration of useful paths, - number and configuration of jammers, - length of the training sequence, - number of sensors, - possible treatment means.

 The analysis of the method presenting the best cost / performance compromise must be made.

 For spatial multiplexing in transmission, the principle of spatial filtering is to perform the reciprocal operation of the spatial filtering in reception, ie instead of separating signals arriving on the basis of antennas according to different azimuths, it is It acts with the same basis to generate a composite signal, each component addressed to a particular mobile propagates in the direction defined by its main director vector.

 Unlike reception filtering where, in order to optimize the signal-to-noise ratio, it is important to take into account all the multipaths of the same source, in emission it is preferable to focus the energy only in the direction of the multipath received the most powerful.

The application of the weights to the signals to be emitted from the different modulators MOD # 1 to &num; 3, 11i, as illustrated in FIG. 4, is written as follows:
x = Wm (46)
with:
m = (m1 (t), m2 (t) ..., mM (t)) signals from the M modulators,
x = (x1 (t), x2 (t) xN (t)) signals transmitted on the N antennas
W vector matrix wi (i = 1 to M).

 The method of adaptation of the coefficients of the spatial filter in reception is theoretically applicable in emission, but due to the residual errors of calibration of the channels in emission and in reception, so-called errors of "copying" are committed which cause errors of pointing with poor positioning of the "holes" in the radiation pattern and the risk of the reception by the mobile of a bad Signal / Scrambler report without possibility of rejection.

 This is why the emission adaptation is carried out according to a slightly different technique called synthetic FAS.

The computation of wi, spatial filter in emission adapted to the mobile i, is obtained by applying to its principal director vector (associated with the most powerful multitrajet) the inverse, either of the estimated correlation matrix of the observations on the N sensors, but of a synthetic correlation matrix of the interfering signals Rji:
wi = R1.â (47)
JI
This matrix is constructed from the vectors of two fictitious jammers positioned on both sides (in azimuth) of each real jammer (each mobile constituting a jammer for all the others).

The advantage of this method is to allow a slight error on the interfering vectors of the jammers by widening the "holes" of the radiation pattern around their estimated directions.

The synthetic correlation matrix used for estimating the weight vector wi (focusing towards the mobile i) is written in the form:

with:
ajk representing the vectors of two fictitious jammers:
ajk = âj (O + k.AS) (49)
in the absence of information on the azimuth of the source O and on the variation of a directional vector with the azimuth of its source (without a calibration table), these vectors are obtained by phase shift of the estimated direction vector ajk = Dk.âj (#) (50)
with Dk diagonal diagonal matrix whose diagonal coefficients are of the form ::
dk (i) = eik. (iN) .aa (51)
Asp is a phase shift of the order of a few degrees to be adapted according to the base of antennas 4,
flj denotes the power of the source j considered jammer in this case, this parameter is fixed at a value proportional to the depth of the "hole" that it is necessary to form in the direction of the mobile j on the radiation pattern (typically 30 dB above the useful signal).

 Oi2 is a term of noise added to adjust the depth of the "hole" in the radiation pattern, its addition also has the effect of making the matrix Rjj invertible and avoid the recourse to a costly calculation of pseudo-inverse matrix.

 The present invention applies more particularly to base stations although certain aspects of the method according to the invention which has just been described can find applications at the level of mobile stations in particular as regards the increase of their reception sensitivity. .

 The DFE equalizer implemented in the present invention can be replaced by an equalizer according to the maximum likelihood principle implementing the Viterbi algorithm.

Claims (18)

 1. A multiplexageletemultiplexing method of digital radio signals organized in frames, comprising sequences known a priori, and transmitted in duplex between at least one base station and several mobiles communicating on the same frequency and time channel by means of a receiver and a multi-channel transmitter (1) integrated in the base station and coupled to an antenna base, characterized in that it consists of:  to estimate (2) the spatial information relating to each mobile, from the signal received by the multi-channel receiver, for the reception and transmission frequencies, by means of source separation methods, by exploiting the known sequences a priori, and from this information,  to isolate (3) in the presence of multipaths in the channel, the respective paths to each mobile having a steering vector whose power is greater than a determined threshold, and to demodulate the signal received from each of the mobiles in connection with the station of basis, to allow spatial demultiplexing, and  - Simultaneously transmit in the direction of the most powerful paths of the isolated paths, the signals to the corresponding mobile, protecting each mobile vis-à-vis the signals transmitted to others by spatial filtering, to allow spatial multiplexing.  2. Method according to claim 1, characterized in that it consists in isolating the main paths by spatial filtering.  3. Method according to claim 2, characterized in that the spatial filtering is implemented by a FAS structure associated with a decision and equalization element.  4. Method according to claim 1, characterized in that it consists in isolating the main paths by spatio-temporal filtering.  5. Method according to claim 4, characterized in that the spatio-temporal filtering is implemented by a FAST-BBT structure associated with a decision and equalization element.  6. Method according to claims 3 and 5, characterized in that the decision and equalization member is implemented by a DFE.  7. Method according to claims 3 and 5, characterized in that the decision and equalization member implements the Viterbi algorithm.  8. Method according to claim 4, characterized in that the spatio-temporal filtering is implemented by a FAST-DFE structure.  9. Method according to any one of claims 1 to 8, characterized in that the estimation of the spatial information is performed in a single-emission configuration, on periodic time periods, arranged in a given frame structure, and allocated specifically to the estimation of the spatial information of each mobile over two consecutive frames in order to take into account the two frequencies of the duplex transmission.  10. Method according to claim 9, characterized in that the estimation of the spatial information is performed from the value taken by a multi-channel detection function of the known sequence a priori, at the optimum timing of synchronization corresponding to the moment of arrival of the most powerful route.  11. Method according to any one of claims 1 to 10, characterized in that it consists in permuting each frame transmission and reception frequencies to allow the acquisition of spatial information on the two frequencies when they are too much. remote so that this information can be transposed from one to the other,  12. Method according to any one of claims 1 to 11, characterized in that it consists to optimize the efficiency of spatial multiplexing and demultiplexing, to manage the allocation of channels at the level of all the frequency resources and temporal data allocated to the base station according to the movements of the mobiles, using a spatial correlation criterion between mobiles integrating the spatial information acquired on the two frequencies used alternately for the duplex transmission.  13. Method according to any one of claims 1 to 12, characterized in that the spatial filtering transmission of the signal to a given mobile is adapted from a synthetic correlation matrix signal noise + scramblers established at from pairs of fictitious jammers arranged spatially on either side of the mobiles towards which the base station must not emit this signal so as to widen the holes "of the radiation pattern towards the jammers and consequently to increase the tolerance of the system to pointing errors resulting from residual errors in the calibration of the transmission and reception channels.  14. A method according to any one of claims 1 to 13, characterized in that it consists in periodically applying a taring procedure and pre-correction amplification on transmission to limit the misdirection in transmission  15. Device for implementing the method according to any one of claims 1 to 14, characterized in that it comprises ::  a transmitting / receiving antenna base (4) coupled to switching and setting means (5),  a multi-channel receiver (6) comprising means of frequency translation, amplification and analog / digital conversion, coupled at the output of the switching and setting means (5),  a spatial demultiplexer (7), coupled at the output of the multi-channel receiver (6), comprising a determined number of spatial or spatio-temporal filters, each spatial or spatio-temporal filter respectively forming a spatial or spatio-temporal channel,  a determined number of demodulators (8i) equal to the number of spatial channels determined per frequency and time channel, coupled at the output of the demultiplexer (7), and delivering on the network of the base stations the data transmitted by the mobiles,  a given number of modulators (here) equal to the number of demodulators (8i), respectively receiving data to be transmitted by the base of antennas (4) intended for mobiles,  a spatial multiplexer (10), coupled at the output of the modulators (they), comprising a determined number of spatial filters equal to the number of modulators (11 i),  a multi-channel transmitter (9) comprising digital-to-analog conversion, frequency conversion and amplification means coupled to the output of the spatial multiplexer (10), and  in that it includes ::  an AMRS digital processing module (12), comprising an AMRS controller (13) and an estimator (14) of the direction vectors of the mobile-related paths communicating with the base station, allowing the updating of the spatial filter weights to the transmission and spatial or space-time filters at the reception, the management of the transmission and reception frequencies, and the controls of the communication and tare means.  Device according to claim 15, characterized in that the calibration means comprise for each channel (i) respectively formed by each antenna of the antenna base (4):  an equalization filter of the transmission signal (22i) coupled at the input of the transmission channel (20i),  an equalization filter (18i) of the reception signal coupled at the output of the reception chain (17i),  an input (25i) and an output (26i) switch of the reception chain (17i), a switch (27i) at the input of the transmission chain (20i) and a two-way switch (23i) at the output of the transmission channel (20i), allowing the device to go through the various operating modes: reception calibration, transmission calibration, transmission / reception,  - a coupler (21 i) arranged between the antenna (Ai) and the two-way switch (23i), for taking a part of the transmission signal to inject it at the input of the reception chain (17i) via the input switch (25i) of the reception chain (17i),  and in that it comprises in common for all the ways (i):  a calibration analyzer (19) for analyzing the various calibration signals arriving at the input (27i) and output (26i) switches respectively of the transmit (20i) and receive (17i) channels, and adapting the equalization filters (22i and 18i) of the transmission and reception channels (20i and 17i).  17. Device according to claim 16, characterized in that a pre-correction filter (28i) of the transmission chain (20i) is arranged between the equalization filter of the transmission channel (20i) and the transmission channel (20i).  18. Device according to any one of claims 15 to 17, characterized in that it further comprises, for the permutation of the transmit and receive frequencies of the same frequency channel, a frequency permutation controller (16i), coupled respectively to the receiver (6) and the transmitter (9) multipath, and the switching and setting means (4).