METHOD AND DEVICE FOR PERFORMING TRANSMISSIONS OF DATA

Abstract

For determining, in a distributed fashion, precoders to be applied for performing transmissions of data between a plurality of transmitters and a plurality of receivers via a global MIMO channel H of a wireless communication system, said precoders jointly forming an overall precoder V, each and every j-th transmitter perform: gathering long-term statistics of CSIT errors incurred by each one of the transmitters; obtaining short-term CSIT related data and building therefrom its own view Ĥ(j) of the global MIMO channel H; determining an estimate {tilde over (V)}(j) of the overall precoder V from the short-term CSIT related data; refining the estimate {tilde over (V)}(j) on the basis of the gathered long-term statistics of CSIT errors so as to obtain a refined precoder {tilde over (V)}(j) that is a view of the overall precoder V from the standpoint of said j-th transmitter, further on the basis of its own view Ĥ(j) of the global MIMO channel H, and further on the basis of a figure of merit representative of performance of said transmissions via the global MIMO channel H.

Description

TECHNICAL FIELD

The present invention generally relates to determining, in a distributed fashion, precoders to be applied for transmitting data between a plurality of transmitters and a plurality of receivers in a MIMO channel-based wireless communication system.

BACKGROUND ART

Wireless communication systems may rely on cooperation in order to improve their performance with regard to their environment. According to one example, such cooperation can be found in a context of a MIMO (Multiple-Input Multiple-Output) channel-based communications network in which node devices, typically access points such as base stations or eNodeBs, cooperate in order to improve overall robustness of communications via the MIMO channel.

So as to perform such cooperation, transmitters of a considered wireless communication system rely on CSI (Channel State Information) related data and/or channel estimation related data for determining a precoder to be applied by said transmitters in order to improve performance of transmissions via the MIMO channel from said transmitters to a predefined set of receivers. Such a precoder is typically determined in a central fashion, and parameters of the determined precoder are then propagated toward said transmitters for further applying said determined precoder during transmissions via the MIMO channel from said transmitters to said receivers.

It would be advantageous, in terms of system architecture and in terms of balance of processing resources usage, to provide a method for enabling determining the precoder parameters in a distributed fashion among the transmitters. However, doing so, CSIT (CSI at Transmitter) mismatch generally appears. This may involve a significant divergence between the precoder parameters computed by the transmitters on their own. The performance enhancement targeted by the cooperation is therefore not as high as expected, since the precoder parameters independently determined by the transmitters involve residual interference that grows with the CSIT mismatch.

It is desirable to overcome the aforementioned drawbacks of the prior art. It is more particularly desirable to provide a solution that allows improving performance of transmissions from a predefined set of transmitters toward a predefined set of receivers in a MIMO-channel based wireless communication system by relying on a precoder determined in a distributed fashion among said transmitters, although CSIT mismatch may exist.

SUMMARY OF INVENTION

To that end, the present invention concerns a method for performing transmissions of data between a plurality of K_t transmitters and a plurality of K_r receivers via a global MIMO channel H=[H₁, . . . ,H_Kr] of a wireless communication system, by determining in a distributed fashion precoders to be applied for performing said transmissions, said precoders being respectively applied by said transmitters and jointly forming an overall precoder V, wherein each and every j-th transmitter among said plurality of K_t transmitters performs: gathering long-term statistics of Channel State Information at Transmitter CSIT errors incurred by each one of the K_t transmitters with respect to the global MIMO channel H, the long-term statistics describing the random variation of the CSIT errors; obtaining short-term CSIT related data and building its own view Ĥ^(j) of the global MIMO channel H; determining an estimate {tilde over (V)}^(j) of the overall precoder V from the obtained short-term CSIT related data; refining the estimate {tilde over (V)}^(j)[{tilde over (V)}₁^(j), . . . ,{tilde over (V)}_Kr^(j)] of the overall precoder Von the basis of the gathered long-term statistics of CSIT errors so as to obtain a refined precoder {tilde over (V)}^(j)[{tilde over (V)}₁^(j), . . . ,{tilde over (V)}_Kr^(j)] is a view of the overall precoder V from the standpoint of said j-th transmitter, further on the basis of its own view Ĥ^(j) of the global MIMO channel H, and further on the basis of a figure of merit representative of performance of said transmissions via the global MIMO channel H; and transmitting the data by applying a precoder that is formed by a part of the refined precoder V^(j) which relates to said j-th transmitter among said plurality of K_t transmitters.

Thus, performance of transmissions via the global MIMO channel of the wireless communication system is improved by relying on a precoder determined in a distributed fashion, although CSIT mismatch may exist. Robustness against CSIT mismatch is thus achieved without needing a central unit to compute the precoder.

According to a particular feature, the figure of merit is a lower bound of a sum rate LBSR^(j) reached via the global MIMO channel H, from the standpoint of said j-th transmitter with respect to its own view Ĥ^(j) of the global MIMO channel H, as follows:

${\mathrm{LBSR}}^{\left(j\right)}=\sum _{k=1}^{K_r}\log {\det \left({\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right)}^{-1}$ $\mathrm{wherein}$ ${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)={}_{\left\{{\Delta }^{\left(1\right)},{\Delta }^{\left(2\right)},\dots ,{\Delta }^{\left(K_t\right)}|{\hat{H}}^{\left(j\right)}\right\}}\left[{\mathrm{MSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right]$

wherein represents the mathematical expectation and, wherein MSE_k^(j) (F₁^j, . . . ,F_Kr^(j)) represents mean square error matrix between the data to be transmitted and a corresponding filtered received vector for a realization of estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT errors.

Thus, the sum of the rate of the receivers served by said transmissions is improved by the determined precoder.

According to a particular feature, the figure of merit is the sum of traces MINMSE^(j), for k=1 to K_r, of EMSE_k^(j) (F₁^j, . . . ,F_Kr^(j)) as follows:

${\mathrm{MINMSE}}^{\left(j\right)}=\sum _{k=1}^{K_r}\mathrm{Trace}\left({\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right)$ $\mathrm{wherein}$ ${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)={}_{\left\{{\Delta }^{\left(1\right)},{\Delta }^{\left(2\right)},\dots ,{\Delta }^{\left(K_t\right)}|{\hat{H}}^{\left(j\right)}\right\}}\left[{\mathrm{MSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right]$

wherein E represents the mathematical expectation and, wherein MSE_k^(j) (F₁^j, . . . ,F_Kr^(j)) represents the mean square error matrix between the data to be transmitted and a corresponding filtered received vector for a realization of estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT errors.

Thus, the average mean square error, as perceived by the receivers is improved by the determined precoder.

According to a particular feature, refining the estimate {tilde over (V)}^(j) of the overall precoder V is performed thanks to a refinement function f(. , .), as well as a set {F_k^(j)} of refinement matrices F_k^(j), k=1 to K_r, in a multiplicative refinement strategy, as follows:

V _k ^(j) =f({tilde over (V)}_k^(j),{tilde over (F)}_k^(j))={tilde over (V)}_k^(j),{tilde over (F)}_k^(j)

Thus, such a multiplicative refinement allows for correcting CSIT mismatch, especially in the context of block diagonalization precoding.

According to a particular feature, the overall precoder V is a block-diagonalization precoder, the transmitters have cumulatively at least as many antennas as the receivers, and refining the estimate {tilde over (V)}^(j) of the overall precoder V thus consists in optimizing the set {F_k^(j)} of the refinement matrices F_k^(j) with respect to the set {{tilde over (V)}_k^(j)} of the matrices {tilde over (V)}_k^(j), which is obtained by applying a Singular Value Decomposition operation as follows:

Ĥ _[k] ^(j) =U _[k] ^(j) [D _[k] ^(j), 0][{tilde over (V)}′_[k]^(j),{tilde over (V)}″_k^(j)]†

wherein Ĥ_[k]^(j) represents a view of an aggregated interference channel estimation Ĥ_[k]^(j) for the k-th receiver among the K_r receivers from the standpoint of said j-th transmitter, with

H _[k] =[H† ₁ , . . . ,H† _k−1 ,H† _k+1 , . . . ,H† _{K r} ]†

wherein {tilde over (V)}_k^(j) is obtained by selecting, according to a predefined selection rule similarly applied by any and all transmitters, a predetermined set of N columns of the matrix {tilde over (V)}″_k^(j) resulting from the Singular Value Decomposition operation, wherein each receiver has a quantity N of receive antennas.

Thus, a distributed block diagonal precoder is made robust to CSIT mismatch.

According to a particular feature, the overall precoder V is an interference aware coordinated beamforming precoder with block-diagonal shape, K_t=K_r, and each transmitter has as a quantity M of transmit antennas equal to a quantity N of receive antennas of each receiver, each transmitter communicates only with a single receiver among the K_r receivers such that k=j,

wherein a sub-matrix W′_k such that:

V _k =E _k W′ _k

is computed as the eigenvector beamforming of the channel matrix defined by E_k^TĤ^(k)E_k, from a Singular Value Decomposition operation applied onto, said channel matrix defined by E_k^TĤ^(k)E_k as follows:

E_k^TĤ^(k)E_k=U′_kD′_kW′_k

wherein E_k is defined as follows:

E _k=[0_M×(k−1)M,I_M×M,0_M×(Kt−k)M]^T

with 0_M×(k−1)M an M×(k−1)M sub-matrix containing only zeros, 0_M×(Kt−k)M an M×(K_t−1)M sub-matrix containing only zeros, and I_M×M an M×M identity sub-matrix.

Thus, a distributed coordinated beamforming precoder is made robust to CSIT mismatch.

According to a particular feature, refining the estimate {tilde over (V)}^(j) of the overall precoder V is performed thanks to a refinement function f(. , .), as well as a set {F_k^(j)} of refinement matrices F_k^(j), k=1 to K_r, in an additive refinement strategy, as follows:

V _k ^(j) =f({tilde over (V)}^(j),F_k^(j))={tilde over (V)}^(j)+F_k^(j)

Thus, an additive refinement allows for correcting CSIT mismatch, especially in the context of regularized zeros forcing precoders.

According to a particular feature, the overall precoder V is a regularized zero-forcing precoder, and the estimate {tilde over (V)}^(j) of the overall precoder V can be expressed as follows:

{tilde over (V)} ^(j)=(Ĥ^(j)†Ĥ^(j)+α^(j)l)⁻¹Ĥ_k^(j)†

wherein α^(j) is a scalar representing a regularization coefficient that is optimized according to statistics of the own view Ĥ^(j) of the global MIMO channel H from the standpoint of said j-th transmitter, and wherein α^(j) is shared by said j-th transmitter with the other transmitters among the K_t transmitters.

Thus, a Regularized Zero Forcing precoder is made robust to CSIT mismatch.

According to a particular feature, refining the estimate {tilde over (V)}^(j) of the overall precoder V is performed under the following power constraint:

Trace((f({tilde over (V)}^(j),F_k^(j)))†f({tilde over (V)}^(j),F_k^(j)))=N

wherein each receiver has a quantity N of receive antennas.

Thus, transmission power is restrained.

The present invention also concerns a device for performing transmissions of data between a plurality of K_t transmitters and a plurality of K_r receivers via a global MIMO channel H=[H₁, . . . ,H_K] of a wireless communication system, by determining in a distributed fashion precoders to be applied for performing said transmissions, said precoders being respectively applied by said transmitters and jointly forming an overall precoder V, wherein said device is each and every j-th transmitter among said plurality of K_t transmitters and comprises: means for gathering long-term statistics of Channel State Information at Transmitter CSIT errors incurred by each one of the k transmitters with respect to the global MIMO channel H, the long-tem statistics describing the random variation of the CSIT errors; means for obtaining short-term CSIT related data and building its own view Ĥ^(j) of the global MIMO channel H; means for determining an estimate {tilde over (V)}^(j) of the overall precoder V from the obtained short-term CSIT related data; means for refining the estimate {tilde over (V)}^(j)=[{tilde over (V)}₁^(j), . . . ,{tilde over (V)}_Kr^(j)] of the overall precoder Von the basis of the gathered long-term statistics of CSIT errors so as to obtain a refined precoder V^(j)=[V₁^(j), . . . ,V_Kr^(j)] that is a view of the overall precoder V from the standpoint of said j-th transmitter, further on the basis of its own view Ĥ^(j) of the global MIMO channel H, and further on the basis of a figure of merit representative of performance of said transmissions via the global MEMO channel H; and means for transmitting the data by applying a precoder that is formed by a part of the refined precoder V^(j) which relates to said j-th transmitter among said plurality of K_t transmitters.

The present invention also concerns a computer program that can be downloaded from a communications network and/or stored on a medium that can be read by a computer or processing device. This computer program comprises instructions for causing implementation of the aforementioned method, when said program is run by a processor. The present invention also concerns an information storage medium, storing a computer program comprising a set of instructions causing implementation of the aforementioned method, when the stored information is read from said information storage medium and run by a processor.

Since the features and advantages related to the communications system and to the computer program are identical to those already mentioned with regard to the corresponding aforementioned method, they are not repeated here.

The characteristics of the invention will emerge more clearly from a reading of the following description of an example of embodiment, said description being produced with reference to the accompanying drawings, among which:

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 schematically represents a wireless communication system in which the present invention may be implemented.

FIG. 2 schematically represents an example of hardware architecture of a communication device, as used in the wireless communication system.

FIG. 3 schematically represents an algorithm for determining, in a distributed fashion, precoders to be applied for transmitting data from a plurality of transmitters toward a plurality of receivers in the wireless communication system.

FIG. 4 schematically represents an iterative algorithm for determining refinement matrices used to refine the precoders.

DESCRIPTION OF EMBODIMENTS

FIG. 1 schematically represents a wireless communication system 100 in which the present invention may be implemented.

The wireless communication system 100 comprises a plurality of transmitters, two 120a, 120b of which being represented in FIG. 1. The wireless communication system 100 further comprises a plurality of receivers, two 110a, 110b of which being represented in FIG. 1. For instance, the transmitters 120a, 120b are access points or base stations of a wireless telecommunications network, and the receivers 110a, 110b are mobile terminals having access to the wireless telecommunications network via said access points or base stations.

The transmitters 120a, 120b cooperate with each other in order to improve performance when performing transmissions from the plurality of transmitters 120a, 120b toward the plurality of receivers 110a, 110b via wireless links 111a, 111b, 111c, 111d. The wireless link 111a represents the transmission channel from the transmitter 120a to the receiver 110a, the wireless link 111b represents the transmission channel from the transmitter 120a to the receiver 110b, the wireless link 111c represents the transmission channel from the transmitter 120b to the receiver 110a, and the wireless link 111d represents the transmission channel from the transmitter 120b to the receiver 110b. The transmitters 120a, 120b are interconnected, as shown by a link 121 in FIG. 1A, so as to be able to exchange long-term statistics about transmission channel observations. The link 121 can be wired or wireless.

The cooperation is achieved by making the transmitters 120a, 120b apply respective precoders when performing said transmissions. Said precoders are determined in a distributed fashion within the wireless communication system so that each transmitter determines the precoder that said transmitter has to apply in the scope of said transmissions. More particularly, each transmitter (identified by an index j among the plurality of transmitters) determines, independently from the other transmitters of said plurality, its own view V^(j) of an overall precoder V that should be cooperatively applied by said plurality of transmitters for performing said transmissions. This aspect is detailed hereafter with respect to FIG. 3.

Herein the quantity of transmitters 120a, 120b in use is denoted K_t, each transmitter having a quantity M of transmit antennas, and the quantity of receivers 110a, 110b in use is denoted K_r, each receiver having a quantity N of receive antennas. The receivers 110a, 110b are configured to simultaneously receive signals from plural transmitters among the K_t transmitters. A global MIMO channel H is thus created between the K_t transmitters and the K_r receivers. The part of the global MIMO channel H which links a j-th transmitter among the K_t transmitters to a k-th receiver among the K_r receivers is represented by an N×M matrix herein denoted H_k,j. One can note that H_k,j is representative a MIMO channel too. The part of the global MIMO channel H that links the k transmitters to the k-th receiver among the K_r receivers is a concatenation of the K_t MIMO channels H_k,j, with j=1 to K_t, and is therefore an N×MK_t matrix herein denoted H_k. One can further note that H_k is representative of a MIMO channel too.

Let's consider a set of symbol vectors s_k. Each symbol vector s_k of length N represents the data that has to be transmitted to the k-th receiver among the plurality of K_r receivers, at a given instant. Let's further denote s the stacked vector s=[s₁^T,s₁^T, . . . ,s_Kr^T]^T that contains all data to be transmitted by the K_t transmitters to the K_r receivers at said given instant, wherein A^T represents the transpose of a vector or matrix A.

Let's further consider the following overall precoder V:

V=[V₁, . . . ,V_Kr]

and further define E_j^TV, with j=1 to K_t, as the part of the overall precoder V to be applied by the j-th transmitter among the K_t transmitters, wherein E_j is an M×NK_r matrix such that E_j=[0_M×(j−1)M,I_M×M,0_M×(Kt−j)M]^T, and wherein V_k, with k=1 to K_r, is the equivalent part of the overall precoder V which has to be applied to reach the k-th receiver among the K_r receivers. Inline with the notations already defined hereinbefore, one should note that V_k^(j) represents hereinafter the view of the precoder equivalent part V_k from the standpoint of the j-th transmitter among the K_t transmitters.

It should be noted that 0_M×(j−1)M in the expression of E_j above represents an M×(j−1)M sub-matrix of E_j containing only zeros, 0_M×(j−1)M represents an M×(K_t−j)M sub-matrix of E_j containing only zeros, and I_M×M represents an M×M identity sub-matrix (could be an M×M identity matrix in other contexts herein).

In a joint processing CoMP (Coordinated Multipoint Transmission) approach, any and all transmitters know entirely the set of the symbol vectors s_k to be transmitted toward the K_r receivers at a given instant.

In a coordinated precoding approach where K_t=K_r, each transmitter among the K_t transmitters communicates with one receiver among the K_r receivers. It means that the j-th transmitter among the K_t transmitters only has to know the symbol vector s_k to be transmitted to the k-th receiver (with k=j) among the K_r receivers with which said j-th transmitter communicates, which implies that the overall precoder V has a block-diagonal shape. In the case where each and every j-th transmitter among the K_t transmitters has to communicates with the k-th receiver among the K_r receivers in such a way that k≠j, reordering of the K_t transmitters with respect to the index j and/or of the K_r receivers with respect to the index k is performed so as to make the overall precoder V have a block-diagonal shape.

Considering the statements here above, a model of the wireless communication system 100 can be expressed as follows:

$y_k=H_k\mathrm{Vs}+n_k=H_kV_ks_k+\sum _{=1,\ne k}^{K_r}H_kV_{}s_{}+n_k$

wherein:

• • y_k represents a symbol vector as received by the k-th receiver among the K_r receivers when the symbol vector s_k has been transmitted to said k-th receiver; and n_k represents an additive noise incurred by said k-th receiver during the transmission of the symbol vector s_k.

It can be noticed that, in the formula above, the term H_kV_ks_k represents the useful signal from the standpoint of the k-th receiver among the K_r receivers and the sum of the terms H_k represents interference incurred by the k-th receiver among the K_r receivers during the transmission of the symbol vector s_k.

A receive filter can be computed from the channel knowledge H_kV by the k-th receiver among the K_r receivers, which may be obtained by a direct estimation if pilots precoded according to the overall precoder V are sent by the concerned transmitter(s) among the K_t transmitters, or by obtaining the overall precoder V by signalling from the concerned transmitter(s) among the K_t transmitters and further by estimating the MIMO channel H_k from pilots sent without precoding on this MIMO channel H_k.

When using a Zero-Forcing receive filter, the k-th receiver among the K_r receivers uses a linear filter T_k defined as follows:

T _k=((H_kV)†H_kV)⁻¹(H_kV)†

When using an MMSE receive filter, the k-th receiver among the K_r receivers uses a linear filter T_k defined as follows:

T _k=((H_kV)†H_kV1)⁻¹(H_kV)†

Then, a decision is made by said k-th receiver on the filtered received vector T_ky_k for estimating the symbol vector s_k.

It has to be noted that, in the case where there is no effective receive filter (for instance when Regularized Zero Forcing is applied by the transmitters), T_k=1.

The K_t transmitters are configured to obtain CSIT (Channel State Information at the Transmitter). CSIT is obtained by each transmitter among the k transmitters from:

• • feedback CSI (Channel State Information) from one more receivers among the K_r receivers, and/or
  • channel estimation performed at said transmitter and using a channel reciprocity property, and/or
  • from such CSI or such channel estimation provided by one or more other transmitters among the K_t transmitters,
  • in such a way that CSI related data and/or channel estimation related data propagation rules within the wireless communication system 100 lead to CSIT errors and moreover to CSIT mismatch among the K_t transmitters (i.e. different CSIT from the respective standpoints of the K_t transmitters), for example due to quantization operations.

One can note that, in addition to quantization operations, disparities in CSI related data effectively received by the K_t transmitters imply that differences in CSIT exist from one transmitter to another among the K_t transmitters, which leads to CSIT mismatch.

The global MIMO channel H can thus be expressed as follows, considering each and every j-th transmitter among the K_t transmitters:

H=Ĥ ^(j)+Δ^(j)

wherein Ĥ^(j) represents a view of the global MIMO channel H from the standpoint of the j-th transmitter among the K_t transmitters, which is obtained by said j-th transmitter from the CSIT obtained by said j-th transmitter, and wherein Δ^(j) represents an estimate error between the effective global MIMO channel H and said view Ĥ^(j) of the global MIMO channel H from the standpoint of said j-th transmitter. In a similar manner, Ĥ_k,i^(j) denotes the view of the MIMO channel H_k,i from the standpoint of said j-th transmitter and Ĥ^(j) denotes the view of the MIMO channel H_k from the standpoint of said j-th transmitter.

Therefore, the view V^(j) of the overall precoder V might then be slightly different from one transmitter to another among the K_t transmitters, due to the CSIT mismatch. The j-th transmitter among the K_t transmitters then extracts, from the view V^(j) of the overall precoder V which has been determined by said j-th transmitter, the precoder E_j^TV^(j) that said transmitter has to apply in the scope of said transmissions. As already mentioned, this is independently performed by each transmitter among the K_t transmitters (j=1 to K_t). Optimization is therefore adequately performed so as to improve the performance of the transmissions from the K_t transmitters to the K_r receivers, in spite of the CSIT mismatch and despite that each j-th transmitter among the K_t transmitters independently determines the precoder E_j^TV^(j) that said transmitter has to apply. This aspect is detailed hereafter with regard to FIG. 3.

FIG. 2 schematically represents an example of hardware architecture of a communication device, as used in the wireless communication system 100. The hardware architecture illustratively shown in FIG. 2 can represent each transmitter 120a, 120b of the wireless communication system 100 and/or each receiver 110a, 110b of the wireless communication system 100.

According to the shown architecture, the communication device comprises the following components interconnected by a communications bus 206: a processor, microprocessor, microcontroller or CPU (Central Processing Unit) 200; a RAM (Random-Access Memory) 201; a ROM (Read-Only Memory) 202; an SD (Secure Digital) card reader 203, or an HDD (Hard Disk Drive) or any other device adapted to read information stored on a storage medium; a first communication interface 204 and potentially a second communication interface 205.

When the communication device is one receiver among the K_r receivers, the first communication interface 204 enables the communication device to receive data from the K_t transmitters via the global MIMO channel H. The second communication interface 205 is not necessary in this case. The first communication interface 204 further enables the communication device to feed back channel state information to one or more transmitter devices among the K_t transmitters.

When the communication device is one transmitter among the K_t transmitters, the first communication interface 204 enables the communication device to transmit data to the K_r receivers, via the global MIMO channel H, cooperatively with the other transmitters among the K_t transmitters. The first communication interface 204 further enables the communication device to receive channel state information fed back by one or more receivers among the K_r receivers. Moreover, the second communication interface 205 enables the communication device to exchange data with one or more other transmitters among the K_t transmitters.

CPU 200 is capable of executing instructions loaded into RAM 201 from ROM 202 or from an external memory, such as an SD card. After the communication device has been powered on, CPU 200 is capable of reading instructions from RAM 201 and executing these instructions. The instructions form one computer program that causes CPU 200 to perform some or all of the steps of the algorithm described herein.

Any and all steps of the algorithms described herein may be implemented in software by execution of a set of instructions or program by a programmable computing machine, such as a PC (Personal Computer), a DSP (Digital Signal Processor) or a microcontroller; or else implemented in hardware by a machine or a dedicated component, such as an FPGA (Field-Programmable Gate Array) or an ASIC (Application-Specific Integrated Circuit).

FIG. 3 schematically represents an algorithm for determining, in a distributed fashion within the wireless communication system 100, estimations of the overall precoder V to be applied for transmitting data from the plurality of transmitters 120a, 120b toward the plurality of receivers 110a, 110b. The algorithm shown in FIG. 3 is independently performed by each transmitter among the K_t transmitters. Let's illustratively consider that the algorithm of FIG. 3 is performed by the transmitter 120a, which is considered as the j-th transmitter among the K_t transmitters.

In a first step S301, the transmitter 120a gathers long-term statistics about the CSIT errors incurred by each one of the K_t transmitters with respect to the global MIMO channel H. The long terms statistics describe the random variation of the CSIT errors, which can for example be the variance of the CSIT errors.

By using a given statistical model of the CSIT errors, for example a centred Gaussian distribution, realizations of CSIT errors can be generated from the gathered long-term statistics for simulating the impact of said CSIT errors. Analytical derivation based on said statistical model and parameterized by said gathered long-term statistics can be performed.

For instance, each j-th transmitter among the K_t transmitters estimates or computes a variance matrix Σ_k,i^(j) associated with the channel estimation error between the MIMO channel estimation Ĥ_k,i^(j) and the effective MIMO channel defined as follows: each coefficient of the variance matrix Σ_k,i^(j) is the variance of the error between the corresponding coefficient of the MIMO channel matrices Ĥ_k,i^(j) and H_k,i. It has to be noted that in this case the channel estimation error is assumed to be independent from one channel coefficient to another. In a variant, a covariance matrix of the vectorization of the difference (on a per-coefficient basis) Ĥ_k,i^(j)−H_k,i between the MIMO channel matrices Ĥ_k,i^(j) and H_k,i is estimated or computed.

When there is no exchange of CSIT information between the transmitters, Ĥ^(j) represents an estimation, by the j-th transmitter among the K_t transmitters, of the global MIMO channel H. Said long term statistics are representative of the error on the CSIT, which can be computed according to a known behaviour divergence of the channel estimation technique in use with respect to the effective considered MIMO channel and according to the effective CSI feedback from the concerned receiver(s) to said j-th transmitter. For example, when pilot symbols are sent in downlink for allowing each k-th receiver among the K_r receivers to estimate the MIMO channel H_k, the resulting estimation error is proportional to the signal to noise plus interference ratio via said MIMO channel H_k, and the corresponding coefficient of proportionality may be computed from the pilot density, such as in the document “Optimum pilot pattern for channel estimation in OFDM systems”, Ji-Woong Choi et al, in IEEE Transactions on Wireless Communications, vol. 4, no. 5, pp. 2083-2088, Sept. 2005. This allows computing statistics relative to the downlink channel estimation error. When channel reciprocity is considered, each j-th transmitter among the K_t transmitters can learn the CSIT from a channel estimation in uplink direction, similar technique as in downlink is used to compute the uplink channel estimation error statistics. When a feedback link is used for obtaining the CSIT at the transmitter from a CSI feedback computed from a channel estimation made by the concerned receiver(s), quantization error statistics on CSI can be estimated in the long term by the concerned receiver(s) and fed back to said j-th transmitter, or said quantization error statistics on CSI can be deduced from analytical models. Indeed, each concerned receiver knows the effective CSI as well as the quantization function, thus the effective quantization error. Said receiver is then able to compute the quantization error statistics over time and is then able to feed them back to said j-th transmitter. For example, the receiver builds an histogram of the quantization error representing the distribution of the quantization error and feeds it back to the j-th transmitter. For example, the receiver and transmitters assume that the quantization error is multivariate Gaussian distributed, and the receiver estimates the mean vector and the covariance variance which are fed back to the j-th transmitter. Any of the above techniques can be combined to obtain said CSIT error statistics associated to the estimation Ĥ^(j) of the global MIMO channel H from the standpoint of the j-th transmitter. Then these long-term statistics can be exchanged between the transmitters, in such a way that each j-th transmitter among the K_t transmitters gathers long-term statistics about the CSIT errors incurred by each one of the K_t transmitters with respect to the global MIMO channel H (which means that all the K_t transmitters share the same long-term statistics).

In another example said long-term statistics are gathered as disclosed in the document “A cooperative channel estimation approach for coordinated multipoint transmission networks”, Qianrui Li et al, IEEE International Conference on Communication Workshop (ICCW), pp. 94-99, 8-12 Jun. 2015, where a combination of channel estimates is performed between transmitter nodes in order to compute the estimation Ĥ_ki^(j) by each j-th transmitter among the K_t transmitters, of the MIMO channel H_k,i, and the combination is then optimized to minimize the mean square error associated with the difference (on a per-coefficient basis) Ĥ_k,i^(j)−H_k,i between the MIMO channel-matrices Ĥ_k,i^(j) and H_k,i. The variance^{matrices Σ}_k,i^(j) ) are thus the result of the combination method described in said document.

Thus, in a particular embodiment, the transmitter 120a gathers the variance matrices Σ_k,i^(j) which entries are the variance of the entries of the estimate error Δ_k,i^(j) between the effective MIMO channel H_k,i and the estimation Ĥ_k,i^(j) of the MIMO channel H_k,i from the standpoint of the j-th transmitter among the K_t transmitters.

Once the step S301 has been performed by each one of the K_t transmitters, all the K_t transmitters share the same long-term statistics about the CSIT errors. The step S301 is performed in an independent process than the process typically in charge of effectively configuring the K_t transmitters so as to transmit the aforementioned set of the symbols vectors s_k.

It can be noted that since the aforementioned statistics about the CSIT errors are by definition long-term data, the latency for ensuring that each one of the K_t transmitters receives said long-term statistics has low importance. Quantization is typically not a limiting factor for transmitting such long-term statistics. On the contrary, the latency for propagating data used by the K_t transmitters so that each transmitter among the K_t transmitters is able to build its own CSIT is of most importance, in order for the wireless communication system 100 to have good reactivity. Quantization with few levels can then be necessary for transmitting such CIST related data and can drastically reduce the quality of the information. By the way, confusion shall be avoided between long-term statistics about CSIT errors received by the transmitter 120a in the step S301 and CSIT related data needed by the transmitter 120a to have its own view Ĥ^(j) of the global MIMO channel H.

In a step S302, the transmitter 120a obtains up-to-date (i.e. short-term) CSIT related data needed by the transmitter 120a to have its own view Ĥ^(j) of the global MIMO channel H. The transmitter 120a preferably shares the CSIT obtained in the step S302 with one or more transmitters among the K_t transmitters, in order to help said one or more transmitters to build their own respective view of the global MIMO channel H.

Once the step S302 is performed by all the K_t transmitters independently (substantially in parallel), the CSIT finally obtained by the K_t transmitters differs from one transmitter to another one among the K_t transmitters, which leads to CSIT mismatch.

In a step S303, the transmitter 120a determines an initial version {tilde over (V)}^(j), from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), of the overall precoder V from the CSIT related data obtained by the transmitter 120a in the step S302. The initial version {tilde over (V)}^(j) of the overall precoder Vis therefore an estimate of the overall precoder V. Since there is CSIT mismatch, this initial version {tilde over (V)}^(j) of the overall precoder V may involve residual interference that grows with the CSIT mismatch.

In a particular embodiment, the type of the overall precoder V and thus of the estimate {tilde over (V)}^(j) of the overall precoder V are both of one precoder type among the followings:

• • block-diagonalization precoders, for coordinated multi-point transmissions with joint processing, as addressed in the document “Cooperative Multi-Cell Block Diagonalization with Per-Base-Station Power Constraints”, Rui Zhang, in IEEE Journal on Selected Areas in Communications, vol. 28, no. 9, pp. 1435-1445, December 2010;
  • interference aware coordinated beamforming precoders, for coordinated multi-point transmissions with coordinated precoding, as addressed in the document “Interference Aware-Coordinated Beamforming in a Multi-Cell System”, Chan-Byoung Chae et al, in IEEE Transactions on Wireless Communications, vol. 11, no. 10, pp. 3692-3703, October 2012; and
  • regularized zero-forcing precoders, for coordinated multi-point transmissions with joint processing, as addressed in the document “A large system analysis of cooperative multicell downlink system with imperfect CSIT”, Jun Zhang et al, in IEEE International Conference on Communications (ICC), pp. 4813-4817, 10-15 Jun. 2012.

A particular embodiment of the present invention for each one of these types of precoders is detailed hereafter.

It has to be noted that the initial version {tilde over (V)}^(j) of the overall precoder V from the CSIT related data obtained by the transmitter 120a (considered as the j-th transmitter among the K_t transmitters) can be determined as indicated in the documents referenced above with respect to each precoder type.

In a step S304, the transmitter 120a refines the initial version {tilde over (V)}^(j) of the overall precoder V according to the CSI error long-term statistics obtained in the step S301, so as to obtain a refined precoder V^(j)=[V₁^(j), . . . ,V_Kr^(j)], which is the view of the overall precoder V from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters).

In a particular embodiment, refining the initial version {tilde over (V)}^(j) of the overall precoder V is performed by the transmitter 120a thanks to a refinement function f(. , .), as well as a set {F_k^(j)} of refinement matrices F_k^(j), k=1 to K_r. More particularly, considering that {tilde over (V)}^(j)=[{tilde over (V)}₁^(j), . . . ,{tilde over (V)}_Kr^(j)], refining the initial version 17(j) of the overall precoder V means refining the sub-matrices {tilde over (V)}^(j), k=1 to K_r, of the initial version {tilde over (V)}^(j) of the overall precoder V, wherein each sub-matrix {tilde over (V)}^(j) is the equivalent within {tilde over (V)}^(j) of the sub-matrix V_k^(j) within V^(j).

Therefore, for each sub-matrix {tilde over (V)}_k^(j), the refinement function f (. , .) and the refinement matrix F_k^(j) can be applied in a multiplicative refinement strategy, such as:

V _k ^(j) =f({tilde over (V)}^(j),F_k^(j))={tilde over (V)}_k^(j)F_k^(j)

or in an additive refinement strategy:

V _k ^(j) =f({tilde over (V)}^(j),F_k^(j))={tilde over (V)}_k^(j)F_k^(j)

preferably under the following power constraint:

Trace((f({tilde over (V)}^(j),F_k^(j)))†f({tilde over (V)}^(j),F_k^(j)))=N

It has to be noticed from the relationships above that the size of each refinement matrix F_k^(j) depends on whether the refinement strategy is additive or multiplicative.

Refining the initial version {tilde over (V)}^(j) of the overall precoder V is further performed as a function of the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), as well as of a figure of merit representative of performance of transmissions from the transmitters to the receivers via the global MIMO channel H, so as to be able to determine optimized version of the set {F_k^(j)} of the refinement matrices F_k^(j), k=1 to K_r. In such a wireless communication system, the figure of merit that is representative of performance of the transmissions via the global MIMO channel H is typically a multi-user performance metric.

It has to be noted that the precoder V^(j) being a refined version of the initial version {tilde over (V)}^(j) of the overall precoder V from the standpoint of each j-th transmitter among the K_t transmitters computed from its own view Ĥ^(j) of the global MIMO channel H, a mismatch exists between the precoders V^(j) independently computed by all the K_t transmitters. Thus, a refinement operation should be designed so as to minimize the impact of the mismatch on the performance characterized by the figure of merit. It can be noted that the transmitters have two types of information for designing the precoder V^(j): first, the local CSIT, which is represented by the view Ĥ^(j) of the global MIMO channel H from the standpoint of each j-th transmitter, and which is exploited to compute the initial version {tilde over (V)}^(j) of the overall precoder V, and the long term statistics on estimate error between the effective global MIMO channel H and said view Ĥ^(j), which are shared between all transmitters and can thus be exploited for said refinement operation. Since a statistics-based refinement is considered, the refinement operation is a statistical method that computes a refined precoder V^(j) out of a set of intermediate random variable {tilde over (V)}^(j) characterizing the possible overall precoder V in view of the previously determined initial version {tilde over (V)}^(j) of the overall precoder V and of the long term statistics on estimate error between the effective global MIMO channel H and said view Ĥ^(j) for each j-th transmitter. Furthermore, the refinement strategy (multiplicative or additive) can be defined in order to be able to statistically correct the initial version {tilde over (V)}^(j) into V^(j), said refinement strategy involving parameters to be optimized so as to statistically reduce the impact of the mismatch on the performance.

Thus, each j-th transmitter can compute the distribution of an intermediate random variable {tilde over (V)}^(j) (as defined hereafter), or generate realizations thereof, associated with the overall precoder V after refinement by all the transmitters, according to the refinement strategy (multiplicative or additive) in use and to the gathered long-term statistics about the CSIT errors, further according to the initial version {tilde over (V)}^(j) of the overall precoder V from the standpoint of said j-th transmitter and of it own view Ĥ^(j) of the global MIMO channel H, as detailed hereafter.

In a first particular embodiment, the figure of merit is a lower bound of a sum rate LBSR^(j) reached via the global MIMO channel H, from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), with respect to its own view Ĥ^(j) of the global MIMO channel H. The sum rate lower bound LBSR^(j) is a function of the set {F_k^(j)}. Considering that the transmitter 120a views the global MIMO channel H as being Ĥ^(j), the sum rate lower bound LBSR^(j) is then defined as follows:

${\mathrm{LBSR}}^{\left(j\right)}=\sum _{k=1}^{K_r}\log {\det \left({\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right)}^{-1}$ $\mathrm{wherein}$ ${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)={}_{\left\{{\Delta }^{\left(1\right)},{\Delta }^{\left(2\right)},\dots ,{\Delta }^{\left(K_t\right)}|{\hat{H}}^{\left(j\right)}\right\}}\left[{\mathrm{MSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right]$

wherein represents the mathematical expectation and, wherein MSE_k^(j)(F₁^(j), . . . ,F_Kr^(j)) represents the mean square error matrix between the symbol vector s_k and the corresponding filtered received vector T_ky_k (as already explained) for a realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT error as obtained in the step S301, for example by considering a centred Gaussian distribution of the CSIT errors, and for the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a. As explained in more details hereafter, computing MSE_k^(j) for all and any k-th receiver among the K_r receivers would allow each considered j-th transmitter among the K_t transmitters to perform the optimization of the considered figure of merit. However, the effective realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) is unknown. An approximation is therefore performed by relying on EMSE_k^(j) instead of MSE_k^(j) thanks to the use of the long-term statistics about the CSIT errors, which have been gathered in the step S301.

The receiver filter T_k may be a function of the overall precoder V and of the global MIMO channel H, which are unknown at the transmitters. But, each j-th transmitter can instead rely on its own view Ĥ^(j) of the global MIMO channel H and realizations of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) matching the long-term statistics gathered in the step S301, so as to obtain the intermediate random variable {tilde over (V)}^(j)) and to obtain its own view T_k^(j) of each receive filter T_k, k=1 to K_r, as described hereafter. As a remark, the intermediate random variable {tilde over (V)}^(j) and the view T_k^(j) of the receive filter T_k are functions of the set tel of the refinement matrices F_k^(j), k=1 to K_r.

Since EMSE_k^(j) (F₁^(j), . . . ,F_K^(j)) may be computed for a fixed set of parameters F₁^(j), . . ,F_Kr^(j), the sum rate lower bound LBSR^(j) may be optimized by randomly defining several candidate sets of matrices to form the set {F_k^(j)} of the refinement matrices F_k^(j), k=1 to K_r, and keeping the candidate set that minimizes the sum rate lower bound LBSR^(j) from the standpoint of the transmitter 120a.

In a preferred embodiment, an optimized sum rate lower bound LBSR^(j) is obtained thanks to an iterative algorithm as detailed hereafter with regard to FIG. 4.

In a second particular embodiment, the figure of merit is the sum of traces, for k=1 to K_r, of EMSE_k^(j)(F₁^(j), . . . ,F_K^(j)), as follows, which leads to a simplified expression MINMSE^(j) thus involving simpler implementation:

${\mathrm{MINMSE}}^{\left(j\right)}=\sum _{k=1}^{K_r}\mathrm{Trace}\left({\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right)$

Therefore, optimization of a figure of merit being a function of EMSE_k^(j), such as the sum rate lower bound LBSR^(j) or the simplified expression MINMSE^(j), according to the mathematical expectation of the realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt), matches the long terms statistics of CSIT error as obtained in the step S301, leads to obtaining the appropriate set {F_k^(j)} of the refinement matrices F_k^{(j), k=}1 to K_r, and then to obtaining the refined precoder V^(j) from said appropriate set {F_k^(j)} of the refinement matrices F_k^(j) (by applying the relevant additive or multiplicative refinement strategy).

In a step S305, the transmitter 120a performs, cooperatively with the other transmitters of the K_t transmitters, the transmission of the set of the symbol vectors s_k toward the K_r receivers. To do so, the transmitter 120a applies the precoder E_j^TV^(j).

The algorithm of FIG. 3 may be applied independently by all the transmitters on a regular basis. The algorithm of FIG. 3 may be applied independently by all the transmitters upon detecting that the global MIMO channel H has changed beyond a predefined threshold. The algorithm of FIG. 3 may be applied independently by all the transmitters before each transmission of data from the K_t transmitters toward the K_r receivers.

Particular embodiment for block-diagonalization precoders

In this particular embodiment, the overall precoder V is a block-diagonalization precoder. It is then assumed that K_tM≥K_rN. By definition of block-diagonalization, considering the k-th receiver among the K_r receivers, the interference induced by the MIMO channels for all other receivers among the K_r receivers is supposed to be eliminated, which means that, ideally:

H_k=0, ∀≠k

Let's denote H_[k] an aggregated interference channel for the k-th receiver among the K_r receivers, which is expressed as follows:

H_[k]=[H₁†, . . . ,H_k−1†,H_k+1†, . . . ,H_Kr†]†

and similarly Ĥ_[k] an aggregated interference channel estimation for the k-th receiver among the K_r receivers, which is expressed as follows:

Ĥ_[k]=[Ĥ₁†, . . . ,Ĥ_k−1†,Ĥ_k+1†, . . . ,Ĥ_Kr†]†

Applying a Singular Value Decomposition (SVD) operation on the expression above of the aggregated interference channel H_[k] results in:

H _[k] =U _[k] [D _[k], 0][V′_[k], V″_k]†

wherein:

• • U_[k] is an N(K_r−1)×N(K_r−1) unitary matrix,
  • D _[k] is an N(K_r−1)×N(K_r−1) diagonal matrix,
  • V′_[k] is an M K_t×N(K_r−1) matrix, and
  • V″_k is an MK_t×MK_t−N(K_r−1) matrix.The size of the part V_k of the overall precoder V which has to be applied for transmitting data toward the k-th receiver is MK_t×N, and V_k is obtained by selecting a predetermined set of N columns of the matrix V″_k. The predetermined set can either be, according to a predefined selection rule, the N first columns of V″_k or the N last columns of V″_k.

Considering the view H(J) of the global MIMO channel H from the standpoint of the j-th transmitter among the K_t transmitters, the expression above becomes:

Ĥ[k] ^(j) U _[k] ^(j) [D _[k] ^(j),0][{tilde over (V)}′_[k]^(j),{tilde over (V)}″_[k]^(j)]†

wherein Ĥ_[k]^(j) represents the view of the aggregated interference channel estimation Ĥ_[k] for the k-th receiver among the K_r receivers from the standpoint of the j-th transmitter among the K_t transmitters,wherein {tilde over (V)}′_[k]^(j) is an MK_t×N(K_r−1) matrix equivalent to V′_[k] when using the estimation Ĥ^(j) instead of the effective global MIMO channel H and {tilde over (V)}″_k^(j) is an MK_t×MK_t−N(K_r−1) matrix equivalent to V″_k when using the estimation Ĥ^(j) instead of the effective global MIMO channel H,and wherein {tilde over (V)}_k^(j) is obtained by selecting a predetermined set of N columns of the matrix {tilde over (V)}″_k^(j) according to the predefined selection rule, the selection rule being similarly applied by any and all transmitters, and wherein {tilde over (V)}_k^(j) is such that the refinement function f (. , .) is used in the aforementioned multiplicative refinement strategy, which means:

V_k^(j)={tilde over (V)}_k^(j)F_k^(j)

where F_k^(j) is a N×N matrix, preferably under the following constraint:

Trace(F_k^(j)F_k^(j))=N

As a result of the precoding strategy, the block-diagonalization property is conserved, which means:

Ĥ_k^(j)V_l^(j)=0, ∀l≠k

However, it has to be noted that the block-diagonalization property is usually not achieved during the transmission on the global MIMO channel H, since a mismatch exists between Ĥ^(j) and H. Thus, if the transmitters use their initial version {tilde over (V)}^(j) of the-precoder for-performing the transmissionsinterference exists between the transmissions towards the receivers. This interference can be reduced by using the statistical knowledge on the long term statistics on estimate error between the effective global MIMO channel H and said view Ĥ^(j), by using the appropriate (multiplicative or additive) refinement strategy.

Therefore, by applying an SVD operation on the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), the matrices {tilde over (V)}′[k_]^(j) and {tilde over (V)}_k^(j) can be determined by the transmitter 120a (considered as the j-th transmitter among the K_t transmitters).

Refining the initial version {tilde over (V)}^(j) of the overall precoder V thus consists in optimizing the set {F_k^(j)} of the refinement matrices F_k^(j) with respect to the set {{tilde over (V)}_k^(j)} of the matrices {tilde over (V)}_k^(j) obtained by the application of the Singular Value Decomposition on the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters). [0093]

First, a system performance metric is derived for a fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) known by the transmitters, and then a statistical analysis on the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt), which are random variables, is applied according to their respective long term statistics gathered at the step S301.

Considering that an MMSE filter is implemented at each one of the K_r receivers for filtering signals received from the K_t transmitters, the expression of the MMSE filter, as computed at the k-th receiver from the perspective of the j-th transmitter and for a fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) is:

$T_k^{\left(j\right)}={\left({\hat{V}}_k^{\left(j\right)}\right)}^{†}{\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)}^{†}{\left(\sum _{=1}^{K_r}\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right){{\hat{V}}_{}^{\left(j\right)}\left({\hat{V}}_{}^{\left(j\right)}\right)}^{†}{\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)}^{†}+I\right)}^{-1}$

wherein {tilde over (V)}_k^(j) is the part of the intermediate random variable {tilde over (V)}^(j) which concerns the k-th receiver among the K_r receivers, by taking into account that the other transmitters among the K_t transmitters also have performed a refinement according to a fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt).

Therefore, each j-th transmitter among the K_t transmitters computes the following, for each and every -th receiver among the K_r (=1 to K_r):

${\hat{V}}_{}^{\left(j\right)}=\left[\begin{array}{c}{E}_1^{†}\left(V_{}^{\left(j\right)}+{\left({\hat{H}}_{\left[\right]}^{\left(j\right)}\right)}^{+}\left({\Delta }_{\left[\right]}^{\left(1\right)}-{\Delta }_{}^{\left(j\right)}\right)V_{}^{\left(j\right)}\right)\\ E_2^{†}\left(V_{}^{\left(j\right)}+{\left({\hat{H}}_{\left[\right]}^{\left(j\right)}\right)}^{+}\left({\Delta }_{\left[\right]}^{\left(2\right)}-{\Delta }_{\left[\right]}^{\left(j\right)}\right)V_{}^{\left(j\right)}\right)\\ ⋮\\ E_{K_t}^{†}\left(V_{}^{\left(j\right)}+{\left({\hat{H}}_{\left[\right]}^{\left(j\right)}\right)}^{+}\left({\Delta }_{\left[\right]}^{\left(K_t\right)}-{\Delta }_{\left[\right]}^{\left(j\right)}\right)V_{}^{\left(j\right)}\right)\end{array}\right]$

and wherein represents the error estimation for the considered -th receiver from the standpoint of the considered j-th transmitter, and

=[Δ₁^(j)†, . . . , , . . . , Δ_Kr^(j)†]†

and represents the view of the MIMO channel from the standpoint of the considered j-th transmitter, and represents an estimation of the aggregated interference channel for the considered receiver from the standpoint of the considered j-th transmitter.

Indeed, it is reminded that:

H=Ĥ ^(j)+Δ^(j)

which then means that, when a fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) is known by the considered j-th transmitter, said j-th transmitter can compute the view Ĥ^(j′) of the global MIMO channel H from the standpoint of any other j′-th transmitter among the K_t transmitters as follows:

Ĥ ^(j′) =Ĥ ^(j)+Δ^(j)−Δ^(j′)

Thus computing ,=1 to K_r, as expressed above, allows then computing T_k^(j), which then allows defining MSE_k^(j) as follows:

$\begin{array}{c}{\mathrm{MSE}}_k^{\left(j\right)}={}_{s_k,n_k}\left[\left(s_k-T_k^{\left(j\right)}y_k\right){\left(s_k-T_k^{\left(j\right)}y_k\right)}^{†}\right]\\ =\left(I-T_k^{\left(j\right)}\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right){\hat{V}}_k^{\left(j\right)}\right){\left(I-T_k^{\left(j\right)}\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right){\hat{V}}_k^{\left(j\right)}\right)}^{†}+\\ {T_k^{\left(j\right)}\left(T_k^{\left(j\right)}\right)}^{†}+\\ T_k^{\left(j\right)}\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)\sum _{\ne k}{{\hat{V}}_{}^{\left(j\right)}\left({\hat{V}}_{}^{\left(j\right)}\right)}^{†}{\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)}^{†}{\left(T_k^{\left(j\right)}\right)}^{†}\end{array}\hspace{1em}$

wherein I is an identity matrix.

Since the effective fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) is unknown at the transmitters in practice, but since the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) are associated with a given probability of occurrence that can be computed from the long term statistics on CSIT error gathered in the step S301, a statistical analysis can be used.

Each j-th transmitter (such as the transmitter 120a) is then able to compute EMSK_k^(j)(F₁^(j), . . . ,F_Kr^(j) by using a Monte Carlo simulation, or by using a numerical integration, on the distribution of Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) in view of the long term statistics on CSIT error gathered in the step S301, further with respect to the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters).

Alternatively a mathematical approximation provides a closed form expression of EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j)), as follows:

$\begin{array}{c}{\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)=(I+{\left({\hat{H}}_k^{\left(j\right)}{\tilde{V}}_k^{\left(j\right)}F_k^{\left(j\right)}\right)}^{†}\times \\ (\sum _{=1,\ne k}^{K_r}{\hat{H}}_k^{\left(j\right)}{\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\hat{H}}_k^{\left(j\right)}{\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}+\\ {I+{\Phi }_k^{\left(j\right)})}^{-1}\\ {{\hat{H}}_k^{\left(j\right)}{\tilde{V}}_k^{\left(j\right)}F_k^{\left(j\right)})}^{-1}\end{array}$ $\mathrm{wherein}$ ${\Phi }_k^{\left(j\right)}=\sum _{=1}^{K_r}\sum _{\begin{array}{c}t=1\\ t\ne j\end{array}}^{K_t}{\hat{H}}_k^{\left(j\right)}E_tE_t^{†}{\hat{H}}_{\left[\right]}^{\left(j\right)+}\left[{{\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}E_tE_t^{†}+{{\Theta }_{\left[\right]}^{\left(j\right)\left(j\right)}\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}\sum _{\begin{array}{c}{t}^{\prime }=1\\ {t}^{\prime }\ne j\end{array}}^{K_t}E_{t^{\prime }}E_{t^{\prime }}^{†}\right]{\hat{H}}_k^{\left(j\right)†}+\mathrm{mdiag}\left(\sum _{t=1,t\ne j}^{K_t}{\Sigma }_k^{\left(t\right)}E_t\mathrm{diag}\left(\sum _{i=1}^{K_r}E_t^{†}{\hat{H}}_{\left[\right]}^{\left(j\right)+}\left({\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}+{\Theta }_{\left[\right]}^{\left(j\right)\left(j\right)}\right){\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}E_t\right)\right)+\mathrm{mdiag}\left({\Sigma }_k^{\left(j\right)}\mathrm{diag}\left(\sum _{=1}^{K_r}{\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}\right)\right)$ $\mathrm{wherein}$ ${\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}=\mathrm{mdiag}\left({\Sigma }_{\left[\right]}^{\left(t\right)}\mathrm{diag}\left({\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}\right)\right)$ $\mathrm{and}$ ${\Sigma }_k^{\left(j\right)}=\left[{\Sigma }_{k,1}^{\left(j\right)},\dots ,{\Sigma }_{k,K_t}^{\left(j\right)}\right]$ $\mathrm{and}$ ${\Sigma }_{\left[k\right]}^{\left(t\right)}={\left[{\left({\Sigma }_1^{\left(t\right)}\right)}^{†},\dots ,{\left({\Sigma }_{k-1}^{\left(t\right)}\right)}^{†},{\left({\Sigma }_{k+1}^{\left(t\right)}\right)}^{†},\dots ,{\left({\Sigma }_{K_r}^{\left(t\right)}\right)}^{†}\right]}^{†}$

and wherein A⁺ is the Moore-Penrose pseudo-inverse of A, mdiag (.) makes a diagonal matrix from a given vector and diag (.) retrieves the diagonal entries of a matrix and stacks them into a vector.

Therefore, optimization of a figure of merit being a function of EMSE_k^(j), such as the sum rate lower bound LBSRM or the simplified expression MINMSE^(j), according to the mathematical expectation of the realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long-term statistics of CSIT error as obtained in the step S301, leads to obtaining the appropriate set {F_k^(j)} of the refinement matrices F_k^(j), k=1 to K_r, and then to obtaining the refined precoder V^(j) from said appropriate set {F_k^(j)} of the refinement matrices F_k^(j) by applying the aforementioned multiplicative refinement strategy.

In a preferred embodiment, an optimized sum rate lower bound LBSRU) is obtained thanks to the iterative algorithm as detailed hereafter with regard to FIG. 4.

Particular embodiment for interference aware coordinated beamforming precoders

In this particular embodiment, it is assumed that the quantity K_t of transmitters is equal to the quantity K_r of receivers, i.e. K_t=K_r, and that the quantity M of transmit antennas is equal to the quantity N of receive antennas, i.e. M=N. Moreover each one of the K_t transmitters communicates only with a single receiver among the K_r. receivers. Considering any k-th receiver among the K_r receivers, the j-th transmitter among the K_t transmitters which communicates with said k-th receiver is such that k=j. In the mathematical expressions used hereafter in the particular embodiment for interference aware coordinated beamforming precoders, the index k (only used hereinbefore for identifying any receiver among the K_r receivers) can be used instead of the index j. Interference aware coordinated beamforming precoding is a sub-case of the block-diagonalization precoding detailed above. Indeed, by considering that the overall precoder V has a block diagonal structure, with K_t blocks of size M, it is considered that each and every k-th transmitter among the K_t transmitters only knows the symbol vector s_k (and not the other symbol vectors s_l, l≠k, that have to be transmitted by the other transmitters among the K_t transmitters), which is precoded by an M×M sub-matrix W′_k such that:

V_k=E_kW′_k

The sub-matrices W′_k, for k=1 to K_r, are obtained by implementing beamforming and/or interference alignment based on the view Ĥ^(k) of the global MIMO channel H from the standpoint of each and every k-th transmitter. For example, the sub-matrices W′_k are computed according to an interference alignment technique, as in the document “Downlink Interference Alignment” Changho Suh et al, IEEE Transactions on Communications, vol. 59, no. 9, pp. 2616-2626, September 2011. In another example, the sub-matrices W′_k are computed as the eigenvector beamforming of the channel matrix defined by E_k^TĤ^(k)E_k, from an SVD operation applied onto said channel matrix by the considered k-th transmitter among the K_t transmitters, such that:

E_k^TĤ^(k)E_k=U′_kD′_kW′_k

wherein:

• • U′_k is an NK_r×NK_r unitary matrix, and
  • D′_k is an NK_r×NK_r diagonal matrix.

The optimization is then very similar to the approach described above with respect to the block-diagonalization precoding.

Then, the approach described above with respect to the block-diagonalization precoding can thus be similarly applied, as follows.

First, an initial version {tilde over (V)}^(j) of the overall precoder V from the standpoint of each j-th transmitter among the K_t transmitters is computed from its own view Ĥ^(j) of the global MIMO channel H, such that the overall precoder V and the initial version {tilde over (V)}^(j) thereof have a block diagonal structure. Each block defined by E_k^T{tilde over (V)}^(j)E_k is related to the precoder used at the k-th transmitter from the standpoint of each j-th transmitter, only for precoding the symbols vector s_k, and is related to the sub-matrice W′_k previously described and determined according to an interference alignment or an the eigenvector beamforming technique.

The intial version {tilde over (V)}_k^(j) is such that the refinement function f (. , .) is used in the aforementioned multiplicative refinement strategy, which means:

V_k^(j)={tilde over (V)}_k^(j)F_k^(j)

where F_k^(j) is a N×N matrix, preferably under the following constraint:

Trace(F_k^(j)†F_k^(j))=N

Each j-th transmitter (such as the transmitter 120a) is then able to compute EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j))) by using a Monte Carlo simulation, or by using a numerical integration, on the distribution of Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT error as obtained in the step S301, further with respect to the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), under the constraint that the refinement matrices F_k^(j) show block-diagonal respective shapes, by using

${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)={\left\{{\Delta }^{\left(1\right)},{\Delta }^{\left(2\right)},\dots ,{\Delta }^{\left(K_t\right)}{\hat{H}}^{\left(j\right)}\right\}}_{}\left[{\mathrm{MSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)\right]$

Alternatively a mathematical approximation provides a closed form expression of EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j)), as follows:

$\begin{array}{c}{\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)=(I+{\left({\hat{H}}_k^{\left(j\right)}{\tilde{V}}_k^{\left(j\right)}F_k^{\left(j\right)}\right)}^{†}\times \\ (\sum _{=1,\ne k}^{K_r}{\hat{H}}_k^{\left(j\right)}{\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\hat{H}}_k^{\left(j\right)}{\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}+\\ {I+{\Phi }_k^{\left(j\right)})}^{-1}\\ {{\hat{H}}_k^{\left(j\right)}{\tilde{V}}_k^{\left(j\right)}F_k^{\left(j\right)})}^{-1}\end{array}$ $\mathrm{wherein}$ ${\Phi }_k^{\left(j\right)}=\sum _{=1}^{K_r}\sum _{\begin{array}{c}t=1\\ t\ne j\end{array}}^{K_t}{\hat{H}}_k^{\left(j\right)}E_tE_t^{†}{\hat{H}}_{\left[\right]}^{\left(j\right)+}\left[{{\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}E_tE_t^{†}+{{\Theta }_{\left[\right]}^{\left(j\right)\left(j\right)}\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}\sum _{\begin{array}{c}{t}^{\prime }=1\\ {t}^{\prime }\ne j\end{array}}^{K_t}E_{t^{\prime }}E_{t^{\prime }}^{†}\right]{\hat{H}}_k^{\left(j\right)†}+\mathrm{mdiag}\left(\sum _{t=1,t\ne j}^{K_t}{\Sigma }_k^{\left(t\right)}E_t^{†}\mathrm{diag}\left(\sum _{i=1}^{K_r}E_t^{†}{\hat{H}}_{\left[\right]}^{\left(j\right)+}\left({\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}+{\Theta }_{\left[\right]}^{\left(j\right)\left(j\right)}\right){\left({\hat{H}}_{\left[\right]}^{\left(j\right)+}\right)}^{†}E_t\right)\right)+\mathrm{mdiag}\left({\Sigma }_k^{\left(j\right)}\mathrm{diag}\left(\sum _{=1}^{K_r}{\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}\right)\right)$ $\mathrm{wherein}$ ${\Theta }_{\left[\right]}^{\left(j\right)\left(t\right)}=\mathrm{mdiag}\left({\Sigma }_{\left[\right]}^{\left(t\right)}\mathrm{diag}\left({\tilde{V}}_{}^{\left(j\right)}{F_{}^{\left(j\right)}\left({\tilde{V}}_{}^{\left(j\right)}F_{}^{\left(j\right)}\right)}^{†}\right)\right)$ $\mathrm{and}$ ${\Sigma }_k^{\left(j\right)}=\left[{\Sigma }_{k,1}^{\left(j\right)},\dots ,{\Sigma }_{k,K_t}^{\left(j\right)}\right]$ $\mathrm{and}$ ${\Sigma }_{\left[k\right]}^{\left(t\right)}={\left[{\left({\Sigma }_1^{\left(t\right)}\right)}^{†},\dots ,{\left({\Sigma }_{k-1}^{\left(t\right)}\right)}^{†},{\left({\Sigma }_{k+1}^{\left(t\right)}\right)}^{†},\dots ,{\left({\Sigma }_{K_r}^{\left(t\right)}\right)}^{†}\right]}^{†}$

Therefore, optimization of a figure of merit being a function of EMSE_k^(j), such as the sum rate lower bound LBSRW or M/NMSEW, according to the mathematical expectation of the realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long-term statistics of CSIT error as obtained in the step S301, leads to obtaining the appropriate set {F_k^(j)} of the refinement matrices F^{U), k=}1 to K_r, and then to obtaining the refined precoder V^(j) from said appropriate set {F_k^(j)} of the refinement matrices F^(j) by applying the aforementioned multiplicative refinement strategy.

In a preferred embodiment, an optimized sum rate lower bound LBSR^(j) is obtained thanks to the iterative algorithm detailed hereafter with regard to FIG. 4.

Particular embodiment for regularized zero-forcing precoders

In this particular embodiment, the estimate {tilde over (V)}^(j) of the overall precoder V from the standpoint of each and every j-th transmitter among the K_t transmitters can be expressed as follows, with respect to each and every k-th receiver among the K_r receivers:

{tilde over (V)} _k ^(j)=(Ĥ^(j)†Ĥ^(j)+α^(j)1)⁻¹Ĥ_k^(j)†

wherein α^(j) is a scalar representing a regularization coefficient allowing to take into account a balance between interference and useful signal after channel inversion, and allowing optimizing the Signal-to-Interference-plus-Noise Ratio (SINR), and wherein α^(j) is optimized according to statistics of the view Ĥ^(j) of the global MIMO channel H from the standpoint of the considered j-th transmitter among the K_t transmitters, and wherein α^(j) is shared by said j-th transmitter with the other transmitters among the k transmitters, and wherein α^(j) is obtained for example as in the document “Regularized Zero-Forcing for Multiantenna Broadcast Channels with User Selection”, Z. Wang et al, in IEEE Wireless Communications Letters, vol. 1, no. 2, pp. 129-132, April 2012, and wherein {tilde over (V)}_k^(j) is such that the refinement function f (. , .) is used in the aforementioned additive refinement strategy, which means:

V _k ^(j) i ={tilde over (V)} ^(j) +F _k ^(j)

wherein F_k^(j) is a MK_t×N matrix.

First, a system performance metric is derived for a fixed realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) known by the transmitters, and then a statistical analysis on the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt), which are random variables, is applied according to their respective long term statistics gathered at the step S301.

{tilde over (V)}_k^(j) can be computed for a fixed realization of Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) and with respect to the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters), as follows:

$\begin{array}{c}{\hat{V}}_k^{\left(j\right)}=\left[\begin{array}{c}{E}_1^{†}V_k^{\left(1\right)}\\ ⋮\\ E_{K_t}^{†}V_k^{\left(K_t\right)}\end{array}\right]\\ =\left[\begin{array}{c}{E}_1^{†}\left({\tilde{V}}_k^{\left(1\right)}+F_k^{\left(1\right)}\right)\\ ⋮\\ E_{K_t}^{†}\left({\tilde{V}}_k^{\left(K_t\right)}+F_k^{\left(K_t\right)}\right)\end{array}\right]\\ =\left[\begin{array}{c}{E}_1^{†}\left({\tilde{V}}_k^{\left(j\right)}+C^{(j{)}^{-1}}\left({\Delta }^{\left(1\right)}-{\Delta }^{\left(j\right)}\right)-2C^{{\left(j\right)}^{-1}}·\mathrm{Re}\left(({\Delta }^{\left(1\right)}-{\Delta }^{\left(j\right)}{)}^{†}{\hat{H}}^{\left(j\right)}\right)·{\tilde{V}}_k^{\left(j\right)}+F_k^{\left(1\right)}\right)\\ E_{K_t}^{†}\left({\tilde{V}}_k^{\left(j\right)}+C^{(j{)}^{-1}}\left({\Delta }^{\left(1\right)}-{\Delta }^{\left(j\right)}\right)-2C^{{\left(j\right)}^{-1}}·\mathrm{Re}\left(({\Delta }^{\left(1\right)}-{\Delta }^{\left(j\right)}{)}^{†}{\hat{H}}^{\left(j\right)}\right)·{\tilde{V}}_k^{\left(j\right)}+F_k^{\left(K_t\right)}\right)\end{array}\right]\end{array}\hspace{1em}$

wherein Re{X} represents the real part of the complex input X,and wherein

C ^{(j) −1} =(Ĥ^(j)†Ĥ^(j)+α^(j)1)⁻¹

which then allows defining MSE_k^(j) as follows:

${\mathrm{MSE}}_k^{\left(j\right)}=\frac{1+{\sum }_{\ne k}{{\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)}^{†}{\hat{V}}_{}^{\left(j\right)}}^2}{1+{\sum }_{}{{\left({\hat{H}}_k^{\left(j\right)}+{\Delta }_k^{\left(j\right)}\right)}^{†}{\hat{V}}_{}^{\left(j\right)}}^2}$

It is reminded that EMSE_k^(j) is defined as follows:

${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)={\left\{{\Delta }^{\left(1\right)},{\Delta }^{\left(2\right)},\dots ,{\Delta }^{\left(K_t\right)}{\hat{H}}^{\left(j\right)}\right\}}_{}\left[{\mathrm{MSE}}_k^{\left(j\right)}\right]$

The transmitter 120a is then able to compute

${\mathrm{EMSE}}_k^{\left(j\right)}\left(F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}\right)$

by using a Monte Carlo simulation, or by using a numerical integration, on the distribution of Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT error as obtained in the step S301, further with respect to the view Ĥ^(j) of the global MIMO channel H from the standpoint of the transmitter 120a (considered as the j-th transmitter among the K_t transmitters).

Therefore, optimization of a figure of merit being a function of EMSE_k^(j), such as the sum rate lower bound LBSR^(j) or MINMSE_k^(j), according to the mathematical expectation of the realization of the estimate errors Δ⁽¹⁾, Δ⁽²⁾, . . . , Δ^(Kt) which matches the long terms statistics of CSIT error as obtained in the step S301, leads to obtaining the appropriate set {F_k^(j)} of the refinement matrices F_k^{(j), k=}1 to K_r, and then to obtaining the refined precoder VU) from said appropriate {F_k^(j)} of the refinement matrices F_k^(j) by applying the aforementioned additive refinement strategy.

In a preferred embodiment, an optimized sum rate lower bound LBSR^(j) is obtained thanks to the iterative algorithm as detailed hereafter with regard to FIG. 4.

FIG. 4 schematically represents an iterative algorithm for determining the refinement matrices F_k^(j) from an optimization of LBSR(j) and thanks to the above descriptions on how to compute EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j)). The algorithm of FIG. 4 is performed by each and every j-th transmitter among the K_t transmitters. Let's illustratively consider that the algorithm of FIG. 4 is performed by the transmitter 120a, considered as the j-th transmitter among the K_t transmitters.

It is considered, when starting executing the algorithm of FIG. 4, that the transmitter 120a knows the matrices {tilde over (V)}_k^(j), Σ_k^(j) and Ĥ_k^(j) for any and all k-th receivers among the K_r receivers.

In a step S401, the transmitter 120a initializes the refinement matrices F_k^(j), for each and every k-th receiver among the K_r receivers. The initialization can be set as random under the following constraint:

Trace((f({tilde over (V)}_k^(j),F_k^(j)))†f({tilde over (V)}_k^(j),F_k^(j)))=N

Alternatively, the refinement matrices F_k^(j) are taken as identity N×N matrices for the block diagonal case, and MK_t×N matrices containing only zeros for the regularized zero forcing case.

In a following step S402, the transmitter 120a computes B_k^(j), for each and every k-th receiver among the K_r receivers, such that:

B _k ^(j)=EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j))

In a following step S403, the transmitter 120a adjusts the refinement matrices F_k^(j), for each and every k-th receiver among the K_r receivers, as follows:

$F_1^{\left(j\right)},\dots ,F_{K_r}^{\left(j\right)}=\underset{A_1,\dots ,A_{K_r}}{\arg \min }\sum _{i=1}^{K_r}\mathrm{Trace}\left(B_k^{\left(j\right)}·{\mathrm{EMSE}}_k^{\left(j\right)}\left(A_1,\dots ,A_{K_r}\right)\right)$

such that the following constraint is preferably met:

Trace((f({tilde over (V)}_k^(j),F_k^(j)))†f({tilde over (V)}_k^(j),F_k^(j)))=N

In a following step S404, the transmitter 120a checks whether convergence has been reached with respect to F_i^(j), . . . ,F_Kr^(j), for each and every k-th receiver among the K_r receivers. If such convergence has been reached, a step S405 is performed in which the algorithm of FIG. 4 ends; otherwise, the step 5402 is repeated, in which B_k^(j) is updated thanks to the values of F₁^(j), . . . ,F_Kr^(j) obtained in the last occurrence of the step S403.

The optimization of MINMSE^(j) can also be done in order to determine the refinement matrices F_k^(j) thanks to the above descriptions on how to compute EMSE_k^(j)(F₁^(j), . . . ,F_Kr^(j)). This leads to a convex optimization problem.

Claims

1-12. (canceled)

13. A method for performing transmissions of data between a plurality of Kt transmitters and a plurality of Kr receivers via a global MIMO channel H=[H1, . . . ,HKr] of a wireless communication system, by determining in a distributed fashion precoders to be applied for performing said transmissions, said precoders being respectively applied by said transmitters and jointly forming an overall precoder V, wherein each and every j-th transmitter among said plurality of Kt transmitters performs: obtaining short-term CSIT related data and building its own view flu) of the global MIMO channel H;determining an estimate {tilde over (V)}(j) of the overall precoder V from the obtained short-term CSIT related data;wherein said j-th transmitter further performs:gathering long-term statistics of Channel State Information at Transmitter CSIT errors incurred by each one of the Kt transmitters with respect to the global MIMO channel H, the long-term statistics describing the random variation of the CSIT errors;refining the estimate {tilde over (V)}(j)=[{tilde over (V)}1(j), . . . ,{tilde over (V)}Kr(j)] of the overall precoder V on the basis of the gathered long-term statistics of CSIT errors so as to obtain a refined precoder {tilde over (V)}(j)=[{tilde over (V)}1(j), . . . ,{tilde over (V)}Kr(j)] that is a view of the overall precoder V from the standpoint of said j-th transmitter, further on the basis of its own view Ĥ(j) of the global MIMO channel H, and further on the basis of a figure of merit representative of performance of said transmissions via the global MIMO channel H; andtransmitting the data by applying a precoder that is formed by a part of the refined precoder V(j) which relates to said j-th transmitter among said plurality of Kt transmitters.

14. The method according to claim 13, wherein the figure of merit is a lower bound of a sum rate LBSR(j) reached via the global MIMO channel H, from the standpoint of said j-th transmitter with respect to its own view Ĥ(j) of the global MIMO channel H, as follows:

15. The method according to claim 13, wherein the figure of merit is the sum of traces M/NMSE(j), for k=1 to Kr, of MESEk(j)(F1(j), . . . ,FKr(j)), as follows:

16. The method according to claim 13, wherein refining the estimate {tilde over (V)}(j) of the overall precoder V is performed thanks to a refinement function f (. , .), as well as a set {Fk(j)} of refinement matrices Fk(j), k=1 to Kr, in a multiplicative refinement strategy, as follows: Vk(j)=f({tilde over (V)}k(j),Fk(j))={tilde over (V)}k(j)Fk(j)

17. The method according to claim 16, wherein the overall precoder V is a block-diagonalization precoder, the transmitters have cumulatively at least as many antennas as the receivers, and in that refining the estimate {tilde over (V)}(j) of the overall precoder V thus consists in optimizing the set {Fk(j)} of the refinement matrices Fk(j) with respect to the set {{tilde over (V)}k(j)} of the matrices {tilde over (V)}k(j), which is obtained by applying a Singular Value Decomposition operation as follows: Ĥ[k](j)=U[k](j)[D[k](j), 0][{tilde over (V)}′[k](j),{tilde over (V)}″k(j)]†

18. The method according to claim 16, wherein the overall precoder V is an interference aware coordinated beamforming precoder with block-diagonal shape, Kt=Kr, and each transmitter has as a quantity M of transmit antennas equal to a quantity N of receive antennas of each receiver, each transmitter communicates only with a single receiver among the Kr receivers such that k=j, wherein a sub-matrix W′k such that: Vk=EkW′k

19. The method according to claim 13, wherein refining the estimate {tilde over (V)}(j) of the overall precoder V is performed thanks to a refinement function f(. , .), as well as a set {Fk(j)} of refinement matrices Fk(j), k=1 to Kr, in an additive refinement strategy, as follows: Vk(j)=f({tilde over (V)}(j),Fk(j))={tilde over (V)}(j)+Fk(j)

20. The method according to claim 19, wherein the overall precoder V is a regularized zero-forcing precoder, and the estimate {tilde over (V)}(j) of the overall precoder V can be expressed as follows: {tilde over (V)}(j)=(Ĥ(j)†Ĥ(j)+α(j)l)−1Ĥk(j)†wherein α(j) is a scalar representing a regularization coefficient that is optimized according to statistics of the own view Ĥ(j) of the global MIMO channel H from the standpoint of said j-th transmitter, and wherein α(j) is shared by said j-th transmitter with the other transmitters among the Kt transmitters.

21. The method according to claim 13, wherein refining the estimate {tilde over (V)}(j) of the overall precoder Vis perfoinied under the following power constraint: Trace((f({tilde over (V)}(j),Fk(j)))†f({tilde over (V)}(j),Fk(j)))=N wherein f (. , .) is a refinement function and Fk(j), k=1 to Kr are refinement matrices, and wherein each receiver has a quantity N of receive antennas.

22. A non-transitory computer readable storage medium, wherein it stores a computer program comprising program code instructions which can be loaded in a programmable device for implementing the method according to claim 13, when the program code instructions are run by the programmable device.

23. A device for performing transmissions of data between a plurality of Kt transmitters and a plurality of Kr receivers via a global MIMO channel H=[H1, . . . ,HKr] of a wireless communication system, by determining in a distributed fashion precoders to be applied for performing said transmissions, said precoders being respectively applied by said transmitters and jointly forming an overall precoder V, wherein said device is each and every j-th transmitter among said plurality of Kt transmitters and comprises a processor configured to: obtain short-term CSIT related data and building its own view {tilde over (H)}(j) of the global MIMO channel H;determine an estimate {tilde over (V)}(j) of the overall precoder V from the obtained short-term CSIT related data;