WO2017116292A1 - Method and receiving node for detecting signals transmitted by multiple users - Google Patents

Abstract

Method and receiving node (302) for detecting signals transmitted by multiple users (300) in data streams on a common radio resource where Multiple-Input- Multiple-Output, MIMO, communication is employed. When the signals S _k transmitted by the respective users k are received (302A) in a combined signal y, a combined channel matrix H _cc comprising an individual channel matrix H _k for each respective user k is obtained. A combined parallelizing matrix Q _P comprising an individual parallelizing matrix Q _k Of each respective user k, is further determined such that the individual parallelizing matrix Q _k of each user k forms an orthogonal space for the other users. An effective received signal y _eff-k is then determined for each user k by applying (302B) the individual parallelizing matrix Q _k of said user k on the combined received signal y. An effective channel matrix H _eff-k is also determined for each user k based on the effective received signal y _eff-k of said user k, and the signal s _k transmitted from each user k is estimated (302C) based on the effective received signal y _eff-k of said user k and the effective channel matrix H _eff-k of said user k.

Description

METHOD AND RECEIVING NODE FOR DETECTING SIGNALS TRANSMITTED

BY MULTIPLE USERS

Technical field

The present disclosure relates generally to a method and a receiving node of a radio network, for detecting signals transmitted by multiple users over a radio interface where Multiple-Input- Multiple-Output, MIMO, communication is employed.

Background

For some years, different types of radio networks for wireless communication have been developed to provide radio access for various wireless devices. The radio networks are constantly improved to provide better coverage and capacity and to meet the demands from subscribers using increasingly advanced services and equipment such as smartphones and tablets, which may require considerable amounts of bandwidth and resources for data transport in the networks. A limiting factor for capacity of a radio network is the amount of available radio resources, e.g. in terms of time, frequency bandwidth and transmit power, and the capacity of a radio network can be improved by more efficient usage of such radio resources.

In this disclosure, the term "user" is used for short to represent any communication entity, commonly referred to as a "User Equipment, UE", which is capable of radio communication with a radio network by sending and receiving radio signals. In this context, a user may be, e.g., a mobile telephone, tablet, laptop computer or Machine-to-Machine, M2M, device. Throughout this disclosure, UE or wireless device could alternatively be used as a synonym for user.

Further, the term "receiving node", is used herein to represent any node that is operative to receive and detect radio signals transmitted by users in a radio network. The receiving node may e.g. control a network entity having radio equipment for receiving such radio signals. The receiving node in this disclosure could also be denoted base station, network node, radio node, e-NodeB, eNB, NB, base transceiver station, access point, etc., depending on the type of network and terminology used. The receiving node in this disclosure is typically a base station or similar but it could also be another user, UE or wireless device, and the disclosure is thus not limited to network nodes belonging to a radio network.

One way of increasing capacity in a network is to employ so-called Multiple-lnput- Multiple-Output, MIMO, communication which means that a user transmits signals from at least two antennas and the receiving node uses multiple antennas for receiving signals. This feature can increase the data throughput by transmitting one data stream from each transmit antenna at the user, thus allowing for multiple simultaneous data streams from the same user. A technique has also been developed for simultaneously receiving and detecting data streams transmitted by more than one user at the same time on the same shared radio resource in terms of time and frequency, referred to as Multi-User MIMO, MU-MIMO, which has been introduced in various standards for telecommunication including Long Term Evolution, LTE, and IEEE802.1 1 . When MU-MIMO is used, the utilization of radio resources can thus be improved considerably. A simplified but illustrative communication scenario where MU-MIMO is employed is shown in Fig. 1 where multiple users 100 transmit at least two data streams from at least two transmit antennas 100A and 100B on a shared radio resource, i.e. more or less simultaneously. A receiving node 1 02 receives all the transmitted data streams by means of multiple receive antennas 102A such that each receive antenna receives all data streams as a combined or total signal y. The users 100 transmit from different locations such that the signals may be received with different power and/or timing at the receiving node 102.

When such multiple signals transmitted by two or more users are received, the receiving node 102 is required to detect and decode the individual signals jointly from the combined or total signal y, referred to as Multiuser Detection, MUD. It is thus a challenge to achieve proper detection and decoding of all signals in a MU- MIMO scenario without too many errors which is typically measured as Bit Error Rate, BER. If the BER is higher than tolerable, extensive retransmission of erroneously detected data may be necessary which effectively reduces or even eliminates the gains in throughput when using MU-MIMO. Techniques have also been developed for achieving simultaneous detection and decoding of signals from multiple users. For example, a maximum-likelihood, ML, detector can be used in a conventional manner to achieve acceptable BER performance. However, the computational complexity of this conventional ML detector is large and increases exponentially with increased number of streams and receiving antennas, thus making the above technique expensive to

implement, in particular for a receiving node having a large antenna array. A linear receiver, particularly a Zero Forcing, ZF, receiver, can be employed instead with less complexity than the ML detector. This receiver employs a ZF detection filter which is designed to eliminate interference between the users' signals.

To enhance the performance for MUD, Successive Interference Cancellation, SIC, technologies may also be used. The Zero Forcing Successive Interference Cancellation, ZF-SIC, receiver is such an advanced nonlinear receiver which can achieve high spectral efficiency with reasonable decoding complexity. However, it is a problem that when a traditional linear ZF receiver is used to eliminate the interference between the users' signals, the impact of noise may become harmful and degrade the signal detection particularly when the number of antennas and data streams is increased such that the noise power is gradually amplified which results in a degraded system performance. Thus, the conventional ZF receiver cannot provide a good performance when a large number of users are active in uplink MIMO, in short UL-MIMO.

A ZF-SIC receiver operates such that when a detected symbol is assumed to be correct, the ZF-SIC receiver successively subtracts the interference contributed by previous streams from the current stream. This process proceeds to next symbol until all symbols are detected. It is a problem with this technique that to decode the k-th user, the streams from all the previous k-1 users must have been correctly decoded to perform the present decoding. Any decoding error made would propagate user by user if one or more previous streams are decoded incorrectly. This leads to degraded system performance and the successive decoding in the ZF-SIC receiver also generates considerable delays when the number of users is high. Summary

It is an object of embodiments described herein to address at least some of the problems and issues outlined above. It is possible to achieve this object and others by using a method and a receiving node as defined in the attached independent claims.

According to one aspect, a method is performed by a receiving node for detecting signals transmitted by multiple users in data streams over a radio interface where Multiple-Input- Multiple-Output, MIMO, communication is employed. In this method, the receiving node receives a combined signal y comprising the signals Sk transmitted by the respective users k, where k=1,..., K, wherein each user transmits one or more data streams on a common radio resource. The receiving node further obtains a combined channel matrix H_cc comprising multiple individual channel matrices where each individual channel matrix /-/_/( Corresponds to a respective user k. The receiving node then determines a combined parallelizing matrix Qp comprising multiple individual parallelizing matrices where each individual parallelizing matrix (_¾ corresponds to a respective user k, such that the individual parallelizing matrix (_¾ of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix Q^ are unitary matrices. The receiving node further determines an effective received signal y_e -k tor each user / by applying the individual parallelizing matrix (_¾ of said user /con the combined received signal y. The receiving node also determines an effective channel matrix H_eff-k tor each user k based on the effective received signal y_e -k O said user k, and estimates, for each user k, the signal Sk transmitted from said user k based on the effective received signal y_ett-k of said user / and the effective channel matrix _en-k of said user k.

According to another aspect, a receiving node is arranged to detect signals transmitted by multiple users in data streams over a radio interface where Multiple- Input- Multiple-Output, MIMO, communication is employed. The receiving node is configured to receive a combined signal y comprising the signals Sk transmitted by the respective users k, where k=1,..., K, wherein each user transmits one or more data streams on a common radio resource. The receiving node is further configured to obtain a combined channel matrix H_cc comprising multiple individual channel matrices where each individual channel matrix /-/_/( Corresponds to a respective user k.

The receiving node is also configured to determine a combined parallelizing matrix QP comprising multiple individual parallelizing matrices where each individual parallelizing matrix (_¾ corresponds to a respective user k, such that the individual parallelizing matrix Q/c Of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix (_¾ are unitary matrices.

The receiving node is further configured to determine an effective received signal e_ff. ( for each user k by applying the individual parallelizing matrix (_¾ of said user / on the combined received signal y. The receiving node is further configured to determine an effective channel matrix H_eff.k tor each user k based on the effective received signal y_ett-k of said user k, and to estimate, for each user k, the signal Sk transmitted from said user k based on the effective received signal y_en-k said user / and the effective channel matrix H_eff.k of said user k.

The above method and receiving node may be configured and implemented according to different optional embodiments to accomplish further features and benefits, to be described below.

A computer program is also provided which comprises instructions which, when executed on at least one processor, cause the at least one processor to carry out the method described above for the receiving node. A carrier containing the above computer program is further provided, wherein the carrier is one of an electronic signal, optical signal, radio signal, or computer readable storage medium.

Brief description of drawings

The solution will now be described in more detail by means of exemplary embodiments and with reference to the accompanying drawings, in which:

Fig. 1 is a communication scenario illustrating a communication scenario where uplink MIMO is employed for multiple users connected to a network node, according to the prior art.

Fig. 2 is a flow chart illustrating a procedure in a receiving node, according to some possible embodiments.

Fig. 3 is a communication scenario illustrating an example of how the solution may be employed, according to further possible embodiments.

Fig. 4 is a flow chart illustrating an example of a procedure when parallel Minimum Mean-Square Error, MMSE, detection is used, according to further possible embodiments.

Fig. 5 is a flow chart illustrating an example of a procedure where Sorted LQ Decomposition, SLQD, is used, according to further possible embodiments.

Fig. 6 is a flow chart illustrating an example of a procedure for performing parallel SLQD, according to further possible embodiments. Fig. 7 is a block diagram illustrating a receiving node in more detail, according to further possible embodiments.

Fig. 8 is a diagram illustrating performance in terms of Bit Error Rate, BER, when the solution is employed as compared to conventional techniques.

Fig. 9 is another diagram illustrating performance in terms of BER when the solution is employed as compared to conventional techniques. Fig. 10 is another diagram illustrating performance in terms of BER when the solution is employed as compared to conventional techniques.

Fig. 1 1 is another diagram illustrating performance in terms of BER when the solution is employed as compared to conventional techniques. Detailed description

Briefly described, a solution is provided to enable reduced complexity and improved accuracy when detecting signals transmitted in data streams on a common radio resource from different users employing MIMO communication. This can be achieved by performing estimation of each individually transmitted signal independently of the signals transmitted from the other users, by applying an individual parallelizing matrix on a combined received signal which comprises all the individually transmitted signals. Each individual parallelizing matrix is determined for a respective user such that it more or less cancels out the signals from the other users when applied on the combined received signal, thereby effectively forming an "orthogonal space" for the respective user. It will be described in more detail below how such individual parallelizing matrices can be determined.

In this solution it is thus not necessary to estimate the individually transmitted signals sequentially, i.e. one by one as in the conventional solutions mentioned above, so that any decoding error in the estimation of an individual signal will not propagate from user to user. Thereby, it is an advantage that the accuracy can be improved by the embodiments described herein. Another advantage is that the accuracy is not necessarily reduced with increased number of users and/or data streams. Another advantage is that it is also possible to perform the estimations basically at the same time in parallel which will reduce latency and delays, particularly when the number of users is high.

The solution will be described herein in terms of various operations and actions performed by a receiving node which receives the combined signal and estimates each individually transmitted signal from the combined signal as follows. The receiving node in this context may be a network node or a wireless device, and the scope of the term "receiving node" as used in this description has been outlined above.

An example of how the solution may be employed will now be described with reference to the flow chart in Fig. 2 illustrating actions of a procedure in a receiving node, for detecting signals transmitted by multiple users over a radio interface where MIMO communication is employed. Reference will also be made to Fig. 3 which illustrates how signals transmitted from multiple users 300 may be received and processed by a receiving node 302 when the procedure of Fig. 2 is executed. This procedure can thus be used to accomplish the basic features outlined above, and some possible but non-limiting embodiments will also be described herein. It is assumed that the number of users is K where K is a positive integer. Throughout this disclosure, the notation k represents anyone of the users 1 ,..., K.

A first action 200 illustrates that the receiving node 302 receives a combined signal y comprising the signals Si ...S^ ...SK transmitted by the respective users k, where k=1, K, wherein each user transmits one or more data streams on a common radio resource. The combined signal y is received by a number of antennas NR in a receiving block 302A, in the figure denoted Base Station, BS or Access Point, AP.

In a next action 202, the receiving node 302 obtains a combined channel matrix H_CC comprising multiple individual channel matrices where each individual channel matrix /-/_/( Corresponds to a respective user k. In general, a channel matrix defines properties of a channel used for communication of signals between a user and an opposite node such as a serving network node. The properties of such a channel are typically dependent on where the user is located and on the current radio conditions between the respective user and its opposite node. Various channel properties, such as channel gain, propagation delay, etc., can be measured or estimated for each user, usually based on a known pilot signal, and the

measurements or estimations can be used for forming the individual channel matrices which are commonly maintained by a node in the network, typically the above-mentioned serving network node.

A further action 204 illustrates that the receiving node 302 determines a combined parallelizing matrix Qp comprising multiple individual parallelizing matrices where each individual parallelizing matrix Qk corresponds to a respective user k, such that the individual parallelizing matrix (_¾ of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix Qk are unitary matrices. Some examples of how the combined parallelizing matrix Qp may be determined will be described in more detail later below.

In a following action 206, the receiving node 302 also determines an effective received signal e^ for each user / by applying the individual parallelizing matrix Q/c Of said user /c on the combined received signal y. In this action, each individual parallelizing matrix Qk ^'\s multiplied with the combined received signal y which produces the effective received signal y_eff.i ... y_eff-k■■■ eff-zcfor the respective user 1 , ../c ...K. Actions 204 and 206 are performed by a parallelization block 302B when obtaining the combined signal yfrom the receiving block 302A.

In another action 208, the receiving node 302 further determines an effective channel matrix _en-k for each user k based on the effective received signal y_ett-k of said user k. A final action 210 illustrates that the receiving node estimates, for each user k, the signal Sk transmitted from said user k based on the effective received signal y_ea-k of said user /c and the effective channel matrix H_eff.k of said user k. Action 210 is performed by a detection block 302C. In this block 302C, a detection filter of each user Ze is applied to the corresponding effective received signal y_ea-k■ The detection filter may in some embodiments be a matrix for linear detections, whereas in other embodiments e.g. applying a nonlinear processing, i.e. Successive Interference Cancellation, SIC, for nonlinear detections.

Various embodiments can be employed by the receiving node when performing one or more of the above actions. In a possible embodiment, determining the combined parallelizing matrix Qp may comprise performing LQ decomposition on submatrices of the inversion of the combined channel matrix H_cc. The term "LQ decomposition" represents a decomposition of a matrix A into a product A = LQ of a lower triangular matrix L and an orthogonal matrix Q. The LQ decomposition in this embodiment may be performed for determining the combined parallelizing matrix Qp either by using a Zero Forcing, ZF, parallelizing process or a Minimum Mean-Square Error, MMSE, parallelizing process, as follows.

In one possible embodiment, when the combined parallelizing matrix Qp is determined in a ZF parallelizing process, an inversion matrix h _Cc ^ the combined channel matrix H_cc is calculated such that the inversion matrix h _cc comprises individual submatrices hT^ (also denoted H ^ in the following description) where each individual submatrix h /( Corresponds to a respective user k. The LQ decomposition of each individual submatrix h k is then performed to obtain an individual parallelizing matrix which is valid for the respective user / and comprised in the combined parallelizing matrix Qp. In another possible embodiment, when the combined parallelizing matrix Qp is determined in an MMSE parallelizing process, a regularized inversion matrix l~Pmmse of the combined channel matrix H_cc is calculated such that the regularized inversion matrix hP^~mm_Se comprises individual submatrices hTmmse.k where each individual submatrix h mmse.k corresponds to a respective user k, and the LQ decomposition of each individual submatrix hTmmse.k is then performed to obtain an individual parallelizing matrix Q_mmse,k which is valid for the respective user k and comprised in the combined parallelizing matrix Qp. In another possible embodiment, the LQ decomposition may be performed by means of a non-linear Sorted LQ Decomposition, SLQD, detector. An example of how this embodiment may be realized will be described later below. In another possible embodiment, the signals transmitted from the respective users / may be estimated in parallel, i.e. more or less at the same time which means that the time until all signals have been estimated can be greatly reduced as compared to the conventional SIC detection technique where the signals must be estimated sequentially.

It will now be described in more detail how the above-mentioned actions and embodiments can be performed and achieved by various calculations and computational operations. First, some conventional procedures for detecting signals in an MU-MIMO scenario will be described.

Compared to a single-antenna system, MIMO technology can provide a significant improvement in system throughput and reliability. MU-MIMO, where the Base Station, BS, equipped with multiple antennas communicates with multiple users simultaneously, has been introduced in various wireless standards, e.g. LTE and 802.1 1 .

The process of decoding all uplink users jointly from a combined signal that includes multiple individual signals is referred to as multiuser detection, MUD. In principle, a maximum-likelihood, ML, detector can be used to achieve a satisfactory performance. However the ML detector is computationally expensive for implementation, in particular for a system with a large number of antennas.

A linear receiver, particularly the zero forcing, ZF receiver, is largely applied in practice as a low-complexity alternative. Mathematically, the received signal is modelled at BS as y = Hs + 11 and it is assumed that the channel matrix

H e C^{NS XNR} is invertible, where N_R is the number of receive antennas and

N_S is the number of transmitted streams. A ZF detection filter is designed to eliminate the interference, that is G_7f = (U^HU) H

A joint estimation of the transmitted signal S can be obtained by s = G_z (Hs + n) = s + G_z n ₍₂₎

2 where the quantity Π denotes a noise term with zero mean and variance &_η . Then, the error covariance matrix is

To enhance the performance for MUD, the nonlinear successive interference cancellation, SIC, process has been introduced. The zero forcing successive interference cancellation, ZF-SIC, is one of such advanced nonlinear receivers, which can achieve high spectral efficiency with reasonable decoding complexity. The procedure of ZF-SIC detection starts with a QR decomposition, which calculates H = QR where Q is a unitary matrix and R is an upper triangular matrix. The effective received signal y _eff is obtained by multiplying the Hermitian transpose of the unitary matrix Q to the received signal as eff = Q " y = Rs + n ₍₄₎ where the quantity n— Q n is the effective noise term. By assuming the detected symbol is correct, the ZF-SIC receiver successively subtracts the interference contributed by previous streams from the current stream. This process proceeds to next symbol until all symbols are detected. Starting the detection from the bottom y_eff,N_s , the ZF-SIC procedure is implemented successively as

where ^ri is the i-th diagonal element of the matrix R and Q(') is a slicing function. Finally, an estimated transmit data vector is ^{s =} l^S\ > ^S2 > " ' > ^SN_S J ■ It can be noted that the detection order may be critical for the ZF-SIC

performance. Thus the Sorted QR decomposition based Detection, SQRD, has been proposed previously. The sorting process is implemented iteratively inside the QR decomposition and the diagonal elements of R is sorted in a descending order from bottom to top. The received signal with stream permutation can therefore be rewritten as y = (HPs + n) ₍₇₎ where P is the permutation matrix and the equivalent channel matrix is

^^■sqr ⁼ H = Q_sqiR_sqr . Similarly, the Hermitian transpose of the unitary matrix

Q_sqr is multiplied to the combined received signal y . The detection procedures of SQRD are the same with ZF-SIC in (5) and (6). The physical meaning of the SQRD is that the strongest stream will always be distinguished first from the remaining streams.

The solution and the embodiments described above provide two parallelizing access procedures for UL MU-MIMO systems. The signals from multiple users can be separated from each other and the effective noise power is not increased by the parallelizing process. The individual users can also be detected in parallel. Furthermore, the individual effective channel matrix of each user is a lower triangular structure after the parallelizing process. The above embodiments comprise a parallel SLQD detector, in which a sorting process is developed by making use of this special structure of the effective channel matrix. Simulations have been performed which show that the proposed procedures can achieve 5 considerable BER gains as compared to the conventional procedures.

In the following description, mathematical models are used to exemplify how the embodiments herein can be achieved, without limitation. Various wireless systems such as LTE, WCDMA, WiFi etc., may benefit from exploiting the ideas covered within this disclosure.

o

In an uplink MU-MIMO system, it is assumed that there are K users each equipped with N _K transmitting antennas, and the BS or access point, AP, is equipped with N_R receiving antennas.

5 For the k -th user, its uplink channel matrix is denoted by e c where r_k < N_k is the number of transmitted data streams from the & -th user and C refers to the complex field. The number of total transmitted data streams

\s N_s . The combined received uplink signal y_u, is

ui = HA + · · · + H,s, + · · · + U_Ks_K + n = H_ccs + n ₍₈₎ 0 where the quantity ^s£ ^e C ^K is the & -th transmitted data vector, n e C^{NR X1} is the noise term, H_cc = [H_{1 ?} - · -H^. , · · -H^] G C ^{R s} is a combined channel matrix that com rises the above-mentioned individual channel matrices, and

S ^— 1»ι ·> ^'

is a combined data vector. At the receiver side, e.g. the receiver node 302 shown in Fig. 3, an estimate of the channel matrix is usually available. A network node such as BS or AP usually has more capability to implement the channel estimation than in a wireless device. With the channel information known at the receiver side, the combined channel matrix can be obtained as

A parallelizing process is implemented in this solution to separate the superposed multi-user signal by applying a parallelizing matrix which can separate the multiple users while not increasing the noise power. Then, each user is detected as if the other users did not exist. As mentioned above, the users can be detected in parallel Some examples of how the proposed parallelizing process can be performed are descripted in detail below.

It was mentioned above that determining the combined parallelizing matrix Qp may comprise performing LQ decomposition on individual submatrices of the inversion of the combined channel matrix H_CC, and that the combined parallelizing matrix Qp may be determined in a ZF parallelizing process. It will now be described how this can be done in more detail.

First, the inversion of the combined channel matrix H_cc comprising the individual channel matrices H₁ ...H^ ... HK ^'\S calculated as,

H+ c- *^rk^xNR II

cc,k , ^ is the -th sub-matrix of " cc that corresp ~onds to the respective user k. For brevity, the c-th sub-matrix of H ^ _mav also be denoted h ^herein.

Given that "-_Cc"-cc ^~ *-N_S , where the quantity I_N is an N_s N_s identity matrix, the following is obtained,

From (11 ), the following relationship can be further derived,

^HI H₄=I_¾, Η Η₇=Ο V/_≠* ₍₁₂₎ Next, the LQ decomposition is performed, i.e. H - L^Q^ _j where the quantity l^_z^ £ is a lower triangular matrix and £_z/,£ ^fc ^ is ar unitary matrix (Q Q_z,£ ⁼ ). Since ~^zf,k is invertible, we have

Q_H. =0 \fj≠k (13)

From (13), it can be seen that the parallelizing matrix Q_z/,¾ forms an orthogonal space for all the other uplink users and the inter-user interferences are cancelled out from the k-h user. Thus, the effective received &-th user's signal is efffk ¹ "eff ,k (14) where the effective noise term ⁿ _eff ,k ^~ Q_z/,¾ⁿ . Because of the parallelizing matrix Q_z ,yt is unitary, the power of the effective noise, E_{n ;} j_{s s}ti 11 the same after the parallelizing process, which is illustrated as

E. (1 5)

The effective channel matrix for the k-th user is obtained as

H eff ,k (16)

Finally, the combined ZF parallelizing matrix Q_/, is obtained as

Because of the unitary matrix Q_z/,£ in (13), the interference between any two users is strictly zero. Therefore the combined effective channel matrix H_eJy j_s

The combined effective received signal y_eff is

Jeff ⁼ pJul ⁼

It can be noted that the parallelizing matrix should be able to separate users from each other as shown in (13) and (18). It will also need to meet the constraint that the power of effective noise is not increased as shown in (15). The parallelizing matrix Q_p is thus obtained by performing the LQ decomposition, although the solution is not limited thereto, and there may be other alternative ways to obtain the parallelizing matrix Q_p , for example by performing QR decomposition, or Singular Value Decomposition, SVD, which is a factorization of a matrix A as A = \∑W^H where U and V are unitary matrices, and ∑ is a diagonal matrix.

It was also mentioned above that the combined parallelizing matrix Qp may be determined in an MMSE parallelizing process. It will now be described how this can be done in more detail. As shown in (12) and (13), the inter-user interference is effectively forced to zero, however, the impact from noise is not taken into account by the above-described ZF parallelizing process. Thus, the system performance could be degraded when the Signal-to-Noise Ratio, SNR, is low such that the noise is a significant factor. A regularized factor is introduced in the MMSE parallelizing process to achieve a tradeoff between inter-user interference and noise. The regularized inversion of the combined channel matrix H_cc j_Sj

H⁺ = (H^,,H_C(. + αI_iVsr¹H XX

mmse cc S ^, c^,c (20)

Considering a case with high SNR, where the noise impact can be neglected, the regularization factor a approaches zero and thus,

^H1, ^H- ^{¾ I}^ (21 ) Similarly, the following relationship can be obtained,

^HI ^H^ ^H+ ≡j » * V^)≠k (22) w c

mmse r^¾XJVs I I

where ^AX _mm„ _t ^c ^ is the -th sub-matrix of ⁿ that corresponds to the res ective user k. Next, the LQ decomposition is performed, i.e.

, where the quantity _{mmse k} e C * j_{s a} |₀wer

Q ^r_k xN_R

mmse,k is an orthogonal matrix. Since L*_mmse,k is invertible, the following can be obtained,

Thus, the unitary matrix Q mmse,k can be used to balance the inter-user interference and the noise. The & -th user's effective received signal is ff,k = Q_mms,k (H_^s + n) = U^_ks_k + n ^leff,k (24) where

⁼ Qmm« ⁿ is the effective channel matrix and effective noise term for the k -th user respectively. Finally, the combined MMSE parallelizing matrix Q_p , the combined effective channel matrix

, and the combined effective received signal _eff are obtained as,

After the above-described uplink ZF or MMSE parallelizing procedure, users can communicate with the receiving node, e.g. BS or AP as if all the other users did not exist. In one case, the conventional linear or non-linear detection method can be applied to the combined effective received signal y _eff . In another case, different detection methods can be applied in parallel to the effective received signals y _ejf ,k from multiple users k. In one embodiment, the MMSE detection filter may be implemented in parallel for each user k as,

It can be assumed that the noise term is Gaussian noise with independent identically distributed entries of zero mean and variance ^σ„ . By utilizing the orthogonally principle that E^Gy _{eff k} S_k )y ^ |— 0 _> the detection filter for the k -th user, in this case denoted ^GMMSE,k , can be obtained as

^GMMSE,k =

(27)

Then the estimated signal for the k -th user can be written as

The procedure of parallel MMSE detection will now be described in more detail with reference to the flowchart in Fig. 4 which illustrates a procedure of the parallel

MMSE detection, where the above-mentioned slicing function. In this procedure the following actions are performed. Action 400: Get the estimation of the combined channel matrix

H_cc = [H_{1 ?} · · -H^ , · · ] G C ^{R 5} and the combined received uplink signal „; ⁼ H_cc ^s + ⁿ , where e C^NRXrk is the -th user's channel matrix. With ^r k being the number of data streams transmitted by the c-th user, the total

K

number of transmitted data streams is ⁼∑ ^rk ,

k=l S = [sf , - - - , s^T _k , - - -s^T _K ]^T e "⁵"¹ is a combined data vector with e C"*^xl being the k -th transmitted data vector, and n e is the noise term.

Action 402: Calculate the regularized channel inversion H as = ⁺ °^ N_s ) , where the parameter OC is the regularization factor for balancing the noise term. Action 404: To get the parallelizing matrix Q_p by performing the LQ decomposition on the submatrices of _mmse , i.e. on

,

H+ ^c II

, ^fc ^ is the c-th submatrix of ⁿ

^mmse,k ^ C ^{k k} j_{s a} lower triangular matrix, Q_mmse,_k ^e C^kXNR is a unitary matrix

that satisfies o V^H„ , O V mmse,k =1 N_R , and mmse,k mmse,K

Action 406: Get the effective received signal y _eff as y _eff = Q_py_ui and the effective channel matrix = Q_pH_cc .

Action 408: Calculate the detection filter in parallel for individual users using the MMSE criterion ^*MMSE,k ⁼ ^- _eff:k^eff,k + ) H ^ _; where the quantity eff,k ^e * * is the -th user's effective channel matrix, and is the -th user's regularization factor for balancing its noise term.

Action 410: Estimate the combined transmitted signal s = [s_i ,---,s_k,---s_K] by performing the quantization for each individual user detection as

¾ ⁼ Q( MMSE,ky eg ,k) in parallel, where _eff,k is the k -t user's effective received signal.

It was further mentioned above that the LQ decomposition may be performed by means of a non-linear Sorted LQ Decomposition, SLQD, detector. An example of how this may be performed will now be described. After the above-described ZF arallelizing process, and considering

= I_¾ , the effective channel is actually a lower triangular matrix, i.e.,

Assuming that with the permutation matrix P ^A _Cc,k , the diagonal elements of τ ^ _lf ¹ _k

P — T are in a descending order from top to bottom and given that _cc,k _{cc k} ^{~ i} r_k , the following can be defined,

^P _CC,_* ^H1,,^H _* ^P _C^ = ^l r_k (30)

Then, the SLQD is developed to fulfill the above design requirement in (30). In Fig. 5, an example of how the sorted LQ decomposition SLQD can be implemented is illustrated. For simplicity and generalization purpose, the unnecessary subscripts are removed. In this procedure the following actions are performed.

Action 500: Initialization stage. The quantity T is the number of transmitted data streams , U is initialized with the channel inversion e initialized as zero matrices, the variable P is used to store the ordered row index and is initialized with the original order p = {1,2, - - - , r} , i is used to index the row with the minimum norm from the remaining vectors and is initialized = 1 . In the following the notation ^■ (or U - ) and q ,· ( or q _m ) is used to denote row vectors of matrices U and Q respectively, and the notation (or l_mj ) is used to denote elements of matrix L .

Action 502: Set j - ί as a loop variable for calculating norms from row to r Action 504: Calculate the norm of a row vector u _} as Norm( j) = J J .

Action 506: Continue the last action 504, until the norm of all row vectors have been calculated.

Action 508: Find the row with the minimum norm and mark it as k _n .

Action 510: Exchange row i with k_min in U , Q , L , p , and Norm .

Action 512: Assign the diagonal element with the norm of U . as l_u = ||u-|| . And then get q ,· by normalizing U - to length one as q, ^{= u}i/ ,i ■

Action 514: Set m = i + l to execute the updating process for the remaining i + 1 to r rows.

Action 516: Perform the updating process by orthogonalizing row vector q _m with regard to q, and subtracting projections generated by q, as

Action 518: Perform the last action 516 until all the remaining rows are updated.

Action 520: Repeat the actions 502 to 518 until all the rows of L and Q are generated.

Action 522: The permutation matrix P is obtained according to the ordered index in p and the ordered L , Q ,and the permutation matrix P are outputted. In this case, as required in (30), the above developed SLQD in Fig.5 is carried out such that P_cc_cc_j,k_tH⁺ _{cc t ,} - L_o o_rr_dd_{zf k} Q ^ord_{zf k} Therefore, the effective channel for SLQD is

where L^{" o1rd}zf ,k is a lower triangular matrix with diagonal elements in a descending order from top to bottom. The effective received signal is o^rdzf,k (32) where the effective noise term is now _eff,k ^~ Qord . In this example

T *^siq,k is used to denote T _o ¹ _{rd k}

Finally, the SIC procedure from stream 1 to stream i=2,...,rk for the -th user can be written as

^hiq,k _l J and

where iq,k_{i t} is the i-th diagonal element of the matrix ~^slq,k . An example of how a procedure of parallel SLQD can be performed, will now be described with reference to the flowchart in Fig. 6. An extension of the parallel SLQD to the MMSE parallelizing process can be made by applying it to the effective channel matrix

= Q mmse, k k _{mmse k} and then following the procedure shown in Fig. 6. This figure illustrates the procedure of the parallel SLQD, where r_k≤ N_k is the number of transmitted data streams of the & -th user, iq,k_i is the i-th diagonal element of the matrix ~^ _siq,k , and Q(') is the slicing function. To simplify notation, the subscript has been left out for the i-th diagonal element of the matrix ~^ _siq,k in the procedure below, so that instead the notation hiq, (i ) is used for the i-th diagonal element of the matrix ^ _siq,k .

Correspondingly, the element iq,k_{ of this matrix is denoted hiq,(i, j) ■ In this procedure the following actions are performed.

Action 600: Initialization stage. Get the ordered L , Q ,and the permutation matrix P from the SLQD obtained in the procedure of Fig. 5. Initialize ^_siq,k ⁼ ^ , and the initial index = 1 .

H

Action 602: Get the effective channel matrix for the k -th user H — QH

Action 604: Get the effective received signal for the

Jeff = -eff^Sk + ^Ueff ■

Action 606: Estimate the signal of the first stream

Action 608: Start the iterative steps by setting i = i + l .

Action 610: Remove the interference of previous i— 1 streams from the current h stream by performing the successive cancellation process as

Action 614: Repeat the step from 608 to 612 until all the streams are estimated.

Action 616: Output the estimation S_k of the transmitted signal ^st of each user k.

The block diagram in Fig. 7 illustrates a detailed but non-limiting example of how a receiving node 700 may be structured to bring about the above-described solution and embodiments thereof. The receiving node 700 may be configured to operate according to any of the examples and embodiments of employing the solution as described above, where appropriate, and as follows. The receiving node 700 is shown to comprise a processor "P", a memory "M" and a communication circuit "C" with suitable equipment for receiving radio signals in the manner described herein. The communication circuit C in the receiving node 700 comprises equipment configured for communication with multiple users 702 e.g. over suitable radio interfaces using a suitable protocol for radio communication depending on the implementation. This communication may be performed when MU-MIMO is employed and multiple users transmit one or more data streams. The solution is however not limited to any specific types of networks, communication technology or protocols.

The receiving node 700 comprises means configured or arranged to perform at least some of the actions 200-210, 400-410, 500-522 and 600-616 of the flow charts in Figs 2 and 4-6, respectively. The receiving node 700 is arranged to detect signals transmitted by the multiple users 702 in data streams over a radio interface where MIMO communication is employed. The receiving node 700 may thus comprise the processor P and the memory M, said memory comprising instructions executable by said processor, whereby the receiving node 700 may be operative as follows.

The receiving node 700 is configured to receive a combined signal y comprising the signals Sk transmitted by the respective users k wherein each user transmits one or more data streams on a common radio resource. This operation may be performed by a receiving module 700A in the receiving node 700, e.g. in the manner described for action 200 above.

The receiving node 700 is further configured to obtain a combined channel matrix H_cc comprising multiple individual channel matrices where each individual channel matrix /-//( Corresponds to a respective user k. This operation may be performed by an obtaining module 700B in the receiving node 700, e.g. as in action 202 above.

The receiving node 700 is also configured to determine a combined parallelizing matrix Qp comprising multiple individual parallelizing matrices where each individual parallelizing matrix (_¾ corresponds to a respective user k, such that the individual parallelizing matrix (_¾ of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix (_¾ are unitary matrices. This operation may be performed by a determining module 700C in the receiving node 700, e.g. as described for action 204 above. The receiving node 700 is further configured to determine an effective received signal for each user kby applying the individual parallelizing matrix (_¾ of said user / on the combined received signal y. This operation may be performed by the determining module 700C, e.g. as described for action 206 above. The receiving node 700 is further configured to determine an effective channel matrix _en-k for each user k based on the effective received signal y_eft-k of said user k. This operation may be performed by the determining module 700C, e.g. as described for action 208 above. The receiving node 700 is further configured to estimate, for each user k, the signal transmitted from said user k based on the effective received signal y_e -k of said user / and the effective channel matrix of said user k. This operation may be performed by an estimating module 700D in the receiving node 700, e.g. as in action 21 0 above. It should be noted that Fig. 7 illustrates various functional modules in the receiving node 700, and the skilled person is able to implement these functional modules in practice using suitable software and hardware. Thus, the functional modules 700A-D in the receiving node 700 may be configured to operate according to any of the features and embodiments described in this disclosure, where appropriate. The functional units 700A-D described above can be implemented in the receiving node 700 by means of program modules of a computer program comprising code means which, when run by the processor P causes the receiving node 700 to perform the above-described actions and procedures. The processor P may comprise a single Central Processing Unit, CPU, or could comprise two or more processing units. For example, the processor P may include a general purpose microprocessor, an instruction set processor and/or related chips sets and/or a special purpose microprocessor such as an Application Specific Integrated Circuit, ASIC. The processor P may also comprise a storage for caching purposes.

Each computer program may be carried by a computer program product in the receiving node 700 in the form of a memory having a computer readable medium and being connected to the processor P. The computer program product or memory M in the receiving node 700 thus comprises a computer readable medium on which the computer program is stored e.g. in the form of computer program modules or the like. For example, the memory M may be a flash memory, a Random-Access Memory, RAM, a Read-Only Memory, ROM, an Electrically Erasable Programmable ROM, EEPROM, or hard drive storage, and the program modules could in alternative embodiments be distributed on different computer program products in the form of memories within the receiving node 700. The solution described herein may be implemented in the receiving node 700 by means of a computer program storage product comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the actions according to any of the above embodiments, where appropriate. By using one or more of the above-described embodiments for detecting signals transmitted by multiple users employing MIMO communication, several advantages can be achieved, which will be explained in more detail below.

In this solution, the multi -user interference can be virtually removed or largely reduced on account of the proposed parallelizing process. This is because each user is detected individually as if the other users did not exist, which can also be made in parallel, i.e. more or less simultaneously. Therefore, the detection procedure described herein can scale up with system dimensions while keeping acceptable performance and delay. The solution would be particularly useful for communication scenarios with a large number of users scheduled at the same resource blocks and when each user transmits multiple streams.

Unlike the downlink transmission, uplink users may use their own transmit power and cross user coordination is usually infeasible for physically distributed users. The parallelizing operations described herein allow individual user detection and therefore, individual power allocation strategy, i.e., so-called "water filling", which is a well-known algorithm for allocating power among different streams and users, can be employed to optimize the power used by the users and for the data streams.

Since the proposed parallelizing operations allow that individual users are detected separately in parallel, different detection operations can be applied to individual users dependent on system parameters such as: Quality of Service, QoS, requirement; user transmit power; distance to serving network node, etc. In contrast, the conventional methods can only detect all users in the same way. Therefore, the embodiments described herein can bring extra flexibility to UL MU- Ml MO detection.

The existing SQRD is not feasible for the UL MU-MIMO detection when uplink users transmit multiple streams. The parallel SLQD described herein may be used instead to overcome this drawback. Compared to procedures like the ZF-SIC or MMSE-SIC, the proposed parallel SLQD procedure can avoid unnecessary delay by detecting users in parallel and it is more robust to error propagation.

Simulation has been carried out to compare the performance of the solution described herein with existing receivers. For simplicity, it is assumed that equal power loading between users and streams is performed. Fig. 8 shows simulation results for the UL MU-MIMO system with a BS or an AP equipped with N^ = 10 receive antennas, and K = 2 users, each user k transmitting r_k = N_k = 5 data streams. This configuration is termed as 10 x (5, 5).

With the proposed ZF parallelizing process, linear parallel MMSE detection can achieve 5.5 dB gains over the conventional ZF detection at BER 10 ² . The proposed parallel SLQD achieves around 2.2 dB and 4 dB gains over the conventional ZF-SIC detection at BER 10 ² and 10 ³ , respectively. Fig. 9 shows that the K = 4 case with the configuration 32 x (8, 8, 8, 8). At BER 10 ² , the ZF parallelizing process with parallel MMSE detection can achieve 6.5 dB gains over the conventional ZF detection, and the ZF parallelizing process with parallel SLQD can achieve 2.5 dB gains over the conventional ZF-SIC detection. In conclusion, the ZF parallelizing process with parallel MMSE process provides the better performance in medium-low SNR region and the ZF parallelizing process with parallel SLQD method outperforms when the SNR is high. The reason is that the proposed ZF parallelizing process is mainly directed to removing the inter-user interference but the noise term has larger impact on system performance when the SNR is low.

From Figs 8 and 9, it can also be verified that the performance of conventional ZF and ZF-SIC degrades when system dimensions increase. In contrast, the proposed detection operations work well with increased number of users and streams.

By taking the noise term into account, the MMSE parallelizing process is developed. Simulation results are illustrated below. Fig. 10 illustrates comparison of BER performances of the proposed methods with MMSE and MMSE-SIC. It can be seen that the proposed MMSE parallelizing process with parallel MMSE detection and the proposed MMSE parallelizing process with parallel SLQD can provide improved performance with around 4.8 dB and 5.9 dB gains over the conventional MMSE and MMSE-SIC, respectively, at BER 10^"3.

In Fig. 1 1 , the system dimensions are increased to involve 200 receiving antennas at the network node and 50 users, each user transmitting 4 data streams. Unlike the conventional ZF detection, the conventional MMSE detection can achieve a balanced performance between inter-user interference and noise. What's more ,the performance of the conventional MMSE detection can scale up with system dimensions. However, a clear advantage over the conventional MMSE detection is still achieved by the embodiments described herein which can achieve around 5.5 dB gains over the conventional MMSE detection at BER 10

The conventional MMSE-SIC can achieve fairly good performance, but the delay among users is a serious problem since all the 200 received streams are detected sequentially. From the delay perspective, therefore, the conventional SIC detections cannot scale up with system dimensions properly. The performance of the proposed MMSE parallelizing with parallel MMSE detection is very close to that of the proposed MMSE parallelizing with parallel SLQD. Thus, it can be concluded that the proposed MMSE parallelizing with parallel MMSE detection can scale up with system dimensions both in terms of performance and delay. It can be noted that the implementation of the proposed procedures do not require any feedback nor feedforward information from or to the multiple users.

While the solution has been described with reference to specific exemplifying embodiments, the description is generally only intended to illustrate the inventive concept and should not be taken as limiting the scope of the solution. For example, the terms "receiving node", "user", "parallelizing matrix", "parallelizing process" and "LQ decomposition" have been used throughout this disclosure, although any other corresponding entities, functions, and/or parameters could also be used having the features and characteristics described here. The solution is defined by the appended claims.

Abbreviation Explanation

AP Access Point

BS Base Station

MIMO Multi-Input Multi-Output ML Maximum-Likelihood

MMSE Minimum Mean-Square Error

MSE Mean Square Error

MUD Multi-User Detection

MU-MIMO Multi-User MIMO SLQD Sorted LQ Decomposition

SQRD Sorted QR Detection

ZF Zero Forcing

ZF-SIC Zero Forcing Successive Interference Cancellation

Claims

1 . A method performed by a receiving node (302) for detecting signals transmitted by multiple users (300) in data streams over a radio interface where Multiple-Input- Multiple-Output, MIMO, communication is employed, the method comprising:

- receiving (200) a combined signal y comprising the signals Sk transmitted by the respective users k, where k=1, K, wherein each user transmits one or more data streams on a common radio resource,

- obtaining (202) a combined channel matrix H_cc comprising multiple individual channel matrices where each individual channel matrix /-/_/( Corresponds to a respective user k,

- determining (204) a combined parallelizing matrix Qp comprising multiple individual parallelizing matrices where each individual parallelizing matrix (_¾ corresponds to a respective user k, such that the individual parallelizing matrix (_¾ of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix (_¾ are unitary matrices,

- determining (206) an effective received signal

for each user k by applying the individual parallelizing matrix (_¾ of said user / on the combined received signal y,

- determining (208) an effective channel matrix H_eff.k tor each user k based on the effective received signal y_ea-k of said user k, and

- estimating (210), for each user k, the signal Sk transmitted from said user k based on the effective received signal y_e -k O said user / and the effective channel matrix H_eff.k of said user k.

2. A method according to claim 1 , wherein determining the combined parallelizing matrix Qp comprises performing LQ decomposition on submatrices of the inversion of the combined channel matrix H_cc.

3. A method according to claim 1 or 2, wherein the combined parallelizing matrix Qp is determined in a Zero Forcing, ZF, parallelizing process by calculating an inversion matrix h _Cc ^ the combined channel matrix H_cc comprising individual submatrices hT ^ where each individual submatrix hT /( Corresponds to a respective user k, and performing the LQ decomposition of each individual submatrix H^ to obtain an individual parallelizing matrix Q_z^ which is valid for the respective user / and comprised in the combined parallelizing matrix Qp.

4. A method according to claim 1 or 2, wherein the combined parallelizing matrix Qp is determined in a Minimum Mean-Square Error, MMSE, parallelizing process by calculating a regularized inversion matrix -T_mmse of the combined channel matrix H_cc comprising individual submatrices h mmse.k where each individual submatrix hTmmse.k corresponds to a respective user k, and performing the LQ decomposition of each individual submatrix H* _mmse,/c ^t0 obtain an individual parallelizing matrix Q_mmSe,k which is valid for the respective user / and comprised in the combined parallelizing matrix Qp.

5. A method according to any of claims 2-4, wherein the LQ decomposition is performed by means of a non-linear Sorted LQ Decomposition, SLQD, detector.

6. A method according to any of claims 1 -5, wherein the signals Sk transmitted from the respective users / are estimated in parallel.

7. A receiving node (700) arranged to detect signals transmitted by multiple users (702) in data streams over a radio interface where Multiple-Input- Multiple- Output, MIMO, communication is employed, wherein the receiving node (700) is configured to: - receive (700A) a combined signal y comprising the signals Sk transmitted by the respective users k , where k=1, K, wherein each user transmits one or more data streams on a common radio resource,

- obtain (700B) a combined channel matrix H_cc comprising multiple individual channel matrices where each individual channel matrix /-//( Corresponds to a respective user k,

- determine (700C) a combined parallelizing matrix Qp comprising multiple individual parallelizing matrices where each individual parallelizing matrix (_¾ corresponds to a respective user k, such that the individual parallelizing matrix (_¾ of each user k forms an orthogonal space for the other users, wherein the combined parallelizing matrix Qp and each individual parallelizing matrix Q^ are unitary matrices,

- determine (700C) an effective received signal e^ for each user k by applying the individual parallelizing matrix (_¾ of said user / on the combined received signal y,

- determine (700C) an effective channel matrix _en-k for each user k based on the effective received signal y_en-k of said user k, and

- estimate (700D), for each user k, the signal Sk transmitted from said user k based on the effective received signal y_eff-k said user / and the effective channel matrix H_eff.k of said user k.

8. A receiving node (700) according to claim 7, wherein the receiving node

(700) is configured to determine the combined parallelizing matrix Qp by performing LQ decomposition on submatrices of the inversion of the combined channel matrix H_cc.

9. A receiving node (700) according to claim 7 or 8, wherein the receiving node (700) is configured to determine the combined parallelizing matrix Qp in a Zero Forcing, ZF, parallelizing process by calculating an inversion matrix hP_cc of the combined channel matrix H_cc comprising individual submatrices h k where each individual submatrix h /( Corresponds to a respective user k, and performing the LQ decomposition of each individual submatrix H^ to obtain an individual parallelizing matrix Q_Zf_tk which is valid for the respective user / and comprised in the combined parallelizing matrix Qp.

1 0. A receiving node (700) according to claim 7 or 8, wherein the receiving node (700) is configured to determine the combined parallelizing matrix Qp in a

Minimum Mean-Square Error, MMSE, parallelizing process by calculating a regularized inversion matrix fmmse of the combined channel matrix H_cc comprising individual submatrices h mmse.k where each individual submatrix hf^~ mmse,k corresponds to a respective user k, and performing the LQ

decomposition of each individual submatrix H* _mmse,/c ^t0 obtain an individual parallelizing matrix Q_mmse,k which is valid for the respective user / and comprised in the combined parallelizing matrix Qp.

1 1 . A receiving node (700) according to any of claims 8-1 0, wherein the receiving node (700) is configured to perform the LQ decomposition by means of a non-linear Sorted LQ Decomposition, SLQD, detector.

1 2. A receiving node (700) according to any of claims 7-1 1 , wherein the receiving node (700) is configured to estimate the signals Sk transmitted from the respective users in parallel.

1 3. A computer program comprising instructions which, when executed on at least one processor, cause the at least one processor to carry out the method according to any one of claims 1 -6.

14. A carrier containing the computer program of claim 13, wherein the carrier is one of an electronic signal, optical signal, radio signal, or computer readable storage medium.