CN103957086A - Achieving method for MU-MIMO precoding - Google Patents

Abstract

The invention discloses an achieving method for MU-MIMO precoding. The method includes the detailed steps that a first precoding matrix is obtained through performing improved QR decomposition on expanding matrixes H of joint channel matrixes HS of all users to eliminate multi-user interference, precoding matrixes of all the users are determined according to a QR decomposition result, and therefore calculation complexity is reduced; then a lattice reduction method is used for obtaining a second precoding matrix better in performance. Compared with a traditional MU-MIMO precoding method, the method is lower in complexity, and meanwhile system performance is improved.

Description

MU-MIMO precoding implementation method

Technical field

The invention belongs to mobile communication technology field, relate to multiple-input and multiple-output (Multiple Input Multiple Output, MIMO) technology wherein, be specifically related to the precoding algorithm in multiuser MIMO (multiuser MIMO, MU-MIMO) system.

Background technology

MIMO, as a kind of technology that can significantly improve wireless communication spectrum efficiency, has received increasing concern under the day by day rare background of frequency spectrum resource.Along with going deep into of multi-antenna technology research, MIMO technology has expanded to the multi-user system of point-to-multipoint from point-to-point single user system.In MU-MIMO system, base station (Base Station, BS) is simultaneously to a plurality of user's transmitted signals, because a plurality of users share same running time-frequency resource, will certainly cause inter-user interference (Multi-user Interference, MUI), greatly reduce the efficiency of transmission of system.In order to meet the requirement of next generation communication system to high data rate communication, the precoding technique that can effectively suppress MUI occurs in a large number.

In descending MU-MIMO system, the precoding algorithm of known preferred is dirty paper code (Dirty Paper Coding, DPC), but because computational complexity is too high, is difficult to use in real system.Therefore the low complex degree precoding algorithm of some suboptimums is suggested, and wherein, block diagonalization (Block Diagonalization, BD) precoding algorithm is widely used a kind of in current MU-MIMO, and the main thought of this algorithm may be summarized to be following 2 points:

Each user utilizes singular value decomposition (Singular Value Decomposition, SVD) to find the null space basis of other user's combined channel matrixes, and the pre-coding matrix that forms oneself with it, eliminates MUI with this, obtains equivalent SU-MIMO channel; Equivalent SU-MIMO channel is carried out to singular value decomposition, and utilizing SVD precoding algorithm optimization system is energy.

Therefore user's pre-coding matrix W can be divided into two parts: W=W ^aw ^b, respectively corresponding above-mentioned 2 points.Although traditional B D-algorithm can be eliminated MUI completely, but need to use the singular value decomposition (SVD) twice with higher computational complexity to ask for each user's pre-coding matrix, increased the complexity of algorithm itself, therefore having limited the use of algorithm in real system, a kind of GZI(Generalized ZF Channel Inversion of low complex degree) precoding algorithm is suggested.

Respectively BD and two kinds of method for precoding of GZI are carried out to brief description below.

(1), BD method for precoding

For convenience of explanation and analyze, suppose in MU-MIMO system, the number of transmit antennas of base station end is N _t, k user's reception antenna number is N _k, total number of users is K, K the total reception antenna of user received number and is the number of transmit antennas N of base station end _tbe more than or equal to the total N of user's reception antenna _r.

To any user k, its channel matrix is H _k, interference channel matrix is ${\tilde{H}}_k={\left[\begin{array}{cccccc}{H}_1^T& ...& H_{k-1}^T& H_{k+1}^T& ...& H_K^T\end{array}\right]}^T,$ Pre-coding matrix is W _k; All K user's combined channel matrix is $H_S={\left[\begin{array}{cccc}{H}_1^T& H_2^T& ...& H_K^T\end{array}\right]}^T,$ Associating pre-coding matrix is W=[W ₁w ₂w _k].

The method comprises the following steps:

Step 1, base station obtains each user's down channel matrix H _k(k=1,2 ..., K).Under time division duplex multiplexing (TDD) pattern, user's channel information can be known by channel reciprocity in base station; Under Frequency Division Duplexing (FDD) multiplexing (FDD) pattern, its channel information can be known by the feedback of user terminal in base station.

Step 2, the interference channel matrix to any user k ${\tilde{H}}_k={\left[\begin{array}{cccccc}{H}_1^T& ...& H_{k-1}^T& H_{k+1}^T& ...& H_K^T\end{array}\right]}^T,$ Carry out SVD decomposition ${\tilde{H}}_k={\tilde{U}}_k{\widetilde{\mathrm{\Σ}}}_k{\left[\begin{array}{cc}{\tilde{V}}_k^{\left(1\right)}& {\tilde{V}}_k^{\left(0\right)}\end{array}\right]}^H,$ Wherein, be left singular matrix, be singular value diagonal matrix, with be respectively front r row and the rear (N of right singular matrix _t-r) row, r is matrix order.From matrix theory knowledge, be kernel orthogonal basis, so can be used as the first of user k pre-coding matrix with this, eliminate inter-user interference, obtain independently equivalent channel matrix

Concerning all K user, the first of its associating pre-coding matrix is $W^a=\begin{array}{c}\left[\begin{array}{cccc}{\tilde{V}}_1^{\left(0\right)}& {\tilde{V}}_2^{\left(0\right)}& ...& {\tilde{V}}_K^{\left(0\right)}\end{array}\right],\end{array}$ With this by many of MU-MIMO channel decomposing parallel equivalent SU-MIMO channels independently: $H_S{W}^a=\mathrm{diag}\{H_1{\tilde{V}}_1^{\left(0\right)},H_2{\tilde{V}}_2^{\left(0\right)},...,H_K{\tilde{V}}_K^{\left(0\right)}\}.$

Step 3, in order to obtain the maximum precoding gain of equivalent channel matrix, carries out SVD decomposition again to equivalent SU-MIMO channel $H_k^{\mathrm{eff}}=U_k{\mathrm{\Σ}}_k{V}_k^H=U_k{\mathrm{\Σ}}_k{\left[\begin{array}{cc}{V}_k^{\left(1\right)}& V_k^{\left(0\right)}\end{array}\right]}^H,$ Wherein, U _kbe left singular matrix, Σ _kbe singular value diagonal matrix, V _kbe right singular matrix, v _kfront N _krow.According to alone family SVD precoding algorithm, choose as pre-coding matrix, can obtain the maximum precoding gain under Single User MIMO system.Therefore the second portion of the pre-coding matrix of user k is concerning all K user, the second portion of its associating pre-coding matrix is $W^b=\mathrm{diag}\{V_1^{\left(1\right)},V_2^{\left(1\right)},...,V_K^{\left(1\right)}\}.$

Step 4, obtains the pre-coding matrix of whole system: W=W ^aw ^b.

(2), GZI method for precoding

The method comprises the following steps:

Step 1, calculates combined channel matrix H _spseudoinverse: wherein, be submatrix, dimension is N _t* N _k;

Step 2 is right carry out QR decomposition: wherein, orthogonal matrix, it is upper triangular matrix.Known according to the character of pseudoinverse ? because invertible matrix, so can be used as the first of user k pre-coding matrix thereby obtain independently equivalent channel matrix $H_k^{\mathrm{eff}}=H_k{\hat{Q}}_k.$

Concerning all K user, the first of its associating pre-coding matrix is $W^a=\left[\begin{array}{cccc}{\hat{Q}}_1& {\hat{Q}}_2& ...& {\hat{Q}}_K\end{array}\right].$

Step 3, in order to obtain the maximum precoding gain of equivalent channel matrix, carries out SVD decomposition again to equivalent SU-MIMO channel $H_k^{\mathrm{eff}}=U_k{\mathrm{\Σ}}_k{V}_k^H=U_k{\mathrm{\Σ}}_k{\left[\begin{array}{cc}{V}_k^{\left(1\right)}& V_k^{\left(0\right)}\end{array}\right]}^H,$ Choose right singular matrix V _kfront N _krow are as the second portion of the pre-coding matrix of user k, concerning all K user, the second portion of its associating pre-coding matrix is $W^b=\mathrm{diag}\{V_1^{\left(1\right)},V_2^{\left(1\right)},...,V_K^{\left(1\right)}\}.$

Step 4, obtains the pre-coding matrix W=W of whole system ^aw ^b.

Although GZI method for precoding decomposes with solving pseudo-inverse operation and QR the SVD decomposition having replaced in the BD method for precoding first step, complexity decreases, but solve pseudoinverse and remain the computing that a kind of complexity is higher, and two kinds of methods all need to carry out SVD decomposition for each user's equivalent channel matrix, so the complexity of GZI method for precoding is still higher.

Summary of the invention

The problems referred to above that exist for prior art, the present invention has designed a kind of MU-MIMO method for precoding based on improved QR decomposition and lattice reduction method, and the method, when reducing traditional MU-MIMO precoding algorithm complex, has improved algorithm performance.

Concrete technical scheme of the present invention is: a kind of MU-MIMO precoding implementation method, specifically comprises the steps:

The all K of step 1. user's combined channel matrix is $H_S={\left[\begin{array}{cccc}{H}_1^T& H_2^T& ...& H_K^T\end{array}\right]}^T,$ To associating channel matrix H _sexpand: $\stackrel{\‾}{H}=\left[\begin{array}{cc}{H}_S& \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right],$ Wherein, α=N _rσ ²/ P _total, σ ²the variance that represents noise, P _totalrepresent the total transmitted power of down link, expression dimension is N _r* N _runit matrix;

Step 2. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}^H=\stackrel{\‾}{Q}\stackrel{\‾}{R}\left[\begin{array}{c}{Q}_{N_T}\\ Q_{N_R}\end{array}\right]\stackrel{\‾}{R},$ Obtaining dimension is N _r* N _rupper triangular matrix with dimension be (N _t+ N _r) * N _runitary matrice and use represent unitary matrice front N _toK, represent unitary matrice rear N _roK;

Step 3. by ${\stackrel{\‾}{H}}^H=\left[\begin{array}{c}{H}_S^H\\ \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right]=\left[\begin{array}{cc}{Q}_{N_T}& \stackrel{\‾}{R}\\ Q_{N_R}& \stackrel{\‾}{R}\end{array}\right]$ Obtain $\sqrt{\mathrm{\α}}{I}_{N_R}=Q_{N_R}\stackrel{\‾}{R},$ And then obtain ${\stackrel{\‾}{R}}_k^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_{N_R},$ ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}={\left({\stackrel{\‾}{R}}^{-1}\right)}^H=\left[\begin{array}{ccccc}{\hat{R}}_1& ...& {\hat{R}}_k& ...& {\hat{R}}_K\end{array}\right],$ Wherein, be submatrix, dimension is N _t* N _k, N _kit is k user's reception antenna number;

Step 4. is obtained orthogonal basis P _k, the first of the pre-coding matrix of user k is arbitrarily with this, eliminate inter-user interference, thereby obtain the equivalent channel matrix after precoding for the first time

Concerning all K user, the first of its associating pre-coding matrix is W ^a=[P ₁p ₂p _k], with this by many of MU-MIMO channel decomposing equivalent SU-MIMO channel independently, H _sw ^a=diag{H ₁p ₁, H ₂p ₂..., H _kp _k;

The equivalent channel matrix of step 5. couple any user k transposition carry out Ge Jiyue and subtract (LR) operation: transformation matrix T _kfor unimodular matrix, i.e. T _kthe value of middle element is that complex integers and determinant meet det (T _k)=± 1;

According to the character of LR, after approximately subtracting, obtain the better equivalent channel matrix of orthogonality can effectively reduce noise scale-up problem.

Equivalent channel matrix after step 6. pair approximately subtracts expand: ${\stackrel{\‾}{H}}_k=\left[\begin{array}{cc}{\tilde{H}}_k^{\mathrm{eff}}& \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right].$

Step 7. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}_k^H={\stackrel{\‾}{Q}}_k{\stackrel{\‾}{R}}_k=\left[\begin{array}{c}{Q}_k^1\\ Q_k^2\end{array}\right]{\stackrel{\‾}{R}}_k,$ Obtaining dimension is N _k* N _kupper triangular matrix with dimension be (N _k+ N _k) * N _kunitary matrice and use represent unitary matrice front N _koK, represent unitary matrice rear N _koK.

Step 8. by ${\stackrel{\‾}{H}}_k^H=\left[\begin{array}{c}{\left({\tilde{H}}_k^{\mathrm{eff}}\right)}^H\\ \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right]=\left[\begin{array}{c}{Q}_k^1{\stackrel{\‾}{R}}_k\\ Q_k^2{\stackrel{\‾}{R}}_k\end{array}\right]$ Obtain $\sqrt{\mathrm{\α}}{I}_{N_k}=Q_k^2{\stackrel{\‾}{R}}_k,$ So ${\stackrel{\‾}{R}}_k^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^2.$ ${\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}={\left({\stackrel{\‾}{R}}_k^{-1}\right)}^H=\frac{1}{\sqrt{\mathrm{\α}}}{\left(Q_k^2\right)}^H.$

Step 9. couple any user k, will as the second portion of pre-coding matrix, $W_k^b=Q_k^1{\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^1{\left(Q_k^2\right)}^H;$

Concerning all K user, the second portion of its associating pre-coding matrix is

$W^b=\mathrm{diag}\{\frac{1}{\sqrt{\mathrm{\α}}}{Q}_1^1{\left(Q_1^2\right)}^H,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_2^1{\left(Q_2^2\right)}^H,...,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_K^1{\left(Q_K^2\right)}^H\}.$

Step 10. is obtained the pre-coding matrix W=W of whole system ^aw ^b.

Beneficial effect of the present invention: method of the present invention is the combined channel matrix H to all users first _sextended matrix carry out improved QR and decompose to obtain first pre-coding matrix to eliminate multi-user interference, and according to the result of QR decomposition, determine each user's pre-coding matrix, reduced the complexity of calculating; Then utilize lattice reduction (LR) method to obtain second more excellent pre-coding matrix of performance, method of the present invention when obtaining more low complex degree, has improved systematic function than traditional MU-MIMO method for precoding.

Accompanying drawing explanation

Fig. 1 is multi-user MIMO system schematic diagram in the embodiment of the present invention.

Fig. 2 is that the complexity of the inventive method and traditional B D and GZI method compares schematic diagram.

Fig. 3 is that the complexity of the inventive method and traditional B D and GZI method compares schematic diagram.

Fig. 4 is that the power system capacity of the inventive method and traditional B D and GZI method compares schematic diagram.

Fig. 5 is that the error rate of the inventive method and traditional B D and GZI method compares schematic diagram.

Embodiment

Below in conjunction with the drawings and specific embodiments, the present invention is described further.Parameter in embodiment does not affect generality of the present invention.

With analysis method for precoding of the present invention, MU-MIMO system is done to following setting: as shown in Figure 1, transmitting antenna sum is more than or equal to the sum of all user's reception antennas, i.e. N for convenience of explanation _t>=N _r; The data fluxion that each user communicates equates with its reception antenna number, and base station end transmit power allocations adopts average power allocation.

The signal that user k receives arbitrarily can be expressed as:

$y_k=H_k{W}_k{s}_k+H_k\underset{i=1,i\≠k}{\overset{K}{\mathrm{\Σ}}}{W}_i{s}_i+n_k$ Formula (1)

In formula (1), H _kk user's N _k* N _tdimension channel matrix, H _kin element separate and to obey average be 0, the multiple Gaussian Profile that variance is 1; W _kk user's N _t* N _kdimension pre-coding matrix; s _kk user's N _kdimension transmitting symbolic vector; n _kn _kthe column vector of dimension, is k user's independent identically distributed additive white Gaussian noise, and variance is σ _k ².

The embodiment of the present invention has been simplified combined channel matrix H _sthe process that pseudoinverse solves, and consider the impact of noise on systematic function, in conjunction with Ge Jiyue, subtract (LR) technology simultaneously, improved systematic function, reduced the complexity of computing.Concrete steps are as follows:

Step 1. pair associating channel matrix H _sexpand: $\stackrel{\‾}{H}=\left[\begin{array}{cc}{H}_S& \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right],$ Wherein, α=N _rσ ²/ P _total, σ ²the variance that represents noise, P _totalrepresent the total transmitted power of down link, expression dimension is N _r* N _runit matrix;

In this step, by associating channel matrix H _sexpand, based on extended matrix the computing that solves of pseudoinverse can be expressed as:

formula (2)

From formula (2), can find out the pseudoinverse of extended matrix upper submatrix can be used as least mean-square error (MMSE) Linear precoding matrix of having considered noise factor.

Step 2. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}^H=\stackrel{\‾}{Q}\stackrel{\‾}{R}\left[\begin{array}{c}{Q}_{N_T}\\ Q_{N_R}\end{array}\right]\stackrel{\‾}{R},$ Obtaining dimension is N _r* N _rupper triangular matrix with dimension be (N _t+ N _r) * N _runitary matrice and use represent unitary matrice front N _toK, represent unitary matrice rear N _roK.

The result that in this step, QR decomposes can be to pseudoinverse solution procedure simplify:

formula (3)

From formula (3), least mean-square error (MMSE) Linear precoding matrix of having considered noise factor, wherein, be submatrix, dimension is N _t* N _k.

Step 3. pair upper triangular matrix conjugate transpose carry out inversion operation, obtain ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}=\left[\begin{array}{cccc}{\hat{R}}_1& {\hat{R}}_2& ...& {\hat{R}}_K\end{array}\right],$ Wherein, be submatrix, dimension is N _t* N _k.

In this step, can be directly to matrix carry out inversion operation, the result that also can utilize extended channel matrices QR to decompose is obtained to further reduce the complexity of algorithm computing.

The formula of decomposing according to extended matrix QR: ${\stackrel{\‾}{H}}^H=\left[\begin{array}{c}{H}_S^H\\ \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right]=\left[\begin{array}{cc}{Q}_{N_T}& \stackrel{\‾}{R}\\ Q_{N_R}& \stackrel{\‾}{R}\end{array}\right],$ Can find out: $\sqrt{\mathrm{\α}}{I}_{N_R}=Q_{N_R}\stackrel{\‾}{R}.$ Can obtain upper triangular matrix thus contrary: upper triangular matrix the inverse matrix of conjugate transpose can be by the conjugate transpose of inverse matrix obtain, ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}={\left({\stackrel{\‾}{R}}^{-1}\right)}^H=\left[\begin{array}{cccc}{\hat{R}}_1& {\hat{R}}_2& ...& {\hat{R}}_K\end{array}\right].$

Step 4. is according to upper triangular matrix the inverse matrix of conjugate transpose ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}=\left[\begin{array}{cccc}{\hat{R}}_1& {\hat{R}}_2& ...& {\hat{R}}_K\end{array}\right]$ And extended channel matrices QR decomposes and to obtain matrix, obtains the first of pre-coding matrix.

In this step, according to upper triangular matrix the inverse matrix of conjugate transpose ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}=\left[\begin{array}{cccc}{\hat{R}}_1& {\hat{R}}_2& ...& {\hat{R}}_K\end{array}\right]$ And extended channel matrices QR decomposes and to obtain matrix, obtains $Q_{N_T}{\left({\stackrel{\‾}{R}}^H\right)}^{-1}=\left[\begin{array}{cccc}{Q}_{N_T}{\hat{R}}_1& Q_{N_T}{\hat{R}}_2& ...& Q_{N_T}{\hat{R}}_K\end{array}\right],$ But due to wherein each submatrix row between and non-orthogonal, also need alignment to carry out orthogonalization, as used Schimidt orthogonalization (GSO) algorithm, obtain orthogonal basis P _k, obtained considering that the first of the pre-coding matrix of respective user k after noise factor is:

Concerning all K user, the first of its associating pre-coding matrix is W ^a=[P ₁p ₂p _k].

Step 5. is in order to obtain the maximum precoding gain of equivalent channel matrix, the equivalent channel matrix to any user k transposition carry out Ge Jiyue and subtract (LR) operation: transformation matrix T _kfor unimodular matrix, i.e. T _kthe value of middle element is that complex integers and determinant meet det (T _k)=± 1.According to the character of LR, after approximately subtracting, obtain the better equivalent channel matrix of orthogonality can effectively reduce noise scale-up problem.

Equivalent channel matrix after step 6. pair approximately subtracts expand: ${\stackrel{\‾}{H}}_k=\left[\begin{array}{cc}{\tilde{H}}_k^{\mathrm{eff}}& \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right].$

Step 7. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}_k^H={\stackrel{\‾}{Q}}_k{\stackrel{\‾}{R}}_k=\left[\begin{array}{c}{Q}_k^1\\ Q_k^2\end{array}\right]{\stackrel{\‾}{R}}_k,$ Obtaining dimension is N _k* N _kupper triangular matrix with dimension be (N _k+ N _k) * N _kunitary matrice and use represent unitary matrice front N _koK, represent unitary matrice rear N _koK.

Step 8. by ${\stackrel{\‾}{H}}_k^H=\left[\begin{array}{c}{\left({\tilde{H}}_k^{\mathrm{eff}}\right)}^H\\ \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right]=\left[\begin{array}{c}{Q}_k^1{\stackrel{\‾}{R}}_k\\ Q_k^2{\stackrel{\‾}{R}}_k\end{array}\right]$ Obtain $\sqrt{\mathrm{\α}}{I}_{N_R}=Q_{N_R}\stackrel{\‾}{R},$ So ${\stackrel{\‾}{R}}_k^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^2.$ ${\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}={\left({\stackrel{\‾}{R}}_k^{-1}\right)}^H=\frac{1}{\sqrt{\mathrm{\α}}}{\left(Q_k^2\right)}^H.$

Step 9. couple any user k, will as the second portion of pre-coding matrix, $W_k^b=Q_k^1{\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^1{\left(Q_k^2\right)}^H.$

Concerning all K user, the second portion of its associating pre-coding matrix is $W^b=\mathrm{diag}\{\frac{1}{\sqrt{\mathrm{\α}}}{Q}_1^1{\left(Q_1^2\right)}^H,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_2^1{\left(Q_2^2\right)}^H,...,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_K^1{\left(Q_K^2\right)}^H\}.$

Step 10. is obtained the pre-coding matrix W=W of whole system ^aw ^b.

Through after step 10, known according to lattice reduction (LR) method, receiving terminal only needs y to received signal to carry out simple linear transformation, just can recover transmitted signal s.

The computational complexity of traditional B D method and the inventive method is analyzed the analysis of complexity of concrete as table 1(traditional B D method below), the analysis of complexity of table 2(GZI method), show the analysis of complexity of 3(the inventive method) as shown in:

Table 1

Table 2

Table 3

Below in conjunction with Fig. 2, Fig. 3, Fig. 4, Fig. 5, complexity, volumetric properties and the error rate (BER) of method for precoding of the present invention and traditional B D and GZI method for precoding are carried out to emulation comparison.

Fig. 2 has compared the complexity of traditional B D method for precoding, GZI method for precoding and the inventive method in the situation that of different user number K.System emulation condition is: each user's reception antenna is counted N _k=2 (k=1,2 ..., K), base station transmit antennas is counted N _t=KN _k, the domain of variation K=2:10 of number of users.

Fig. 3 has compared at user's reception antenna and has counted N _kthe complexity of traditional B D method for precoding, GZI method for precoding and the inventive method in different situations.System emulation condition is: number of users K=3, base station transmit antennas is counted N _t=KN _k, the domain of variation N of each user's reception antenna number _k=2:8.Can find out, traditional BD method for precoding, owing to having used the SVD that complexity is higher to decompose, causes this algorithm complex very high, and constantly increases along with the increase of number of users.Based on pseudoinverse, solve the complexity that the GZI method for precoding decomposing with QR has reduced algorithm to a certain extent.And algorithm of the present invention is on the basis of above-mentioned two kinds of methods, in conjunction with extended matrix QR, decompose and lattice reduction (LR) technology, effectively reduce the complexity of algorithm.

Fig. 4 is under the condition of different signal to noise ratios, the volumetric properties comparison diagram of traditional B D method for precoding, GZI method for precoding and the inventive method.Simulated conditions is set as: base station transmit antennas is counted N _t=4, each user's reception antenna number is 2, and number of users K=2 can find out, the inventive method has than traditional BD and the higher power system capacity of GZI method for precoding.

Fig. 5 is under the condition of different signal to noise ratios, the error rate (BER) comparison diagram of traditional B D method for precoding, GZI method for precoding and the inventive method.Simulated conditions is set as: base station transmit antennas is counted N _t=4, each user's reception antenna number is 2, number of users K=2.Can find out, the inventive method has than traditional BD and the more excellent bit error rate performance of GZI method for precoding, has further confirmed the inventive method when reducing computational complexity, can elevator system performance.

Claims (2)

1. a MU-MIMO precoding implementation method, specifically comprises the steps:

The all K of step 1. user's combined channel matrix is $H_S={\left[\begin{array}{cccc}{H}_1^T& H_2^T& ...& H_K^T\end{array}\right]}^T,$ To associating channel matrix H _sexpand: $\stackrel{\‾}{H}=\left[\begin{array}{cc}{H}_S& \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right],$ Wherein, α=N _rσ ²/ P _total, σ ²the variance that represents noise, P _totalrepresent the total transmitted power of down link, expression dimension is N _r* N _runit matrix;

Step 2. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}^H=\stackrel{\‾}{Q}\stackrel{\‾}{R}\left[\begin{array}{c}{Q}_{N_T}\\ Q_{N_R}\end{array}\right]\stackrel{\‾}{R},$ Obtaining dimension is N _r* N _rupper triangular matrix with dimension be (N _t+ N _r) * N _runitary matrice and use represent unitary matrice front N _toK, represent unitary matrice rear N _roK;

Step 3. by ${\stackrel{\‾}{H}}^H=\left[\begin{array}{c}{H}_S^H\\ \sqrt{\mathrm{\α}}{I}_{N_R}\end{array}\right]=\left[\begin{array}{cc}{Q}_{N_T}& \stackrel{\‾}{R}\\ Q_{N_R}& \stackrel{\‾}{R}\end{array}\right]$ Obtain $\sqrt{\mathrm{\α}}{I}_{N_R}=Q_{N_R}\stackrel{\‾}{R},$ And then obtain ${\stackrel{\‾}{R}}^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_{N_R},$ ${\left({\stackrel{\‾}{R}}^H\right)}^{-1}={\left({\stackrel{\‾}{R}}^{-1}\right)}^H=\left[\begin{array}{ccccc}{\hat{R}}_1& ...& {\hat{R}}_k& ...& {\hat{R}}_K\end{array}\right],$ Wherein, be submatrix, dimension is N _t* N _k, N _kit is k user's reception antenna number;

Step 4. is obtained orthogonal basis P _k, the first of the pre-coding matrix of user k is arbitrarily thereby obtain the equivalent channel matrix after precoding for the first time

Concerning all K user, the first of its associating pre-coding matrix is W ^a=[P ₁p ₂p _k], by MU-MIMO channel decomposing, be many independently equivalent SU-MIMO channels, H _sw ^a=diag{H ₁p ₁, H ₂p ₂..., H _kp _k;

The equivalent channel matrix of step 5. couple any user k transposition carry out Ge Jiyue and subtract (LR) operation: transformation matrix T wherein _kfor unimodular matrix, i.e. T _kthe value of middle element is that complex integers and determinant meet det (T _k)=± 1;

Equivalent channel matrix after step 6. pair approximately subtracts expand: ${\stackrel{\‾}{H}}_k=\left[\begin{array}{cc}{\tilde{H}}_k^{\mathrm{eff}}& \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right];$

Step 7. pair extended channel matrices conjugate transpose carry out QR decomposition: ${\stackrel{\‾}{H}}_k^H={\stackrel{\‾}{Q}}_k{\stackrel{\‾}{R}}_k=\left[\begin{array}{c}{Q}_k^1\\ Q_k^2\end{array}\right]{\stackrel{\‾}{R}}_k,$ Obtaining dimension is N _k* N _kupper triangular matrix with dimension be (N _k+ N _k) * N _kunitary matrice and use k represents unitary matrice front N _koK, represent unitary matrice rear N _koK;

Step 8. by ${\stackrel{\‾}{H}}_k^H=\left[\begin{array}{c}{\left({\tilde{H}}_k^{\mathrm{eff}}\right)}^H\\ \sqrt{\mathrm{\α}}{I}_{N_k}\end{array}\right]=\left[\begin{array}{c}{Q}_k^1{\stackrel{\‾}{R}}_k\\ Q_k^2{\stackrel{\‾}{R}}_k\end{array}\right]$ Obtain $\sqrt{\mathrm{\α}}{I}_{N_k}=Q_k^2{\stackrel{\‾}{R}}_k,$ So ${\stackrel{\‾}{R}}_k^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^2,$ ${\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}={\left({\stackrel{\‾}{R}}_k^{-1}\right)}^H=\frac{1}{\sqrt{\mathrm{\α}}}{\left(Q_k^2\right)}^H;$

Step 9. couple any user k, will as the second portion of pre-coding matrix, that is:

$W_k^b=Q_k^1{\left({\stackrel{\‾}{R}}_k^H\right)}^{-1}=\frac{1}{\sqrt{\mathrm{\α}}}{Q}_k^1{\left(Q_k^2\right)}^H;$

Concerning all K user, the second portion of its associating pre-coding matrix is:

$W^b=\mathrm{diag}\{\frac{1}{\sqrt{\mathrm{\α}}}{Q}_1^1{\left(Q_1^2\right)}^H,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_2^1{\left(Q_2^2\right)}^H,...,\frac{1}{\sqrt{\mathrm{\α}}}{Q}_K^1{\left(Q_K^2\right)}^H\};$

Step 10. is obtained the pre-coding matrix W=W of whole system ^aw ^b.

2. according to the said a kind of MU-MIMO precoding implementation method of claim 1, it is characterized in that, step 4 is specifically obtained by Schimidt orthogonalization (GSO) method orthogonal basis P _k.