CN102916786B - Multiple input multiple output pre-coding codebook generating method based on lattice basis reduction - Google Patents

Abstract

The invention discloses a multiple input multiple output pre-coding codebook generating method based on lattice basis reduction, and the method comprises the following steps of: estimating a channel through a receiving terminal to obtain a channel matrix, determining the time L of a lattice basis reduction, the characteristic parameter lambda m of a reduction matrix and the discrete domain of n according to the system performance demand or supported feedback bit number; traversing each obtained characteristic parameter value to obtain a plurality of matrixes, and calculating to select the reduction matrix Di in each reduction and obtaining the value index of the characteristic parameter of the selected Di; performing corresponding lattice basis reduction operation on the channel matrix; gradually executing the reduction until the channel matrix cannot be reduced or executing all L times of reduction; and performing continued product on all Di to obtain a codebook matrix, and forming the quantifying information of the codebook matrix through the value indexes of the characteristic parameters of the Di. The multiple input multiple output pre-coding codebook generating method disclosed by the invention is used for solving the problem that the traditional pre-coding scheme cannot enable transmission data stream to obtain full mark set gain and avoiding the complexity of a sending terminal.

Description

A kind of multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue

Technical field

The present invention relates to wireless communication field, particularly relate to a kind of multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue.

Background technology

Multiple-input and multiple-output (Multiple Input Multiple Output, MIMO) technology, because it can improve power system capacity exponentially, thus becomes one of key technology in future communication systems.Precoding technique is distinctive a kind of adaptive technique in multiaerial system, can select suitable pre-coding matrix according to the symbolic vector being changed to transmission of channel condition, and the speed in conjunction with each data flow of making a start controls, and effectively improves link throughput.

Consider that mimo system has N _troot transmitting antenna and N _rroot reception antenna, transmitting terminal transmission l (l≤min (N _r, N _t)) individual data flow, adopt precoding processing, then system model can be expressed as: y=HPx+n.Wherein, $y={\left[\begin{array}{cccc}{y}_1& y_2& ...& y_{N_r}\end{array}\right]}^T$ It is Received signal strength vector; represent N _r× N _tchannel matrix; X=[x ₁x ₂x _l] ^tsend signal phasor; $n={\left[\begin{array}{cccc}{n}_1& n_2& ...& n_{N_r}\end{array}\right]}^T$ Be the zero-mean white Gauss noise vector on reception antenna, variance is σ ²; it is pre-coding matrix.

Based in the precoding mode of code book, pre-coding matrix P obtains at receiving terminal.Receiving terminal chooses the matrix that makes systematic function optimum according to concrete channel conditions from a matrix codebook set determined, then the index of this matrix in codebook set is fed back to transmitting terminal, as just adopted in this way in LTE system.Here for the thinking that the design of codebook set is mostly based on quantizing the unitary space.

And based in the precoding mode of non-code book, pre-coding matrix obtains at transmitting terminal.More common algorithm has based on singular value decomposition (SVD), based on ZF(ZF)/MMSE(Minimum Mean Square Error) criterion etc., this class algorithm is when transmitting terminal obtains channel condition information, suitable decomposition or process are carried out to channel information matrix, and then obtains corresponding transmitting terminal pre-coding matrix P.

Precoding based on ZF and MMSE criterion realizes simple, but can not obtain diversity gain; And the precoding of decomposing based on SVD is when transmitting multithread, often first-class diversity gain is all different, and between each stream, performance difference is large, needs the selection carrying out coded modulation scheme separately.Based on the precoding mode of code book, existing scheme is mostly be decomposed into theoretical foundation based on SVD, so effect is similar with it.

Ge Jiyue subtracts, and (LR, Lattice Reduction) effectively can improve the singularity of matrix, make each row of matrix orthogonal as much as possible and wait mould, thus reduce matrix respectively arrange between correlation.Channel matrix H can be resolved into the matrix H of a nearly orthogonal by mathematical LR implementation algorithm _redbe the modular matrix T(of complex integers and determinant with a value be ± 1), as shown in Equation 1:

H=H _redT （1）

H _redidentical grid space can be formed with H, but H _redeach row between angle closer to the mould of orthogonal and each row also close to equal, namely with H _redwords channel singular as equivalent channel matrix can be improved, and each at this moment transmitted data flow can obtain identical diversity gain, and performance difference is little, can avoid the selection of the coded modulation scheme of multithread.

Ideally, if transmitting terminal knows the information of ideal communication channel, can using unimodular matrix T corresponding for matrix H as pre-coding matrix to improve the singularity of equivalent channel matrix, now equivalence system model be:

y=βHTx+n=βH _redx+n （2）

Wherein, T is precoding codebook, and β is the normalization coefficient of T, to ensure transmitting power normalizing.

And when transmitting terminal does not know channel matrix H, need receiving terminal will to be carried out LR decomposition by the multi-antenna channel matrix after channel estimating, obtain matrix T, then T to be quantized and to transmitting terminal feedback quantization bit; Transmitting terminal carries out such as formula 2 based on the precoding processing of LR sending signal according to feedback quantization bit generating pre-coding matrix T.So LR is used for codebook precoding design must have simple code book generating mode efficiently.

In the algorithm that Ge Jiyue subtracts, famous LLL algorithm is subject to extensive use because it has binomial low complex degree.By successively about subtracting the singularity improving matrix with swap operation.But LLL algorithm, can not ensure that each walks the singularity about subtracting and exchange rear matrix and is all better than initial matrix, when matrix dimensionality is larger, complexity can be very high.

Therefore, the technical scheme that the multiple-input, multiple-output precoding codebook subtracted based on Ge Jiyue of a kind of improvement of current needs generates solves the problems referred to above.

Summary of the invention

Technical problem to be solved by this invention is to provide a kind of multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue, solve in traditional pre-coding scheme the problem that transmitting data stream can not be made to obtain full diversity gain, the complexity simultaneously also can effectively avoiding transmitting terminal to carry out multi-stream modulation coding mode selection bringing.

In order to solve the problem, the invention provides a kind of multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue, comprising:

After receiving terminal obtains channel matrix H by channel estimating, determine to carry out that Ge Jiyue subtracts about subtracts the feature parameter"λ" that number of times L peace treaty subtracts matrix according to the performance requirement of system or the number of bits of feedback of system support _{m, n}discrete codomain;

After receiving terminal obtains multiple matrix by each characteristic ginseng value that traversal obtains, therefrom select about to subtract matrix D when about subtracting by calculating at every turn _iand obtain selected D _ithe value index of characteristic parameter; Then corresponding Ge Jiyue reducing is carried out to channel matrix H; Progressively perform and about subtract until channel matrix H can not about subtract or execute and all about be kept to for L time only; To all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition codebook matrix T quantitative information, complete the generation of precoding codebook.

Further, said method also can comprise:

After described receiving terminal obtains multiple matrix by each characteristic ginseng value that traversal obtains, therefrom select about to subtract matrix D when about subtracting by calculating at every turn _iand obtain selected D _ithe value index of characteristic parameter; Then corresponding Ge Jiyue reducing is carried out to channel matrix H, progressively perform and about subtract until channel matrix H about can not subtract again or execute whole L time about subtracts; To all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition codebook matrix T quantitative information, complete the step of the generation of precoding codebook, comprising:

Step a, initialization unimodular matrix T=I, l=1;

Step b, receiving terminal get unit matrix I, for each feature parameter"λ" in the set of characteristic parameter value _m,nneutral element position in element traversal I on all off-diagonals about subtracts matrix D to form _iafter, calculate each and about subtract Δ corresponding to matrix _{m, n}if, all Δs _{m, n}be all zero, then make l=l+L; Otherwise select maximum Δ _{m, n}what value was corresponding about subtracts matrix D _maxas the unimodular matrix about subtracted specifically, and perform T=TD _max, l=l+1, wherein Δ _{m, n}be the decrease of the S (H) after an iteration, S (H) is the parameter of orthogonality being used for weighing matrix in Seysen mode;

If step c, receiving terminal judge l>L, then to all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition T quantitative information, complete the generation of precoding codebook; Otherwise, perform step b.

Further, said method also can comprise: described receiving terminal determines the feature parameter"λ" about subtracting matrix _m,ndiscrete codomain step in, the method for exhaustion, greedy algorithm and tree search method are adopted to the selection of (m, n).

Further, said method also can comprise:

Described receiving terminal, for channel matrix H, determines according to the performance requirement of system or the number of bits of feedback of system support the feature parameter"λ" about subtracting matrix _m,nthe step of discrete codomain, comprising:

Described receiving terminal obtains feature parameter"λ" by following formula _m,ncodomain,

${\mathrm{\λ}}_{m,n}=\mathrm{round}\left[0.5(\frac{{\tilde{a}}_n^T{\tilde{a}}_m^{*}}{{\left|\right|{\tilde{a}}_m\left|\right|}^2}-\frac{{\tilde{h}}_n^H{\tilde{h}}_n}{{\left|\right|{\tilde{h}}_n\left|\right|}^2})\right],$ Wherein be m row, be n-th line, t _mthe m row of T.

Further, said method also can comprise:

Described receiving terminal calculates Δ corresponding to each matrix _m,nstep, comprising:

Described receiving terminal obtains Δ by following formula _{m, n},

${\mathrm{\Δ}}_{m,n}=2{\left|\right|{\tilde{h}}_m\left|\right|}^2{\left|\right|{\tilde{a}}_n^T\left|\right|}^2,$

Wherein, described Δ _{m, n}be the decrease of the S (H) after an iteration, be the parameter of orthogonality being used for weighing matrix in Seysen mode, the less representing matrix of S (H) value is more close to orthogonal; Wherein h _nn-th row of representing matrix H, it is matrix n-th line.

Compared with prior art, application the present invention, carries out LR quantization algorithm by the multi-antenna channel matrix estimated receiving terminal, generates the least possible quantization bit, transmitting terminal according to feedback quantization bit to the precoding processing sending signal and carry out based on LR, to improve the performance of mimo system; The present invention utilizes matrix L R to decompose to regard as and multiplely about subtract the feature of matrix product and about subtract the characteristic distributions of matrix element, appropriate design precoding codebook, save the quantization bit of feedback, make system under unusual channel, have better robustness and transmitting terminal complexity can be reduced.

Accompanying drawing explanation

Fig. 1 is the block diagram of codebook precoding of the present invention.

Fig. 2 is the flow chart of the multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue of the present invention.

Fig. 3 is that example of the present invention is at N _t=N _rwhen=2, Ge Jiyue subtracts the middle probability density histogram (being wherein 2 antenna configurations) about subtracting number of times.

Fig. 4 is that example of the present invention is at N _t=N _rwhen=2, LR precoding under different feedback bits with the performance comparison figure (being wherein 2 antenna configurations) of ZF precoding.

Fig. 5 is that example of the present invention is at N _t=N _rwhen=4, Ge Jiyue subtracts the middle probability density histogram (being wherein 4 antenna configurations) about subtracting number of times.

Fig. 6 is that example of the present invention is at N _t=N _rwhen=4, LR precoding under different feedback bits with the performance comparison figure (being wherein 4 antenna configurations) of ZF precoding.

Embodiment

Below in conjunction with the drawings and specific embodiments, the invention will be further described.

First the symbol related in the present invention and notation are described, wherein, () ^hthe complex-conjugate transpose of representing matrix, 1 and 0 represents all 1's matrix and null matrix respectively, || || _frepresent the F norm of getting matrix, represent the maximum integer being not more than a, () _ii-th row of representing matrix, e _irepresentation unit vector, its i-th element is 1, and other element is 0.

The present invention improves mainly for codebook precoding mode, and the operating process of codebook precoding as shown in Figure 1.Core of the present invention is the quantization algorithm of pre-coding matrix (i.e. unimodular matrix T), by analyzing the constitutive characteristic of unimodular matrix and a large amount of emulation, quantization algorithm reasonable in design.Adopt Seysen mode to carry out Ge Jiyue in the present invention to subtract, each step of Seysen mode about subtracts rear matrix singularity and all makes moderate progress, and therefore effectively can be reduced the complexity of LR algorithm by the number of times controlling about to subtract.

Be N for dimension _r× N _tchannel matrix H, seysen mode passes through parameter weigh the orthogonality of matrix.Wherein h _nn-th row of representing matrix H, it is matrix h ^hn-th line.For any H, there is S (H)>=N _t, when and if only if H is unitary matrix, equal sign is set up.The less representing matrix of S (H) value is more close to orthogonal.The principle of Seysen mode is exactly adopt the mode of iteration to obtain base to H continuous renewal based on S (H) parameter until can not little again till.Its idiographic flow is as follows.

In Seysen mode, first initialization: in each iteration, need to find a pair sequence number (m, n) (1≤m, n≤N _tand m ≠ n) and a Gaussian integer λ _{m, n}(namely real part and imaginary part are all the plural numbers of integer) upgrades and T:

$\begin{array}{c}{\tilde{h}}_m\←{\tilde{h}}_m+{\mathrm{\λ}}_{m,n}{\tilde{h}}_n,\\ {\tilde{a}}_n^T\←{\tilde{a}}_n^T-{\mathrm{\λ}}_{m,n}{\tilde{a}}_m^T,\\ t_m\←t_m+{\mathrm{\λ}}_{m,n}{t}_n\end{array}---\left(3\right)$

Wherein be m row, be n-th line, t _mthe m row of T.Through type ${\mathrm{\λ}}_{m,n}=\mathrm{round}\left[0.5(\frac{{\tilde{a}}_n^T{\tilde{a}}_m^{*}}{{\left|\right|{\tilde{a}}_m\left|\right|}^2}-\frac{{\tilde{h}}_n^H{\tilde{h}}_n}{{\left|\right|{\tilde{h}}_n\left|\right|}^2})\right]$ (4) λ can be obtained _{m, n}:

${\mathrm{\λ}}_{m,n}=\mathrm{round}\left[0.5(\frac{{\tilde{a}}_n^T{\tilde{a}}_m^{*}}{{\left|\right|{\tilde{a}}_m\left|\right|}^2}-\frac{{\tilde{h}}_n^H{\tilde{h}}_n}{{\left|\right|{\tilde{h}}_n\left|\right|}^2})\right]---\left(4\right)$

S (H) decrease after an iteration is:

${\mathrm{\Δ}}_{m,n}=2{\left|\right|{\tilde{h}}_m\left|\right|}^2{\left|\right|{\tilde{a}}_n^T\left|\right|}^2---\left(5\right)$

Wherein the parameter of orthogonality being used for weighing matrix in Seysen mode, the less representing matrix of S (H) value more close to orthogonal, parameter it is all the parameter before renewal.Seysen mode constantly performs formula (3) and (4) upgrade matrix until matrix Δ _{m, n}in all elements be all zero.Seysen mode can adopt the methods such as the method for exhaustion, greedy algorithm and tree search to the selection of (m, n).

Main feature of the present invention is: by Seysen mode, is quantized by the unimodular matrix T of channel matrix after Ge Jiyue subtracts.Seysen mode is progressively about subtracted by channel matrix, until about subtracted, about reducing can represent T=D with matrix product at every turn ₁d ₂d _l, L represents and about subtracts number of times.Utilize this feature of Seysen mode, can by restriction Ge Jiyue subtract in the number of times L(that about subtracts namely about subtract matrix D _inumber) save quantization bit.Again to about subtracting matrix D _irationally quantize, quantization bit is fed back to transmitting terminal, carry out pre-encode operation.

Seysen mode is about subtracting in iteration at every turn, needs a Gaussian integer λ _m,nupgrade matrix, therefore matrix T comprises λ by a series of _m,nthe product about subtracting matrix to form, namely often occur once about to subtract and be just multiplied by one and about subtract matrix, λ _m,nvalue uniquely to determine each characteristic parameter about subtracting matrix, and this about subtracts matrix and is:

Feature parameter"λ" _{m, n}be positioned at matrix D _ithe capable and n-th line of m.About subtract matrix and insert feature parameter"λ" by the element position on the off-diagonal of a unit matrix _{m, n}obtain, it is also a unimodular matrix, so T=D ₁d ₂d _l.D _i(1≤i≤L) represents by similar and different feature parameter"λ" _m,nwhat form about subtracts matrix.Based on the formation feature of above unimodular matrix, can be had an appointment by the institute of transmission formation T and subtract matrix D _ireplace transmission matrix T.About subtract the characteristic ginseng value that matrix only needs position on coded quantization off-diagonal, and T matrix needs all elements value on all positions of coded quantization matrix.Such as

$T=\left[\begin{array}{cccc}1& 0& -1& j\\ 0& 1& -1& 0\\ -1& 0& 2& -j\\ 0& 0& 0& 1\end{array}\right]=\left[\begin{array}{cccc}1& & & \\ & 1& & \\ -1& & 1& \\ & & & 1\end{array}\right]\left[\begin{array}{cccc}1& & & j\\ & 1& & \\ & & 1& \\ & & & 1\end{array}\right]\left[\begin{array}{cccc}1& & & \\ & 1& & -1\\ & & 1& \\ & & & 1\end{array}\right]\left[\begin{array}{cccc}1& & -1& \\ & 1& & \\ & & 1& \\ & & & 1\end{array}\right]---\left(7\right)$

The precoding codebook (see formula 7) of mimo system is received: T=D for one 44 ₁d ₂d ₃d ₄.As can be seen from formula 7, the element span of matrix T is that { 1 ,-1,2, j ,-j}, codomain is 5, therefore hypothesis matrix T and matrix D _ithe codomain of element span be 5.Now carry out quantification to unimodular matrix T and have two kinds of modes, a kind of is 16 elements of direct quantization matrix T, and the quantization bit at this moment needed is: individual bit; Another kind is that quantification 4 about subtracts matrix D _iin the characteristic ginseng value of 12 elements except diagonal, the quantization bit at this moment needed is: individual bit.Therefore quantize than quantizing simpler to T direct coding about subtracting matrix coder.The present invention just adopts the coding method quantizing about to subtract matrix to design precoding codebook.

As the above analysis, the bit number required for unimodular matrix T quantification is:

$B=L\×{\log }_2^{(N_t(N_t-1)\×C)}---\left(8\right)$

L about subtracts matrix D _inumber (namely about subtracting number of times), C is feature parameter"λ" _m,nthe codomain size of value, N _tit is number of transmit antennas.After receiving terminal quantizes unimodular matrix, quantization bit is fed back to transmitting terminal (namely receiving terminal carries out Ge Jiyue and subtracts, and after obtaining pre-coding matrix, re-quantization feeds back to transmitting terminal), transmitting terminal carries out pre-encode operation.Obviously, the quantization bit of unimodular matrix T is more, and close to real Ge Jiyue subtractive energy, but the number of bits of feedback can supported when system is prescribed a time limit, and just needs rationally to quantize code book.As can be seen from formula 8, about can be subtracted the codomain size C of number of times L and characteristic parameter by restriction, effectively save feedback bits.

In practice, matlab software can be used to carry out subtracting based on the Ge Jiyue of Seysen mode, then to about subtracting the feature parameter"λ" produced in process to a large amount of random channel matrix produced _m,nvalue is added up, and statistics is in table 1 and table 2.As can be seen from these two tables, λ _{m, n}be worth stepped distribution, the λ that modulus value is equal _m,nprobability density is close, a formation ladder, and the λ that modulus value is less _m,nprobability density is larger.Under 2 antenna configurations, the cumulative frequency of first ladder is the cumulative frequency of 0.6048, first ladder and second ladder is 0.8304.Under 4 antenna configurations, the cumulative frequency of first ladder is the cumulative frequency of 0.9352, first ladder and second ladder is 0.9991.Therefore the characteristic ginseng value integrated distribution produced in iteration each time in Seysen mode is in a very little scope.For this feature, the unimodular matrix that transmission restriction about subtracts matrix number and characteristic parameter span just can obtain good Ge Jiyue subtractive energy.

As shown in Figure 2, the multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue of the present invention, comprising:

After step 210, receiving terminal obtain channel matrix H by channel estimating, determine to carry out about subtracting number of times L and uniquely can determine each discrete codomain about subtracting the characteristic parameter of matrix required for Ge Jiyue reducing according to the performance requirement of system or the number of bits of feedback of system support

Particularly, determine according to the performance requirement of system or the number of bits of feedback of system support, from adding up the feature parameter"λ" obtained _m,nvalue peace treaty subtracts (see table 1, table 2, Fig. 3, Fig. 5) in number of times, determines that size is the feature parameter"λ" of C _m,nnamely the number L(that value set E peace treaty subtracts matrix D performs the number of times about subtracted), wherein m, n and L are natural number, 1≤m, n≤N _tand m ≠ n; .

Step 220, receiving terminal are undertaken after traversal obtains multiple matrix by each characteristic ginseng value chosen, and therefrom select about to subtract matrix D when about subtracting by calculating at every turn _iand obtain selected D _ithe value index of characteristic parameter; Then corresponding Ge Jiyue reducing is carried out to channel matrix, progressively perform and about subtract until meet exit criteria;

Wherein exit criteria is: matrix about can not subtract again or execute whole L time and about subtract.

Particularly, comprising: step 2201, first, initialization unimodular matrix T=I, l=1;

Step 2202, get unit matrix I, for each λ in characteristic parameter value set E _m,nneutral element position in element traversal I on all off-diagonals about subtracts matrix D to form _i, then calculate the Δ that each matrix is corresponding _{m, n}if, all Δs _{m, n}be all zero, then make l=l+L; Otherwise select maximum Δ _{m, n}what value was corresponding about subtracts matrix D _maxas the unimodular matrix about subtracted specifically, and perform T=TD _max, l=l+1;

If step 2202 l>L, perform step 230; Otherwise perform step 2202.

Step 230, to all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition T quantitative information, complete the generation of precoding codebook.

Particularly, the quantitative information of codebook matrix T is fed back to transmitting terminal by receiving terminal, and transmitting terminal, according to quantitative information generating pre-coding matrix, carries out pre-encode operation.

Below by l-G simulation test, the present invention will be described.

First for 2 antenna configurations, table 1 about subtracts feature parameter"λ" in matrix _m,nvalue distribution situation.In order to contrast the performance under different characteristic parameter codomain, setting two kinds of codomains and emulating: E={i ,-i, 1 ,-1}, E={i ,-i, 1 ,-1 ,-1-i, 1-i, 1+i ,-1+i}, codomain size is respectively C=4 and C=8.As can be seen from the figure, about subtract the maximum probability (0.6) that number of times is 1, about subtract the probability next (0.22) that number of times is 2, therefore set two kinds and about subtract matrix number: L=1 and L=2.In this several situation, the bit number that quantizing unimodular matrix T needs is respectively: C=4, L=1, B ₁=log ₂ ^{(2 × 4)}=3 bits; C=8, L=1, B ₂=log ₂ ^{(2 × 8)}=4 bits; C=4, L=2, B ₃=2 × log ₂ ^{(2 × 4)}=6 bits; C=8, L=2, B ₄=2 × log ₂ ^{(2 × 8)}=8 bits.

Under table 1:2 antenna configuration, Ge Jiyue subtracts the feature parameter"λ" of middle unimodular matrix _{m, n}value frequency statistics table

Table 1 is that example of the present invention is at N _t=N _rwhen=2, Ge Jiyue subtracts the feature parameter"λ" of middle unimodular matrix _m,nthe probability density (PDF) of value and the statistical form of accumulated probability density (CDF).

Figure is the performance comparison figure of above-mentioned four kinds of LR precodings and ZF precoding under 2 antenna configurations.As can be seen from the figure, obvious quantization bit is more, and the performance of LR precoding is better.When about subtracting matrix number L and being identical, the performance of C=4 and C=8 is more or less the same, and therefore selects the characteristic parameter codomain E={i of C=4 ,-i, and 1 ,-1} to save feedback bits.C=4 is 10 in the error rate ^-3time, about subtract the performance boost of the Performance Ratio L=2 of matrix number L=1 about 2dB, feedback bits adds 3 bits simultaneously.2dB performance gain is very considerable, therefore should consider that the feedback bits of increase by 3 bit is to obtain the performance gain of 2dB, namely selects about to subtract matrix number L=2.

At N _t=N _runder the antenna configuration of=4, feature parameter"λ" _m,nvalue reaches 0.9352 in the cumulative probability density of first ladder, and major part is all distributed in (see table 2) in the first ladder.Again because find in the result of 2 antenna configurations, characteristic parameter codomain size is that the performance of C=8 and C=4 is more or less the same, and therefore under 4 antenna configurations, the codomain only setting a kind of characteristic parameter emulates: E={i, i, 1 ,-1}, C=4.As can be seen from the figure, about subtract the maximum probability (0.18) that number of times is 5, about subtract the probability next (0.16) that number of times is 4, therefore setting the number about subtracting matrix is L=4 and L=5.Under both of these case, the quantification of unimodular matrix T needs B respectively ₁=4 × log ₂ ^{(12 × 4)}=23 bits and B ₂=5 × log ₂ ^{(12 × 4)}=28 bits.

Under table 2:4 antenna configuration, Ge Jiyue subtracts the feature parameter"λ" of middle unimodular matrix _m,nvalue frequency statistics table

Table 2 is that example of the present invention is at N _t=N _rwhen=4, Ge Jiyue subtracts the feature parameter"λ" of middle unimodular matrix _{m, n}the probability density (PDF) of value and the statistical form of accumulated probability density (CDF).

Figure is the performance comparison figure of two kinds of LR precodings and ZF precoding.As seen from the figure, the performance about subtracting the Performance Ratio L=4 of matrix number L=5 slightly well, is 10 in the error rate ^-3time performance boost be about 1dB, but feedback bits adds 5 bits.Therefore, for saving feedback bits, can select about to subtract matrix number L=4.For different systems, available above-mentioned quantization method designs best L value and C value, quantizes unimodular matrix.

The above; be only the present invention's preferably embodiment, but protection scope of the present invention is not limited thereto, any people being familiar with this technology is in the technical scope disclosed by the present invention; the change that can expect easily or replacement, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection range of claim.

Claims (3)

1. the multiple-input, multiple-output precoding codebook generation method subtracted based on Ge Jiyue, is characterized in that, comprising:

After receiving terminal obtains channel matrix H by channel estimating, according to the number of bits of feedback that the performance requirement of system or system are supported, determine to carry out that Ge Jiyue subtracts about subtracts the feature parameter"λ" that number of times L peace treaty subtracts matrix _m,ndiscrete codomain;

After receiving terminal obtains multiple matrix by each characteristic ginseng value that traversal obtains, therefrom select about to subtract matrix D when about subtracting by calculating at every turn _iand obtain selected D _ithe value index of characteristic parameter; Then corresponding Ge Jiyue reducing is carried out to channel matrix H; Progressively perform and about subtract until channel matrix H can not about subtract or execute and all about be kept to for L time only; To all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition codebook matrix T quantitative information, complete the generation of precoding codebook, wherein,

Step a, initialization unimodular matrix T=I, l=1;

Step b, receiving terminal get unit matrix I, for each feature parameter"λ" in the set of characteristic parameter value _m,nneutral element position in element traversal I on all off-diagonals about subtracts matrix D to form _iafter, calculate each and about subtract △ corresponding to matrix _m,nif, all △ _m,nbe all zero, then make l=l+L; Otherwise select maximum △ _m,nwhat value was corresponding about subtracts matrix D _maxas the unimodular matrix about subtracted specifically, and perform T=TD _max, $\tilde{H}=\tilde{H}{D}_{\max },$ L=l+1, wherein ${\mathrm{\Δ}}_{m,n}=2{\left|\right|{\tilde{h}}_m\left|\right|}^2{\left|\right|{\tilde{a}}_n^T\left|\right|}^2$ Be the decrease of the S (H) after an iteration, it is the parameter of orthogonality being used for weighing matrix in Seysen mode; Wherein, the less representing matrix of S (H) value more close to orthogonal, h _nn-th row of representing matrix H, it is matrix n-th line, be m row, be n-th line, N _tfor total columns of matrix H;

If step c, receiving terminal judge l>L, then to all D _icarry out continued product and obtain codebook matrix T, and use D _icharacteristic parameter value index composition T quantitative information, complete the generation of precoding codebook; Otherwise, perform step b.

2. the method for claim 1, is characterized in that,

Comprise further: described receiving terminal determines about to subtract the feature parameter"λ" of matrix _m,ndiscrete codomain step in, the method for exhaustion, greedy algorithm or tree search method are adopted to the selection of (m, n).

3. the method for claim 1, is characterized in that,

Described receiving terminal, for channel matrix H, determines according to the performance requirement of system or the number of bits of feedback of system support the feature parameter"λ" about subtracting matrix _m,nthe step of discrete codomain, comprising:

Described receiving terminal obtains feature parameter"λ" by following formula _m,ncodomain,

${\mathrm{\λ}}_{m,n}=\mathrm{round}\left[0.5(\frac{{\tilde{a}}_n^T{\tilde{a}}_m^{*}}{{\left|\right|{\tilde{a}}_m\left|\right|}^2}-\frac{{\tilde{h}}_n^H{\tilde{h}}_m}{{\left|\right|{\tilde{h}}_n\left|\right|}^2})\right],$ Wherein be m row, be n-th line.