CN103825679B - 3D (3-Dimensional) MU-MIMO precoding method based on pseudo codebooks - Google Patents

Abstract

The invention discloses a 3D (3-Dimensional) MU-MIMO precoding method based on pseudo codebooks. A uniform surface antenna array is adopted at a base station end, and DFT codebooks are stored at the base station end and a user end simultaneously; optimal precoding vectors are selected in a horizontal dimension and a vertical dimension respectively according to a 3D channel matrix at the user end, reference numbers are fed back to the base station end, the optimal precoding vectors in the horizontal dimension and the vertical dimension are recovered by a base station, and sub-optimal precoding vectors in the horizontal dimension and the vertical dimension with the maximum similarity with the optimal precoding vectors in the horizontal dimension and the vertical dimension are computed respectively. Point multiplication conversion is carried out respectively, and new optimal precoding vectors in the horizontal dimension and the vertical dimension are constructed, thus a precoding matrix is generated. The adopted thought of pseudo codebooks can be matched with the current channel information better, and the system performance is obviously promoted under the condition that the system feedback quantity is not increased.

Description

A kind of 3d mu-mimo method for precoding based on pseudo- code book

Technical field

The present invention relates to field of wireless communications, relate generally to lte-advanced technical field.

Background technology

With the fast development of mobile Internet, business datum amount sharp increase, bring greatly to prior wireless network Adjustment pressure, for communication system, increase power system capacity, reduce interference be still most important developing goal.Especially It is the core technology mimo technology as lte-advanced and lte air interface, not increasing bandwidth and can send out Improve power system capacity, by using multipath transmisstion characteristic and the random fading of many antennas and channel in the case of penetrating power Characteristic, can effectively improve transmission quality, and promoting transmission rate.

Under actual practical environment, the embodiment of mimo technical advantage depend heavilys on mimo channel model characteristic.Base In this, more of paramount importance is exactly to set up the channel model that effectively can reflect true mimo working environment.However, it is traditional Mimo channel model mostly set up in two-dimentional (2dimension, 2d) plane, 2d mimo does not make full use of well Space resources, if simple, from the point of view of electromagnetic wave propagation mechanism, this model only considers the horizontal transmission of spatial electromagnetic ripple, for The vertical dimension of spatial electromagnetic ripple is propagated and is left intact, therefore, this channel model imperfection.For this reason, academia includes work Industry has carried out the mimo Channel Modeling work based on 3d in a large number.3d mimo channel model requires to consider azimuth simultaneously and bows The elevation angle, considers the horizontal and vertical propagation of spatial electromagnetic ripple, therefore, channel model is more accurate simultaneously.In view of 3d The advantage of mimo technology, industrial quarters includes the company such as high pass, A Lang, docomo, Nuo Xi and has specially set up the advanced sky towards 4g Middle interface (advanced radio interface technologies for4g systems, artist4g) working group, should Working group points out: the 3d mimo technology based on active array will improve Cell Edge User as future mobile communication system and gulp down The key technology of the amount of telling data transfer rate.Meanwhile, 3gpp, in order to meet the technical need of lte-advanced, starts The technical work of lte r10 version, i.e. lte-advanced.It is complete that the standard of lte-advanced r10 works in March, 2011 Become, with respect to other versions of lte, in r10, introduce more new technique, additionally, the evolution version r11 of lte-advanced r10 Complete with December, 2012.3gpp will start to start the technical specification standard work of lte-advanced r12 from the end of the year in 2012 Make, formally using 3d mimo technology as lte-advanced r12 technical specification core.Taking this as an opportunity, has attracted domestic Substantial amounts of scholar, engineer are added in 3d mimo key technology and standardized research work outward.

In lte-advanced system, include selection of codebook design, code book etc. based on the precoding technique of 3d channel. In the design of code book, dft code book, random code book, Jim Glassman code book etc. belong to most basic precoding codebook.The selection of code book The i.e. selection to optimum precoding vectors, is based on different performance indications criterions.Common optimum precoding vectors are chosen accurate Then include minimum singular value criterion, maximum capacity criterion etc..Under the pre-coding system based on code book, will design first Code book be stored in transmitting terminal and receiving terminal, receiving terminal receives signal, obtains current channel condition information by channel estimation (channel state information, csi), then utilizes csi, in stored local code book, based on precoding choosing Take criterion to select optimum precoding vectors, and the sequence number corresponding in the codebook of optimum precoding vectors is fed back to base station End, base station end recovers optimum precoding vectors by Receive sequence number, and builds pre-coding matrix, and input data is carried out at precoding Launch after reason.

In 3d mimo multi-user system, increasing with number of users, user distribution zones of different in cell, including Center of housing estate and cell edge.Traditional precoding technique with only the channel information of level dimension, and cannot tie up vertical User is made a distinction, this is for the utilization of information and imperfection.Therefore, use to lift center of housing estate and cell edge simultaneously The capacity at family and user performance, need simultaneously using level dimension with vertical dimension channel information to realize 3d mimo precoding.

Content of the invention

The present invention is directed to the deficiency of said method, proposes a kind of 3d mu-mimo method for precoding based on pseudo- code book. In lte-advanced system, according to the feature of 3d mimo multi-user system, choose suitable dft code book and be stored concurrently in base station With user side, base station adopts uniform surface aerial array；Receiving terminal, according to 3d channel matrix, is chosen in level dimension and vertical dimension respectively Optimum precoding vectors, and level dimension is fed back as value of feedback with vertical dimension optimum precoding vectors sequence number in corresponding code book Go back to base station, base station goes out level dimension and vertical dimension optimum precoding vectors according to the SN recovery receiving, and calculates and level Dimension optimum precoding vectors similarity highest level dimension suboptimum precoding vectors, and the optimum precoding vectors phase with vertical dimension Like degree highest vertical dimension suboptimum precoding vectors, finally using pseudo- code book thought to level dimension and vertical dimension optimum precoding Vector converts again, and then generates pre-coding matrix.The present invention can be utilized respectively water under conditions of using identical feedback quantity Channel information on mean latitude degree and vertical dimensions, improves the overall performance of system.The method particularly as follows:

A kind of 3d mu-mimo method for precoding based on pseudo- code book, code book is stored concurrently in base station end and user side； User side, according to 3d channel matrix, is chosen optimum precoding vectors in horizontal dimensions with vertical dimensions respectively, and is corresponded to Call number feeds back to base station end, and base station end recovers level dimension and vertical dimension optimum precoding vectors according to call number, counts respectively Calculate and level dimension and vertical dimension optimum precoding vectors similarity highest level dimension and vertical dimension suboptimum precoding vectors, point Do not carry out dot product conversion, build new level dimension and vertical dimension optimum precoding vectors, and then form pre-coding matrix to transmission Data carries out pretreatment.

User side is prelisted code vector based on maximizing minimum singular value criterion and carry out optimum in horizontal dimensions and vertical dimensions The selection of amount, specifically includes: according to equivalent channel h w_iMinimum non-zero singular value λ_min{h f_x, traversal code book is according to formula: $f_{h\_\mathrm{opt}}=\arg \underset{1\≤l\≤l}{\max }{\mathrm{\λ}}_{\min }\left\{\begin{array}{cc}h& f_i\end{array}\right\},f_{v\_\mathrm{opt}}=\arg \underset{1\≤l\≤l}{\max }{\mathrm{\λ}}_{\min }\left\{\begin{array}{cc}h& f_j\end{array}\right\}$ (l is number of codewords in code book) selects optimum code word f_{h_opt}And f_{v_opt}As horizontal dimensions and vertical dimensions optimum precoding vectors.Select the optimum precoding vectors with level dimension f_h-optSimilarity highest precoding vectors tie up suboptimum precoding vectors f ' as level_h-opt, select to prelist with vertical dimension optimum Code vector f_v-optSimilarity highest precoding vectors are as vertical dimension suboptimum precoding vectors f '_v-opt, according to formula cos θ= f_x-opt/f′_x-opt=| (f_x-_opt)^hf′_x-opt| (x represents horizontal dimensions or vertical dimensions, and that is, for h or v, x represents that level is tieed up respectively Degree or vertical dimensions, θ is optimum precoding vectors f_x-optWith suboptimum precoding vectors f '_x-optBetween angle, (f_x-opt)^hFor f_x-optConjugate transpose) know, when θ get over hour f_x-optWith f '_x-optSimilarity is higher, and precoding vectors when selecting θ minimum are made For suboptimum precoding vectors, you can obtain level dimension suboptimum precoding vectors f '_h-optWith vertical dimension suboptimum precoding vectors f′_v-opt.The optimum precoding vectors of level dimension and suboptimum precoding vectors (f_h-optf′_h-opt) and vertical dimension optimum prelist code vector Amount and suboptimum precoding vectors (f_v-optf′_v-opt) become two phases it is to be understood that replicating the dft storing in system code book Same code book 1 and code book 2, in horizontal dimensions, f_h-optCan regard as and obtain optimum precoding vectors, f ' from code book 1_h-optCan Regard as and obtain suboptimum precoding vectors, in the vertical dimension., f from code book 2_v-optCan regard as and obtain optimum from code book 1 and prelist Code vector, f '_v-optCan regard as and obtain suboptimum precoding vectors from code book 2.Again to f_h-optWith f '_h-optIt is dot product conversion: f_h =f_h-opt.*f′_h-optThe f of gained_hReplace f_h-optBecome the new optimum precoding vectors of level dimension；To f_v-optWith f '_v-optDo a little Take advantage of conversion: f_v=f_v-opt.*f′_v-opt, the f that obtains_vReplace f_v-optAs new vertical dimension optimum precoding vectors.

User side is based on and maximizes minimum singular value criterion calculated level dimension and vertical dimension optimum precoding vectors.User side Feedback information comprises level dimension value of feedback and vertical dimension value of feedback, and wherein level dimension value of feedback is coupling present level dimension channel letter The optimum precoding vectors of breath call number in the codebook, vertical dimension value of feedback is the optimum that coupling current vertical ties up channel information Call number in corresponding code book for the precoding vectors.

In the present invention, base station end adopts uniform surface aerial array, and user side is according to 3d channel matrix respectively in horizontal dimensions With the selection that optimum precoding vectors are carried out on vertical dimensions, then two vector manipulative indexings number are fed back to base station end, base Stand end according to call number recovery level dimension and vertical dimension optimum precoding vectors, and using puppet code book thought, level is tieed up optimum Precoding vectors and level dimension suboptimum precoding vectors carry out dot product conversion, simultaneously to vertical dimension optimum precoding vectors with vertical Dimension suboptimum precoding vectors are also carried out dot product conversion, construct new level dimension and vertical dimension optimum precoding vectors, generate whole Body pre-coding matrix.Pseudo- code book thought employed in the present invention can make institute's structure in the case of not increasing system feedback amount The pre-coding matrix built can preferably state current channel letter by being utilized respectively level dimension and vertical dimension upper signal channel information Breath, improves systematic function simultaneously.

Brief description

Fig. 1 is the system block diagram of 3d mimo method for precoding under multi-user proposed by the present invention；

Fig. 2 is the precoding basic operation flow chart based on code book；

Fig. 3 is 3d pre-coding scheme operational flowchart in the present invention；

Fig. 4 is pseudo- code book thought operation chart in the present invention；

Fig. 5 be winner channel under " a1 " scene method for precoding of the present invention with dft codebook precoding in identical feedback quantity Lower bit error rate performance comparison diagram.

Specific embodiment

Fig. 1 show the 3d mu-mimo method for precoding system block diagram based on pseudo- code book for the present invention, adopts in base station end Uniform surface antenna array structure, the wherein line number of uniform surface aerial array are represented with n, the columns m of uniform surface aerial array Represent that user side receives sky it is ensured that each transmitting antenna can process the information in horizontal dimensions and vertical dimensions simultaneously Line number is r.The signal that i-th user receives can be expressed as:

$y_i=h_i{w}_i{x}_i+h_i\underset{j=1,j\&notequal;i}{\overset{i}{\mathrm{\σ}}}{w}_j{x}_j+n_i,i=\mathrm{1,2},...,i$

Wherein h_iw_ix_iThe useful signal receiving for receiving terminal,Believe from other for what i user was subject to Number interference, n_iRepresent noise signal, be additive white Gaussian noise, obey cn (0, n₀) distribution.

User side obtains 3d channel information matrix h by channel estimation_n×m×r, carry out precoding selection, distinguish in the codebook Select the optimum precoding vectors of vertical dimension and level dimension, and two vector manipulative indexings number fed back base station end, base station end according to The call number receiving recovers level dimension and vertical dimension optimum precoding vectors, selects to tie up optimum precoding vectors phase with level Tie up suboptimum precoding vectors like degree highest vector as level, select and vertical dimension optimum precoding vectors similarity highest Precoding vectors, as vertical dimension suboptimum precoding vectors, in horizontal dimensions, are tieed up optimum precoding vectors and are tieed up with level to level Suboptimum precoding vectors carry out dot product conversion, draw the new optimum precoding vectors of level dimension, in the same manner in the vertical dimension., to vertical Straight optimum precoding vectors of tieing up carry out dot product conversion with vertical dimension suboptimum precoding vectors, draw new vertical dimension optimum precoding Vector.And then generation pre-coding matrix carries out precoding processing to input data.

Base station end, according to received feedback information, recovers level dimension and vertical dimension optimum precoding vectors respectively, Utilize pseudo- code book thought, the optimum precoding vectors of level dimension and suboptimum precoding vectors (f simultaneously_h-optf′_h-opt) and vertical dimension Optimum precoding vectors and suboptimum precoding vectors (f_v-optf′_v-opt) it is to be understood that can be by the dft code book stored by base station It is copied into two content identical code books 1 and code book 2, in horizontal dimensions, f '_h-optCan regard as by the level selecting in code book 1 The optimum precoding vectors of dimension, f '_h-optThe optimum precoding vectors of level dimension selecting from code book 2 can be regarded as；In the vertical dimension., f_v-optThe vertical dimension optimum precoding vectors obtaining from code book 1, f ' can be regarded as_v-optObtain from code book 2 vertical can be regarded as Dimension suboptimum precoding vectors.Again to f_h-optWith f '_h-optIt is dot product conversion: f_h=f_h-opt.*f′_h-optThe f of gained_hReplace f_h-optBecome For the new optimum precoding vectors of level dimension, precoding vectors f optimum to vertical dimension_v-optWith vertical dimension suboptimum precoding vectors f′_v-optDo dot product conversion and obtain new vertical dimension precoding vectors f_v.The new optimum precoding vectors f of level dimension_hHang down with new Straight dimension precoding vectors f_vThe level of being dimension pre-coding matrix w_hWith vertical dimension pre-coding matrix w_v, and then generate pre-coding matrix, Precoding processing is carried out to input data.

Fig. 2 is the precoding basic operation flow chart based on code book, in lte-advanced system, input data, and enter Row channel coding modulation, code book is stored concurrently in base station end and user side, and user side carries out precoding processing, user side according to work as Front csi, travels through all code words in the codebook, selects the code word mating current csi the most as optimum precoding vectors, by sky Between be multiplexed by its in the codebook corresponding call number feed back to base station end, base station end carries out channel estimation and calculates feedback information, Base station end recovers optimum precoding vectors according to the call number receiving, demodulation, and then generates pre-coding matrix to input number According to carrying out precoding processing.

Fig. 3 show 3d pre-coding scheme operational flowchart in the present invention, implements step as follows:

Step 21, base station adopts uniform surface aerial array, not only can receive level dimension channel information, and for vertical Dimension channel information is also fully used.

Step 22, the structure of code book, common precoding codebook has dft code book, random code book, Jim Glassman code book etc., Because random code book, Jim Glassman code book need to complete by offline emulation or search, complexity is higher, simultaneously dft code Originally be also precoding codebook specified in 3gpp standard, therefore the present invention can using dft code book as precoding codebook, below with As a example dft code book, the enforcement of the present invention is described in further detail, as follows in a kind of applicable dft code book:

According to above-mentioned dft code book, determine a suitable dft code book such as: f=[f₁,f₂,...,f_l], l represents code in code book Number of words mesh, and it is stored in transmitting-receiving two-end, as level dimension code book and vertical dimension code book.

Step 23, user side is based on and maximizes minimum singular value criterion, in level dimension code book with vertical dimension code book respectively Select corresponding optimum precoding vectors, as a example rotating optimum precoding vectors below based on minimum singular value criterion, specifically It is bright as follows: traversal codebook selecting optimum code word is optimum precoding vectors,

$f_{h\_\mathrm{opt}}=\arg \underset{1\≤l\≤l}{\max }{\mathrm{\λ}}_{\min }\left\{\begin{array}{cc}h& f_i\end{array}\right\}$

$f_{v\_\mathrm{opt}}=\arg \underset{1\≤l\≤l}{\max }{\mathrm{\λ}}_{\min }\left\{\begin{array}{cc}h& f_j\end{array}\right\}$

Wherein, f_iCode word in expression level dimension code book, f_jRepresent code word in vertical dimension code book, h f_iWith h f_jRepresent respectively Level dimension and vertical dimension equivalent channel, λ_min{h f_iAnd λ_min{h f_jRepresent h f respectively_iWith h f_jMinimum non-zero singular value, f_{h_opt}With f_{v_opt}The optimum precoding vectors therefrom selecting after representing the corresponding code book of traversal.Again level dimension optimum is prelisted code vector Amount f_{h_opt}With vertical dimension optimum precoding vectors f_{v_opt}Call number in corresponding code book feeds back to base station as feedback information End.

Step 24, base station end, according to the feedback information receiving, according to dft code book, recovers corresponding level dimension optimum Precoding vectors f_h-optWith vertical dimension optimum precoding vectors f_v-opt, and then find optimum with level dimension from dft code book respectively Precoding vectors f_h-optWith vertical dimension optimum precoding vectors f_v-optDegree of approximation highest precoding vectors tie up suboptimum as level Precoding vectors f '_h-optWith vertical dimension suboptimum precoding vectors f '_v-opt, degree of approximation highest determines optimum precoding vectors f_x-optWith suboptimum precoding vectors f '_x-optBetween angle theta minimum, by cos θ=f_x-opt/f′_x-opt=| (f_x-opt)^hf′_x-opt| (x Represent horizontal dimensions or vertical dimensions, that is, x represents horizontal dimensions or vertical dimensions respectively for h or v) understand, prelist with optimum Code vector f_x-optThe minimum precoding vectors of angle theta make | (f_x-opt)^hf′_x-opt| maximum, that is, horizontal peacekeeping vertical dimension suboptimum is pre- Coded vector can be obtained by below equation:

$f_{h-\mathrm{opt}}^{\′}=\underset{1\≤i\≤c,i\&notequal;h-\mathrm{opt}}{\arg \max }\left\{\right|{\left(f_{h-\mathrm{opt}}\right)}^h{f}_i\left|\right\}$

$f_{v-\mathrm{opt}}^{\′}=\underset{1\≤j\≤c,j\&notequal;v-\mathrm{opt}}{\arg \max }\left\{\right|{\left(f_{v-\mathrm{opt}}\right)}^h{f}_j\left|\right\}$

Wherein, c represents number of codewords in corresponding code book, f_i, f_jRepresent code word (i.e. f in corresponding code book_iRepresent i-th in code book OK, f_jRepresent jth row in code book), f '_h-optMake for traversal code book | (f_h-opt)^hf_i| maximum level dimension suboptimum prelists code vector Amount, f '_v-optMake for traversal code book | (f_v-opt)^hf_j| maximum vertical dimension suboptimum precoding vectors.The physics meaning of above two formulas Justice is, in all row of dft code book, selects to make the maximum row (column) of target function value secondary as level dimension (vertical dimension) Excellent precoding vectors f '_x-opt.Obtain level dimension suboptimum precoding vectors f ' according to above-mentioned computational methods_h-optWith vertical dimension suboptimum Precoding vectors f '_v-opt.

Step 25, according to the level dimension suboptimum precoding vectors f ' obtaining_h-optWith vertical dimension suboptimum precoding vectors f′_v-optLevel is tieed up with optimum precoding vectors f_h-optWith vertical dimension optimum precoding vectors f_v-optConvert, according to Fig. 4 Pseudo- code book map function schematic diagram, the optimum precoding vectors of level dimension and suboptimum precoding vectors (f_h-optf′_h-opt) and hang down Straight dimension optimum precoding vectors and suboptimum precoding vectors (f_v-optf′_v-opt) it is to be understood that can will be stored in system Dft code book is copied into two identical code books 1 and code book 2, in horizontal dimensions, f_h-optCan regard as and obtain from code book 1 The optimum precoding vectors of level dimension, f '_h-optThe level dimension suboptimum precoding vectors obtained from code book 2 can be regarded as, vertical In dimension, f_v-optThe vertical dimension optimum precoding vectors obtained from code book 1, f ' can be regarded as_v-optCan regard as from code book 2 In obtain vertical dimension suboptimum precoding vectors.Again to f_h-optWith f '_h-optIt is dot product conversion a: f_h=f_h-opt.*f′_h-optObtained F_hReplace f_h-optBecome the optimum precoding vectors of level dimension.Using same method, class is made to vertical dimension optimum precoding vectors Convert (f like dot product_v=f_v-opt.*f′_v-opt), the f obtaining_vReplace f_v-optAs vertical dimension optimum precoding vectors.New level The optimum precoding vectors f of dimension_hWith vertical dimension optimum precoding vectors f_vThe level of being dimension pre-coding matrix w_hWith vertical dimension precoding Matrix and w_v, by w_hWith w_vDo dot product extended arithmetic, generation 3d pre-coding matrix w:

W=[w'_h；w'_h；...w'_h].×[w_v；w_v；...w_v]

Finally pretreatment is carried out to transmission data.

Fig. 5 show under " a1 " (indoor office) scene method for precoding of the present invention with dft codebook precoding in phase With bit error rate performance comparison diagram under feedback quantity.Emulation is using " a1 " (the indoor office/residential) under winner Scene, under dft codebook precoding scheme, base station and user side store a code book being made up of dft matrix, and base station end is according to one As pre-encode operation based on code book to process, do not do any conversion.As shown in Figure 5, compared with dft codebook precoding scheme, Encoding scheme of the present invention has obvious advantage in error rate of system performance.

3d mu-mimo method for precoding proposed by the present invention, respectively to level dimension channel information and vertical dimension channel information Be used, and according to pseudo- code book thought, respectively level is tieed up optimum precoding vectors and level dimension suboptimum precoding vectors with And vertical dimension optimum precoding vectors carry out dot product conversion with vertical dimension suboptimum precoding vectors, final structure entirety precoding square Battle array, it has higher matching degree for current 3d mimo channel information state, and has bright for the overall performance of system Aobvious lifting.

Claims (2)

1. a kind of 3d mu-mimo method for precoding based on pseudo- code book is it is characterised in that be stored concurrently in base station end by code book And user side；User side, according to 3d channel matrix, chooses optimum precoding vectors respectively in horizontal dimensions with vertical dimensions, will Horizontal dimensions optimum precoding vectors feed back to base station with vertical dimensions optimum call number in corresponding code book for the precoding vectors End, base station end recovers level dimension and vertical dimension optimum precoding vectors according to call number, calculates respectively and ties up with level and hang down Straight optimum precoding vectors similarity highest level of tieing up is tieed up and vertical dimension suboptimum precoding vectors, carries out dot product conversion respectively, Build new level dimension and vertical dimension optimum precoding vectors, and then obtain pre-coding matrix and carry out pretreatment to sending data；

Described calculate respectively and level dimension and vertical dimension optimum precoding vectors similarity highest level dimension and vertical dimension time Excellent precoding vectors are: find precoding vectors f optimum with level dimension from code book respectively_h-optPrelist code vector with vertical dimension optimum Amount f_v-optDegree of approximation highest precoding vectors tie up suboptimum precoding vectors f ' as level_h-optWith vertical dimension suboptimum precoding Vector f '_v-opt, degree of approximation highest determines optimum precoding vectors f_x-optWith suboptimum precoding vectors f '_x-optBetween angle theta Little, by cos θ=f_x-opt/f′_x-opt=| (f_x-opt)^hf′_x-opt| understand, with optimum precoding vectors f_x-optMinimum pre- of angle theta Coded vector makes | (f_x-opt)^hf′_x-opt| maximum, that is, horizontal peacekeeping vertical dimension suboptimum precoding vectors can be obtained by below equation :

$f_{h-opt}^{\′}=\underset{1\≤i\≤c,i\&notequal;h-opt}{\mathrm{argmax}}\left\{\right|{\left(f_{h-opt}\right)}^h{f}_i\left|\right\}$

$f_{v-opt}^{\′}=\underset{1\≤j\≤c,j\&notequal;v-opt}{\mathrm{argmax}}\left\{\right|{\left(f_{v-opt}\right)}^h{f}_j\left|\right\}$

Wherein, c represents number of codewords in corresponding code book, f_i, f_jRepresent code word in corresponding code book, i.e. f_iRepresent the i-th row in code book, f_j Represent jth row, f ' in code book_h-optMake for traversal code book | (f_h-opt)^hf_i| maximum level dimension suboptimum precoding vectors, f′_v-optMake (f for traversal code book_v-opt)^hf_j| maximum vertical dimension suboptimum precoding vectors, x represents horizontal dimensions or vertical Dimension, that is, x represent horizontal dimensions or vertical dimensions, (f respectively for h or v_x-opt)^hFor f_x-optConjugate transpose, according to above-mentioned calculating Method obtains level dimension suboptimum precoding vectors f '_h-optWith vertical dimension suboptimum precoding vectors f ' v_-opt；

The new level dimension of described structure and vertical dimension optimum precoding vectors specifically include: the dft storing in system code book is divided Solve as two identical code books 1 and code book 2, in horizontal dimensions, tieed up as level by selecting optimum precoding vectors in code book 1 Optimum precoding vectors f_h-opt, select optimum precoding vectors to tie up suboptimum precoding vectors f ' as level from code book 2_h-opt, To f_h-optWith f '_h-optIt is dot product conversion: f_h=f_h-opt.*f′_h-optThe f of gained_hReplace f_h-optThe level dimension becoming new is optimum pre- Coded vector；In the vertical dimension., optimum precoding vectors are obtained from code book 1 as vertical dimension optimum precoding vectors f_v-opt, obtain suboptimum precoding vectors from code book 2 as vertical dimension suboptimum precoding vectors f '_v-opt, to f_v-optWith f '_v-opt It is dot product conversion: f_v=f_v-opt.*f′_v-opt, the f that obtains_vReplace f_v-optAs new vertical dimension optimum precoding vectors.

2. method according to claim 1 is it is characterised in that user side is based on maximum in horizontal dimensions with vertical dimensions Change the selection that minimum singular value criterion carries out optimum precoding vectors, specifically include: according to equivalent channel h w_iMinimum non-zero Singular value travels through code book, according to formula: Select optimum code word f_{h_opt}And f_{v_opt}As horizontal dimensions and vertical dimensions optimum precoding vectors, wherein, l is code in code book Number of words mesh.