[background technique]
Massive MIMO technology can be obviously improved the performance and capacity of 5G wireless communication system.Massive mimo system
Dual-mode antenna it is in a large number, in systems use digital beam forming technology, will lead to system cost of implementation and energy consumption be very high,
Obstruction is brought for the application of Massive MIMO beamforming technique.
In order to overcome this problem, there is Massive MIMO mixed-beam figuration scheme (i.e. part connection framework),
Compared to digital beam forming technology, system cost of implementation and energy consumption can be reduced using mixed-beam figuration technology, i.e., in part
It connecting in framework, each rf chain is only connected with some antennas, although system sacrifices the gain of part wave beam forming,
It is the implementation complexity for greatly reducing hardware, still, in part connection framework, some algorithms based on encoder design,
Performance loss is than more serious.
As shown in Figure 1, being Massive mimo system downlink communication mode under partially connected architecture.At this
In system, transmitting terminal is equipped with NtRoot antenna, receiving end are equipped with single antenna, send N from transmitting terminalsData flows to receiving end.For
The transmission of realization multiple data stream, transmitting terminal are equipped withRf chain connects L on every rf chaintRoot antenna, receiving end are equipped with
Single rf chain.
Assuming that array antenna model is ULA, the spacing of antenna is the half of wavelength.Define dimension vector fi, and dimension
Vector fiIn each element meetSo entire simulation pre-coding matrix of transmitting terminal is
In order to simplify following description, index of definition collection Ω1, by FRFThe position of neutral element forms, then as (i, j) ∈ Ω1
When,Wherein, i and j respectively indicates the line number and columns of simulation pre-coding matrix,Indicate simulation precoding square
The element of i-th row jth column in battle array, in transmitting end structure, digital precode matrix FBBDimension beTransmitting terminal
Total power constraint passes through normalization FBBRealize, meet | | FRFFBB| |=Ns。
Consider a narrow band blocks flat fading channel, y=HF can be expressed as by receiving signalRFFBBS+n, y are that dimension is
Nr× 1 reception vector, H are that dimension is Nr×NtChannel matrix, NrIndicate that the antenna amount of receiving end, s are to send symbol arrow
Amount, meets the mathematics phase Expression dimension is Ns×NsUnit vector, n is noise vector, is obeyed only
The vertical Gaussian Profile with distribution, mean value 0, variance σ2。
In above-mentioned part connection framework, since analog domain pre-coding matrix is permanent mould, so optimization problem right and wrong
Convex, therefore, the optimization problem for solving transmitting terminal is extremely complex, needs to consume a large amount of time and can just solve the complete of optimization problem
The optimal pre-coding matrix of office, this results in increasing the energy loss of existing part connection architecture system, and greatly improves
Network delay.
[summary of the invention]
The object of the present invention is to provide a kind of extensive multiple-input and multiple-output mixed-beam figuration calculations based on part connection
Method improves the performance of algorithm, is close to the performance of digital beam-forming system under the premise of cost and energy consumption control,
Reduce the energy loss of part connection architecture system.
The invention adopts the following technical scheme: a kind of extensive multiple-input and multiple-output mixed-beam based on part connection is assigned
Shape algorithm, comprising the following steps:
Given part connects the optimal public without constraint precoder, primary simulation pre-coding matrix and convergence of architecture system
Difference, and initial number pre-coding matrix is calculated according to without constraint precoder and primary simulation pre-coding matrix, and then calculate
Initial error out;
When initial error is less than or equal to convergence tolerance, by primary simulation pre-coding matrix and initial number precoding square
Battle array connects the optimal simulation pre-coding matrix and optimal digital precode matrix of architecture system as part;
When initial error is greater than convergence tolerance, it is iterated calculating as target to minimize error, until error is less than
Or it is equal to convergence tolerance, or complete maximum number of iterations, corresponding simulation pre-coding matrix and digital pre-coding matrix are made
For optimal simulation pre-coding matrix and optimal digital precode matrix;
Part connection architecture system is generated and issued according to optimal simulation pre-coding matrix and optimal digital precode matrix
Mixed-beam.
Further, optimal to be obtained without constraint precoder by following steps:
The channel state information that architecture system is connected according to part, obtains channel matrix
Wherein, NcIndicate group variety number, NpIndicate the number of path in every cluster, αilIndicate the l articles biography in the i-th cluster reflector
The gain factor in defeated path obeys the multiple Gauss distribution that mean value is 0, variance is 1, for (i, l) strip diameter, θr,ilWith
Respectively indicate the azimuth for leaving angle and pitch angle, θt,ilWithAzimuth and the pitch angle of angle of arrival are respectively indicated,WithRespectively indicate azimuth angle thetar,ilWithPitching angle thetat,ilWithCorresponding receiving array
Response and emission array response;
Pass through H=U Σ VHSingular value decomposition is carried out to channel matrix H, unit matrix V is obtained, is proposed from unit matrix V
Its preceding NsColumn obtain matrix V1;Pass throughDiagonal matrix Γ is calculated, and passes through F*=V1Γ obtains optimal pre- without constraint
Encoder F*;Wherein, U is unit matrix, and Σ is diagonal matrix.
Further, pass throughObtain initial number pre-coding matrixAnd pass throughObtain initial error ε0。
Further, the detailed process of iterative calculation are as follows:
Set objective functionConversion | | F*-FRFFBB||F
It can obtainWherein, TnFor F*In n-th of matrix-block, fnIndicate composition FRFIn
N-th of matrix-block,Indicate FBBLine n vector;
Then to objective function optimize for
Wherein, tn,mIt indicatesM column,It indicatesM-th of element,The F obtained for kth time iterationRF
In n-th of matrix-block,To indicateM-th of element phase increment, It is
F after k iterationBBLine n vector;It indicatesThe average value of phase increment;
By to obtaining simulation pre-coding matrix after the objective function optimizationAnd from which further follow that digital precode square
Battle arrayAnd error εk;
By error εkIt is compared with convergence tolerance, whenWhen, by corresponding simulation pre-coding matrixAnd number
Pre-coding matrixAs optimal simulation pre-coding matrix and optimal digital precode matrix;WhenWhen, repeat on
Step is stated, until error is less than or equal to convergence tolerance, or maximum number of iterations is completed, by corresponding simulation pre-coding matrixWith digital pre-coding matrixAs optimal simulation pre-coding matrix and optimal digital precode matrix.
The beneficial effects of the present invention are: the present invention is based on Massive MIMO mixed-beam shaping systems, to promote frequency spectrum
Efficiency is target, mixing pre-coding matrix is designed using a kind of alternative optimization algorithm based on matrix decomposition, to channel matrix
Singular value decomposition is carried out, design is optimal to design final precoding by alternative optimization without constraint digital precode device and synthesizer
Device improves the performance of algorithm, is close to the property of digital beam-forming system under the premise of controlling cost and energy consumption as far as possible
Can, solve the problems, such as performance loss, and effective verifying has been carried out to the performance of algorithm.
[specific embodiment]
The following describes the present invention in detail with reference to the accompanying drawings and specific embodiments.
A kind of extensive multiple-input and multiple-output mixed-beam forming algorithm based on part connection is disclosed in the present invention, is obtained
Take base station to the channel state information between all user terminalsWherein, HkIndicate base station to use
Fading information between the k of family, K are the total number of users of cell.Optimal digital precode matrix based on singular value decomposition design base station end
Digital precode matrix and simulation pre-coding matrix are designed by alternative optimization with synthesizer matrix.Information data transmission is opened
Begin, sends signal and first pass around a digital precoder FBB, then pass throughA radio frequency link, usingA simulation
Precoder, later LtSignal is sent to wireless channel simultaneously by root antenna.In receiving end, a simulation precoder receives channel
In signal, by a rf chain, using a digital precoder WBBReceive signal.
As shown in Fig. 2, method of the invention specifically:
Given part connects the optimal public without constraint precoder, primary simulation pre-coding matrix and convergence of architecture system
Difference, and initial number pre-coding matrix is calculated according to without constraint precoder and primary simulation pre-coding matrix, and then calculate
Initial error out.
It is optimal to be obtained without constraint precoder by following steps:
The channel state information that architecture system is connected according to part, obtains channel matrix
Wherein, NcIndicate group variety number, NpIndicate the number of path in every cluster, αilIndicate the l articles biography in the i-th cluster reflector
The gain factor in defeated path obeys the multiple Gauss distribution that mean value is 0, variance is 1, for (i, l) strip diameter, θr,ilWith
Respectively indicate the azimuth for leaving angle and pitch angle, θt,ilWithAzimuth and the pitch angle of angle of arrival are respectively indicated,WithRespectively indicate azimuth angle thetar,ilWithPitching angle thetat,ilWithCorresponding receiving array
Response and emission array response;
Pass through H=U Σ VHSingular value decomposition is carried out to channel matrix H, unit matrix V is obtained, is proposed from unit matrix V
Its preceding NsColumn obtain matrix V1;Pass throughDiagonal matrix Γ is calculated, and passes through F*=V1Γ obtains optimal pre- without constraint
Encoder F*;Wherein, U is unit matrix, and Σ is diagonal matrix.
Pass throughObtain initial number pre-coding matrixAnd pass throughObtain initial error ε0。
When initial error is less than or equal to convergence tolerance, by primary simulation pre-coding matrix and initial number precoding square
Battle array connects the optimal simulation pre-coding matrix and optimal digital precode matrix of architecture system as part.
When initial error is greater than convergence tolerance, it is iterated calculating as target to minimize error, until error is less than
Or it is equal to convergence tolerance, or complete maximum number of iterations, corresponding simulation pre-coding matrix and digital pre-coding matrix are made
For optimal simulation pre-coding matrix and optimal digital precode matrix.
The detailed process of iterative calculation are as follows:
Set objective functionIteration optimization two
Parameter, as simulation pre-coding matrix and digital pre-coding matrix.
Conversion | | F*-FRFFBB||FIt can obtainWherein, TnFor F*In
N-th of matrix-block, fnIndicate composition FRFIn n-th of matrix-block,Indicate FBBLine n vector, it is maximum in the present invention
The number of iterations can be set as Ku=100, convergence tolerance is set as
When then being optimized to objective function, due to that should meet in the systemSo in optimization process
In, temporarily remove normalization constraint
Simplified optimization problem are as follows:
The kth time iteration of definition mixing precoding isAssuming that known initialThenClosed solutions beConversely, when knownWhen, it can update
It is updatingWhen, following derivation is first done, F is worked asBBTo timing, definitionI-th column vector be vi, then can be with
It obtainsIn view of FRFStructure, then derivation process are as follows:
Wherein, TnIt is F*N-th of block matrix, dimension Lt×Ns。
Then above-mentioned optimization problem can be converted toA subproblem, wherein n-th of subproblem be
In order to solve optimization problem, it is assumed thatA small range search pairIt is updated, definesM-th yuan
Element phase beIt can be expressed asSo,It can be expressed as
Wherein,It indicatesThe increment of m-th of element phase, whenWhen very little, Taylor expansion approximation is done
Wherein,It is a vector, symbolIt is defined as Hadamard product, optimization problem can be reconstructed into
Above formula optimization problem is a convex quadratic objective function, and has been contemplated that permanent modular constraint, and above formula is based on approximationCome what is rebuild, so, only whenIt is just able to satisfy when very little, thereforeConstraint must add,
So optimization problem is converted into
It is a convex optimization problem at this time, once it obtainsIt is obtained with
Further definitionM-th of element beM be classified as tn,m, above formula can derive are as follows:
Optimization problem, which can be converted into, solves LtA subproblem, wherein m-th of subproblem is
By to obtaining simulation pre-coding matrix after the objective function optimizationAnd from which further follow that digital precode square
Battle arrayAnd error εk;
By error εkIt is compared with convergence tolerance, whenWhen, by corresponding simulation pre-coding matrixAnd number
Pre-coding matrixAs optimal simulation pre-coding matrix and optimal digital precode matrix;WhenWhen, repeat on
Step is stated, until error is less than or equal to convergence tolerance, or maximum number of iterations is completed, by corresponding simulation pre-coding matrixWith digital pre-coding matrixAs optimal simulation pre-coding matrix and optimal digital precode matrix.
Part connection architecture system is generated and issued according to optimal simulation pre-coding matrix and optimal digital precode matrix
Mixed-beam.
Verify embodiment one:
Architecture system is connected for part of the invention, corresponding spectrum efficiency is
Wherein,It is signal-to-noise ratio, P indicates the average transmission power between base station and mobile station, Σ1It represents to angular moment
First dimension of battle array Σ is Ns×NsBlock, be defined as
When what is sent in the channel is Gauss symbol, then spectrum efficiency is
Wherein,It is Ns×NsUnit vector, P is average transmission power, NsIt is number of data streams, RnIt is noise association side
Poor matrix, Rn=σ2FRFFBB, σ expression variance, HkIt is channel matrix;
Mixing Precoding Design can be obtained by the optimal solution of the optimization problem of one transmitting terminal of solution, this optimization
The target of problem is that spectrum efficiency maximizes, i.e.,
Wherein,Be simulate pre-coding matrix in the i-th row, jth column element,It is digital precode matrix
In the i-th row, jth column element.
System uses the narrowband the Saleh-Valenzuela group variety channel model of geometry.The aerial array of base station and user terminal
It is all made of uniform straight line array, its array response vector is
If the receiving array response in the channel matrix of above formula is replaced with emission array response using uniform straight line array
αULA(θ), wherein N indicates that array number in line array, λ indicate that carrier wavelength, d indicate the spacing between antenna, and angle/arrive is left in θ expression
Up to the attitude angle at angle.
In emulation part, millimeter wave cluster channel model is not only considered, it is also contemplated that Rayleigh channel model.Believe in Rayleigh fading
In road model, normalize channel matrix H in each element obey independent identically distributed mean value be 0, variance σ2Multiple height
This distribution.
Numerical result shows that the performance of mixing method for precoding of the invention can be close to be assigned in high-dimensional digital wave beam
The performance of shape system, and it is much higher than the performance of analog beam shaping system.The present invention simulates the gains such as Precoding Design is based on
Transmission, digital precode matrix design are based on singular value decomposition.Final mixing precoder design is based on alternative optimization, connects it
Nearly optimal no constraint matrix, so that performance of the system performance close to digital beamforming algorithm.
Verify embodiment two:
Verified in the present embodiment by Matlab emulation the spectrum efficiency of the alternative optimization algorithm based on matrix decomposition with
The advantage of other algorithm performances.
The antenna number that sending and receiving end is equipped in the embodiment is 128 and 1, and the number of rf chain is all 4, number of data streams
It is 2, using the millimeter wave channel model of ULA Array Model, each parameter occurrence is as described in Table 1 in the model.
Parameter assignment in 1 embodiment of table
As shown in figure 3, which show the spectrum efficiency comparison results of many algorithms.The performance of analog beam shaping system is most
The performance of difference, digital beam forming system is best.The performance of algorithm proposed by the invention is substantially better than analog beam figuration
Performance, and better than with algorithm performance in the prior art, closest to the performance of digital beam-forming system.
As shown in Figure 4, it is shown that influence of the number of data streams to algorithm spectrum efficiency.Compare be number of data streams be 1,
2, the spectrum efficiency of 4 and 8 lower proposed algorithms, as number of data streams increases, the spectrum efficiency of system increases, system performance
It becomes better and better.