CN106599551A - Rapid adaptive beam-forming algorithm applied to array antenna soccer robot - Google Patents

Abstract

The invention provides a rapid adaptive beam-forming algorithm applied to an array antenna soccer robot. The rapid adaptive beam-forming algorithm is characterized by comprising the following steps: preprocessing a training sample set, and eliminating an expected signal; estimating a covariance matrix by using preprocessed data, and performing RGS orthogonalization on a column vector of the covariance matrix to construct an interference subspace; and orthogonally protecting a corresponding static weight vector to the interference subspace to obtain an adaptive weight vector. Compared with other adaptive beam-forming algorithms, the rapid adaptive beam-forming algorithm has the advantages of high beam pointing accuracy, high output SINR (Signal Interference Noise Ratio) and high beam pointing error robustness, and is a practical and rapid adaptive beam-forming algorithm being robust for the cancellation phenomenon of the expected signal.

Description

A kind of quick self-adapted beamforming algorithm for array antenna Soccer robot

Technical field

The present invention relates to a kind of utilization array antenna assist people carries out the algorithm of target recognition.

Background technology

Current most of Soccer robots can not in dense fog, rain, compete in the environment such as night, in order to be able to allow robot foot Ball match accessible development in 24 hours, needs to carry out target recognition using array antenna assist people.

Traditional array antenna when train snapshot data do not contain desired signal when, conventional SMI algorithms to Beam steering error, Ro-vibrational population equal error is insensitive, but unstable to low snap error.During low snap, conventional SMI algorithms can be because of noise The corresponding little eigenvalue shake in space, causes pattern distortion, minor level to be raised.And for actual radar system, often It is required that adaptive beam-forming algorithm has relatively low operand and faster convergence rate.In order to improve conventional adaptive beam To the robustness of low snap error and the rapidity of realization, Chinese scholars propose many of good performance formation algorithm in succession ADBF algorithms.Diagonal loading (LSMI) algorithm is reduced by with the addition of diagonal add-in on the object function of conventional SMI algorithms The impact of noise disturbance, so as to overcoming low snap error.But the maximum deficiency of LSMI methods is the optimum of diagonal loading amount Value is difficult to determine.Error amount construction between SMI (PFM-SMI) algorithm adaptive weights with function is deteriorated and static weights Deteriorate function, while ensureing that array output SINR is maximum, meet some quadratic constraints, and then try to achieve optimum weight vector. PFM-SMI algorithms avoid to a certain extent secondary lobe caused by low snap error and raise, but the method operand is big, and exist Problem of parameter selection.Orthogonal projection algorithm carries out Eigenvalues Decomposition and constructs interference signal sky by interference noise covariance matrix Between, then constraint steering vector is made into rectangular projection to interference space, obtain self adaptation weight vector.Due to empty by interference Between ask for self adaptation power, therefore the algorithm avoids impact of the noise disturbance to self adaptation weight vector, will not make adaptive beam Figure secondary lobe is raised.But the algorithm is decomposed due to characteristics of needs value, therefore operand is larger, while depositing when interference space is constructed In the determination problem of interference number.

Above-mentioned these sane ADBF algorithms more or less all come with some shortcomings, and all exist one it is quick realize ask Topic.In recent years, Gram-Schmidt (GS) orthogonalization algorithm of the proposition such as Hung is a kind of Fast Subspace projection algorithm, the calculation Method can preferably reconstruct interference space in higher dry making an uproar than under, the Fast Convergent characteristic with subspace projection algorithm, And its computational complexity is little, is easy to Project Realization, it is particularly suited for that array number is more, interference source number is less and strong interferers Occasion, so as to widely be paid close attention to.

The content of the invention

The purpose of the present invention is to propose to a kind of be based on the orthogonalized quick self-adapted beamforming algorithms of GS.

In order to achieve the above object, the technical scheme is that there is provided a kind of for array antenna Soccer robot Quick self-adapted beamforming algorithm, it is characterised in that comprise the following steps：

Step 1, pretreatment is carried out to training sample set, reject desired signal, it is assumed that the fast umber of beats of k moment array receiveds It is x (k), then l-th data component x in x (k) according to vector_lCan be expressed as, l=1,2 ..., N：

In formula (1), s_iRepresent i-th steering vector, β_{L, i}I-th phase-modulation of l-th snap is represented, the inside includes DOA information, n_lRepresent the noise vector of l-th snap；

To x_lX is obtained as data prediction_l',

Step 2, using pretreated data estimation covariance matrix, and the column vector to covariance matrix carries out RGS Orthogonalization constructs interference space；

Step 3, corresponding static weight vector is obtained into self adaptation weight vector to interference space as rectangular projection.

Preferably, the step 2 is comprised the following steps：

Step 2.1, write formula (2) as matrix form, had：

X ' (k)=Bx (k) (3), in formula (3)：

X ' (k)=[x₁', x₂' ..., x_N-1′]^TRepresent preprocessed data vector；

B represents blocking matrix

The covariance matrix of x ' (k) is expressed as

Step 2.2, orthogonalization adaptive threshold Δ ' (k) for being calculated covariance matrix GS orthogonal algorithms

The noise variance approximate representation of preprocessed data vector x ' (k) is：

In formula (6), σ_nRepresent noise variance；

Covariance matrixNoise varianceIt is approximately：

Then there is orthogonalization adaptive threshold Δ ' (k) to be：

In formula, U_iAnd U_i' represent rightColumn vector carry out the orthogonal vectors that RGS orthogonalizations are obtained, i represents i ＆ lt Iteration；

Step 2.3, the orthogonalization procedure of covariance matrix GS orthogonal algorithms are expressed as：

Preferably, in the step 3：

The self adaptation weight vector of covariance matrix GS orthogonal algorithms is w_MRGS, then have：

In formula (10),Represent the interference source number that covariance matrix GS orthogonal algorithms judge, w_qRepresent static weight vector.

Compared to other adaptive beam-forming algorithms, algorithm proposed by the present invention has beam-pointing accuracy height, output The advantage such as SINR is high and strong to Beam steering error robustness, is that one kind sane to desired signal cancellation phenomenon is practical and quick Adaptive beam-forming algorithm.

Description of the drawings

Fig. 1 is the algorithm process schematic flow sheet of the present invention；

Fig. 2 (a) compares for the inventive algorithm and RGS algorithms adaptive direction figure under strong interference environment；

Fig. 2 (b) compares for the inventive algorithm and RGS algorithms adaptive direction figure under weak jamming environment.

Specific embodiment

To become apparent the present invention, hereby with preferred embodiment, and accompanying drawing is coordinated to be described in detail below.

For there is a problem of that desired signal will cause conventional RGS algorithm performances degradation in covariance matrix, propose Covariance matrix GS orthogonalization algorithms (MRGS algorithms) based on data prediction.The algorithm enters first to training sample set Row pretreatment, rejects desired signal；Then pretreated data estimation covariance matrix is utilized, and to the row of covariance matrix Vector carries out RGS orthogonalizations construction interference space；Finally corresponding static weight vector is made into rectangular projection to interference space Obtain self adaptation weight vector.In addition, the present invention will also enter for MRGS algorithms to the orthogonalization adaptive threshold of preprocessed data Row amendment, to estimate interference space exactly.

The snapshot data vector for assuming k moment array receiveds is x (k), then l-th data component x in x (k)_lCan represent For, l=1,2 ..., N：

In formula (1), s_iRepresent i-th steering vector, β_{L, i}I-th phase-modulation of l-th snap is represented, the inside includes DOA information, n_lRepresent the noise vector of l-th snap；

To x_lX is obtained as data prediction_l',

By formula (2) as can be seen that to x_lAs the x that data prediction is obtained_l' in comprise only interference and noise signal composition, in advance Processing procedure serves the effect for rejecting desired signal.Therefore, write formula (2) as matrix form, had：

X ' (k)=Bx (k) (3), in formula (3)：

X ' (k)=[x₁', x₂' ..., x_N-1′]^TRepresent preprocessed data vector；

B represents blocking matrix

The covariance matrix of x ' (k) is expressed as

Comprise only interference and noise signal composition.It is rightWhen carrying out RGS orthogonalizations, orthogonalization adaptive threshold Need to be improved, accurately to estimate interference space.

The noise variance approximate representation of preprocessed data vector x ' (k) is：

In formula (6), σ_nRepresent noise variance；

Covariance matrixNoise varianceIt is approximately：

Orthogonalization adaptive threshold Δ ' (k) for being then calculated MRGS algorithms is：

In formula, U_iAnd U_i' represent rightColumn vector carry out the orthogonal vectors that RGS orthogonalizations are obtained, i represents i ＆ lt Iteration；

Therefore, the orthogonalization procedure of MRGS algorithms can be expressed as follows：

So, the self adaptation weight vector w of MRGS algorithms_MRGSCan be calculated as follows：

In formula (10),Represent the interference source number that covariance matrix GS orthogonal algorithms judge, w_qStatic weight vector is represented, Generally expect steering vector.

Fig. 1 gives the handling process schematic diagram of MRGS algorithms.MRGS algorithms proposed by the present invention are to training snapshot data Vector carries out data prediction, solves the problems, such as the signal cancellation that conventional RGS algorithms occur；Estimate covariance matrix column is sweared Amount carries out RGS orthogonalizations, and can reduce noise disturbance using more sample informations affects.Therefore, contain in training sample The application scenario of desired signal, MRGS algorithms proposed by the present invention are strictly a kind of quick and sane Adaptive beamformer side Method.

The performance of MRGS algorithms proposed by the present invention will be verified by emulation experiment below.In emulation, it is assumed that array number is₁6 omnidirectional's array element constitutes equidistant even linear array, array element distance d=λ/2, the meansquaredeviationσ of noise_n=1.Now, blocking matrix B InOutput SINR is that 100 Monte Carlo Experiment emulation is averagely obtained, and beam pattern is 1 illiteracy Special Carlow experiment simulation is obtained.

Experiment：MRGS algorithms and RGS algorithm adaptive direction figures compare.Suppose there is 1 desired signal and 3 interference letters Number, to respectively θ₀=0 °, θ_i=-28 °, 17 °, 41 °, input signal-to-noise ratio SNR=0dB, fast umber of beats K=20.Adaptive beam As shown in Fig. 2 Fig. 2 (a) is the simulation result in the case of strong jamming, dry making an uproar be 30dB than INR for the simulation result of figure；And Fig. 2 B () is the simulation result in the case of weak jamming, INR is 10dB.

From Fig. 2 (a) as can be seen that because desired signal power is relatively weak, therefore not carried out as interference by RGS algorithms Suppress.Now, RGS algorithms and MRGS algorithms can form null in interference radiating way, and in desired signal direction maximum wave beam is formed Sensing, but the null depth ratio RGS algorithm of MRGS algorithms at least depth 15dB, performance is more preferable.From Fig. 2 (b) as can be seen that due to this When desired signal power it is relatively strong, RGS algorithms are suppressed desired signal as interference, not only without in desired signal side To maximum beam position is formed, the null of certain depth, beam pattern severe exacerbation are defined on the contrary；And the beam pattern of MRGS algorithms Keep good, null will not be formed in desired signal direction, this is mainly the effect of data prediction.It can be seen that, do in certain Disturb in power bracket, MRGS algorithms can keep good wave beam performance, and RGS algorithms performance will be serious under weak jamming environment Deteriorate.

Claims (3)

1. a kind of quick self-adapted beamforming algorithm for array antenna Soccer robot, it is characterised in that including following Step：

Step 1, pretreatment is carried out to training sample set, reject desired signal, it is assumed that the snapshot data arrow of k moment array receiveds Measure as x (k), then l-th data component x in x (k)_lCan be expressed as, l=1,2 ..., N：

$x_l=\underset{i=0}{\overset{N}{\mathrm{\Σ}}}{s}_i\·e^{{\mathrm{j\β}}_{l,i}}+n_l---\left(1\right)$

In formula (1), s_iRepresent i-th steering vector, β_{L, i}I-th phase-modulation of l-th snap is represented, the inside contains DOA Information, n_lRepresent the noise vector of l-th snap；

To x_lX is obtained as data prediction_l',

$\begin{array}{c}{x}_l^{\′}=x_l-e^{-j\left({\mathrm{\β}}_{l+1,0}-{\mathrm{\β}}_{l,0}\right)}{x}_{l+1}\\ =\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{s}_i{e}^{{\mathrm{j\β}}_{l,i}}\left(1-e^{j\[\left({\mathrm{\β}}_{l+1,i}-{\mathrm{\β}}_{l,i}\right)-\left({\mathrm{\β}}_{l+1,0}-{\mathrm{\β}}_{l,0}\right)\]}\right)+n_l-n_{l+1}{e}^{-j\left({\mathrm{\β}}_{l+1,0}-{\mathrm{\β}}_{l,0}\right)}\end{array}---\left(2\right);$

Step 2, using pretreated data estimation covariance matrix, and it is orthogonal to carry out RGS to the column vector of covariance matrix Change construction interference space；

Step 3, corresponding static weight vector is obtained into self adaptation weight vector to interference space as rectangular projection.

2. a kind of quick self-adapted beamforming algorithm for array antenna Soccer robot as claimed in claim 1, its It is characterised by, the step 2 is comprised the following steps：

Step 2.1, write formula (2) as matrix form, had：

X ' (k)=Bx (k) (3), in formula (3)：

X ' (k)=[x₁', x₂' ..., x_N-1′]^TRepresent preprocessed data vector；

B represents blocking matrix

$B={\left[\begin{array}{cccccc}1& -e^{-j\left({\mathrm{\β}}_{2,0}-{\mathrm{\β}}_{1,0}\right)}& ...& 0& 0& 0\\ 0& 1& -e^{-j\left({\mathrm{\β}}_{3,0}-{\mathrm{\β}}_{2,0}\right)}& ...& 0& 0\\ ...& ...& ...& ...& ...& ...\\ 0& 0& 0& 0& 1& -e^{-j\left({\mathrm{\β}}_{N,0}-{\mathrm{\β}}_{N-1,0}\right)}\end{array}\right]}_{\left(N-1\right)\×N}---\left(4\right)$

The covariance matrix of x ' (k) is expressed as

$\begin{array}{c}{\hat{R}}^{\′}\left(k\right)=\frac{1}{(N-1)}\underset{k=1}{\overset{N-1}{\Σ}}{x}^{\′}\left(k\right)x^{\′H}\left(k\right)\\ =\frac{1}{\left(N-1\right)}\underset{k=1}{\overset{N-1}{\Σ}}Bx\left(k\right){\left(Bx\right(k\left)\right)}^H\\ =B\hat{R}{B}^H\end{array}---\left(5\right)$

Step 2.2, orthogonalization adaptive threshold Δ ' (k) the preprocessed data arrow for being calculated covariance matrix GS orthogonal algorithms Amount x ' (k) noise variance approximate representation be：

${\mathrm{\σ}}_n^{\′2}\≈(1+|-e^{-j({\mathrm{\β}}_{l+1,0}-{\mathrm{\β}}_{l,0})}{|}^2){\mathrm{\σ}}_n^2=2{\mathrm{\σ}}_n^2---\left(6\right)$

In formula (6), σ_nRepresent noise variance；

Covariance matrixNoise varianceIt is approximately：

Then there is orthogonalization adaptive threshold Δ ' (k) to be：

$\left\{\begin{array}{c}{\mathrm{\Δ}}^{\′}\left(k\right)={\left(1.5{{\mathrm{\σ}}^{\′}}_R\right)}^2(N-1)(1+\underset{i=1}{\overset{k-1}{\mathrm{\Σ}}}{\mathrm{\β}}_{ki}^{\′2})\\ {\mathrm{\β}}_{ki}^{\′2}=\frac{|\left({\hat{R}}^{\′}\right(k),U_i){|}^2}{|U_i^{\′}{|}^2}\end{array}\right.---\left(8\right)$

In formula, U_iAnd U_i' represent rightColumn vector carry out the orthogonal vectors that RGS orthogonalizations are obtained, i represents ith iteration；

Step 2.3, the orthogonalization procedure of covariance matrix GS orthogonal algorithms are expressed as：

$\left\{\begin{array}{c}{U}_1=\frac{{\hat{R}}^{\′}\left(1\right)}{\left|\right|{\hat{R}}^{\′}\left(1\right)\left|\right|}\\ U_k^{\′}={\hat{R}}^{\′}\left(k\right)-\underset{l=1}{\overset{k-1}{\mathrm{\Σ}}}{U}_l^H{\hat{R}}^{\′}\left(k\right)U_l\\ U_k=\frac{U_k^{\′}}{\left|\right|U_k^{\′}\left|\right|}\end{array}\right.,2\≤k\≤N-1---\left(9\right).$

3. a kind of quick self-adapted beamforming algorithm for array antenna Soccer robot as claimed in claim 2, its It is characterised by, in the step 3：

The self adaptation weight vector of covariance matrix GS orthogonal algorithms is w_MRGS, then have：

$w_{MRGS}=w_q-\underset{i=1}{\overset{{\hat{P}}^{\′}}{\Σ}}{U}_i^{\′H}{w}_q{U}_i^{\′}---\left(10\right)$

In formula (10),Represent the interference source number that covariance matrix GS orthogonal algorithms judge, w_qRepresent static weight vector.