CN102664666A - Efficient robust self-adapting beam forming method of broadband - Google Patents

Abstract

The present invention provides an efficient broadband robust adaptive beamforming method, which is used in the field of wireless communication. The method is as follows: perform FFT transformation on the received data of the array, and obtain the received data at different frequency points and the covariance of the received data at each frequency point Matrix, select the center frequency point as the reference frequency point; use the spread operator idea, respectively divide the covariance matrix of each frequency point and the covariance matrix of the center frequency point into matrix blocks, and obtain the spread operator of each frequency point and the center frequency point point propagation operator; construct a focusing transformation matrix to focus the propagation operators of different frequency points on the same reference frequency point; obtain the final propagation operator estimation and noise subspace; combine the feature space idea to construct the weight of the broadband beamforming algorithm vector for robust wideband adaptive beamforming. Compared with the traditional coherent signal subspace algorithm, the present invention does not require any singular value or eigenvalue decomposition, and does not require diagonal loading technology. It has better performance in environments with low snapshots and strong expected signals, especially in Under the condition that there is a certain error in the estimation of the expected signal, the improved method has stronger robustness and reduces the complexity at the same time.

Description

An Efficient Broadband Robust Adaptive Beamforming Method

technical field

The invention relates to the technical fields of radar, sonar and wireless communication, in particular to an efficient broadband robust adaptive beamforming method.

Background technique

Digital beamforming (DBF) uses digital signal processing technology to weight the signals received by the antenna array, and adjusts the weights of each array element so that the signals can be received effectively. This is because the weights of each array element form the array weight vector, and the weight vector directly determines the direction diagram of the adaptive array, that is, determines the receiving effect of useful signals. Effective reception of useful signals includes two aspects: one is to align the main lobe of the array pattern (the direction of the maximum gain of the array antenna) with the direction of the desired signal; the other is to align the zero point of the pattern with the interference signal for effective suppression. When the traditional narrowband-based beamforming algorithm is used for broadband signals, it will lead to great changes in the direction of the beam and the width of the main lobe at different frequency points, causing deviations. Therefore, in the case of a large number of broadband signal sources, how to perform broadband beamforming has become a research hotspot in wireless communication.

The wideband DBF method that appeared early is based on the incoherent signal subspace method (ISM), which decomposes the wideband signal into several subbands, and directly performs narrowband processing on each subband, that is, the signal correlation matrix of each subband. In narrowband beamforming, the weighted values of all subbands are arithmetically averaged or geometrically averaged, and finally the wideband signal DBF is obtained. Usually this kind of processing algorithm can't get satisfactory results, the main reason is that the calculation is heavy and the coherent signal source cannot be estimated.

The coherent signal subspace method (Coherent Signal Subspace Method, CSM) is an effective algorithm, which was first proposed by Wang and Kaveh (Wang, M Kaveh. "Coherent signal-subspace processing for the detecting and estimation of angles of arrival of multiple wideband sources," IEEE Transactions onASSP, vol.33, no.4, 1985, pp.823-831.). The basic idea of this type of algorithm is to focus the signal space on the non-overlapping frequency points in the frequency band to the reference frequency point, and obtain the data covariance matrix of a single frequency point after focusing, and then use narrow-band technology to perform DBF. This algorithm can solve the problem of coherence Signal source problem. After the CSM algorithm, many CSM algorithms under different constraints have been proposed, such as TCT (Bilateral Correlation Transform) algorithm, SST (Signal Subspace Transform) algorithm, RSS (Rotated Signal Subspace) algorithm and so on. The above algorithms all need to use eigenvalue or singular value decomposition when constructing the focusing matrix, and the calculation amount is o(M ³ ), which will lead to difficulties in real-time processing. In addition, this type of method uses a classic Capton beamformer or a minimum variance distortionless response (Minimum Variance Distortionless Response, MVDR) beamformer when implementing beamforming. This type of beamformer needs to know the orientation information of the desired signal in advance. If the direction of arrival of the desired signal is not estimated accurately, the main beam will point to a wrong direction. At this time, the optimal weight will suppress the desired signal as interference , so as to form a null in the real direction of incoming wave, so the performance is poor in the environment with high signal-to-noise ratio. Diagonal loading technology can improve the performance of beamforming to a certain extent, but how to accurately select the amount of diagonal loading is still a technical problem.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a broadband robust adaptive beamforming method based on propagation operators and feature spaces. The present invention combines the propagation operator (PM) with the broadband coherent signal subspace method, and derives the PM-based broadband signal subspace method. No eigenvalue decomposition is required, and the calculation amount is about o(PM ² ). It can be seen that the complexity of the PM algorithm is about o(P/M) of the TCT focusing method. Therefore, when the number of array elements is large, the method based on the propagation operator will greatly reduce the amount of computation. Finally, the idea of eigenspace adaptive beamforming algorithm is applied to PM, and the broadband robust adaptive beamforming is realized. Compared with the traditional method, the present invention has better performance in the environment of low snapshot and strong expected signal, especially under the condition that there is a certain error in the estimation of the expected signal, the improved method has stronger robustness.

The present invention is realized through the following technical solutions, and the method steps are as follows:

1) Perform FFT transformation on the received data of the array to obtain the data on J frequency points, and then obtain the covariance matrix of the received data at each frequency point respectively, and select the center frequency point _f0 as the reference frequency point;

2) Using the propagation operator idea, divide the covariance matrix R _j of each frequency point and the covariance matrix R ₀ of the central frequency point into matrix blocks, and obtain the propagation operator P _j of each frequency point and the propagation of the central frequency point Operator P ₀ ;

3) Use P _j and P ₀ to construct the focusing matrix T _j of each frequency point, and focus the propagation operators of different frequency points on the same reference frequency point;

构造Q和Q₀，即噪声子空间；4) Get the final propagation operator estimate

Construct Q and Q ₀ , the noise subspace;

5) Combined with the idea of feature space, the weight vector w _PM-BESB of the broadband beamforming algorithm is constructed to realize robust broadband adaptive beamforming.

Each step of the present invention is described in further detail below:

Described step 1), concrete realization is as follows:

则第k次快拍(k＝1，...，N)第j个频点的阵列接收数据为Consider the case of a uniform linear array with M omnidirectional array elements, and the array element spacing is half the wavelength of the center frequency. The P broadband far-field signals are represented by Incident to the array, white Gaussian noise independent of the source signal is appended. The data received by the array is uniformly sampled and divided into K non-overlapping blocks, each containing N sampling points. Perform N-point FFT on each block, and then select J frequency points for subsequent processing. Define each discrete frequency point

Then the array received data of the k-th snapshot (k=1,...,N) at the j-th frequency point is

x _j,k = A _j (θ)s _j,k +n _j,k

Here A _j (θ)=[a _j (θ ₁ ),..., a _j (θ _P )] represents a (M×P)-dimensional array flow pattern matrix, s _{j, k} is a (P×1)-dimensional The signal vector of n _{j, k} is the noise vector of (M×1) dimension. Selecting the center frequency point f ₀ as the focus frequency, the array received data of the k-th snapshot frequency point f ₀ can be written as x _0,k =A ₀ (θ)s _0,k +n _0,k . Therefore, the covariance matrix of the data received by the array at f _j and f ₀ frequency points is:

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; R</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="h\; R"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">h</figure-callout></span><figure-callout\; id="0"\; label="h\; R"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">\; </figure-callout>}}& \\ {\mathrm{<span\; class="notranslate">\; </span>}}_{}{\mathrm{<span\; class="notranslate">R</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; K</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; x</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; k</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}\end{array}\right.$

Described step 2), concrete realization is as follows:

The covariance matrix R _j of each frequency point array and the covariance matrix R ₀ of the central frequency point array are divided into blocks respectively:

R _j = [G _j , H _j ] = [G _j , G _j P _j ]

R ₀ =[G ₀ , H ₀ ]=[G ₀ , G ₀ P ₀ ]

Both G _j and G ₀ here are (M×P) dimensional matrices, and both H _j and H ₀ are M×(MP) dimensional matrices. Using the propagation operator idea, the covariance matrix is divided into blocks to obtain the propagation operator of each frequency point:

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; \&DoubleRightArrow;</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$

$<span\; class="notranslate">\; \&DoubleRightArrow;</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}$

Similarly, the propagation operator P ₀ of the center frequency point can be obtained.

Described step 3), concrete realization is as follows:

Using the propagation operator P _j of each frequency point and the propagation operator P ₀ of the center frequency point, define the focusing matrix, and transform the propagation operators of different frequency points to the reference frequency point:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}$

Then each focusing transformation matrix is:

${\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}$

Described step 4), concrete realization is as follows:

It is not difficult to obtain the following formula by using the knowledge of matrix theory:

${\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="G\; j\; P"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">G</figure-callout></span><figure-callout\; id="0"\; label="G\; j\; P"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">\; </figure-callout>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{}_{}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

And then there are:

${\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; <figure-callout\; id="0"\; label="G\; j\; P"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">G</figure-callout></span><figure-callout\; id="0"\; label="G\; j\; P"\; filenames="HSA00000697442900021.png,HSA00000697442900022.png"\; state="\{\{state\}\}">\; </figure-callout>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{}_{}{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}$

Finally, the estimation of the propagation operator of a single frequency point after focusing is obtained:

${\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; P</span>}}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; J</span>}}<span\; class="notranslate">\; [</span>\underset{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; J</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; G</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}{\mathrm{<span\; class="notranslate">\; T</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}}^{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; ]</span>$

构造M×(M-P)维的矩阵Q，且满足：Combined with the idea of propagation operator, the estimation of the propagation operator using a single frequency point

Construct a matrix Q of M×(MP) dimension, and satisfy:

$\mathrm{<span\; class="notranslate">\; Q</span>}<span\; class="notranslate">\; =</span>\left(\begin{array}{c}{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; P</span>}}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}\\ <span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; m</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; )</span>}\end{array}\right)$

And the Q criterion is orthogonalized to obtain Q ₀ =Q(Q ^H Q) ^(-1/2) , which is equivalent to the noise subspace of the reference frequency point.

Described step 5), concrete realization is as follows:

这里的R₀指的是参考频点的协方差矩阵，a(θ₀)表示的是期望信号的导向矢量，θ₀是期望信号的DOA。最后利用权矢量实现稳健的宽带自适应波束形成。Using the idea of eigenspace adaptive beamforming algorithm and the focusing theory of propagation operator, the weight vector of wideband eigenspace adaptive beamforming (PM-BESB) algorithm is obtained

Here R ₀ refers to the covariance matrix of the reference frequency point, a(θ ₀ ) represents the steering vector of the desired signal, and θ ₀ is the DOA of the desired signal. Finally, the robust broadband adaptive beamforming is realized by using the weight vector.

The present invention has the following advantages: 1) Compared with the traditional coherent signal subspace algorithm, the present invention does not require any singular value or eigenvalue decomposition, which reduces the computational complexity; 2) The present invention does not require diagonal loading technology, which facilitates practical Processing; 3) The present invention has better performance in environments with low snapshots and strong expected signals, especially under the condition that there is a certain error in the expected signal estimation, the improved method has stronger robustness.

Description of drawings

Fig. 1 is the algorithm flow chart of the present invention

Figure 2 shows the relationship between the output SINR and the number of snapshots in the frequency domain

Figure 3-a shows the relationship between the beamforming performance of the TCT-SMI algorithm and the SNR (not diagonally loaded)

Figure 3-b shows the relationship between the beamforming performance of the TCT-SMI algorithm and the SNR (20dB diagonal loading)

Figure 3-c shows the relationship between beamforming performance and SNR of the PM-BESB algorithm

Figure 3-d shows the relationship between the output SINR and SNR of the two algorithms

Figure 4-a shows the relationship between the beamforming performance of the TCT-SMI algorithm and the DOA estimation error (not diagonally loaded)

Figure 4-b shows the relationship between the beamforming performance of the TCT-SMI algorithm and the DOA estimation error (0dB diagonal loading)

Figure 4-c shows the relationship between the beamforming performance of the TCT-SMI algorithm and the DOA estimation error (20dB diagonal loading)

Figure 4-d shows the relationship between the beamforming performance of the PM-BESB algorithm and the DOA estimation error

Figure 4-e shows the relationship between the output SINR of the two algorithms and the DOA estimation error

Detailed ways

The example is for a uniform line array composed of 10 omnidirectional array elements. Three broadband signals, in which the desired signal θ ₀ =20°, two interference signals respectively from θ ₁ =-15° and θ ₂ =50°, INR=40dB. The relative bandwidth of the signal is 40% (1.8GHZ-2.7GHZ), the center frequency is 2.25GHZ, and the array element spacing is half the wavelength of the center frequency. Divide the received sampling data into snapshots segments (that is, the number of snapshots in the frequency domain is snapshots), each segment has 64 points (that is, the number of FFT points in each segment), and select 11 frequency points. Comparing the performance of the PM-BESB-based beamformer with the wideband SMI beamformer based on Bilateral Correlation Transform (TCT) focusing and the SMI beamformer with diagonal loading technique (DL-SMI), the results are averaged from 100 independent experiments value. In this example, the performance of the PM-ESB algorithm and the traditional MVDR beamforming is compared under three conditions of different frequency domain snapshots, different expected signal-to-noise ratios, and errors in the DOA estimation of the expected signal. The specific implementation process is as follows:

1) The impact of frequency domain snapshots on performance

The desired signal SNR=20dB is fixed, and the output signal-to-interference-noise ratio of the two algorithm arrays under the condition that the number of snapshots in the frequency domain ranges from 10 to 100 is investigated, as shown in FIG. 1 . It can be seen that the PM-BESB algorithm has almost the same output SINR as the TCT-SMI algorithm with 20dB diagonal loading, and the performance of TCT-SMI without diagonal loading is relatively poor.

2) Impact of expected signal SNR on performance

The number of snapshots in the frequency domain is fixed at 50, and the relationship between the performance of the two beamforming methods and the SNR of the desired signal is investigated. Figure 2-a shows the performance of the TCT-SMI algorithm without diagonal loading. It can be seen that the SMI algorithm has good performance when the SNR is low (SNR=0dB). However, as the SNR increases, the sidelobe gradually increases. When SNR=20dB, a null begins to appear in the direction of the main lobe θ ₀ =20°, indicating that the desired signal cancellation has occurred at this time. Figure 2-b shows the performance of the TCT-SMI algorithm under the condition of 20dB diagonal loading. It can be seen that the beamforming is very stable. PM-ESB algorithm under the condition of low signal-to-noise ratio (SNR=0dB), the beam sidelobe level is relatively high, the main reason is that in the case of low signal-to-noise ratio, the error of constructing the propagation operator and the focusing transformation matrix is relatively large. However, under the condition of high signal-to-noise ratio, it has a more robust performance. From Figure 2-c, it can be seen that the PM-BESB algorithm changes with the SNR, and its beam pattern is relatively stable. Figure 2-d shows the relationship between the output SINR and SNR of the two methods.

3) Impact of expected signal DOA error on performance

The number of snapshots in the fixed frequency domain is 50, and the desired signal SNR=10dB. The real DOA of the desired signal is θ ₀ =20°, but due to the existence of systematic errors and noises, the estimated DOA varies between ±5°. Figure 3-(a～d) shows the relationship between the wavenumber formation performance of the two methods and the DOA estimation error of the expected signal, and Figure 3-e shows the relationship between the output SINR of the two methods and the DOA estimation error of the expected signal. It can be seen that, The performance of the TCT-SMI algorithm without diagonal loading decreases rapidly with the increase of the angle estimation deviation. Diagonal loading can greatly improve the robustness of the SMI algorithm, but it still cannot overcome the impact of the expected signal estimation error on the main lobe direction of the beam. Impact. It can be seen from the results that the beamforming performance and output SINR of the PM-BESB beamformer hardly change with the DOA estimation error.

The traditional broadband MVDR beamformer based on coherent signal subspace focusing algorithm has poor beamforming performance under the condition of no diagonal loading. Diagonal loading can improve the beam performance to a certain extent, but it is still very sensitive to the error of DOA estimation of the desired signal, which is mainly reflected in the deviation of the main lobe of the beam. It should be pointed out that in practice, the desired signal is generally unknown, which makes accurate estimation of the direction of arrival of the desired signal very important for the performance of broadband beamforming. In addition, the selection of the diagonal loading in practice is often related to the signal characteristics and the surrounding environment, which restricts the flexibility of processing. The present invention does not require any singular value or eigenvalue decomposition, which reduces the computational complexity; in addition, the present invention does not require any diagonal loading measures, and exhibits better performance in environments with low snapshots and strong expected signals. Especially under the condition that there is a certain error in the estimation of the expected signal, the improved method has stronger robustness.

Claims (9)

1. the sane adaptive beam in broadband formation method efficiently is characterized in that may further comprise the steps:

1) the array received data is carried out the FFT conversion, obtain J the data on the frequency, and then obtain the covariance matrix that each frequency receives data respectively, choose center frequency point f ₀Frequency as a reference;

2) utilize propagation operator thought, respectively with the covariance matrix R of each frequency _jCovariance matrix R with center frequency point ₀Carry out the partitioning of matrix, obtain the propagation operator P of each frequency _jPropagation operator P with center frequency point ₀

3) utilize P _jAnd P ₀Construct the focussing matrix T of each frequency _j, the propagation operator of different frequent points is focused on the same reference frequency point;

4) obtaining final propagation operator estimates

Structure Q and Q ₀, i.e. noise subspace;

5) combine feature space thought, structural wideband beamforming algorithm weight vector w _PM-BESB, realize that sane wideband adaptive wave beam forms.

2. a kind of broadband efficiently according to claim 1 sane adaptive beam formation method is characterized in that, said step 1), and concrete the realization as follows:

Consideration has the situation of the even linear array of M omnidirectional's array element, and array element distance is the half-wavelength of centre frequency.P broadband far-field signal incides on this array with

respectively, additional and source signal independent Gaussian white noise.With the data uniform sampling of array received and be divided into the piece of K non-overlapping copies, every comprises N sampled point.Each piece is carried out N point FFT, choose J frequency then and carry out subsequent treatment.Define then the k time snap (k=1 of each discrete frequency ; ..., N) the array received data of j frequency do

x _j，k＝A _j(θ)s _j，k+n _j，k

Here A _j(θ)=[a _j(θ ₁) ..., a _j(θ _P)] expression (the array flow pattern matrix of the dimension of M * P), s _{J, k}Be the signal phasor of (P * 1) dimension, n _{J, k}It is the noise vector of (M * 1) dimension.

3. a kind of broadband efficiently according to claim 2 sane adaptive beam formation method is characterized in that, after obtaining each frequency reception data, chooses center frequency point f ₀As focusing on frequency, then the k time snap frequency f ₀The array received data can be written as x _{0, k}=A ₀(θ) s _{0, k}+ n _{0, k}So f _jAnd f ₀The covariance matrix of array received data is on the frequency:

$\left\{\begin{array}{c}{R}_j=\frac{1}{K}{\mathrm{\&Sigma;}}_{k=1}^K{x}_{j,k}{x}_{j,k}^H\\ R_0=\frac{1}{K}{\mathrm{\&Sigma;}}_{k=1}^K{x}_{0,k}{x}_{0,k}^H\end{array}\right.$

4. a kind of broadband sane adaptive beam formation method efficiently according to claim 1 is characterized in that said step 2), the concrete realization as follows:

Respectively with each frequency array covariance matrix R _jAnd center frequency point array covariance matrix R ₀Carrying out piecemeal can get:

R _j＝[G _j，H _j]＝[G _j，G _jP _j]

R ₀＝[G ₀，H ₀]＝[G ₀，G ₀P ₀]

The G here _jAnd G ₀All be (dimension of M * P) matrix, H _jAnd H ₀It all is the dimension of M * (M-P) matrix.

5. a kind of broadband sane adaptive beam formation method efficiently according to claim 4 is characterized in that, utilizes propagation operator thought, and covariance matrix is carried out the propagation operator that piecemeal obtains each frequency:

$G_j{P}_j=H_j\&DoubleRightArrow;G_j^H{G}_j{P}_j=G_j^H{H}_j$

$\&DoubleRightArrow;P_j={\left(G_j^H{G}_j\right)}^{-1}{G}_j^H{H}_j$

In like manner can draw the propagation operator P of center frequency point ₀

6. a kind of broadband sane adaptive beam formation method efficiently according to claim 1 is characterized in that, said step 3), and concrete the realization as follows:

Utilize the propagation operator P of each frequency _jPropagation operator P with center frequency point ₀, the definition focussing matrix transforms to the propagation operator of different frequent points on the reference frequency point:

$T_j{P}_j^H=P_0^H$

Then each focusing transform matrix is:

$T_j=P_0^H{P}_j{\left(P_j^H{P}_j\right)}^{-1}$

7. a kind of broadband efficiently according to claim 1 sane adaptive beam formation method is characterized in that, said step 4), and concrete the realization as follows:

Utilize matrix theory knowledge to be not difficult to obtain following formula:

$H_j{T}_j^H=G_j{P}_j{T}_j^H=G_j{\left(T_j{P}_j^H\right)}^H=G_j{P}_0$

And then have:

$G_j^H{H}_j{T}_j^H=G_j^H{G}_j{P}_0$

The estimation of the propagation operator of single frequency after finally obtaining focusing on:

${\hat{P}}_0=\frac{1}{J}\left[\underset{j=1}{\overset{J}{\mathrm{\&Sigma;}}}{\left(G_j^H{G}_j\right)}^{-1}{G}_j^H{H}_j{T}_j^H\right]$

8. a kind of broadband efficiently according to claim 7 sane adaptive beam formation method; It is characterized in that; In conjunction with propagation operator thought; Utilize the matrix Q of dimension of estimation

the structure M of the propagation operator of single frequency * (M-P), and satisfy:

$Q=\left(\begin{array}{c}{\hat{P}}_0\\ -I_{(M-P)}\end{array}\right)$

And, obtain Q with the Q orthonormalization ₀=Q (Q ^HQ) ^(1/2), be equivalent to the noise subspace of reference frequency point.

9. a kind of broadband efficiently according to claim 1 sane adaptive beam formation method is characterized in that, said step 5), and concrete the realization as follows:

Utilize the focusing of feature space adaptive beam-forming algorithm thought and propagation operator theoretical, obtain broadband characteristics spatially adaptive wave beam and form (PM-BESB) algorithm weight vector

The R here ₀Refer to the covariance matrix of reference frequency point, a (θ ₀) what represent is the steering vector of desired signal, θ ₀Be the DOA of desired signal.Utilize weight vector to realize that sane wideband adaptive wave beam forms at last.

Images