WO2018045567A1

WO2018045567A1 - Robust stap method based on array manifold priori knowledge having measurement error

Info

Publication number: WO2018045567A1
Application number: PCT/CN2016/098599
Authority: WO
Inventors: 阳召成; 全桂华; 黄建军; 黄敬雄
Original assignee: 深圳大学
Priority date: 2016-09-09
Filing date: 2016-09-09
Publication date: 2018-03-15

Abstract

A robust STAP method based on array manifold priori knowledge having measurement error comprises the steps of: S1, obtaining a clutter space-time steering vector set according to a given error range; S2, searching for an important space-time steering vector in the clutter space-time steering vector set, and calculating an eigenvalue and an eigenvector of the important space-time steering vector; and S3, obtaining a clutter covariance matrix according to the eigenvalue and the eigenvector of the important space-time steering vector, and obtaining a filter weight vector according to the clutter covariance matrix. Errors unavoidably exist in obtaining array manifold knowledge, which directly causes processing performance limitation of STAP based on the array manifold knowledge. Compared with the STAP method based on array manifold knowledge in the prior art, the method reduces requirements for the accuracy of priori knowledge, has a robust characteristic for the priori knowledge having certain errors, and can avoid a process of clutter covariance matrix inversion in designing a filter, thereby achieving the objective of reducing the computation complexity of a system.

Description

A robust STAP method based on prior knowledge of array manifold with measurement error

Technical field

The invention belongs to the field of radar signal processing, and more particularly to a robust STAP method based on prior knowledge of array manifolds with measurement errors.

Background technique

In traditional Space-Time Adaptive Processing (STAP), such as Principal Components (PC), joint domain localized (JDL) algorithm, and feature subspace method based on cross-spectral scales (Cross-Spectral Metric, CSM), Multistage Winer Filter (MWF) algorithm, Joint Iterative Optimal Rank Reduction Adaptive Filter (JIOAF) method, etc. The number of training samples required is reduced to 2 times the dimensionality after dimensionality reduction or 2 times the wavelet rank after the reduced rank. Parametric Adaptive Matched Filter (PAMF) based on multi-channel vector autoregressive (AR) model reduces the number of required training samples from twice the system space-time degree of freedom to the AR model order. 2 times. However, these methods have a large number of training samples compared to the non-uniform clutter environment. .

Recently proposed array-based manifold knowledge STAP technology, using prior knowledge such as carrier height, speed, operating frequency, pulse repetition frequency (PRF), array antenna pointing, etc., to estimate clutter covariance in real environment. The matrix, and then the corresponding space-time filter, is designed to implement clutter suppression and moving target detection. Compared with the traditional STAP algorithm, this kind of algorithm greatly reduces the number of training samples required in the non-uniform environment of clutter, and can better adapt to complex and variable environments to achieve moving targets. Detection, showing superior performance. However, this type of algorithm requires high computational complexity, and its performance depends on the accuracy of prior knowledge, which will be detrimental to the application of the algorithm in practical systems.

Summary of the invention

Aiming at the defects of the prior art, the object of the present invention is to provide a robust STAP method based on prior knowledge of array manifolds with measurement errors, which aims to solve the measurement error of the array manifold knowledge obtained in the prior art. A problem that has a large impact on STAP performance.

The present invention provides a robust STAP method based on prior knowledge of array manifolds with measurement errors, including the following steps:

S1: obtaining a clutter space-time steering vector set according to a given error range;

S2: using a plurality of training samples, searching for an important space-time steering vector in the clutter space-time steering vector set, and calculating feature values and feature vectors of the important space-time steering vector;

S3: Obtain a clutter covariance matrix according to the feature value of the important space-time steering vector and the feature vector, and obtain an adaptive filter weight vector according to the clutter covariance matrix.

Further, in step S1, it is specifically:

S11: obtaining a maximum range of Doppler frequency error of a single clutter block |Δf _i,k |=Δf _max,i,k ; where i is a distance fuzzy number index, i=1,...,N _a , N _a is the number of fuzzy distance loops, k is the index of the number of discrete clutter blocks, k=1, . . . , N _c , N _c are the number of discrete clutter scatterers;

S12: Obtain a clutter block Doppler frequency f _i,k ∈[f′ _i,k −Δf _max,i,k ,f′ _i,k +Δf _max according to a Doppler frequency error range of a single clutter block _{, i, k} ], and construct the Doppler frequency subspace according to the clutter block Doppler frequency, and uniformly divide the Doppler frequency subspace of the clutter into N _f aliquots

Where f' _i,k is the Doppler frequency of a single clutter scatterer calculated based on prior knowledge, Δf _max,i,k is the maximum error range of the single clutter scatterer Doppler frequency, and g is discrete The Doppler frequency subspace index, N _f is the number of discrete Doppler frequency subspaces, g = 1, ..., N _f , f _{i, k, g} are discrete Doppler frequencies;

S13: obtaining a clutter space-time steering vector according to the azimuth angle φ _{i,k of the} single clutter block, the pitch angle θ _i,k and the Doppler frequency f _i,k,g

Where v _t (f _i,k,g ) is a time domain steering vector, and v _s (φ _i,k ,θ _i,k ) is a spatial domain steering vector;

S14: All the spurious space-time steering vectors v(φ _i,k , θ _i,k , f _i,k,g ) are formed into a set to form a clutter space-time steering vector set Φ.

Further, in step S2, it is specifically:

(2.1) Initialization: sample set B ₀ = [x ₁ , ..., x _L ], initial space-time steering vector γ ₀ = φ, termination condition: p _max , ε.

(2.2) Obtain the first important space-time steering vector as

γ ₁ ={(i ₁ ,k ₁ ,g ₁ )}, and its eigenvector is

Calculate its eigenvalue

And the residual vector is b _{l; 1} = b _{l; 0 -} λ _{l; 1} u _{c; 1} , l = 1, ..., L, p = 2;

(2.3) When the conditions are met

And when p-1 ≤ p _max , the following iterative process is performed:

(a)

(b)

(c)

(d) b _{l; p} = b _{l; p-1 -} λ _{l; p} u _{c; p} , l = 1, ..., L;

(e) p=p+1;

(2.4) When the iteration is terminated, the pth important space-time steering vector γ=γ _{p is obtained} , and the corresponding feature vector U _c =[u _c;1 ,...,u _c;p ] and the eigenvalue

Where L is the number of training samples obtained by the distance unit of interest and the neighboring distance unit; ||·|| _∞ is a l _∞ norm; p _max is the maximum number of iterations; ε is a normal number, indicating the iteration residual termination condition .

Further, in step S3, the estimated true clutter covariance matrix is

or

Where p is the number of iterations when finding the most important space-time steering vector; u _c;q is the qth eigenvector, p is the number of iterations,

Is the qth eigenvector (the qth average eigenvalue).

Further, the weight vector of the adaptive filter

among them,

To receive thermal noise power estimates, and

or

Diag(·) is a diagonal matrix.

In the present invention, due to the inevitable error in the acquisition of the array manifold knowledge, the processing performance of the STAP based on the array manifold knowledge is directly limited. Because the prior art lacks performance in considering the accuracy of array manifold knowledge, compared with the existing STAP method based on array manifold knowledge, the present invention reduces the requirement for the accuracy of prior knowledge, and the first is a certain error. The knowledge has robust characteristics, and avoids the process of inverting the clutter covariance matrix in the process of designing the filter, thus achieving the purpose of reducing the computational complexity of the system.

DRAWINGS

FIG. 1 is a flowchart of implementing a robust STAP method based on prior knowledge of array manifolds with measurement errors according to an embodiment of the present invention; FIG.

Fig. 2 is a graph showing the relationship between the SINR performance of the classical radar I system and the target Doppler frequency under different errors. Δψ _m is the yaw angle error and Δv _pm is the carrier speed error, where: (a) is the error Δψ _m =0.5° and Δv _pm =1m/s, and (b) is the error Δψ _m =1° and Δv _pm = 1m/s, (c) is the error Δψ _m =2.5° and Δv _pm =1m/s, (d) is the error Δψ _m =0.5° and Δv _pm =2m/s, and (e) is the error Δψ _m =0.5° And Δv _pm =3m/s, (f) is the error Δψ _m =0.5° and Δv _pm =4m/s;

Figure 3 is a graphical representation of the relationship between the SINR performance of the Classical Radar II system and the target Doppler frequency for different errors. Δψ _m is the yaw angle error, and Δv _pm is the carrier speed error, where (a) is the error Δψ _m =0.5° and Δv _pm =1m/s, and (b) is the error Δψ _m =1° and Δv _pm = 1m/s, (c) is the error Δψ _m =2.5° and Δv _pm =1m/s, (d) is the error Δψ _m =0.5° and Δv _pm =2m/s, and (e) is the error Δψ _m =0.5° And Δv _pm =3m/s, (f) is the error Δψ _m =0.5° and Δv _pm =4m/s;

4 is a schematic diagram showing the relationship between the SINR performance and the detection distance in the forward side view and the forward view direction of the radar I system; wherein (a) is the positive side view direction and (b) is the forward view direction;

5 is a schematic diagram showing the relationship between the SINR performance and the detection distance in the forward side view and the forward view direction of the radar II system; wherein (a) is the positive side view direction and (b) is the forward view direction;

6 is a graph of related system parameters of the classical radar I and II systems of the present invention;

7 is a schematic diagram of SINR performance curves under different methods; (a) is a relationship between SINR loss (SINRLoss) and the number of training samples, and (b) is a relationship between SINR loss (SINRLoss) and target Doppler frequency;

Fig. 8 is a graph showing the relationship between the detection probability Pd and the SINR under different methods.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

The invention relates to the field of radar signal processing, in particular to moving object detection and clutter suppression direction. A robust STAP method based on the prior knowledge of array manifold with measurement error is proposed. The prior knowledge of a certain error range, such as carrier speed and yaw angle, is formed to form a clutter space-time steering vector in real environment. The set, and then a small number of training samples to estimate the clutter covariance matrix in the real environment, and finally design a robust STAP filter with low computational complexity, so as to achieve the purpose of clutter suppression and target detection.

In order to reduce the influence of prior knowledge of measurement error on the performance of STAP method and the high computational complexity when estimating covariance matrix, a robust STAP method based on prior knowledge of array manifold with measurement error is proposed. .

Under ideal conditions, the space-time snapshot model with clutter and noise can be expressed as: x=Vσ+n;

For the guidance dictionary of clutter, v(φ _i,k , θ _i,k ,f _i,k ) is the steering vector of the ikth clutter block, f _i,k =2v _p cosθ _i,k sin(φ _{i , k} + ψ) / λ _c (v _p , λ _c , ψ respectively for carrier speed, wavelength and yaw angle) are Doppler frequencies, φ _{i, k} , θ _{i, k} , respectively azimuth, pitch angle. σ is the complex amplitude of all clutter blocks, n is the mean zero, and the variance is

Gaussian white noise.

In practice, the prior knowledge we obtained, such as carrier speed and yaw angle, is subject to error, which will lead to deviations from the assumed clutter space-time steering vector in practice. The invention is based on the inaccurate array manifold prior knowledge, and designs a robust STAP method based on the prior knowledge of array manifolds with measurement errors. The core idea of the invention is to estimate the clutter covariance matrix in the real environment by using the inaccurate array manifold prior knowledge in the clutter model, and design a robust space-time filter to realize clutter suppression and target detection.

A robust STAP method based on prior knowledge of array manifolds with measurement errors provided by an embodiment of the present invention specifically includes the following steps:

(1) using a priori knowledge of a given error range to form a clutter space-time steering vector set in a real environment;

Suppose platform velocity and the actual yaw system were measured v _'p, ψ', and there is _{_{v 'p ∈ [v p -Δv}} pm, v p + Δv pm], ψ'∈ [ψ-Δψ _m , ψ + Δ ψ _m ] respectively obey the uniform distribution, the error of the carrier speed and the yaw angle are Δv _p = v' _p - v _p , Δ ψ = ψ ' - 分别, and | Δv _p | ≤ Δv _pm And |Δψ|≤Δψ _m . Thus, the Doppler frequency error range for a single clutter block can be calculated as:

Therefore, the actual clutter block Doppler frequency f _i,k ∈[f' _i,k -Δf _max,i,k ,f' _i,k +Δf _max,i,k ] can constitute a Doppler Frequency subspace. Evenly divide the Doppler frequency subspace of the clutter into N _f aliquots

Therefore, according to the azimuth angle φ _{i,k of the} single clutter block, the elevation angle θ _i,k and the true Doppler frequency f _i,k,g , the real clutter space-time steering vector can be obtained.

among them,

Representing the Kronocker product, v _t (f _i,k,g ), v _s (φ _i,k ,θ _i,k ) are the time domain and spatial domain steering vectors of the ikth clutter block, respectively. Direct all space-timed dictionaries v(φ _i,k ,θ _i,k ,f _i,k,g )i=1,...,N _a ,k=1,....,N _c ,g =1,...,N _f forms a clutter space-time steering vector set Φ.

(2) From the obtained clutter space-time steering vector set, find the most important space-time steering vector and calculate the corresponding feature value and feature vector;

This step is the core idea of the present invention. According to the training sample x of a certain distance unit, a method similar to orthogonal matching pursuit is proposed, which is selected from the obtained clutter space-time steering vector set Φ. The important space-time steering vector is calculated, and the corresponding feature vector U _c =[u _c;1 , u _c;2 ,...] and the eigenvalue λ=[λ ₁ ,λ ₂ ,...] ^{T are} calculated.

The specific method flow is as follows:

(2.1) Initialization: Set the initial space-time steering vector to γ ₀ = φ, and the residual vector b ₀ = x

(2.2) When the number of iterations p=1: find the first most important vector from the set Φ

Calculate the corresponding eigenvector

And eigenvalue

After the first iteration, the residual vector b ₁ =b ₀ -z ₁ u _c;1 .

(2.3) When the number of iterations is p≥2: Assuming that the p-1th important space-time steering vector is found to be γ _p-1 and the residual vector is b _p-1 , then the p-th important space-time steering vector is

Calculate the corresponding eigenvector

And eigenvalues

(2.4) Termination condition: when the number of iterations reaches the set iteration extremum (ie p≥p _max ) or satisfies the condition ||Φ ^H b _p || _∞ ≤ ε (where ||·|| _∞ is l _∞ norm When ε is a normal number, the above iterative process needs to be terminated. Therefore, the most important clutter space-time steering vector finally obtained is γ=γ _p , and the corresponding eigenvector is U _c =[u _c;1 , u _c;2 ,...u _c;p ], and the eigenvalue is λ = [λ ₁ , λ ₂ , ... λ _p ] ^T .

In order to improve the accuracy of the selected space-time steering vector, the present invention can obtain a plurality of training samples by using the distance of interest unit and the adjacent distance unit, select the most important space-time steering vector γ, and calculate the feature vector and the feature value. Suppose that the sample matrix consisting of L training samples is X=[x ₁ ,...,x _L ], and the residual matrix is B=[b ₁ ,...,b _L ], eigenvalue matrix

Find γ using multiple samples, calculate eigenvector U _c and eigenvalue

The method process is:

(2.2.1) Initialization: B ₀ = [x ₁ , ..., x _L ], γ ₀ = φ, termination condition: p _max , ε.

(2.2.2) The first vector γ:

b _l;1 =b _l;0 -λ _l;1 u _c;1 , l=1,...,L,p=2.

(2.2.3) Satisfying the conditions

And when p-1 ≤ p _max , the following iterative process is performed

(a)

(b)

(c)

(d)b _l;p =b _l;p-1 -λ _l;p u _c;p ,l=1,...,L,

(e) p=p+1.

(2.2.4) γ=γ _p , U _c =[u _c;1 ,...,u _c;p ], and

among them

(3) Estimate the clutter covariance matrix and design a robust space-time filter.

Using the obtained feature vector U _c of the most important steering vector and the eigenvalue λ (or

To estimate the true clutter covariance matrix, and design a robust space-time filter to achieve clutter suppression and target detection.

The estimated clutter covariance matrix is

or

Therefore, the space-time adaptive processing of the weight vector of the adaptive filter is

among them

To receive thermal noise power estimates, and

or

In the above solution of the filter weight vector, the process of inverting the covariance matrix of the clutter is avoided, which has the advantages of reducing the computational complexity of the system and saving the actual cost compared with the traditional STAP filter weight vector solution.

In order to further illustrate a robust STAP method based on prior knowledge of array manifold with measurement error provided by an embodiment of the present invention, the present invention is compared with the prior art to illustrate the beneficial effects of the present invention; as follows:

Since the present invention requires to know a priori knowledge of the airborne velocity v _p and the error range of the yaw angle ψ, the portion from the lower error rate and different onboard yaw angle, signal to interference noise ratio of the analysis system (SINR) and The relationship of the Doppler frequency of the target is compared to existing methods.

It can be seen from Fig. 2 and Fig. 3 that the present invention has more comparison with the LSE method (STAP filter designed by using the space-time guided vector formed by the measured airborne speed and yaw angle) under different measurement errors. Good performance. This is due to the existence of measurement errors. The space-time steering vector formed in the LSE method does not represent the true steering vector, so that the estimated clutter covariance matrix is not accurate enough, thus reducing the performance of clutter suppression. However, the method of the present invention can maintain good performance for various errors and exhibits good robustness, because the proposed method takes measurement errors into consideration in the assumed space-time steering vector, to some extent. Contains or approximates the real clutter subspace.

As can be seen from FIG. 4 and FIG. 5, the method of the invention has better performance and better robustness than the LSE method under the condition that the prior knowledge has measurement error. At the same time, the existence of the distance ambiguity problem has little effect on the SINR performance of the two methods. Although the SINR performance of the present invention drops by 1-2 dB under high pulse repetition frequency radar due to distance blur, under the premise of high computational complexity It is acceptable. Figure 6 shows the relevant system parameters of the classical radar I, II system of the present invention.

In order to fully embody the advantages of the present invention, this section will compare SINR performance and PD performance indicators with other methods. The main comparison methods are: 4×3 JDL method, PAMF method, CSMIECC method, Stoica scheme, KAPE method, PAMF method based on array manifold knowledge.

It can be concluded from Fig. 7(a) that the present invention can obtain a performance lower than the optimal space-time filter performance by -2 dB using a single training sample; the method based on the array manifold knowledge has a comparison with the conventional STAP method. Better accuracy and convergence, because the former uses prior knowledge to calculate the covariance matrix of the clutter. From the SINR performance and the target Doppler relationship graph in Fig. 7(b), it can be concluded that the method based on prior knowledge can exhibit better performance in a small number of samples or even a single sample, and the present invention compares other methods. Have a better advantage. As can be seen from FIG. 8, the present invention and the method based on array manifold prior knowledge have better target detection performance than the conventional STAP method.

In summary, the method proposed in the present invention considers the measurement error of the array manifold knowledge in the assumed space-time steering vector, oversamps the constructed clutter subspace to form a real hybrid waveguide vector, and selects the most important space. The time-directed vector can obtain the clutter subspace more accurately than the LSE method. At the same time, the present invention can obtain better clutter suppression effect under a single training than the conventional SATP method, and has better SINR performance and target detection performance than the existing array manifold knowledge (presence error) STAP method. .

Those skilled in the art will appreciate that the above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and scope of the present invention, All should be included in the scope of protection of the present invention.

Claims

A robust STAP method based on prior knowledge of array manifolds with measurement errors, comprising the steps of:

S1: obtaining a clutter space-time steering vector set according to a given error range;

S2: using a plurality of training samples, searching for an important space-time steering vector in the clutter space-time steering vector set, and calculating feature values and feature vectors of the important space-time steering vector;

S3: Obtain a clutter covariance matrix according to the feature value of the important space-time steering vector and the feature vector, and obtain an adaptive filter weight vector according to the clutter covariance matrix.
The robust STAP method according to claim 1, wherein the step S1 is specifically:

S11: obtaining a maximum range of Doppler frequency error of a single clutter block |Δf i,k |=Δf max,i,k ; where i is a distance fuzzy number index, i=1,...,N a , N a is the number of fuzzy distance loops, k is the index of the number of discrete clutter blocks, k=1, . . . , N c , N c are the number of discrete clutter scatterers;

S12: Obtain a clutter block Doppler frequency f i,k ∈[f′ i,k −Δf max,i,k f′ i,k +Δf max,i according to a Doppler frequency error range of a single clutter block , k ], and construct the Doppler frequency subspace according to the clutter block Doppler frequency, and uniformly divide the Doppler frequency subspace of the clutter into N f aliquots

Where f' i,k is the Doppler frequency of a single clutter scatterer calculated based on prior knowledge, Δf max,i,k is the maximum error range of the single clutter scatterer Doppler frequency, and g is discrete The Doppler frequency subspace index, N f is the number of discrete Doppler frequency subspaces, g = 1, ..., N f , f i, k, g are discrete Doppler frequencies;

S13: obtaining a clutter space-time steering vector according to the azimuth angle φ i,k of the single clutter block, the pitch angle θ i,k and the Doppler frequency f i,k,g

Where v t (f i,k,g ) is a time domain steering vector, and v s (φ i,k ,θ i,k ) is a spatial domain steering vector;

S14: All the spurious space-time steering vectors v(φ i,k , θ i,k , f i,k,g ) are formed into a set to form a clutter space-time steering vector set Φ.
The robust STAP method according to claim 1 or 2, wherein the step S2 is specifically:

(2.1) Initialization: sample set B 0 = [x 1 , ..., x L ], initial space-time steering vector γ 0 = φ, termination condition: p max , ε.

(2.2) Obtain the first important space-time steering vector as

γ 1 ={(i 1 ,k 1 ,g 1 )}, and its eigenvector is

Calculate its eigenvalue
And the residual vector is

b l;1 =b l;0 -λ l;1 u c;1 ,l=1,...,L,p=2;

(2.3) When the conditions are met
And when p-1 ≤ p max , the following iterative process is performed:

(d) b l; p = b l; p-1 - λ l; p u c; p , l = 1, ..., L;

(e) p=p+1;

(2.4) When the iteration is terminated, the pth important space-time steering vector γ=γ p is obtained , and the corresponding feature vector U c =[u c:l ,...,u c:p ] and the eigenvalue

Where L is the number of training samples obtained by the distance unit of interest and the neighboring distance unit; ||·|| ∞ is a l ∞ norm; p max is the maximum number of iterations; ε is a normal number, indicating the iteration residual termination condition .
The robust STAP method of claim 1 wherein in step S3, the estimated real clutter covariance matrix is
or
Where p is the number of iterations when finding the most important space-time steering vector; u c; q is the qth eigenvector, p is the number of iterations, λ q
Is the qth eigenvector (the qth average eigenvalue).
A robust STAP method according to claim 4, wherein the weight vector of said adaptive filter

among them,
To receive thermal noise power estimates, and
or
Diag(·) is a diagonal matrix.