Subspace signal detection method and system suitable for scene with insufficient training samples

Abstract

The invention belongs to the technical field of multi-channel signal detection, and particularly relates to a subspace signal detection method and system suitable for a scene with insufficient training samples, and the method comprises the steps: firstly constructing a target signal matrix, an interference signal matrix and a training sample matrix, and carrying out the skew symmetry processing of the constructed matrixes; constructing a splicing matrix by using the target signal matrix and the interference signal matrix; constructing an orthogonal projection matrix by using the interference signal matrix and the splicing matrix after the skew symmetry processing; using the transformed sampling covariance matrix to construct a quasi-whitening matrix for whitening processing; constructing a detection statistic by using the whitened matrix; and determining a detection threshold by using a false alarm probability preset value, comparing the detection statistic with the detection threshold, and judging whether a target exists or not. The detector designed by the invention integrally realizes interference suppression, signal accumulation and constant false alarm characteristics, and compared with the existing method, the method has the advantage that the detection performance is improved in a scene in which training samples are insufficient.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-channel signal detection, and in particular relates to a subspace signal detection method and system suitable for scenarios with insufficient training samples. Background Art

[0002] In airborne radar multi-channel systems, the problem of target adaptive detection in scenarios with insufficient training samples has always been a key and difficult problem in the field of space-time adaptive detection (STAD). In actual scenarios, the reasons for insufficient training samples are complex and diverse, such as the multipath effect of target echoes and the non-uniform fluctuations of the detection environment. In existing research results, many methods have emerged to address the problem of insufficient training samples, such as directly reducing the demand for training samples through data dimensionality reduction and rank reduction, and indirectly reducing the detector's demand for training samples by using the prior information of the noise covariance matrix.

[0003] It should be pointed out that most of the existing adaptive detection methods for insufficient training samples are proposed under the assumption that there is no interference. However, in actual scenarios, interference signals are often ubiquitous in the detection environment, such as internal interference caused by the system itself and malicious interference released by the enemy. The existence of these interferences will further lead to the problem of insufficient training samples. In the field of STAD, interference is usually modeled as existing in a certain known subspace, but with unknown coordinates. The impact of subspace interference on the radar detection system is omnidirectional. When it enters the radar system from the main lobe direction, it will raise the threshold of target detection and cause missed detection; when it enters the radar detection system from the side lobe, it may appear in the form of a false target, causing a false alarm. In addition, the appearance of subspace interference will cause a sharp decrease in the number of available training samples in the detection background, that is, it is difficult to obtain sufficient training samples that are independent and identically distributed with the samples to be detected, which makes the performance of the existing space-time adaptive detector drop sharply.

[0004] For the problem of insufficient training samples caused by subspace interference, the existing method is to integrate the subspace interference information into the detection model and design the detector based on the generalized likelihood ratio test (GLRT) criterion. However, the GLRT detector can only achieve good detection performance in some scenarios, and does not consider the problem of reduced number of training samples caused by subspace interference. Based on this consideration, the skew-symmetric prior information of background clutter in the present invention is integrated into the detection model to reduce the demand for training samples; then, based on the Rao criterion, a skew-symmetric space-time adaptive detector with excellent performance suitable for scenarios with insufficient training samples is designed. Summary of the invention

[0005] The purpose of the present invention is to provide a subspace signal detection method and system suitable for scenarios with insufficient training samples, which can integrate interference suppression, signal accumulation and constant false alarm characteristics, and improve the detection performance in scenarios with insufficient training samples compared with existing methods.

[0006] The technical solution adopted by the present invention is as follows:

[0007] A subspace signal detection method suitable for a scenario with insufficient training samples comprises the following steps:

[0008] Step 1: Construct the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected, where the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected are represented by A, J, S and X respectively, and the dimensions of A, J, S and X are N×p, N×q, N×N and N×K respectively, where (p+q)≤N, K represents the extended dimension of the target, N represents the number of system channels, and the unitary matrices A and J constitute the signal subspace and interference subspace <j>, the sampling covariance matrix is ​​expressed as S = YY H , Y represents the training sample matrix, symbol (·) H represents conjugate transpose;

[0009] Step 2: Use the target signal matrix A and the interference signal matrix J to construct a concatenated matrix B with a dimension of N×(p+q). The unitary matrix can construct the subspace ;

[0010] Step 3: Construct a skew-symmetric transformation matrix M with a dimension of N×N, and then use the skew-symmetric transformation matrix M to perform skew-symmetric transformation on A, J, B and S to obtain the corresponding matrix with skew-symmetric characteristics;

[0011] Step 4: Use the sampling covariance matrix S after skew symmetric transformation to perform quasi-whitening on matrices X, A, J, and B, and obtain the quasi-whitening matrices and

[0012] Step 5: Use the quasi-whitening matrix Constructing the orthogonal projection matrix of the interference signal And the corresponding complementary projection matrix At the same time, using the quasi-whitening matrix Construct the orthogonal projection matrix of the stitching matrix B And the corresponding complementary projection matrix

[0013] Step 6: Use quasi-whitening to detect the data matrix Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix Construct test statistics;

[0014] Step 7: Determine the detection threshold using the detection statistic and the preset value of false alarm probability;

[0015] Step 8: Compare the detection statistic with the detection threshold and determine whether the target exists.

[0016] Furthermore, in step 5, the quasi-whitening interference signal matrix The constructed orthogonal projection matrix and complementary projection matrix are expressed as and Using quasi-whitening splicing matrix The constructed orthogonal projection matrix and complementary projection matrix are expressed as and in(·) -1 represents the inverse of the matrix, I N is the identity matrix of dimension N×N.

[0017] Furthermore, in step 1, the constructed target signal matrix A and interference signal matrix J are both column-full rank unitary matrices, and the unitary matrices A and J are linearly independent.

[0018] Furthermore, in step 3, the constructed skew-symmetric transformation matrix M is expressed as:

[0019]

[0020] in, F N is a permutation matrix with dimension N×N, whose anti-diagonal element value is equal to 1, and the remaining position elements are equal to 0; the skew-symmetric transformation matrix M is used to perform skew-symmetric transformation on the target signal matrix, interference signal matrix and sampling covariance matrix, respectively, which are expressed as and

[0021] Furthermore, in step 4, the sampling covariance matrix S after skew symmetric transformation is used to perform quasi-whitening on the matrices X, A, J and B, and the obtained quasi-whitened data matrix to be detected, quasi-whitened target signal matrix, quasi-whitened interference signal matrix and quasi-whitened splicing matrix are respectively expressed as and

[0022] Furthermore, in step 6, the quasi-whitening of the data matrix to be detected is performed. Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix The constructed detection statistic is expressed as:

[0023]

[0024] where tr(·) represents the trace of the matrix, I2 represents a 2×2 dimensional unit matrix. Further, in step 7, the detection threshold determined by using the detection statistic and the preset value of the false alarm probability is:

[0025] η=t(n)

[0026] Wherein, η represents the detection threshold, n = [Qμ], Q is the number of Monte Carlo simulations, μ represents the false alarm probability value set by the system, [·] represents the rounding operation, and t(n) is the sequence;

[0027]

[0028] The nth largest value when arranged from largest to smallest, X(k) represents the k-th realization k=1, 2,…,Q of the data to be detected containing only interference and thermal noise components.

[0029] A subspace signal detection system suitable for scenarios with insufficient training samples, comprising the following modules:

[0030] Signal data matrix construction module: used to construct target signal matrix, interference signal matrix and training sample matrix, and use target signal matrix and interference signal matrix to construct splicing matrix;

[0031] Sampling covariance matrix construction module: used to construct the sampling covariance matrix using training samples;

[0032] Skew-symmetric transformation module: used to perform skew-symmetric transformation on target signal matrix, interference signal matrix and training sample matrix using skew-symmetric transformation matrix;

[0033] Data quasi-whitening module: used to perform quasi-whitening processing on the target signal matrix, the interference signal matrix and the data matrix to be detected, and obtain the quasi-whitened target signal matrix, the quasi-whitened interference signal matrix and the quasi-whitened data matrix to be detected respectively;

[0034] Detection statistic construction module: used to construct the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix, and construct the detection statistic using the quasi-whitened data matrix to be detected, the quasi-whitened target signal matrix, the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix;

[0035] Detection threshold calculation module: used to determine the detection threshold according to the preset value of false alarm probability;

[0036] Target decision module: used to compare the detection statistic with the detection threshold and determine whether the target exists.

[0037] A subspace signal detection method suitable for scenarios with insufficient training samples is applied to detect radar targets under the condition that the signal-to-noise ratio is higher than 17dB.

[0038] The technical effects achieved by the present invention are:

[0039] A subspace signal detection method and system suitable for a scene with insufficient training samples of the present invention utilizes a skew-symmetric transformation matrix M to perform skew-symmetric transformation processing on a target signal matrix, an interference signal matrix and a sampling covariance matrix, thereby reducing the demand for training samples of the detection method of the present invention, and by utilizing a sampling covariance matrix S to perform quasi-whitening processing on a data matrix to be detected, a target signal and an interference signal matrix, the detector proposed by the present invention is integrated to realize interference suppression, signal accumulation and constant false alarm characteristics, and compared with the existing method, the detection performance is improved in a scene with insufficient training samples, and the target detection problem in a scene with insufficient training samples can be solved, and compared with the existing detection method based on the GLRT criterion, the radar target detection performance is improved in a scene with insufficient training samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] < / j> Figure 1 It is a schematic flow chart of the method of the present invention;

[0041] Figure 2 It is a system structure framework diagram of the present invention;

[0042] Figure 3 is the detection probability of the target under different signal-to-noise ratios SNR of the present invention;

[0043] Figure 4 It is the detection probability of the target under different SNRs in the scenario where the training samples are insufficient. DETAILED DESCRIPTION

[0044] In order to make the purpose and advantages of the present invention more clearly understood, the present invention is specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present invention, and does not strictly limit the scope of protection of the specific claims of the present invention.

[0045] In this technical solution, it is assumed that the target signal s and the interference signal h are N×1 dimensional vectors, and both are located in the subspace formed by the unitary matrices A and J, and the corresponding subspace coordinates are φ and ψ, that is, the target signal and the interference signal can be expressed as:

[0046] s=Aφ Formula (1)

[0047] and

[0048] h=Jψ Formula (2)

[0049] Among them, A is an N×p-dimensional column full rank matrix, J is an N×q-dimensional column full rank matrix, and φ and ψ are vectors of dimensions p×1 and q×1, respectively.

[0050] Let n be the clutter and noise components in the data matrix x to be detected, and satisfy the mean value of 0 N×1 , a complex Gaussian distribution with covariance matrix R. Assume that there are L independent and identically distributed training samples x l , l = 1, 2, ..., L, which contains clutter and noise components n e,l , and also obeys the mean of 0 N×1 , a complex Gaussian distribution with covariance matrix R. In summary, the binary hypothesis detection problem in the subspace interference environment can be expressed as:

[0051]

[0052] Among them, H0 means that the data to be detected only contains subspace interference signals and clutter, and H1 means that the data to be detected contains target signals, subspace interference signals and clutter components.

[0053] In order to reduce the detector's demand for training samples and make it more adaptable to the subspace interference detection environment, the skew-symmetric transformation matrix M is used to perform skew-symmetric transformation on each signal component in equation (3). The structure of the skew-symmetric transformation matrix M is designed as:

[0054]

[0055] in, I N is the identity matrix with dimension N×N; F N is a permutation matrix with dimension N×N, whose anti-diagonal element value is equal to 1 and the remaining position elements are equal to 0. After skew symmetric transformation, the detection model in formula (3) can be expressed as:

[0056]

[0057] Among them, Φ=[Re(φ),jIm(φ)], Ψ=[Re(ψ),jIm(ψ)], X=[x e ,x o ], Re(·) and Im(·) represent the real and imaginary part operations respectively, x e and x o The expressions are:

[0058]

[0059] in(·) * Indicates conjugation.

[0060] Embodiment 1:

[0061] like Figure 1-Figure 4 As shown, the present invention provides a subspace signal detection method suitable for a scenario with insufficient training samples, comprising the following steps:

[0062] Step 1: Construct the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected, where the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected are represented by A, J, S and X respectively, and the dimensions of A, J, S and X are N×p, N×q, N×N and N×K respectively, where (p+q)≤N, K represents the extended dimension of the target, N represents the number of system channels, and the unitary matrices A and J constitute the signal subspace and interference subspace <j>, the sampling covariance matrix is ​​expressed as S = YY H , Y represents the training sample matrix, symbol (·) H represents conjugate transpose;

[0063] Step 2: Use the target signal matrix A and the interference signal matrix J to construct a concatenated matrix B with a dimension of N×(p+q). The unitary matrix can construct the subspace ;

[0064] Step 3: Construct a skew-symmetric transformation matrix M with a dimension of N×N, and then use the skew-symmetric transformation matrix M to perform skew-symmetric transformation on A, J, B and S to obtain the corresponding matrix with skew-symmetric characteristics;

[0065] Step 4: Use the sampling covariance matrix S after skew symmetric transformation to perform quasi-whitening on matrices X, A, J, and B, and obtain the quasi-whitening matrices and

[0066] Step 5: Use the quasi-whitening matrix Constructing the orthogonal projection matrix of the interference signal And the corresponding complementary projection matrix At the same time, using the quasi-whitening matrix Construct the orthogonal projection matrix of the stitching matrix B And the corresponding complementary projection matrix

[0067] Step 6: Use quasi-whitening to detect the data matrix Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix Construct test statistics;

[0068] Step 7: Determine the detection threshold using the detection statistic and the false alarm probability preset value, where the false alarm probability preset value is 10 -4 ;

[0069] Step 8: Compare the detection statistic with the detection threshold and determine whether the target exists.

[0070] In step 1, the constructed target signal matrix A and interference signal matrix J are both full-rank unitary matrices, and the unitary matrices A and J are linearly independent. The target signal subspace can be constructed using matrices A and J respectively.< / j> and interference signal subspace <j>, and then interference suppression is performed through the uncorrelated characteristics between the two subspaces.

[0071] In step 3, the constructed skew-symmetric transformation matrix M is expressed as:

[0072]

[0073] in, F N is a permutation matrix with dimension N×N, whose anti-diagonal element value is equal to 1, and the remaining position elements are equal to 0. The skew-symmetric transformation matrix M is used to perform skew-symmetric transformation on the target signal matrix, interference signal matrix and sampling covariance matrix, which are expressed as and Using the matrix after skew symmetric transformation and The designed detector can make effective use of prior information.

[0074] Specifically, in step 4, the sampling covariance matrix S after skew symmetric transformation is used to perform quasi-whitening on the matrices X, A, J and B. The obtained quasi-whitened data matrix to be detected, quasi-whitened target signal matrix, quasi-whitened interference signal matrix and quasi-whitened splicing matrix are respectively expressed as and The above quasi-whitening process can effectively reduce the influence of white noise on the detector.

[0075] Specifically, in step 5, the quasi-whitened interference signal matrix The constructed orthogonal projection matrix and complementary projection matrix are expressed as and Using quasi-whitening splicing matrix The constructed orthogonal projection matrix and complementary projection matrix are expressed as and in(·) -1 represents the inverse of the matrix, I N is the identity matrix of dimension N×N.

[0076] Specifically, in step 6, the quasi-whitening of the data matrix to be detected is performed Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix The constructed detection statistic is expressed as:

[0077]

[0078] where tr(·) represents the trace of the matrix, I2 represents a 2×2 dimensional unit matrix. Specifically, in step 7, the detection threshold determined by using the detection statistic and the preset value of the false alarm probability is:

[0079] η=t(n)

[0080] Where η represents the detection threshold, n = [Qμ], Q is the number of Monte Carlo simulations, μ represents the false alarm probability value set by the system, [·] represents the rounding operation, and t(n) is the sequence

[0081]

[0082] The nth largest value when arranged from largest to smallest, (·) H represents the conjugate transpose transformation, X(k) represents the k-th realization k=1, 2,…,Q of the data to be detected containing only interference and thermal noise components.

[0083] Embodiment 1:

[0084] See also< / j> Figure 2 As shown, the present invention provides a subspace signal detection system suitable for a scenario with insufficient training samples based on Example 1, including the following modules:

[0085] Signal data matrix construction module: used to construct target signal matrix, interference signal matrix and training sample matrix, and use target signal matrix and interference signal matrix to construct splicing matrix;

[0086] Sampling covariance matrix construction module: used to construct the sampling covariance matrix using training samples;

[0087] Skew-symmetric transformation module: used to perform skew-symmetric transformation on target signal matrix, interference signal matrix and training sample matrix using skew-symmetric transformation matrix;

[0088] Data quasi-whitening module: used to perform quasi-whitening processing on the target signal matrix, the interference signal matrix and the data matrix to be detected, and obtain the quasi-whitened target signal matrix, the quasi-whitened interference signal matrix and the quasi-whitened data matrix to be detected respectively;

[0089] Detection statistic construction module: used to construct the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix, and construct the detection statistic using the quasi-whitened data matrix to be detected, the quasi-whitened target signal matrix, the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix;

[0090] Detection threshold calculation module: used to determine the detection threshold according to the preset value of false alarm probability;

[0091] Target judgment module: used to compare the detection statistic with the detection threshold and determine whether the target exists.

[0092] Embodiment 3:

[0093] like Figure 1-4 As shown, this embodiment discloses a simulation experiment based on the above embodiment to further illustrate the effect of the present invention. The simulation experiment process is as follows:

[0094] Assume that the number of radar system channels is N = 12, and the default value of false alarm probability is 1×10 -4 , the element at the (i, j)th position of the covariance matrix R of the clutter and noise is set to R(i, j) = 0.95 |i-j| , i, j = 1, 2,…, N, where the symbol |·| indicates taking the absolute value. Figure 3 The detection probability of the target under different signal-to-noise ratios (SNRs) of the proposed method is given, where the subspace dimensions of the target signal and the interference signal are p=4 and q=4, respectively, the number of training samples is L=2N, the interference-to-noise ratio (INR) is 5dB, and the INR and SNR are:

[0095] INR=ψ H J H R -1 Jψ

[0096] and

[0097] SNR=φ H A H R -1 Aφ

[0098] from Figure 3 It can be seen that when the signal-to-noise ratio is higher than 17dB, the detection probability of the proposed method for the target is higher than 90%; in addition, compared with the existing GLRT criterion detector, the proposed detection method has better detection performance. Figure 4 The detection probability of the target under different SNRs in the scenario with insufficient training samples is given, where p=3, q=2, the number of training samples L=N, and the interference-to-noise ratio INR=10dB.

[0099] From the results, it can be seen that in the scenario with insufficient training samples, the performance of the proposed Rao criterion detector is significantly better than the existing GLRT detector.

[0100] In summary, the present invention utilizes a skew-symmetric transformation matrix M to perform skew-symmetric transformation processing on a target signal matrix, an interference signal matrix, and a sampling covariance matrix, thereby reducing the demand for training samples for the detection method of the present invention. By utilizing a sampling covariance matrix S to perform quasi-whitening processing on a data matrix to be detected, a target signal, and an interference signal matrix, the detector proposed in the present invention is integrated to achieve interference suppression, signal accumulation, and constant false alarm characteristics. Compared with existing methods, the detection performance is improved in scenarios with insufficient training samples, and the target detection problem in scenarios with insufficient training samples can be solved. Compared with the existing detection method based on the GLRT criterion, the radar target detection performance is improved in scenarios with insufficient training samples.

[0101] The above is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications should also be considered as the protection scope of the present invention. The structures, devices and operating methods not specifically described and explained in the present invention shall be implemented according to the conventional means in the art unless otherwise specified and limited.

Claims

1. A subspace signal detection method suitable for scenarios with insufficient training samples, characterized by: The following steps are involved: Step 1: Construct the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected, where the target signal matrix, interference signal matrix, sampling covariance matrix and data matrix to be detected are represented by A, J, S and X respectively, and the dimensions of A, J, S and X are N×p, N×q, N×N and N×K respectively, where (p+q)≤N, K represents the extended dimension of the target, N represents the number of system channels, and the unitary matrices A and J constitute the signal subspace and interference subspace <j>, the sampling covariance matrix is ​​expressed as S = YY H , Y represents the training sample matrix, symbol (·) H represents conjugate transpose;< / j> Step 2: Use the target signal matrix A and the interference signal matrix J to construct a concatenated matrix B with a dimension of N×(p+q). The unitary matrix can construct the subspace ; Step 3: Construct a skew-symmetric transformation matrix M with a dimension of N×N, and then use the skew-symmetric transformation matrix M to perform skew-symmetric transformation on A, J, B and S to obtain the corresponding matrix with skew-symmetric characteristics; Step 4: Use the sampling covariance matrix S after skew symmetric transformation to perform quasi-whitening on matrices X, A, J, and B, and obtain the quasi-whitening matrices and Step 5: Use the quasi-whitening matrix Constructing the orthogonal projection matrix of the interference signal And the corresponding complementary projection matrix At the same time, using the quasi-whitening matrix Construct the orthogonal projection matrix of the stitching matrix B And the corresponding complementary projection matrix Step 6: Use quasi-whitening to detect the data matrix Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix Construct test statistics; Step 7: Determine the detection threshold using the detection statistic and the preset value of false alarm probability; Step 8: Compare the detection statistic with the detection threshold and determine whether the target exists.

2. The subspace signal detection method applicable to a scenario with insufficient training samples according to claim 1, characterized in that: In step 5, the quasi-whitening interference signal matrix is ​​used The constructed orthogonal projection matrix and complementary projection matrix are expressed as and Using quasi-whitening splicing matrix The constructed orthogonal projection matrix and complementary projection matrix are expressed as and in(·) -1 represents the inverse of the matrix, I N is the identity matrix of dimension N×N.

3. The subspace signal detection method applicable to a scenario with insufficient training samples according to claim 1, characterized in that: In step 1, the constructed target signal matrix A and interference signal matrix J are both column-full rank unitary matrices, and the unitary matrices A and J are linearly independent.

4. The subspace signal detection method applicable to a scenario with insufficient training samples according to claim 1, characterized in that: In step 3, the constructed skew-symmetric transformation matrix M is expressed as: in, F N is a permutation matrix with dimension N×N, whose anti-diagonal element value is equal to 1, and the remaining position elements are equal to 0; the skew-symmetric transformation matrix M is used to perform skew-symmetric transformation on the target signal matrix, interference signal matrix and sampling covariance matrix, respectively, which are expressed as and 5. The subspace signal detection method applicable to a scenario with insufficient training samples according to claim 1, characterized in that: In step 4, the sampling covariance matrix S after skew symmetric transformation is used to perform quasi-whitening on matrices X, A, J and B, and the obtained quasi-whitened data matrix to be detected, quasi-whitened target signal matrix, quasi-whitened interference signal matrix and quasi-whitened splicing matrix are respectively expressed as and 6. The subspace signal detection method applicable to a scenario with insufficient training samples according to claim 1, characterized in that: In step 6, the quasi-whitening data matrix to be detected Quasi-whitened target signal matrix Orthogonal projection matrix of interference signal and the concatenation matrix The orthographic projection matrix The constructed detection statistic is expressed as: where tr(·) represents the trace of the matrix, I2 represents the 2×2 dimensional identity matrix.

7. The subspace signal detection method applicable to the scenario with insufficient training samples according to claim 1, characterized in that: In step 7, the detection threshold determined by using the detection statistic and the preset value of the false alarm probability is: η=t(n) Wherein, η represents the detection threshold, n = [Qμ], Q is the number of Monte Carlo simulations, μ represents the false alarm probability value set by the system, [·] represents the rounding operation, and t(n) is the sequence; The nth largest value when arranged from largest to smallest, (·) H represents the conjugate transpose transformation, X(k) represents the k-th realization k=1, 2,…,Q of the data to be detected containing only interference and thermal noise components.

8. A subspace signal detection system suitable for a scenario with insufficient training samples, using the subspace signal detection method according to any one of claims 1 to 7, characterized in that: Includes the following modules: Signal data matrix construction module: used to construct target signal matrix, interference signal matrix and training sample matrix, and use target signal matrix and interference signal matrix to construct splicing matrix; Sampling covariance matrix construction module: used to construct the sampling covariance matrix using training samples; Skew-symmetric transformation module: used to perform skew-symmetric transformation on target signal matrix, interference signal matrix and training sample matrix using skew-symmetric transformation matrix; Data quasi-whitening module: used to perform quasi-whitening processing on the target signal matrix, the interference signal matrix and the data matrix to be detected, and obtain the quasi-whitened target signal matrix, the quasi-whitened interference signal matrix and the quasi-whitened data matrix to be detected respectively; Detection statistic construction module: used to construct the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix, and construct the detection statistic using the quasi-whitened data matrix to be detected, the quasi-whitened target signal matrix, the orthogonal projection matrix of the interference signal and the orthogonal projection matrix of the splicing matrix; Detection threshold calculation module: used to determine the detection threshold according to the preset value of false alarm probability; Target decision module: used to compare the detection statistic with the detection threshold and determine whether the target exists.

9. Application of a subspace signal detection method suitable for a scenario with insufficient training samples as recited in any one of claims 1 to 7 to detect radar targets under a condition where the signal-to-noise ratio is higher than 17 dB.

Images