Radar adaptive detection method based on improved large-dimensional precision matrix estimation

Abstract

The invention relates to the technical field of radar signal processing, in particular to a radar adaptive detection method based on improved large-dimensional precision matrix estimation. The problem that existing radar detection is low in accuracy is solved. The method comprises the steps of collecting radar data and preprocessing the radar data to obtain preprocessed radar data; determining a to-be-detected unit and a reference unit according to the preprocessed radar data; constructing a sample covariance matrix according to the reference unit, and then inversing the sample covariance matrix to obtain a sample precision matrix; obtaining an improved precision estimation matrix according to the sample precision matrix; and judging whether a target exists in the distance unit of the place where the unit to be detected is located according to the improved precision estimation matrix. By using the method, the detection performance of a radar system under the condition that the dimension N and the sample size K are in the same order of magnitude can be remarkably improved. The problem of low radar detection accuracy caused by poor detection performance of a self-adaptive detector in a large-dimensional asymptotic system in the existing method is solved.

Description

Technical Field

[0001] The invention relates to the technical field of radar signal processing, and in particular to a radar adaptive detection method based on improved large-dimensional precision matrix estimation. Background Art

[0002] Radar target detection in a strong clutter (plus noise) environment has always been one of the core issues in the field of radar signal processing. With the advancement of electronic technology, various new radar systems continue to emerge, especially phased array radars have developed rapidly. Thanks to the successful description of multidimensional non-Gaussian signals by spherically invariant random vectors, the coherent processing of radar echo signal vectors has become a research hotspot in the past three decades. Whether it is multidimensional Gaussian or non-Gaussian clutter, the precision matrix (i.e., the inverse of the covariance matrix) is an important parameter for describing the signal amplitude and correlation characteristics. If the clutter precision matrix is ​​known in advance, the optimal detector can be derived by the Neyman-Pearson (NP) criterion. However, in real scenarios, the prior information of the clutter (or noise) precision matrix is ​​usually not available in advance. At this time, it is necessary to estimate the real clutter precision matrix through some auxiliary data near the unit to be detected. This type of detector is called an adaptive detector.

[0003] The design problem of optimal adaptive detectors has attracted extensive attention from researchers in the radar field over the past four decades. Improving the detection capability of adaptive detectors in complex environments is currently a research focus in the field of radar target detection. Precision matrix estimation is of great significance to adaptive detectors, and a sufficiently accurate precision matrix is ​​the key to improving the performance of adaptive detectors.

[0004] In the current radar signal processing process, the precision matrix estimation is based on the classical asymptotic system, that is, the low-dimensional large sample case (in the classical asymptotic system, it is generally assumed that the dimension N is fixed and the number of samples K tends to infinity, that is, N is fixed and K tends to infinity). For example, in the Gaussian clutter background, the sample precision matrix (also the maximum likelihood estimate) is often used to estimate the true clutter covariance matrix.

[0005] However, with the emergence of phased array radars and the use of large-scale antenna arrays, as well as due to the limitations of data storage space and the real-time requirements of signal processing, the sample size K and sample dimension N of the observed data in actual radar signal processing are at the same order of magnitude. In addition, in some special scenarios, due to the inhomogeneity of clutter, the sample size K may be smaller than the sample dimension N. For example, in a skywave over-the-horizon radar, a coherent accumulation cycle often contains 128 or 256 samples, that is, N = 128 or 256, but due to the inhomogeneity of the skywave over-the-horizon radar clutter, there are only 40-60 auxiliary data available for the estimation of the clutter covariance matrix in the distance dimension, that is, K = 40~60. This situation is called a large-dimensional asymptotic system, that is, the dimension N and the sample size K tend to infinity at the same time, and the ratio of the two tends to a constant (when N, K→∞, N / K→c∈(0,1)). In a large-dimensional asymptotic system, the sample precision matrix estimate is not an unbiased consistent estimate of the true covariance matrix.

[0006] The inconsistency of the precision matrix estimation ultimately leads to a significant loss in detector performance. Taking the generalized likelihood ratio detector as an example, when K = 2N, the performance of the GLRT detector has a 3dB signal-to-noise ratio loss compared to the optimal detector, which is unacceptable in practical scenarios.

[0007] Therefore, in the existing radar detection method in a large-dimensional asymptotic system (ie, when N and K are substantially equivalent in magnitude), there is a problem of low radar detection accuracy due to poor detection performance of the radar adaptive detector. Summary of the invention

[0008] The purpose of the present invention is to solve the problem that the existing radar detection method has a low radar detection accuracy due to poor detection performance of the radar adaptive detector in a large-dimensional asymptotic system (i.e., when N and K are substantially equivalent in order of magnitude). A radar adaptive detection method based on improved large-dimensional precision matrix estimation is invented, comprising the following steps:

[0009] S1: Initialize radar parameters; collect radar data and preprocess the radar data to obtain preprocessed radar data;

[0010] S2: Determine the unit to be detected and the reference unit according to the preprocessed radar data;

[0011] S3: Construct a sample covariance matrix based on the reference unit, and then invert the sample covariance matrix to obtain the sample accuracy matrix;

[0012] S4: According to the sample precision matrix, the eigenvalues ​​and eigenvectors of the sample precision matrix are obtained.

[0013] S5: construct an improved accuracy estimation matrix based on the radar parameters set in S1 and the eigenvalues ​​and eigenvectors of the sample accuracy matrix obtained in S4;

[0014] S6: constructing a test statistic of the unit to be tested according to the improved precision estimation matrix;

[0015] S7: Perform detection according to the test statistic of the unit to be detected and the set radar parameters to obtain a detection result.

[0016] The beneficial effects of the present invention are:

[0017] The detection method of the present invention can significantly improve the detection performance of the radar system when the dimension N and the sample size K are in the same order of magnitude. -3 When the detection probability is 0.9, the signal-to-noise ratio gain of the present invention is improved by more than 3dB compared with the traditional method, which solves the problem of low radar detection accuracy caused by poor detection performance of the adaptive detector in the existing radar detection method in a large-dimensional asymptotic system (that is, when N and K are basically equivalent in order of magnitude). BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the implementation process of the present invention;

[0019] Figure 2 It is a schematic diagram of the detection performance curve of the present invention. DETAILED DESCRIPTION

[0020] Specific implementation method 1: Combination Figure 1 The invention is described, comprising:

[0021] S1: Initialize radar parameters; collect radar data and preprocess the radar data to obtain preprocessed radar data;

[0022] S2: Determine the unit to be detected and the reference unit according to the preprocessed radar data;

[0023] S3: Construct a sample covariance matrix based on the reference unit, and then invert the sample covariance matrix to obtain the sample accuracy matrix;

[0024] S4: According to the sample precision matrix, the eigenvalues ​​and eigenvectors of the sample precision matrix are obtained.

[0025] S5: construct an improved accuracy estimation matrix based on the radar parameters set in S1 and the eigenvalues ​​and eigenvectors of the sample accuracy matrix obtained in S4;

[0026] S6: constructing a test statistic of the unit to be tested according to the improved precision estimation matrix;

[0027] S7: Perform detection according to the test statistic of the unit to be detected and the set radar parameters to obtain a detection result.

[0028] The detection method of the present invention can significantly improve the detection performance of the radar system when the dimension N and the sample size K are in the same order of magnitude. -3 When the detection probability is 0.9, the signal-to-noise ratio gain of the present invention is improved by more than 3dB compared with the traditional method, which solves the problem of low radar detection accuracy caused by poor detection performance of adaptive detectors in large-dimensional asymptotic systems (i.e., when N and K are basically equivalent in order of magnitude) of existing radar detection methods.

[0029] Specific implementation method 2: The difference between this implementation method and specific implementation method 1 is that in S1, radar parameters are initialized; radar data is collected and preprocessed to obtain preprocessed radar received data; the specific process is:

[0030] S1.1: Set radar system parameters,

[0031] S1.2: The radar scans the target area according to the set parameters. Each pulse signal emitted by the radar is detected by the radar's N a Antenna array elements are used to collect and process radar echo data;

[0032] The radar echo data includes X range unit radar echo data; X is a positive integer;

[0033] The directivity of a single antenna is limited. To meet the requirements of various applications, two or more single antennas working at the same frequency are fed and arranged in space according to certain requirements to form an antenna array, also called an antenna array. The antenna radiating units that constitute the antenna array are called antenna array elements.

[0034] In radar detection, a range unit is the smallest area into which the radar system divides the received signal according to the distance. Each range unit corresponds to the spatial interval from the radar to a certain distance (or target).

[0035] S1.3: preprocessing the radar echo data to obtain preprocessed radar data;

[0036] The other steps and parameters are the same as those in the first specific implementation.

[0037] Specific implementation method three: This implementation method is different from the specific implementation method one in that:

[0038] The specific process of radar system parameters in S1.1 is as follows:

[0039] Set the number of radar transmitted pulses to N p , Np is a positive integer;

[0040] Set the radar's false alarm probability P fa ;

[0041] The other steps and parameters are the same as those in the first to second embodiments.

[0042] Specific implementation method 4: This implementation method is different from specific implementation methods 1 to 4 in that:

[0043] The specific process of preprocessing the radar echo data in S1.3 to obtain the preprocessed radar data is as follows:

[0044] For a range unit in the radar echo data, N p ×N a The data is rearranged by column to obtain an N-dimensional column vector, where N = N p ×N a ; Finally, X N-dimensional column vectors are obtained, and the X N-dimensional column vectors are used as preprocessed radar data;

[0045] Rearrange by columns. For example, if the original data dimension is 2×3, it can be expressed as:

[0046]

[0047] Rearranged N = 6-dimensional data representation

[0048]

[0049] The other steps and parameters are the same as those in the first to third embodiments.

[0050] Specific implementation method 5: This implementation method is different from specific implementation methods 1 to 4 in that:

[0051] In S2, the unit to be detected and the reference unit are determined according to the preprocessed radar data; the specific process is:

[0052] S2.1: Determine the distance unit where the unit to be detected is located, and take the N-dimensional column vector corresponding to this distance unit as the unit to be detected y 0 ;

[0053] S2.2: Take K / 2 distance units before and after the distance unit to be detected, and obtain K distance units. Take the K N-dimensional column vectors corresponding to the selected K distance units as reference units [y 1 ,y 2 ,...,y K ],

[0054] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0055] Specific implementation method 6: This implementation method is different from specific implementation methods 1 to 5 in that:

[0056] In S3, a sample covariance matrix is ​​constructed according to the reference unit, and then the sample covariance matrix is ​​inverted to obtain a sample precision matrix, which is expressed as follows:

[0057]

[0058] in(·) H represents conjugate transpose; represents the sample covariance matrix, represents the sample precision matrix; y i Represents the N-dimensional feature vector corresponding to the i-th distance unit; i = [1, 2, 3..., K]; Represents y i The conjugate transpose of ; the other steps and parameters are the same as those in any one of the specific implementation modes one to five.

[0059] Specific implementation method 7: This implementation method is different from specific implementation methods 1 to 6 in that:

[0060] In S4, the eigenvalues ​​and eigenvectors of the sample precision matrix are obtained according to the sample precision matrix. The specific process is as follows:

[0061] Sample precision matrix Perform QR decomposition to obtain the sample precision matrix The eigenvalues ​​of , and the eigenvectors corresponding to the eigenvalues;

[0062] The sample precision matrix The N eigenvalues ​​of The N eigenvectors corresponding to the N eigenvalues ​​are expressed as in(·) T represents transpose;

[0063] QR decomposition is a method of decomposing a matrix into an orthogonal matrix Q and an upper triangular matrix R.

[0064] QR decomposition can be used to calculate the eigenvalues ​​of a matrix, and the eigenvalues ​​can be gradually approximated by repeatedly performing QR decomposition, which is well known to those skilled in the art.

[0065] The other steps and parameters are the same as those in Specific Embodiments 1 to 6.

[0066] Specific implementation eight: This implementation differs from specific implementations one to seven in that:

[0067] In S5, an improved precision estimation matrix is ​​constructed according to the radar parameters set in S1 and the eigenvalues ​​and eigenvectors of the sample precision matrix obtained in S4. The specific process is as follows:

[0068] S5.1: According to S1, set the radar parameters to obtain the rank r of the clutter component in the radar echo; it can be expressed as:

[0069] r=N p +N a -1

[0070] S5.2: Based on the rank of the clutter component in the radar echo, the eigenvalues ​​and eigenvectors of the sample precision matrix obtained in S4, an improved precision estimation matrix is ​​constructed. The formula is:

[0071]

[0072] represents the estimated noise power; and Represents the sample precision matrix The kth and jth sample eigenvalues ​​are sorted from large to small, where k and j represent the sorting indexes for the sum; Represents the improved clutter component eigenvalue.

[0073] The other steps and parameters are the same as those in Specific Embodiments 1 to 7.

[0074] Specific implementation method 9: This implementation method is different from specific implementation methods 1 to 8 in that:

[0075] In S6, the test statistic T of the unit to be detected is constructed according to the improved precision estimation matrix, which is expressed as follows:

[0076]

[0077] Where s represents the N×1 steering vector, s H represents the conjugate transpose of s, y 0 Indicates the unit to be detected. Represents y 0 The conjugate transpose of ; || means taking the modulus of a complex number;

[0078] The steering vector s is used to indicate the direction or target in signal processing. It is set artificially according to the actual situation. The steering vector s is generally defined as a unit vector in a certain direction.

[0079] The other steps and parameters are the same as those in Specific Embodiments 1 to 8.

[0080] Specific implementation method 10: This implementation method is different from specific implementation methods 1 to 9 in that:

[0081] The specific process of judging whether there is a target in the distance unit where the unit to be detected is located according to the test statistic T of the unit to be detected and the set detection threshold τ in S7 is:

[0082] If T ≥ τ, τ represents the set detection threshold, then the unit to be detected y 0 There is a target within the distance unit;

[0083] If T < τ, it means that the unit to be detected y 0 There is no target within the distance unit:

[0084] The detection threshold τ is set according to the false alarm probability P of the radar. fa The conversion of the detection threshold and the false alarm probability by looking up a table is well known to those skilled in the art.

[0085] The other steps and parameters are the same as those in Specific Embodiments 1 to 9.

[0086] Combining the simulation analysis of specific implementation methods one to ten

[0087] Radar system configuration: In this embodiment, the radar system is configured to transmit N p = 16 pulses, each pulse consists of N a =4 antenna array elements receive. Therefore, your data dimension for each distance unit is N = N p ×N a =64.

[0088] Reference unit selection: Select data of K=128 distance units around the unit to be detected as reference units.

[0089] False alarm probability setting: Set the false alarm probability to P fa =10 -3 .

[0090] The radar system is started to scan the target area and collect a data set containing multiple range cells.

[0091] The 16×4 data in each distance unit are rearranged column by column into a 64-dimensional column vector.

[0092] Select a distance unit from the data set as the unit to be detected y 0 .

[0093] The data of 128 distance units around the unit to be detected are selected as reference units [y 1 ,y 2 ,...,y KUsing 128 reference cell data, the 64×64-dimensional sample covariance matrix is ​​obtained according to the following formula:

[0094]

[0095] in(·) H represents the conjugate transpose.

[0096] The sample covariance matrix Inverse to get the sample precision matrix

[0097] Using QR decomposition, we get The 64 eigenvalues ​​of and its corresponding eigenvector in(·) T Indicates transpose.

[0098] According to the theoretical formula r = N p +N a -1, calculate the rank r of the clutter component in the echo. The 64 eigenvalues ​​of are subjected to nonlinear shrinkage:

[0099]

[0100] Among them, N=64, K=128.

[0101] The improved precision matrix estimation result is obtained according to the following formula:

[0102]

[0103] Where N = 64. Using the improved precision matrix estimation results, the generalized likelihood ratio test statistic T is constructed according to the following formula:

[0104]

[0105] According to the set false alarm probability P fa =10 -3 , the desired detection threshold τ is obtained through Monte Carlo simulation experiments.

[0106] Compare the test statistic T and the threshold τ. If T ≥ τ, it means that the distance unit y to be detected 0 There is a target; if T < τ, it means the unit to be detected y 0 No target exists.

[0107] The detection performance curve of this embodiment is plotted on Figure 2 In which the threshold τ is 10 5The Monte Carlo simulation experiment was performed, and the clutter rank r was set to 8. It can be seen that when N and K have the same order of magnitude, the detection performance of the present invention is significantly improved compared with the prior art (generalized likelihood ratio detector based on sample precision matrix): when the false alarm probability is 10 -3 , when the detection probability is 0.9, compared with the existing technology based on sample precision matrix, the signal-to-noise ratio gain of the present invention is more than 3dB;

[0108] The above only describes the preferred embodiments of the present invention. It should be understood that the present invention is not limited to the above specific implementation methods. Although the present invention has been disclosed as above in the preferred embodiments, it is not used to limit the present invention. Any technician familiar with the profession can make some changes or modify the technical contents disclosed above into equivalent embodiments with equivalent changes without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.

Claims

1. A radar adaptive detection method based on improved large-dimensional precision matrix estimation, characterized in that: The following steps are involved: S1: setting radar parameters; collecting radar data and preprocessing the radar data to obtain preprocessed radar data; S2: Determine the unit to be detected and the reference unit according to the preprocessed radar data; S3: Construct a sample covariance matrix based on the reference unit, and then invert the sample covariance matrix to obtain the sample accuracy matrix; S4: According to the sample precision matrix, the eigenvalues ​​and eigenvectors of the sample precision matrix are obtained. S5: construct an improved accuracy estimation matrix based on the radar parameters set in S1 and the eigenvalues ​​and eigenvectors of the sample accuracy matrix obtained in S4; S6: constructing a test statistic of the unit to be tested according to the improved precision estimation matrix; S7: Perform detection according to the test statistic of the unit to be detected and the set radar parameters to obtain a detection result.

2. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 1 is characterized in that: In the S1, radar parameters are set; radar data is collected and preprocessed to obtain preprocessed radar receiving data; the specific process is as follows: S1.1: Set radar system parameters, S1.2: The radar scans the target area according to the set parameters. Each pulse signal emitted by the radar is detected by the radar's N a Antenna array elements perform acquisition and processing; Get radar echo data; The radar echo data includes X range unit radar echo data; X is a positive integer; S1.3: Preprocess the radar echo data to obtain preprocessed radar data.

3. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 2 is characterized in that: The specific process of setting radar system parameters in S1.1 is as follows: Set the number of radar transmitted pulses to N p , N p is a positive integer; Set the radar's false alarm probability P fa .

4. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 3 is characterized in that: The specific process of preprocessing the radar echo data in S1.3 to obtain the preprocessed radar data is as follows: For a range unit in the radar echo data, N p ×N a The data is rearranged by column to obtain an N-dimensional column vector, where N = N p ×N a ; Finally, X N-dimensional column vectors are obtained, and the X N-dimensional column vectors are used as preprocessed radar data.

5. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 4 is characterized in that: In S2, the unit to be detected and the reference unit are determined according to the preprocessed radar data; the specific process is: S2.1: Determine the distance unit where the unit to be detected is located, and take the N-dimensional column vector corresponding to this distance unit as the unit to be detected y0; S2.2: Take K / 2 distance units before and after the distance unit where the unit to be detected is located, and obtain K distance units. The K N-dimensional column vectors corresponding to the selected K distance units are used as reference units. The K reference units are represented as [y1, y2, ..., y K ].

6. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 5, characterized in that: In S3, a sample covariance matrix is ​​constructed according to the reference unit, and then the sample covariance matrix is ​​inverted to obtain a sample precision matrix, which is expressed as follows: in(·) H represents conjugate transpose; represents the sample covariance matrix, represents the sample precision matrix; y i Represents the N-dimensional feature vector corresponding to the i-th distance unit; i = [1, 2, 3..., K]; Represents y i The conjugate transpose of .

7. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 6 is characterized in that: In S4, the eigenvalues ​​and eigenvectors of the sample precision matrix are obtained according to the sample precision matrix. The specific process is as follows: Sample precision matrix Perform QR decomposition to obtain the sample precision matrix The eigenvalues ​​of and the eigenvectors corresponding to the eigenvalues; The sample precision matrix The N eigenvalues ​​of The N eigenvectors corresponding to the N eigenvalues ​​are expressed as in(·) T Indicates transpose.

8. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 7 is characterized in that: In S5, an improved precision estimation matrix is ​​constructed according to the radar parameters set in S1 and the eigenvalues ​​and eigenvectors of the sample precision matrix obtained in S4. The specific process is as follows: S5.1: According to S1, set the radar parameters to obtain the rank r of the clutter component in the radar echo; it can be expressed as: r=N p +N a -1 S5.2: Construct an improved precision matrix based on the rank of the clutter component in the radar echo and the eigenvalues ​​and eigenvectors of the sample precision matrix obtained in S4 The formula is: represents the estimated noise power; represents the improved clutter component eigenvalue, and Represents the sample precision matrix The kth and jth sample eigenvalues ​​are sorted from large to small, where k and j are sorting numbers.

9. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 8, characterized in that: In S6, the test statistic of the unit to be detected is constructed according to the improved precision estimation matrix, which is expressed as follows: Where T represents the test statistic of the unit to be tested, s represents the steering vector, and s H represents the conjugate transpose of s, y0 represents the unit to be detected, represents the conjugate transpose of y0; || represents the modulo processing.

10. The radar adaptive detection method based on improved large-dimensional precision matrix estimation according to claim 9, characterized in that: In S7, detection is performed according to the test statistic of the unit to be detected and the set radar parameters to obtain the detection result. The specific process is as follows: If T≥τ, τ represents the set detection threshold, it means that there is a target in the distance unit where the unit to be detected y0 is located; If T < τ, it means that there is no target in the distance unit where the unit to be detected y0 is located: The detection threshold τ is set according to the false alarm probability P of the radar. fa Look up the table for conversion.

Images