An adaptive interference suppression method based on feature subspace decision

Abstract

The application aims to provide an adaptive interference suppression method based on feature subspace decision, comprising the following steps: calculating a full-rank sample covariance matrix; decomposing N feature values distributed from large to small and corresponding feature vectors to obtain a main subspace formed by feature vectors corresponding to the first D large feature values and a noise subspace formed by feature vectors corresponding to N-D small feature values; finding target signal components in the main subspace and the noise subspace, if any, finding and deleting the same, and combining the remaining two subspaces to form an interference and noise subspace; reconstructing the covariance matrix, and finally realizing strong interference signal suppression and accurate estimation of weak target direction by calculating the spatial spectrum. The application has higher calculation efficiency without calculating the power of interference and target signals, has less environmental mismatch influence and higher robustness, is suitable for target sound source motion scenes, and has strong engineering practical value as confirmed by actual sea trial data.

Description

TECHNICAL FIELD

[0001] The present application relates to a kind of underwater acoustic array signal processing method, specifically underwater sonar array signal processing method. BACKGROUND

[0002] In complex underwater environment, the weak target signal received by hydrophone array is often submerged by strong interference and background noise, making the weak target bearing estimation particularly difficult. In order to better achieve accurate estimation of the bearing of weak target signal, an adaptive strong interference suppression method is needed to improve the signal-to-noise ratio and high resolution of the target signal.

[0003] At present, scholars at home and abroad have carried out in-depth research and analysis on the problem of strong interference suppression, and put forward many practical technical solutions. Literature 1 [Matrix filter design for passive sonar interference suppression[J]. Acoustical Society of America Journal, 2004, 115(6): 3010-3020.] and literature 2 [Matrix spatial pre-filtering target bearing estimation[J]. Acta Acustica, 2007, 32(2): 7.] proposed a matrix spatial matrix filter based interference suppression method, which can ensure that the signals in the passband range pass without distortion and suppress the interference and noise in the stopband range by designing a matrix filter in the spatial domain. However, this method requires the general bearing of the target signal to be known, and it is difficult to select the bearing range of the passband, stopband and transition band. In addition to the matrix spatial filter type interference suppression method, scholars have also proposed some subspace type interference suppression methods.In 2004, Brian F. Harrison proposed an eigencomponent association method for adaptive interference suppression in document 3 [The eigencomponent association method for adaptive interference suppression[J]. Journal of the Acoustical Society of America, 2004, 115(5): 2122-2128.], which can maintain good interference suppression performance under the condition of known target signal direction, but cannot process moving target signals and does not give a clear decision threshold; document 4 [Eigenanalysis-based adaptive interference suppression[J]. Acta Acustica, 2013, 38(3): 9.], document 5 [Eigenanalysis-Based Adaptive Interference Suppression and Its Application in Acoustic Source Range Estimation[J]. IEEE Journal of Oceanic Engineering, 2015, 40(4): 903-916.], and document 6 [Eigenanalysis-based adaptive interference suppression for underwater target estimation[J]. Journal of the Acoustical Society of America, 2017, 142(4): 2728-2728.] proposed an eigenanalysis-based adaptive interference suppression method, which only needs to know the general direction of the target, and can well realize the suppression of strong interference through an adaptive decision condition, but the method does not have a clear power ratio and decision threshold condition.

[0004] Therefore, in actual use, a processing method capable of suppressing strong interference and realizing accurate estimation of the direction of a weak target signal is needed. SUMMARY

[0005] The purpose of the present application is to provide an eigen-subspace decision-based adaptive interference suppression method applicable to underwater strong interference suppression and underwater weak target direction estimation.

[0006] The purpose of the present application is achieved as follows:

[0007] The eigen-subspace decision-based adaptive interference suppression method of the present application is characterized in that:

[0008] (1) A horizontal array consisting of N hydrophones evenly distributed at a spacing of d is placed in seawater at a set depth. From different directions, Θ=[θ1,θ2,…,θ K At position [ ], K narrowband signals are received, including one target signal and K-1 interference signals, with the target signal having the weakest signal-to-noise ratio. The signal model received by the array can be written as:

[0009] x(t)=A(τ)s(t)+n(t),t=t1,…,t T

[0010] A(τ)=e -j2πfτ τ=dsinθ k / c, f is the signal frequency, c is the underwater speed of sound, s(t) = [s1(t),…,s k [(t)] represents the incident signal component, n(t) represents Gaussian white noise, and θ k ∈(0°,180°), k=1,2,…,K;

[0011] The sample covariance matrix is ​​estimated using sampled data with a finite number of snapshots of L.

[0012]

[0013] The superscript H indicates conjugate transpose. It is an N×N complex matrix;

[0014] (2) By using the sample covariance matrix Performing eigenvalue decomposition yields:

[0015]

[0016] Where λ1,λ2,…,λ N and e1, e2, ..., e N The sample covariance matrix is ​​shown below. The eigenvalues ​​and eigenvectors obtained by eigenvalue decomposition are mutually orthogonal;

[0017] The eigenvalues ​​are distributed in descending order as follows:

[0018] λ1>…>λ d >λ d+1 >…>λ N

[0019] Based on the distribution of eigenvalues, the eigenvectors corresponding to the top D largest eigenvalues ​​form the principal subspace U. Z The eigenvectors corresponding to the remaining ND small eigenvalues ​​form the noise subspace U. N , respectively represented as:

[0020] U Z = [e1, e2, …, e D ], U N = [e D+1 , e2, …, e N ] ;

[0021] (3) According to the beam forming CBF method, the target signal direction θ0is estimated, and its direction is selected, which is set as [θ S1 , θ S2 ], which satisfies 0°≤θ S1 < θ0< θ S2 ≤180°, the grid size of the direction interval is 0.1°, there are Q=(θ S2 -θ S1 ) / 0.1+1 values, and Q steering vectors are constructed, that is:

[0022]

[0023] Wherein, A(θ Si ) is an N×1 complex vector, and [θ S1 , θ S2 ] interval can form a steering vector matrix for beam scanning:

[0024] A = [A1, A2, …, AQ] Q

[0025] Wherein, A is an N×Q complex matrix, A i is an N×1 complex vector, and i∈[1, …, Q];

[0026] In the estimated target signal range, a decision factor is constructed, and the target signal components in the principal subspace and noise subspace are found by traversing the Q steering vectors between the N eigenvectors, and the decision factor is expressed as follows:

[0027]

[0028] Wherein, DF i represents the result of the eigenvector in the subspace traversed by the steering vector corresponding to the general direction of the target signal; A(θ k ) represents the steering vector corresponding to the general direction of the target signal, which is an N×Q complex matrix, and e(n) represents the N eigenvectors obtained by eigen decomposition of the sample covariance matrix Each eigenvector is an N×1 complex matrix, and the superscript H represents conjugate transpose;

[0029] According to the relationship between the target signal steering vector and the corresponding eigenvector, the decision threshold is set as ​Wherein N is the number of array elements, gamma is the decision coefficient, indicating the correlation degree between the target signal steering vector and the corresponding characteristic vector;

[0030] Each characteristic vector in the main subspace and the noise subspace is traversed and judged by the decision factor, and is separated, so that the interference and noise subspace U I+N and the signal subspace U S :

[0031] If DF i < η, then e(n) ∈ U I+N ; if DF i ≥ η, then e(n) ∈ U S ;

[0032] (4) The obtained interference and noise subspace U I+N is removed from the sample covariance matrix , and the sample covariance matrix is reconstructed, that is:

[0033]

[0034] Wherein, P ⊥ is the orthogonal projection matrix of the interference and noise subspace, that is I is the unit matrix, and the superscript H is the conjugate transpose;

[0035] The reconstructed sample covariance matrix is substituted into the spatial power spectrum estimation of CBF, so that the spatial spectrum output result is obtained:

[0036]

[0037] Wherein, ω(θ)=A(θ) / N, θ ∈ [0°, 180°].

[0038] The application can also include:

[0039] 1. The value range of N in step (1) is 8-256, the array element spacing is 0.25-16 m, the value range of the snapshot number L is 100-1000, and the underwater sound speed is 1500 m / s.

[0040] 2. The value of D in step (2) is 2-7, which is the number of interference and target signals.

[0041] 3. The value range of gamma in step (3) is 0.5-1, the target range is selected according to the estimated target direction θ0, and the value is [θ S1 , θ S2 ] ∈ θ0-5°< θ0< θ0+5°.

[0042] 4. In step (3):

[0043] If DF i ≥η, it indicates that there is eigenvector corresponding to target signal in the main subspace and noise subspace, at this time, the eigenvector is the eigenvector corresponding to the target signal;

[0044] If DF i <η, it indicates that there is no eigenvector corresponding to target signal in the main subspace and noise subspace, that is, only the eigenvector corresponding to the interference and noise signal, only the eigenvector corresponding to the interference signal and noise signal in the main subspace and noise subspace.

[0045] The advantage of the present application is that: the present application utilizes the property that the signal eigenvector is the standard basis of the signal steering vector and the concept of subspace orthogonal projection, combined with the definition of the constructed decision factor and decision threshold, through the reconstruction processing of the sample covariance matrix SCM, realizes strong interference suppression and accurate estimation of weak target direction, compared with the existing subspace type suppression method, (1) without calculating the power of interference and target signal, the calculation efficiency is higher; (2) less affected by environmental mismatch, higher robustness; (3) suitable for target sound source motion scene, actual sea trial data confirms its effectiveness, has strong engineering practical value. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The flowchart of the present application;

[0047] Figure 2 The spatial spectrum estimation diagram of the present application method, ECA method and conventional beam forming CBF;

[0048] Figure 3 When the target signal moves, the spatial spectrum estimation diagram of the present application method, ECA method and conventional beam forming CBF;

[0049] Figure 4 When the estimated target direction deviates from the actual target direction by 3°, the spatial spectrum estimation diagram of the present application method, ECA method and conventional beam forming CBF;

[0050] Figure 5 The processing result diagram of the conventional beam forming CBF method on actual sea trial data;

[0051] Figure 6 The interference processing result diagram of the present application on actual sea trial data. DETAILED DESCRIPTION

[0052] The present application will be described in more detail below with examples combined with the drawings:

[0053] Combined Figures 1-6 , the present application is a kind of based on eigen-subspace decision self-adapting interference suppression method, including the following steps:

[0054] Step 1: Based on the far-field interference and narrowband target signal received by the uniform horizontal array, a signal model is formed, and the full-rank sample covariance matrix SCM is calculated with a sufficient number of snapshots.

[0055] Step 2: Perform eigenvalue decomposition on the sample covariance matrix SCM to decompose it into N eigenvalues ​​distributed from largest to smallest and their corresponding eigenvectors, thereby obtaining the principal subspace spanned by the eigenvectors corresponding to the first D largest eigenvalues ​​and the noise subspace spanned by the eigenvectors corresponding to the ND smallest eigenvalues.

[0056] Step 3: Roughly estimate the target signal azimuth using the conventional beamforming (CBF) method. Based on this, set the approximate azimuth range of the target. Construct a decision factor and use the steering vector composed of the approximate azimuth of the target signal to traverse among N eigenvectors to find the target signal components in the principal subspace and noise subspace. If they exist, find and delete them. The remaining two subspaces are merged to form an interference plus noise subspace.

[0057] Step 4: Orthogonally project the interference and noise signal subspace without signal components, and subtract it using the identity matrix. Then reconstruct the covariance matrix SCM. Finally, by calculating the spatial spectrum, strong interference signals are suppressed and the weak target's azimuth is accurately estimated.

[0058] The specific process of this invention is as follows: Figure 1 As shown, the specific steps are as follows:

[0059] Step 1: The uniform horizontal array has N elements and d element spacing. N ranges from 8 to 256, and the element spacing ranges from 0.25 to 16 m. From different directions, Θ = [θ1, θ2, ..., θ...] K If the array receives K narrowband signals at position [ ], including one target signal and K-1 interference signals, and the target signal has the weakest signal-to-noise ratio, then the signal model received by the array can be written as:

[0060] x(t)=A(τ)s(t)+n(t),t=t1,…,t T (1)

[0061] In the formula, A(τ)=e -j2πfτ τ=dsinθ k / c, where: f is the signal frequency, c is the underwater sound speed, typically taken as 1500 m / s, s(t) = [s1(t),…,s k [(t)] represents the incident signal component, n(t) represents Gaussian white noise, and θ k ∈(0°,180°), k=1,2,…,K.

[0062] The sample covariance matrix is ​​estimated using sampled data with a finite number of snapshots of L.

[0063]

[0064] The superscript H indicates conjugate transpose. It is an N×N complex matrix, and the number of snapshots L ranges from 100 to 1000;

[0065] Step 2: By analyzing the sample covariance matrix By performing eigenvalue decomposition, we can obtain:

[0066]

[0067] Where λ1,λ2,…,λ N and e1, e2, ..., e N The sample covariance matrix is ​​shown below. The eigenvalues ​​and eigenvectors obtained by eigenvalue decomposition are mutually orthogonal.

[0068] The eigenvalues ​​are distributed in descending order as follows:

[0069] λ1>…>λ d >λ d+1 >…>λ N (4)

[0070] Based on the distribution of eigenvalues, the eigenvectors corresponding to the top D largest eigenvalues ​​form the principal subspace U. Z The eigenvectors corresponding to the remaining ND small eigenvalues ​​form the noise subspace U. N They can be represented as:

[0071] U Z =[e1,e2,…,e D ], U N =[e D+1 ,e2,…,e N (5)

[0072] Where D takes values ​​from 2 to 7, it is generally the number of information sources.

[0073] Step 3: Estimate the target signal azimuth θ0 using the conventional beamforming (CBF) method, and select its approximate azimuth, which can be set as [θ0]. S1 ,θ S2 ]∈θ0-5°<θ0<θ0+5°, satisfying 0°≤θ S1 <θ0<θ S2 ≤180°, the grid size for the azimuth interval is 0.1°, and the total value is Q=(θ S2 -θ S1) / 0.1+1 values. Q steering vectors are constructed, i.e.

[0074]

[0075] where A(θ Si ) is an N x 1 complex vector, [θ S1 , θ S2 ] is the interval in which beam scanning can be formed, and A is an N x Q complex matrix.

[0076] A = [A1, A2, …, AQ] (7) Q

[0077] where A is an N x Q complex matrix, A i is an N x 1 complex vector, and i ∈ [1, …, Q];

[0078] The eigenvector corresponding to the target signal is the standard basis of the target signal steering vector, and the eigenvector can be considered as a normalized form of the steering vector. Since the N eigenvectors are mutually orthogonal, the target signal steering vector multiplied by other eigenvectors is approximately 0, but when the target steering vector is multiplied by the corresponding eigenvector, the value is maximum. Therefore, in the general range of the estimated target signal, a decision factor is constructed, and the target signal components in the principal subspace and the noise subspace are found by traversing the Q steering vectors among the N eigenvectors. The decision factor formed is expressed as follows:

[0079]

[0080] where DF i represents the result of the steering vector corresponding to the general direction of the target signal traversing the eigenvectors in the subspace; A(θ k ) represents the steering vector corresponding to the general direction of the target signal, which is an N x Q complex matrix, and e(n) represents the sample covariance matrix N eigenvectors are obtained by eigen decomposition, and each eigenvector is an N x 1 complex matrix. The superscript H represents conjugate transposition.

[0081] According to the relationship between the target signal steering vector and the corresponding eigenvector, the decision threshold can be set as where N is the number of array elements, γ is a decision coefficient, and γ represents the degree of correlation between the target signal steering vector and the corresponding eigenvector. The value of γ ranges from 0.5 to 1.

[0082] (1) If DF i ≥ η, it indicates that there is an eigenvector corresponding to the target signal in the principal subspace and the noise subspace, so the eigenvector at this time can be considered as the eigenvector corresponding to the target signal;

[0083] (2) If DF​i <η, represents the eigenvectors corresponding to no target signals in the main subspace and noise subspace, that is, only the eigenvectors corresponding to interference signals and noise signals in the main subspace and noise subspace, which can be considered as only the eigenvectors corresponding to interference signals and noise signals in the main subspace and noise subspace.

[0084] Each eigenvector in the main subspace and noise subspace is traversed and judged by a decision factor, and is separated, so that the interference and noise subspace U I+N and the signal subspace U s are obtained.

[0085] If DF i < η, then e(n) ∈ U I+N ; if DF i ≥ η, then e(n) ∈ U s .

[0086] Step 4: The obtained interference and noise subspace U I+N is removed from the sample covariance matrix , and the sample covariance matrix is reconstructed, that is:

[0087]

[0088] , wherein P ⊥ is the orthogonal projection matrix of the interference and noise subspace, that is, I is the unit matrix, that is, the elements on the diagonal line from the upper left corner to the lower right corner (referred to as the main diagonal line) are all 1, and the superscript H is the conjugate transpose.

[0089] The reconstructed sample covariance matrix is substituted into the spatial power spectrum estimation of the CBF to obtain the spatial spectrum output result:

[0090]

[0091] , wherein ω(θ) = A(θ) / N, θ ∈ [0°, 180°].

[0092] Embodiment

[0093] According to the specific implementation process of the present application, the specific implementation process is as shown in Figure 1 . The steps are as follows:

[0094] Step 1: A M = 20 element uniform linear array is used to receive signals, the element spacing is 6.25, and the spacing is half wavelength; the signal sampling frequency is f s= 1000 Hz, the signal center frequency is 120 Hz, the receiving array receives 4 incident narrowband sound source signals, 3 of which are strong interference signals and 1 of which is a weak target signal, the signal directions are -40°, -20°, 10° and 30°, the sound source signal at 10° is a weak target signal, the signal-to-noise ratio SNR is 3 dB, and the direction moves from 10° to 15°, the strong interference signal-to-noise ratio is 22 dB, 20 dB and 18 dB respectively, the sound speed is 1500 m / s, the background noise is additive Gaussian white noise, and the number of snapshots is 100. According to the 4 narrowband signals received by the uniform linear array, the covariance matrix of the signals is calculated by the array

[0095] Step 2: Perform eigenvalue decomposition on the sample covariance matrix , and the eigenvalues and eigenvectors of the sample covariance matrix , λ1, λ2, …, λ 20 and e1, e2, …, e 20 are obtained According to the eigenvalues and the size distribution of the eigenvalues, the eigenvectors are divided into a main subspace U Z and a noise subspace U N , which are respectively represented as:

[0096] U Z = [e1, e2, …, e4], U N = [e5, e6, …, e N ] (1)

[0097] Step 3: According to the conventional beam forming CBF method, the target signal direction 10° is estimated, and the general direction is selected as [5°, 15°], the grid size of the direction interval is 0.1°, and there are Q = 101 values. 101 steering vectors are constructed, that is: Then the steering vector matrix A of beam scanning is constructed.

[0098] Through the traversal of the 101 steering vectors among the 20 eigenvectors, the eigenvector corresponding to the target steering vector is found, so a decision factor is constructed to find the target signal component in the main subspace and the noise subspace. The decision factor is expressed as follows:

[0099]

[0100] According to the relationship between the target signal steering vector and the corresponding eigenvector, γ is 0.9, that is, the decision threshold is η = 4.05. Through the decision factor, it can be obtained that in the angle range [5°, 15°], there is a weak target signal component in the main subspace, and the remaining 3 interference signal components in the noise subspace and the main subspace combine to form an interference noise subspace, that is, UI+N = U Z-S + U N , a total of 19 eigenvectors.

[0101] Step 4: interference and noise subspace U I+N , the original sample covariance matrix is reconstructed by using the orthogonal projection matrix, and the reconstructed sample covariance matrix

[0102]

[0103] wherein, I is the unit matrix, and the superscript H is the conjugate transpose.

[0104] The reconstructed sample covariance matrix is substituted into the spatial power spectrum estimation of CBF to obtain the spatial spectrum output result:

[0105]

[0106] wherein, ω (θ) = A (θ) / 20, θ ∈ [0°, 180°].

[0107] Figure 2 indicates the spatial spectrum diagram when the target signal is stationary, and the interference suppression comparison results of the conventional beam forming CBF method, the ECA method and the method described in the application.

[0108] Figure 3 indicates the spatial spectrum diagram when the target signal moves from 10° to 15°, and the interference suppression comparison results of the conventional beam forming CBF method, the ECA method and the method described in the application.

[0109] Figure 4 indicates the spatial spectrum diagram when the estimated target direction deviates from the actual target direction by 3°, and the interference suppression comparison results of the conventional beam forming CBF method, the ECA method and the method described in the application.

[0110] Figure 5 indicates the processing result of the conventional beam forming CBF method on the actual sea trial data.

[0111] Figure 6 indicates the processing result of the method described in the application on the interference suppression of the actual sea trial data.

[0112] Figure 3 and Figure 4The results show that in a strong interference environment, a weak target signal is submerged by a sidelobe under a conventional beam forming CBF and is difficult to identify. The ECA method needs to know the target direction due to its algorithm defects, and if the target direction changes, the weak target signal will be suppressed as interference, and the weak target direction cannot be accurately estimated. The method described in the application is not affected by the target motion and the left and right swing of the horizontal array caused by the sea water flow, can not only suppress the strong interference signal, but also can accurately estimate the weak target signal direction.

[0113] Figure 5 and Figure 6 are the results of processing actual sea trial data by the conventional beam forming CBF method and the method described in the application. Through comparative analysis of the two, it can be seen that the method described in the application can well suppress the strong interference signal and part of the noise, and embodies the effectiveness and feasibility of the method described in the application.

Claims

1. An adaptive interference suppression method based on feature subspace decision, characterized by: (1) will be by Each hydrophone is according to A horizontal array composed of evenly spaced elements is placed in seawater at a set depth, viewed from different directions. Receiving A narrowband signal, containing one target signal and With -1 interference signal and the target signal having the weakest signal-to-noise ratio, the signal model received by the array can be written as: , , For signal frequency, The speed of sound underwater. For the incident signal component, It is Gaussian white noise. , ; The sample covariance matrix is ​​obtained using a finite number of snapshots. The data is estimated based on the sampled data under the given conditions, i.e. The superscript H indicates conjugate transpose. It is an N×N complex matrix; (2) By using the sample covariance matrix Performing eigenvalue decomposition yields: in, and The sample covariance matrix is ​​shown below. The eigenvalues ​​and eigenvectors obtained by eigenvalue decomposition are mutually orthogonal; The eigenvalues ​​are distributed in descending order as follows: Based on the distribution of eigenvalues, the eigenvectors corresponding to the top D largest eigenvalues ​​are used to form the principal subspace. The remaining The eigenvectors corresponding to the small eigenvalues ​​form the noise subspace. , respectively represented as: ， ； Where D represents the number of interfering and target signals; (3) Predict the azimuth of the target signal based on the beamforming CBF method. Select its location and set it as [location]. ,satisfy The grid size for the azimuth interval is 0.1°, totaling... Values, construct One guide vector, namely: ， in, Let N×1 be a complex vector. Within the interval, a steering vector matrix can be formed for beam scanning: in, For Complex matrices Let N×1 be a complex vector. ; Within the range of the predicted target signal, a decision factor is constructed, and then... A steering vector traverses among N eigenvectors to find the target signal component within the principal subspace and noise subspace. The resulting decision factor is represented as follows: = ( )， in, This represents the result of the guide vector corresponding to the approximate orientation of the target signal traversing the eigenvectors within the subspace. The steering vector representing the approximate orientation of the target signal is: Complex matrices, Represents the sample covariance matrix The eigenvalues ​​obtained by eigenvalue decomposition are N eigencomponents, each eigenvector being an N×1 complex matrix, with the superscript H indicating the conjugate transpose; Based on the relationship between the target signal steering vector and its corresponding feature vector, the decision threshold is set as follows: Where N is the number of array elements, It is the decision coefficient, which represents the degree of correlation between the target signal steering vector and the corresponding feature vector; The decision factor is used to traverse and decide each feature vector in the principal subspace and the noise subspace, and then the features are separated to obtain the interference and noise subspaces. and signal subspace : if ,but ;if ,but ; (4) The obtained interference and noise subspace is obtained by orthogonal projection. From the sample covariance matrix Remove from the middle and reconstruct the sample covariance matrix. ,Right now: in, Let be the orthogonal projection matrix of the interference and noise subspaces, i.e. , It is the identity matrix, and the superscript H is the conjugate transpose; Reconstruct the sample covariance matrix Substituting the spatial power spectrum estimation of CBF, we obtain the spatial spectrum output: in, = N, [0°, 180°].

2. The adaptive interference suppression method based on feature subspace decision according to claim 1, characterized in that: In step (1), the value of N ranges from 8 to 256, the element spacing ranges from 0.25 to 16m, the number of snapshots L ranges from 100 to 1000, and the underwater sound speed is 1500m / s.

3. The adaptive interference suppression method based on feature subspace decision according to claim 1, characterized in that: In step (2), the value of D is 2 to 7.

4. The adaptive interference suppression method based on feature subspace decision according to claim 1, characterized in that: In step (3) The value ranges from 0.5 to 1, based on the estimated target location. Select target range, value .

5. The adaptive interference suppression method based on feature subspace decision according to claim 1, characterized in that: In step (3): if , indicating that there are eigenvectors corresponding to the target signal in the principal subspace and the noise subspace, and the eigenvectors at this time are the eigenvectors corresponding to the target signal; if , indicating that there are no feature vectors corresponding to the target signal in the principal subspace and the noise subspace, that is, there are only feature vectors corresponding to the interference and noise signals in the principal subspace and the noise subspace.

Images