Far-field near strong and weak target orientation estimation method suitable for near-field strong interference condition

Abstract

The invention discloses a far-field near strong and weak target orientation estimation method suitable for a near-field strong interference condition, and belongs to the technical field of orientation estimation. According to the method, grid point division is carried out on the airspace range of the far and near fields, a far and near field mixed source dictionary set is constructed, noise power is suppressed through covariance reconstruction, the signal-to-noise ratio of weak target signals is improved, weak target signal components are prevented from being compressed in the iteration process, covariance fitting is adopted to apply different norm constraints to the signals and noise, and the robustness of the weak target signals is improved. The sparsity of the signal power is improved, and the phenomenon that the noise power preempts the zero value is effectively inhibited. The problem that in the prior art, when strong interference exists in a near field, far-field strong and weak target orientation recognition precision is low is solved.

Description

Technical Field

[0001] This invention belongs to the field of orientation estimation technology, specifically relating to an orientation estimation method for near-field strong and weak targets in the far field under conditions of strong near-field interference. Background Technology

[0002] Limited by the size and load capacity of the small platform itself, it is difficult to construct an effective vibration isolation and noise reduction environment for such systems. Therefore, their array signal processing is easily affected by strong near-field interference such as platform self-noise. Under these circumstances, the performance of traditional DOA estimation algorithms based on the far-field plane wave assumption deteriorates sharply, specifically manifested in a decrease in angular resolution and DOA estimation accuracy, while multiple sets of spurious peaks will appear in the far-field spatial spectrum.

[0003] When far-field and near-field signals coexist, the academic community mainly focuses on the Fresnel region approximation model, aiming to reduce the computational complexity of the algorithm by performing a Taylor expansion of the two-dimensional steering vector in the near-field range through the Fresnel region approximation model. A typical algorithm is the fourth-order cumulant decomposition method proposed by Liang et al., which achieves near-field source angle-distance parameter decoupling by constructing a fourth-order cumulant matrix. Its theoretical positioning accuracy is close to the Cramér-Rao Lower Bound (CRLB). Wang further extended this framework and achieved far-field and near-field signal separation by constructing an orthogonal subspace. The above algorithms have similar processing ideas and also have similar disadvantages: (1) They rely on the second-order Taylor expansion approximation of the Fresnel region steering vector, which significantly reduces the ability to handle non-Fresnel region near-field interference such as platform self-noise; (2) In order to avoid orientation ambiguity, the algorithm relies on the array element spacing being less than or equal to Wavelength pair array; (3) The step-by-step processing strategy leads to error accumulation, and the overall performance deteriorates sharply in complex environments.

[0004] Li Chenmu, Qiu Longhao, and others proposed a near-field mixed source localization method based on sparse reconstruction. This method utilizes the correlation between plane waves and spherical waves to divide the near and far fields, avoiding errors introduced by the Fresnel zone approximation model. Simultaneously, the algorithm directly processes the near-field mixed source signals by constructing a near-field overcomplete dictionary, avoiding the accumulation of errors. However, the sparse constraint form used in this algorithm is similar to L1-SVD, both achieving spatial spectral sparsity by compressing weak signal components during the iterative process. When strong nearby interference exists, this sparse constraint mechanism forces the weak target signal to shrink towards zero, causing the weak target signal to be completely masked. In this case, common interference suppression methods such as spatial matrix filtering and constructing interference blocking matrices cannot be effectively preprocessed due to angular proximity. To address these issues, a method is proposed that can identify the location of strong and weak targets in the far field under conditions of strong near-field interference. Summary of the Invention

[0005] The present invention aims to solve the problem of low accuracy in identifying the azimuth of strong and weak targets in the far field when there is strong interference in the near field, and proposes a method for estimating the azimuth of nearby strong and weak targets in the far field under strong interference conditions.

[0006] A method for estimating the azimuth of nearby strong and weak targets in the far field under conditions of strong near-field interference includes the following steps:

[0007] Step 1: Determine the spatial range of the far field and near field, and perform discrete grid point division for the near field and far field to obtain the initial grid point set for the far field. and the initial grid point set in the near field ;

[0008] Step 2: Construct a near-field hybrid source dictionary set A based on the near-field and far-field mesh partitioning;

[0009] Step 3: Receive array signals and obtain the covariance matrix based on the array signals. For the covariance matrix Reconstruct the matrix to obtain the matrix. ;

[0010] Step 4: Initialize the signal power and noise power using the periodogram method;

[0011] Step 5: Update the fitted covariance matrix ;

[0012] Step 6: Update signal power and noise power ;

[0013] Step 7: Extract the far-field signal power from the signal power. Calculate tolerance deviation ;

[0014] Step 8: Repeat steps 5 to 7 until the maximum number of iterations is reached or the tolerance deviation is less than or equal to the threshold.

[0015] Step 9: Obtain the far-field spatial spectrum Find the far-field spectrum The peak value in the result is used as the target orientation detection result.

[0016] The near-far field hybrid source dictionary set A in step two is:

[0017]

[0018] in, The far-field orientation is discretized into grid points as the steering vector of the k-th field source. Discretize the near-field orientation into grid points The near-field distance is uniformly discretized into grid points. , This indicates the number of grid points in the near-field partition. , , Let represent the M-dimensional identity matrix.

[0019] Step three involves the covariance matrix. Reconstruct the matrix to obtain the matrix. The process is as follows:

[0020] The covariance matrix Perform eigenvalue decomposition:

[0021] ;

[0022] in, ,Right now It is the covariance matrix The former A diagonal matrix composed of large eigenvalues It is from the front The signal subspace is composed of the eigenvectors corresponding to the large eigenvalues; ,Right now It is composed of the covariance matrix After A diagonal matrix composed of small eigenvalues For the after The noise subspace is composed of eigenvectors corresponding to small eigenvalues, and the superscript H indicates the conjugate transpose.

[0023] Signal covariance matrix In the matrix Small values ​​superimposed on the main diagonal The reconstructed covariance matrix for: , Let represent the M-dimensional identity matrix.

[0024] The formula for initializing the signal power and noise power in step four is as follows:

[0025] ;

[0026] in, The reconstructed covariance matrix is... Let be the steering vector of the k-th field source, and when k takes the value [1, N], be the initialized signal power. When k takes the value [N+1, N+M], it represents the initial noise power. .

[0027] The formula for updating the fitted covariance matrix R in step five is as follows:

[0028] ;

[0029] Where A is the near-far field mixed source dictionary set, and P is:

[0030] .

[0031] The process of step six is ​​as follows:

[0032] Define covariance fitting criteria Applying power to the signal Norm constraints are applied to noise power. Norm constraints, and according to Construct the cost function:

[0033]

[0034] Among them, matrix To fit the covariance matrix, Calculated from the array received signal. , For array to receive signals, Where N is the number of snapshots, M is the number of grid cells, and M is the number of array elements. Let k be the power; when k is in the range [1, N], it represents the signal power; when k is in the range [N+1, N+M], it represents the noise power. For adaptive weights, , ;

[0035] The cost function is solved using the gradient descent method to obtain the signal weights and noise weights. Differentiating the signal weights and noise weights yields the iterative formulas for the signal power and noise power.

[0036] ;

[0037] ;

[0038] use and The formula updates the signal power and noise power.

[0039] The process of step seven is as follows:

[0040] Take the far-field signal power from the updated signal power. The tolerance deviation is calculated using the following formula. :

[0041]

[0042] j represents the number of iterations. This represents the far-field signal power after the (j+1)th iteration update. This represents the far-field signal power after the j-th iteration update. Represents the norm.

[0043] The beneficial effects of this invention are:

[0044] This invention divides the spatial domain into grid points for both near and far fields, constructing a hybrid near-field source dictionary. Noise power is suppressed through covariance reconstruction, improving the signal-to-noise ratio of weak target signals and avoiding compression of weak target signal components during iteration. Covariance fitting applies different norm constraints to the signal and noise, improving signal power sparsity and effectively suppressing noise power preemption of zero values. In complex environments with strong near-field interference and strong far-field interference, it can effectively identify the location of both strong and weak far-field targets. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the array receiving model of the present invention;

[0046] Figure 2 The image shows the far-field time-azimuth history of the FN-MSBL algorithm.

[0047] Figure 3 This is a far-field time-azimuth history diagram of the FN-SPICE algorithm.

[0048] Figure 4 This is a far-field time-azimuth history diagram of the present invention. Detailed Implementation Specific implementation method one:

[0050] This implementation method is a method for estimating the azimuth of nearby strong and weak targets in the far field under conditions of strong near-field interference, combined with... Figure 1 The specific process will be explained.

[0051] Step 1: Determine the spatial range of the far field and near field, and perform discrete grid point division for the near field and far field to obtain the initial grid point set for the far field. and the initial grid point set in the near field .

[0052] The array receiving signal model with both far-field and near-field signal sources is as follows: Figure 1 As shown, a coordinate system is constructed with the center of the array as the origin. A near-field source and Multiple far-field sources are incident on an M-element uniform linear array, with the element spacing being half a wavelength. .

[0053] With the center of the array as the origin, the first The location of a near-field source can be represented as... , , and the first The distances between each array element and the origin of the coordinate system are respectively expressed as: , ,Depend on Figure 1 achievable and The geometric relationship between them is:

[0054] ;

[0055] in, Indicates the first Each array element corresponds to Axis coordinates, i.e. , Therefore, we can obtain the first... The steering vectors corresponding to the near-field sources are:

[0056]

[0057] According to the plane wave assumption in the far field, at this time Since the incident angles of the far-field source received by each array element are consistent, only the orientation of the far-field source is estimated. At this time, the first... The steering vectors of the far-field sources are:

[0058]

[0059] In summary, the array received signal at time t is:

[0060] .

[0061] in, The near-field array manifold represents the signal received by the array at time t. The far-field array manifold is ; Here is the near-field signal vector. ; For far-field signal vectors, ; This indicates the received additive noise. .

[0062] Assuming that the sound sources are uncorrelated, the noise is uncorrelated with the sound sources, and the noise received by each array element is uncorrelated, a simplified model is established based on these assumptions. The received signal model corresponding to L snapshots is as follows:

[0063] ;

[0064] in, , , The covariance matrix of the received signal corresponding to L snapshots is:

[0065]

[0066] in, This represents the power corresponding to the near-field signal. The power corresponding to the far-field signal. This represents the noise power received by each array element.

[0067] Near-field orientation discretized into grid points The near-field distance is uniformly discretized into grid points. Therefore, the discretized near-field network is ,in, The number of grid points in the near field. Similarly, the far-field azimuth is discretized into... .

[0068] Obtain the initial grid point set in the far field and the initial grid point set in the near field .

[0069] Step 2: Construct a near-field mixed source dictionary set A based on the near-field and far-field grid division.

[0070] make , Rewritten as:

[0071] ;

[0072] The source dictionary set A for near and far fields is:

[0073]

[0074] in, The steering vector of the k-th field source. Let represent the M-dimensional identity matrix.

[0075] for:

[0076] .

[0077] Step 3: Receive array signals and obtain the covariance matrix based on the array signals. The covariance matrix Perform eigenvalue decomposition:

[0078] ;

[0079] in, ,Right now It is composed of the covariance matrix The former A diagonal matrix composed of large eigenvalues It is from the front The signal subspace is composed of the eigenvectors corresponding to the large eigenvalues; ,Right now It is composed of the covariance matrix After A diagonal matrix composed of small eigenvalues For the after The noise subspace is composed of eigenvectors corresponding to small eigenvalues, and the superscript H indicates the conjugate transpose.

[0080] Signal covariance matrix In the matrix Small values ​​superimposed on the main diagonal The reconstructed covariance matrix for: , Let represent the M-dimensional identity matrix.

[0081] Step 4: Initialize the signal power and noise power using the periodogram method, with the following formula:

[0082] ;

[0083] in, The reconstructed covariance matrix is... Let be the steering vector of the k-th field source, and when k takes the value [1, N], be the initialized signal power. When k takes the value [N+1, N+M], it represents the initial noise power. .

[0084] Step 5: Calculate the results in Step 4. Substitute into the formula Update the fitted covariance matrix .

[0085] Step 6: Update signal power and noise power .

[0086] Define covariance fitting criteria Applying power to the signal Norm constraints are applied to noise power. Norm constraints, and according to Construct the cost function:

[0087]

[0088] Among them, matrix To fit the covariance matrix, Calculated from the array received signal. , For array to receive signals, Where N is the number of snapshots, M is the number of grid cells, and M is the number of array elements. Let k be the power; when k is in the range [1, N], it represents the signal power; when k is in the range [N+1, N+M], it represents the noise power. For adaptive weights, , ;

[0089] The cost function is solved using the gradient descent method to obtain the signal weights and noise weights. Differentiating the signal weights and noise weights yields the iterative formulas for the signal power and noise power.

[0090] ;

[0091] ;

[0092] use and The formula updates the signal power and noise power.

[0093] Step 7: Extract the far-field signal power from the updated signal power. Calculate tolerance deviation :

[0094] ;

[0095] Where j represents the iteration number, This represents the far-field signal power after the (j+1)th iteration update. This represents the far-field signal power after the j-th iteration update. Represents the norm.

[0096] Step 8: Repeat steps 5 to 7 until the maximum number of iterations is reached or the tolerance deviation is less than or equal to the threshold.

[0097] Step 9: Obtain the far-field spatial spectrum Find the far-field spectrum The peak value in the result is used as the target orientation detection result.

[0098] The same lake test data were processed using the FN-MSBL algorithm, the FN-SPICE algorithm, and the proposed method, respectively, and the results are as follows: Figure 2 , 3 As shown in Figure 4, this method demonstrates a significant advantage in multi-target resolution in complex environments where strong near-field interference and strong far-field targets coexist, and can effectively distinguish between strong and weak targets in the vicinity of the far field.

Claims

1. A method for estimating the azimuth of nearby strong and weak targets in the far field under conditions of strong near-field interference, characterized in that, Includes the following steps: Step 1: Determine the spatial range of the far field and near field, and perform discrete grid point division for the near field and far field to obtain the initial grid point set for the far field. and the initial grid point set in the near field ; Step 2: Construct a near-field hybrid source dictionary set A based on the near-field and far-field mesh partitioning; Step 3: Receive array signals and obtain the covariance matrix based on the array signals. For the covariance matrix Reconstruct the matrix to obtain the matrix. ; Step 4: Initialize the signal power and noise power using the periodogram method; Step 5: Update the fitted covariance matrix ; Step 6: Update signal power and noise power ; Step 7: Extract the far-field signal power from the signal power. Calculate tolerance deviation ; Step 8: Repeat steps 5 to 7 until the maximum number of iterations is reached or the tolerance deviation is less than or equal to the threshold. Step 9: Obtain the far-field spatial spectrum Find the far-field spectrum The peak value in the result is used as the target orientation detection result.

2. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 1, characterized in that, The near-far field hybrid source dictionary set A in step two is: in, The far-field orientation is discretized into grid points as the steering vector of the k-th field source. Discretize the near-field orientation into grid points The near-field distance is uniformly discretized into grid points. , This indicates the number of grid points in the near-field partition. , , Let represent the M-dimensional identity matrix.

3. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 2, characterized in that, Step three involves the covariance matrix. Reconstruct the matrix to obtain the matrix. The process is as follows: The covariance matrix Perform eigenvalue decomposition: ； in, ,Right now It is composed of the covariance matrix The former A diagonal matrix composed of large eigenvalues It is from the front The signal subspace is composed of the eigenvectors corresponding to the large eigenvalues; ,Right now It is composed of the covariance matrix After A diagonal matrix composed of small eigenvalues For the after The noise subspace is composed of eigenvectors corresponding to small eigenvalues, and the superscript H indicates the conjugate transpose. Then the signal covariance matrix In the matrix Small values ​​superimposed on the main diagonal The reconstructed covariance matrix is ​​obtained. for: , Let represent the M-dimensional identity matrix.

4. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 3, characterized in that, The formula for initializing the signal power and noise power in step four is as follows: ； in, The reconstructed covariance matrix is... Let be the steering vector of the k-th field source, and when k takes the value [1, N], be the initialized signal power. When k takes the value [N+1, N+M], it represents the initial noise power. .

5. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 4, characterized in that, The formula for updating the fitted covariance matrix R in step five is as follows: ； Where A is the near-far field mixed source dictionary set, and P is: 。 6. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 5, characterized in that, The process of step six is ​​as follows: Define covariance fitting criteria Applying power to the signal Norm constraints are applied to noise power. Norm constraints, and according to Construct the cost function: Among them, matrix To fit the covariance matrix, Calculated from the array received signal. , For array to receive signals, Where N is the number of snapshots, M is the number of grid cells, and M is the number of array elements. Let k be the power; when k is in the range [1, N], it represents the signal power; when k is in the range [N+1, N+M], it represents the noise power. For adaptive weights, , ; The cost function is solved using the gradient descent method to obtain the signal weights and noise weights. Differentiating the signal weights and noise weights yields the iterative formulas for the signal power and noise power. ； ； use and The formula updates the signal power and noise power.

7. The method for estimating the azimuth of near-field strong and weak targets under near-field strong interference conditions as described in claim 6, characterized in that, The process of step seven is as follows: Take the far-field signal power from the updated signal power. The tolerance deviation is calculated using the following formula. : ； Where j represents the iteration number, This represents the far-field signal power after the (j+1)th iteration update. This represents the far-field signal power after the j-th iteration update. Represents the norm.

Images