DOA estimation method and system of random linear array

Abstract

The invention provides a DOA (direction of arrival) estimation method and device of a random linear array. The DOA estimation method comprises the following steps: establishing an observation model representing a relationship between direct output of a sensor unit and a source signal, and carrying out compressed sampling on the direct output of each unit of the linear array by applying a random demodulator; constructing a linear observation relationship between the sampling signal of the random demodulator and a decomposition coefficient matrix of the source signal matrix on a certain sparse base, obtaining an estimated value of the decomposition coefficient matrix by using a sparse sampling reconstruction method, and further reconstructing to obtain an estimated value of an original source signal matrix; and calculating the DOA spectrum and estimating the DOA angle of the target or source according to the peak value of the spectrum. According to the invention, the technical problems of low estimation precision and low resolution caused by large design difficulty of a sensor array, heavy data acquisition burden and single analysis signal are solved.

Description

Technical Field

[0001] This invention relates to the field of radio or sensor array signal processing, and more specifically to a method and system for estimating the DOA of random linear arrays. Background Technology

[0002] Direction of Arrival (DOA) estimation for targets or radiation sources is a common problem in radio or sensor array signal processing. The most common sensor arrays use a linear array of uniformly arranged sensor elements, followed by DOA estimation using methods such as Capon beamforming and Multiple Subspace Classification (MUSIC). For example, the existing patent application "Grid Iterative ESPRIT, a Scalable Fast Algorithm for Estimating 2D DOA of Uniform Circular Arrays" includes: using a spatial grid, employing cyclic compensation and iterative estimation using the classic ESPRIT algorithm; the algorithm can estimate the DOA values ​​of (M-1) signals using an M-element uniform circular array; this algorithm can simultaneously compensate for error factors including mutual coupling, channel inconsistencies, and radiation patterns; and the algorithm provides unbiased and asymptotically consistent estimation of the 2D DOA of spatial signals. However, in the aforementioned existing schemes, the uniformly linearly arranged sensor elements must satisfy the Nyquist spatial sampling rate to avoid spatial ambiguity, which requires the spacing between adjacent elements to not exceed half the sensor's operating wavelength. However, excessively small cell spacing increases direct energy coupling between cells and is unsuitable for situations where sensor cell sizes are too large. Therefore, the use of non-uniformly arranged sensor arrays has gradually been proposed. A typical example is a random linear array, where sensor cells are arranged linearly but non-uniformly with intervals greater than or equal to half the wavelength. Random linear arrays employ an unconventional uniform linear array configuration, possessing the characteristic of sparsely arranged sensor cells, which can reduce the size of the sensor and data acquisition channels. For example, the existing invention patent application document CN110244272A, entitled "Direction of Arrival Estimation Method Based on Rank-One Denoising Model," includes the following steps: establishing a radar received signal model and determining the measurement matrix of the received signal; constructing a signal covariance matrix based on a rank-one denoising model based on the measurement matrix of the received signal; sparsely reconstructing the signal covariance matrix based on the rank-one denoising model to obtain a sparse signal vector; and estimating the direction of arrival of the target source using an alternating grid optimization algorithm based on the sparsely reconstructed signal vector. The existing scheme, under unknown non-uniform noise conditions, is based on a rank-one denoising model. It eliminates non-uniform noise by redesigning the covariance matrix of the dimensionality-reduced signal, then uses vectorization operations to obtain the equivalent source vector, and finally determines the direction of arrival (DOA) of the signal. Although the aforementioned existing techniques and DOA estimation algorithms such as MUSIC can also handle non-uniform arrays, they do not effectively utilize the sparsity and randomness of spatial sampling of sensor arrays, and cannot sufficiently improve the accuracy and resolution of DOA estimation for targets or radiation sources.

[0003] In summary, existing technologies suffer from technical problems such as the difficulty in designing sensor arrays, the heavy burden of data acquisition, and the low estimation accuracy and resolution due to the single type of signal being analyzed. Summary of the Invention

[0004] The technical problem to be solved by this invention is: how to solve the technical problems of high difficulty in sensor array design, heavy data acquisition burden, and low estimation accuracy and resolution due to single analysis signal.

[0005] This invention solves the above-mentioned technical problems by employing the following technical solution: A method for estimating the DOA of a random linear array includes:

[0006] S1. Based on the configuration parameters and operating parameters of the random linear array, establish an observation model that directly relates the sensor unit's output to the source signal, where the sensor unit's direct output represents the random linear array.

[0007] S2. Directly output the data from each sensor unit in the random linear array and use a random demodulator for compressed sampling.

[0008] S3. Construct the sampled signal and source signal matrices of the random demodulator, and decompose the linear observation relationship between the coefficient matrices on a sparse basis;

[0009] S4. Using compressed sensing reconstruction, obtain the estimated value of the decomposed coefficient matrix.

[0010] S5. Based on the estimated value And with sparse basis, the estimated value of the source signal matrix is ​​obtained by solving.

[0011] S6. Based on the estimated value of the source signal matrix The estimated value of the DOA angle is obtained by solving the problem.

[0012] This invention reduces the design complexity of sensor arrays by using fewer sensors and sampling channels to obtain higher-resolution DOA spectra. Furthermore, the direct reconstruction of the original signal facilitates further analysis and estimation of other useful information in the source signal. The DOA estimation method of this invention is applicable to random linear arrays, reducing the design complexity of sensor arrays while using fewer sensors and sampling channels, thus lowering the data acquisition burden.

[0013] In a more specific technical solution, in S1, for a random linear array containing M sensor units, the DOA angle range of the random linear array is... Divide the data into N evenly spaced DOA angles {θ1, θ2, ..., θ3}. N};

[0014] Based on the configuration parameters and operating parameters of the random linear array, an observation model is established:

[0015]

[0016] In the formula, matrix X represents the direct output signal matrix of the random linear array sensor unit, with a dimension of M×L, and L represents the number of time-domain snapshots; source signal matrix S represents the source signal matrix corresponding to the N DOA angles, with a dimension of N×L; and matrix A represents the matrix composed of the guide vectors of the linear array at the N DOA angles as column vectors, with a dimension of M×N.

[0017] In a more specific technical solution, S1 uses the following logic to express the expression corresponding to the preset angle θ∈{θ1,θ2,...,θ N Array steering vector at}

[0018]

[0019] In the formula, the superscript [·] T This represents the transpose of a vector. This represents the position vector of each sensor unit in the array relative to the reference unit. This represents the wavenumber vector.

[0020] In a more specific technical solution, in S2, compressed sampling is performed to obtain M signal vectors with a dimension of K×1. Where m = 1, ..., M, and K is a positive integer less than the number of time-domain snapshots L.

[0021] In a more specific technical solution, S3 involves constructing a random demodulator sampling signal. And the linear observation relationship between the source signal matrix S and the decomposition coefficient matrix α on the sparse basis Ψ, where the sparse basis includes: Fourier basis, wavelet basis and Gabor basis:

[0022]

[0023] In the formula, matrix Y is the M compressed sampled signal vectors. A (MK)×1 dimensional column vector, m=1,...,M, formed by stacking the columns sequentially; matrix f=[vec(α T )] LN×1 It is a column vector of dimension (LN)×1, where matrix α represents the decomposition coefficient matrix of the source signal matrix S on the sparse basis matrix Ψ of dimension N×N:

[0024] S=αΨ T

[0025] In the formula, the superscript [·] T represents the matrix transpose; the operator vec(·) represents the operator that takes elements from the input matrix and arranges them into a new column vector; matrix N represents the additive white Gaussian noise matrix; matrix H is the observation matrix associated with the steering vector matrix A, the sparse basis matrix Ψ, and the random demodulator, and the dimension of matrix H is (MK)×(LN).

[0026] In a more specific technical solution, the observation matrix H is expressed using the following logic:

[0027]

[0028] In the formula, Let Ψ represent the row vector formed by the i-th row elements of the direct output signal matrix X of the random linear array sensor unit in S1, and let Ψ represent the selected sparse basis matrix, where Φ1=Φ2=…Φ M Let L represent the equivalent sampling matrix of the random demodulator applied to the output of each sensor unit. Its dimension is K×L, where L represents the number of time-domain snapshots of the direct output signal of the sensor unit, and K is a positive integer less than the number of time-domain snapshots L.

[0029] This invention uses sparse sampling to estimate the DOA of a random linear array, and can obtain high-precision, high-resolution DOA estimation results by utilizing a sparse sampling reconstruction algorithm.

[0030] This invention uses a random demodulator for sampling after each sensor unit. The equivalent observation matrix Φ of the random demodulator satisfies the RIP property and is incoherent with typical time-frequency dictionaries such as Gabor dictionaries, making it suitable as a sampling observation system for sparse sampling reconstruction.

[0031] In a more specific technical solution, S4 uses an l1-norm regularized sparse sampling reconstruction method to solve the matrix equations: Obtain the estimated value of the reconstructed vector f. Then according to f = [vec(α) T )] LN×1 Based on the relationship, we obtain an estimate of the decomposition coefficient matrix α.

[0032] In a more specific technical solution, in S5, the estimated value of the source signal matrix S is obtained using the following logic:

[0033]

[0034] In the formula, Ψ represents the sparse basis matrix.

[0035] The DOA estimation method of this invention uses a sparse sampling reconstruction algorithm to directly reconstruct the original signal. In addition to obtaining the DOA information of the target or radiation source, it can further analyze other information such as frequency in the source signal. Moreover, compared with other DOA estimation methods, the accuracy and resolution of DOA estimation are improved.

[0036] In a more specific technical solution, S6 involves estimating the source signal matrix of dimension N×L. By summing the squares of the elements along the rows of the matrix, we obtain a vector of dimension N×1. As a DOA estimation spectrum;

[0037] The N′ angles θ corresponding to one or more peaks in the DOA estimation spectrum p1 ,θ p2 ...,θ pN′ ∈{θ1,θ2,...,θ N}, serving as the target angle of arrival and the source angle of arrival.

[0038] In more specific technical solutions, the DOA estimation system for random linear arrays includes:

[0039] The direct relationship observation model construction module is used to establish an observation model that directly relates the sensor unit output to the source signal based on the configuration parameters and operating parameters of the random linear array. The random linear array is represented by the direct output of the sensor unit.

[0040] The compressed sampling module is used to directly output each sensor unit in the random linear array and perform compressed sampling using a random demodulator. The compressed sampling module is connected to the direct relationship observation model construction module.

[0041] The linear observation relationship acquisition module is used to construct the sampled signal and source signal matrix of the random demodulator, and decompose the linear observation relationship between the coefficient matrices on the sparse basis. The linear observation relationship acquisition module is connected to the compressed sampling module.

[0042] The coefficient matrix estimation module is used to process the estimated values ​​of the coefficient matrix obtained by compressed sensing reconstruction. The decomposition coefficient matrix estimation module is connected to the linear observation relationship acquisition module;

[0043] The source signal matrix estimation module is used to estimate the source signal matrix based on the estimated values. And with sparse basis, the estimated value of the source signal matrix is ​​obtained by solving. The source signal matrix estimation module is connected to the decomposition coefficient matrix estimation module;

[0044] The angle calculation module is used to calculate the angle based on the estimated value of the source signal matrix. The DOA angle estimate is obtained by solving the problem, and the angle solution module is connected to the source signal matrix estimation module.

[0045] The present invention has the following advantages over the prior art:

[0046] This invention reduces the design complexity of sensor arrays by using fewer sensors and sampling channels to obtain higher-resolution DOA spectra. Furthermore, the direct reconstruction of the original signal facilitates further analysis and estimation of other useful information in the source signal. The DOA estimation method of this invention is applicable to random linear arrays, reducing the design complexity of sensor arrays while using fewer sensors and sampling channels, thus lowering the data acquisition burden.

[0047] This invention uses sparse sampling to estimate the DOA of a random linear array, and can obtain high-precision, high-resolution DOA estimation results by utilizing a sparse sampling reconstruction algorithm.

[0048] This invention uses a random demodulator for sampling after each sensor unit. The equivalent observation matrix Φ of the random demodulator satisfies the RIP property and is incoherent with typical time-frequency dictionaries such as Gabor dictionaries, making it suitable as a sampling observation system for sparse sampling reconstruction.

[0049] The DOA estimation method of this invention uses a sparse sampling reconstruction algorithm to directly reconstruct the original signal. In addition to obtaining the DOA information of the target or radiation source, it can further analyze other information such as frequency in the source signal. Moreover, compared with other DOA estimation methods, the accuracy and resolution of DOA estimation are improved.

[0050] This invention solves the technical problems of high difficulty in designing sensor arrays, heavy data acquisition burden, and low estimation accuracy and resolution due to the single analysis signal in the prior art. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the basic steps of the DOA estimation method for a random linear array according to Embodiment 1 of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1

[0054] like Figure 1 As shown, the DOA estimation method for random linear arrays provided by this invention includes the following basic steps:

[0055] S1. Based on the configuration and operating parameters of the random linear array, establish an observation model representing the direct relationship between the sensor unit output of the random linear array and the source signal.

[0056] In this embodiment, for a random linear array containing M elements, its DOA angle range is... Divide the data into N evenly spaced DOA angles {θ1, θ2, ..., θ3}. N Based on the configuration and operating parameters of the random linear array, an observation model is established to represent the relationship between the direct output of its sensor unit and the source signal. Where matrix X represents the direct output signal matrix of the random linear array sensor unit, with a dimension of M×L, and L represents the number of time-domain snapshots; matrix S represents the source signal matrix corresponding to the N DOA angles, with a dimension of N×L; and matrix A represents the matrix composed of the guide vectors of the linear array at the N DOA angles as column vectors, with a dimension of M×N.

[0057] The configuration parameters of the random linear array are: the position vector of each sensor element in the array relative to the reference element.

[0058] i = 1, ..., M; the reference unit is preferably the first sensor unit of the array, i.e., r1 = 0;

[0059] The operating parameters of the random linear array are wavenumber vectors. Its amplitude is λ represents the operating wavelength; the direction of the wavenumber vector is the array beam scanning center, preferably the array normal direction;

[0060] The term corresponds to a certain angle θ∈{θ1,θ2,...,θ N The array steering vector expression at} is: Among them, the superscript [·] T This represents the transpose of a vector. This represents the position vector of each sensor unit in the array relative to the reference unit. Represents the wavenumber vector;

[0061] Specifically, consider a linear array containing M sensor elements, for example, a radar antenna array, where the sensor elements are antenna elements, numbered sequentially along the linear array as 1, 2, ..., M-1, M. Let the first sensor element (numbered 1) be the reference element, and the position vector of each element relative to the reference element be... Where i = 1, ..., M, r1 = 0. Assuming all elements have the same directionality, the baseband signal emitted by a radiation source with an azimuth angle of arrival (DOA) of θ, after being measured, received, and orthogonally demodulated by the reference element, is represented as s(t), where t represents time. Then, the signal received by the sensor element numbered i is represented as...

[0062]

[0063] in, Denotes the wavenumber vector, whose magnitude is λ represents the working wavelength; the direction of the wavenumber vector is the scanning center of the array beam, preferably the array normal direction.

[0064] The output vector is composed of the measurement signals of all elements in the array at time t.

[0065]

[0066] in, Let [·] be the guide vector of the array at angle θ. T This indicates the transpose of a matrix or vector.

[0067] So, when there are a total of N baseband signals sn(t) at different DOA angles θ n When a linear array is simultaneously illuminated by radiation sources n = 1, ..., N, the output snapshot vector composed of the measurement signals of all array elements at time t can be expressed as:

[0068] x(t)=A(θ)s(t)+n(t)

[0069] Among them, A(θ)=[a(θ1),...,a(θ N )] represents the corresponding N angles θ n The guiding vectors a(θ1), ..., a(θ) of n = 1, ..., N N The matrix s(t) consists of M×N elements. s(t) = [s1(t),...,s...] N (t)] T Let N baseband signals s1(t),...,s be generated at time t. N A column vector consisting of (t).

[0070] S2. For the direct output of each unit of the linear array, a random demodulator is used for compressed sampling;

[0071] In this embodiment, the direct output of each unit of the linear array is compressed and sampled using a random demodulator to obtain M compressed and sampled signal vectors of dimension K×1. Where m = 1, ..., M, and K is a positive integer less than the number of time-domain snapshots L;

[0072] S3. Construct the sampled signal and source signal matrices of the random demodulator, and decompose the linear observation relationship between the coefficient matrices on a certain sparse basis.

[0073] In this embodiment, a random demodulator sampling signal is constructed. The linear observation relationship between the source signal matrix S and the coefficient matrix α decomposed on a certain sparse basis Ψ. Where matrix Y represents M compressed sampled signal vectors. A (MK)×1 dimensional column vector, m=1,...,M, formed by stacking the columns sequentially; matrix f=[vec(α T )] LN×1 It is a column vector of dimension (LN)×1, where matrix α represents the decomposition coefficient matrix of the source signal matrix S (of dimension N×L) on a sparse basis matrix Ψ of dimension N×N, i.e., S=αΨ T , superscript [·] T The operator vec(·) represents the matrix transpose; the operator vec(·) represents the operator that takes elements from the input matrix and arranges them into a new column vector; matrix N represents the additive white Gaussian noise matrix, which consists of M independent and identically distributed noise molecules with zero mean and variance σ. 2 Gaussian stochastic processes The samples with a single point count of K are stacked in columns; matrix H is the observation matrix associated with the steering vector matrix A, the sparse basis matrix Ψ, and the random demodulator, and its dimension is (MK)×(LN);

[0074] The sparse basis can be a Fourier basis, a wavelet basis, or preferably a Gabor basis;

[0075] The expression for the observation matrix H is as follows:

[0076]

[0077] in, Ψ represents the row vector formed by the i-th row elements of the direct output signal matrix X of the random linear array sensor unit in step S1, and Ψ represents the selected sparse basis matrix, Φ1=Φ2=…Φ M The equivalent sampling matrix of the random demodulator applied to the output of each sensor unit is represented by K×L, where L represents the number of time-domain snapshots of the direct output signal of the sensor unit, and K is a positive integer less than the number of time-domain snapshots L.

[0078] Specifically, the most common sensor array uses a linear array of uniformly arranged sensor elements, and then applies the array for multiple times t1, t2, ..., t... LThe measured output snapshot signals x(t1), x(t2),..., x(t L ) are used for DOA estimation by methods such as Capon beamforming and multiple subspace classification (MUSIC). Therefore, the use of non-uniformly arranged sensor arrays has been gradually proposed. A typical sensor array called a random linear array means that the sensor units are linearly but non-uniformly randomly arranged at intervals greater than or equal to half a wavelength. The random linear array adopts an unconventional uniform linear array configuration and has the characteristic of sparse arrangement of sensor units, which can greatly reduce sensors and data acquisition channels. For conventional DOA estimation algorithms such as beamforming and MUSIC, the non-uniform sampling characteristics of the random linear array will cause grating lobes and ambiguities in the estimation results, making it not fully applicable to this configuration. However, the sparse sampling reconstruction theory can handle it easily.

[0079] According to the sparse sampling reconstruction theory, in order to obtain a signal f = Ψx that can be represented by a certain sparse basis matrix Ψ and a compressible signal x, where f, x ∈ R N , Ψ ∈ R N×N , it is not necessary to directly measure f. Instead, the measurement value y = Φf can be indirectly obtained through a linear sampling observation system represented by an observation matrix Φ of M×N dimensions (where M < N), and then f can be accurately reconstructed by the method of l1-norm regularization. For a signal f that can be sparsely represented, a matrix Φ that satisfies the restricted isometry property (RIP) and its represented sampling observation system can be designed, and the pressure at the data acquisition end can be reduced through signal processing methods. A common sparse signal sampling system is called a random demodulator, which multiplies the signal by a pseudo-random +1 and -1 sequence and then integrates it along a time window, and then samples the integration result at the end of the time interval. It has been proven that the equivalent observation matrix Φ of the random demodulator satisfies the restricted isometry property (RIP) and is incoherent with typical time-frequency dictionaries such as the Gabor dictionary, and is suitable as a sampling observation system for sparse sampling reconstruction. Therefore, the technical patent of this invention uses a random demodulator for sampling after each sensor unit.

[0080] S4. Obtain the estimated value of the decomposition coefficient matrix using the compressive sensing reconstruction method

[0081] In this embodiment, the sparse sampling reconstruction method with l1-norm regularization is used to solve the matrix equation to obtain the estimated value of the reconstructed vector f Then, according to f = [vec(α T )] LN×1 the estimated value of the decomposition coefficient matrix α is obtained

[0082] S5. Solve according to the adopted sparse basis to obtain the estimated value of the source signal matrix;

[0083] In this embodiment, the estimated value of the decomposition coefficient matrix obtained in step S4 is used... Using the sparse basis employed, the estimated value of the source signal matrix S is obtained. Where Ψ represents the sparse basis matrix.

[0084] Step S6, based on the estimated value of the source signal matrix The method to obtain the DOA angle estimate is as follows: estimate the source signal matrix of dimension N×L. By summing the squares of the elements along the rows of the matrix, we obtain a vector of dimension N×1. As a DOA estimation spectrum, the N′ angles θ corresponding to one or more peaks of this spectrum p1 ,θ p2 ...,θ pN′ ∈{θ1,θ2,...,θ N} represents the estimated angle of arrival of the target or source.

[0085] In summary, this invention reduces the design complexity of sensor arrays by using fewer sensors and sampling channels to obtain higher-resolution DOA spectra. Furthermore, the direct reconstruction of the original signal facilitates further analysis and estimation of other useful information in the source signal. The DOA estimation method of this invention is applicable to random linear arrays, reducing the design complexity of sensor arrays while using fewer sensors and sampling channels, thus lowering the data acquisition burden.

[0086] This invention uses sparse sampling to estimate the DOA of a random linear array, and can obtain high-precision, high-resolution DOA estimation results by utilizing a sparse sampling reconstruction algorithm.

[0087] This invention uses a random demodulator for sampling after each sensor unit. The equivalent observation matrix Φ of the random demodulator satisfies the RIP property and is incoherent with typical time-frequency dictionaries such as Gabor dictionaries, making it suitable as a sampling observation system for sparse sampling reconstruction.

[0088] The DOA estimation method of this invention uses a sparse sampling reconstruction algorithm to directly reconstruct the original signal. In addition to obtaining the DOA information of the target or radiation source, it can further analyze other information such as frequency in the source signal. Moreover, compared with other DOA estimation methods, the accuracy and resolution of DOA estimation are improved.

[0089] This invention solves the technical problems of high difficulty in designing sensor arrays, heavy data acquisition burden, and low estimation accuracy and resolution due to the single analysis signal in the prior art.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of DOA estimation for a random linear array, characterized in that, The method comprises: S1, according to the configuration parameters and working parameters of the random linear array, an observation model of the direct output of the sensor unit directly related to the source signal is established, wherein the direct output of the sensor unit represents the random linear array; S2, for each of the direct output of the sensor unit in the random linear array, a random demodulator is used for compressed sampling; S3, a linear observation relationship between coefficient matrices is decomposed on a sparse basis by constructing a random demodulator sampling signal and a source signal matrix; S4, using a compressive sensing reconstruction method, processing the obtained estimate of the decomposition coefficient matrix S5、according to the estimated value and the sparse basis, to obtain an estimated value of the source signal matrix S6、according to the estimated value of the source signal matrix Solving the DOA angle estimation value.

2. The DOA estimation method of random linear array according to claim 1, characterized in that, In the S1, for the random linear array containing M sensor units, the DOA angle range of the random linear array is divided into N uniform sampling DOA angles {θ1, θ2,..., θN} by equal intervals , and the DOA angle range of the random linear array is divided into N uniform sampling DOA angles {θ1, θ2,..., θN} by equal intervals N} According to the configuration parameters and working parameters of the random linear array, the observation model is established: In the formula, the matrix X represents the direct output signal matrix of the sensor unit of the random linear array, and the dimension is M*L, L represents the time domain snapshot number; the source signal matrix S represents the source signal matrix corresponding to the divided N DOA angles, and the dimension is N*L; the matrix A represents a matrix composed of the steering vector of the linear array at the N DOA angles as the column vector, and the dimension is M*N.

3. The method of DOA estimation for random linear array according to claim 1, characterized in that, In the S1, the following logic is used to express the array steering vector corresponding to a preset angle θ∈{θ1, θ2,..., θN} at the array: N} at the array: where the superscript [·] T denotes vector transposition, i = 1,..., M denotes the position vector of each sensor element of the array relative to a reference element, denotes the wave number vector.

4. The method of DOA estimation for random linear array according to claim 1, characterized in that, In the S2, by performing the compressed sampling, M signal vectors with a dimension of Kx1 are obtained wherein m = 1,...,M, and K is a positive integer less than the time domain snapshot number L.

5. The method of DOA estimation for random linear array according to claim 1, characterized in that, In the S3, the random demodulator sampling signal is constructed m = 1,...,M, and a linear observation relationship between the source signal matrix S and the decomposition coefficient matrix α on the sparse basis Ψ, wherein the sparse basis comprises: a Fourier basis, a wavelet basis and a Gabor basis: where the matrix Y is an M compressed signal vectors successively stacked (MK) x 1 dimensional column vectors, m = 1,..., M; matrix f = [vec(a T ) LN×1 is an (LN) x 1 dimensional column vector, where the matrix a represents the decomposition coefficient matrix of the source signal matrix S on a N x N dimensional sparse basis matrix Y: S = a Ψ T In the formula, the superscript [·] T represents the matrix transpose; the operator vec(·) represents the operator that takes elements from the input matrix and arranges them into a new column vector; matrix N represents the additive white Gaussian noise matrix; matrix H is the observation matrix associated with the steering vector matrix A, the sparse basis matrix Ψ, and the random demodulator, and the dimension of matrix H is (MK)×(LN).

6. The DOA estimation method of random linear array according to claim 5, characterized in that, The following logic is used to express the observation matrix H: wherein i = 1,..., M represents the i-th row element of the direct output signal matrix X of the random linear array sensor unit in the S1, Ψ represents the sparse basis matrix selected, Φ1= Φ2=... Φ M represents the equivalent sampling matrix of the random demodulator applied to the output of each sensor unit, and has a dimension of K x L, L represents the time domain snapshot number of the direct output signal of the sensor unit, and K is a positive integer less than the time domain snapshot number L.

7. The method of DOA estimation for random linear array according to claim 1, characterized in that, In S4, the matrix equation is solved by using a sparse sampling reconstruction method of l1-norm regularization: obtaining an estimate of the reconstructed vector f According to the relation f = [vec(a T )] LN×1 obtaining the estimate of the decomposition coefficient matrix a 8. The method of DOA estimation for random linear array according to claim 1, characterized in that, In S5, the following logic is used to obtain the estimated value of the source signal matrix S: In the formula, Ψ represents a sparse basis matrix.

9. The method of DOA estimation for random linear array according to claim 1, characterized in that, said estimate of the source signal matrix of dimension N x L along the direction of the matrix rows, obtaining the accumulated value of the sum of the squares of the elements, resulting in a vector of dimension N x 1 as a DOA estimation spectrum; N' angles θ p1 ,θ p2 ...,θ pN′ ∈ {θ1,θ2...,θ N} corresponding to one or more peaks of the DOA estimation spectrum are selected as target angles of arrival, source angles of arrival.

10. A DOA estimation system of a random linear array, characterized by, The system comprises: A direct relationship observation model construction module is configured to establish an observation model of the direct output of the sensor unit directly related to the source signal according to the configuration parameters and working parameters of the random linear array, wherein the direct output of the sensor unit represents the random linear array; A compressed sampling module is configured to use a random demodulator to perform compressed sampling on each of the direct output of the sensor unit in the random linear array, and the compressed sampling module is connected with the direct relationship observation model construction module; A linear observation relationship acquisition module is configured to construct a linear observation relationship between coefficient matrices on a sparse basis by constructing a random demodulator sampling signal and a source signal matrix, and the linear observation relationship acquisition module is connected with the compressed sampling module; a decomposition coefficient matrix estimation module, configured to use a compressed sensing reconstruction method to process to obtain an estimated value of the decomposition coefficient matrix The decomposition coefficient matrix estimation module is connected with the linear observation relationship acquisition module. a source signal matrix estimation module configured to obtain an estimated value of a source signal matrix according to the estimated value of the decomposition coefficient matrix and the sparse basis and the sparse basis The source signal matrix estimation module is connected with the decomposition coefficient matrix estimation module. an angle solving module for solving the DOA angle estimation value according to the estimated value of the source signal matrix solving the DOA angle estimation value, the angle solving module being connected with the source signal matrix estimation module.

Images