Sparse array direction of arrival estimation method based on optimized array manifold dictionary

Abstract

The invention particularly relates to a sparse array DOA (direction of arrival) estimation method based on an optimized array manifold dictionary, and the method comprises the steps: firstly carrying out the layout optimization of an initial uniform area array through combining a genetic algorithm, constructing an eight-element sparse array structure, and maintaining a directional diagram which is close to a full array while greatly reducing the number of array elements; performing energy screening on multichannel observation signals acquired by the sparse array, and extracting a dominant frequency band to improve an effective signal-to-noise ratio; and an optimized array manifold dictionary is constructed based on a phase compensation mechanism and an amplitude weighting mechanism, and the internal correlation of the dictionary is reduced by minimizing off-diagonal elements of a dictionary Gramer matrix. And on the basis, sparse reconstruction is carried out on the optimized dictionary by using an orthogonal matching pursuit algorithm, so that a direction-of-arrival estimation result of the sound source signal is obtained. According to the scheme, inherent direction finding fuzziness and grating lobes of the sparse array are effectively suppressed, the direction estimation precision is improved, and high reliability and high robustness are still kept under the condition that the number of array elements is remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of array signal processing and target positioning, and particularly relates to a sparse array direction of arrival estimation method based on an optimized array manifold dictionary. BACKGROUND

[0002] Traditional direction-of-arrival (DOA) estimation methods have significant performance bottlenecks in complex noise environments and few snapshots. Subspace decomposition-based algorithms rely on high signal-to-noise ratio and sufficient snapshots, while noise interference in low-energy frequency bands of wideband signals easily leads to estimation bias. Although sparse representation-based algorithms have low signal-to-noise ratio robustness, the existing array manifold dictionary has high redundancy and strong coherence between atoms, resulting in low sparse reconstruction efficiency. In addition, low-energy frequency bands are easily contaminated by noise, and traditional methods are difficult to dynamically select dominant frequency bands, resulting in the uneven energy distribution characteristics of multiple frequency points of wideband signals not being effectively utilized.

[0003] On this basis, in actual application process, to adapt to hardware constraints, the 7x7 uniform planar array in the system needs to be reduced to 8 channels for data processing. However, improper channel selection can affect the direction finding ambiguity of sparse arrays and the problem of grating lobe effect is serious. For example, the direction finding ambiguity area of the maximum element spacing eight-element sparse array can reach ±8°, while that of the full array is only ±1.5°.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present application, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The present application provides a sparse array direction of arrival estimation method based on an optimized array manifold dictionary, a computer-readable storage medium, and a computer program product, which can effectively overcome the defects in the prior art.

[0006] Other characteristics and advantages of the present application will become apparent from the following detailed description, or will be learned by practice of the present application.

[0007] According to a first aspect of the present application, a sparse array direction of arrival estimation method based on an optimized array manifold dictionary is provided, the method comprising: optimizing the layout of the initial uniform planar array in combination with a genetic algorithm to obtain an eight-element sparse array layout structure; acquiring an observation signal using the eight-element sparse array layout structure; wherein the observation signal is a multi-channel time-domain signal collected when the eight-element sparse array receives a sound source signal; Energy screening is performed on the observation signal to obtain a dominant frequency band; and in combination with the dominant frequency band and the eight-element sparse array layout structure, a target array manifold dictionary is constructed; Based on the orthogonal matching pursuit algorithm, the target array manifold dictionary is sparsely reconstructed to obtain a sparse vector corresponding to the sound source signal; and peak value searching is performed on the sparse vector to obtain the direction of arrival estimation of the sound source signal.

[0008] According to a second aspect of the present application, a computer readable storage medium is provided, which comprises a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to perform the above-mentioned sparse array direction of arrival estimation method based on an optimized array manifold dictionary.

[0009] According to a third aspect of the present application, a computer program product is provided, which comprises a computer program, and when the computer program is executed by a processor, the above-mentioned sparse array direction of arrival estimation method based on an optimized array manifold dictionary is implemented.

[0010] According to a fourth aspect of the present application, an electronic device is provided, which comprises: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to implement the above-mentioned sparse array direction of arrival estimation method based on an optimized array manifold dictionary via execution of the executable instructions.

[0011] The embodiment of the present application provides a sparse array direction of arrival estimation method based on an optimized array manifold dictionary, first, genetic algorithm is combined to select and optimize the geometry layout of initial uniform planar array elements, an eight-element sparse array structure meeting platform installation constraints and having low sidelobe and narrow main lobe characteristics is constructed, and the number of array elements is greatly reduced while the directivity pattern performance close to that of a full array is maintained; in terms of signal processing, the energy of the multi-channel observation signal collected by the sparse array is selected, and the dominant frequency band is extracted to improve the effective signal-to-noise ratio; then, the optimized array manifold dictionary is constructed based on the phase compensation mechanism and the array element amplitude weighting mechanism of the array element position, the correlation within the dictionary is reduced by minimizing the off-diagonal elements of the dictionary Gram matrix, and the angular resolution capability of sparse reconstruction is significantly improved; on this basis, the orthogonal matching pursuit algorithm is used for sparse reconstruction of the optimized dictionary, and the sparse vector corresponding to the sound source direction is obtained; then, the peak value of the sparse vector is searched, and the direction of arrival estimation result of the sound source signal is directly obtained through the position with the maximum amplitude. The scheme of the present application not only effectively suppresses the inherent direction finding ambiguity and grating lobe of the sparse array, improves the direction estimation precision, but also maintains high reliability and high robustness under the condition that the number of array elements is significantly reduced, and is suitable for high-precision sound source positioning tasks of space-limited platforms such as shipborne, buoy and shore-based platforms. The technical problems of serious direction finding ambiguity, sidelobe lifting and unstable sparse reconstruction caused by uneven distribution of array elements in the traditional sparse array in the prior art are solved.

[0012] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0013] The drawings incorporated into the specification and constituting a part of the specification show embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained from these drawings without creative labor for those skilled in the art.

[0014] Figure 1 The flow chart of the sparse array direction of arrival estimation method based on the optimized array manifold dictionary is schematically shown in the exemplary embodiment of the present application; Figure 2 The cooperative target sonar system receiving array optimization design schematic diagram of the sparse array direction of arrival estimation method based on the optimized array manifold dictionary is schematically shown in the exemplary embodiment of the present application; Figure 3 The schematic diagram of determining the eight-element sparse array layout structure based on the genetic algorithm of the sparse array direction of arrival estimation method based on the optimized array manifold dictionary is schematically shown in the exemplary embodiment of the present application; Figure 4AA schematic diagram of a 7x7 uniform planar array in the prior art; Figure 4B A schematic diagram of the azimuth direction finding ambiguity corresponding to a test signal received by a 7x7 uniform planar array in the prior art; Figure 4C A schematic diagram of the elevation direction finding ambiguity corresponding to a test signal received by a 7x7 uniform planar array in the prior art; Figure 5A A schematic diagram of an 8-element extended array sparse array in the prior art; Figure 5B A schematic diagram of the azimuth direction finding ambiguity corresponding to a signal received by an 8-element extended array sparse array in the prior art; Figure 5C A schematic diagram of the elevation direction finding ambiguity corresponding to a signal received by an 8-element extended array sparse array in the prior art; Figure 6A A schematic diagram of an 8-element sparse array based on an optimized array manifold dictionary of a sparse array direction of arrival estimation method in an exemplary embodiment of the present application; Figure 6B A schematic diagram of the azimuth direction finding ambiguity corresponding to a signal received by an 8-element sparse array based on an optimized array manifold dictionary of a sparse array direction of arrival estimation method in an exemplary embodiment of the present application; Figure 6C A schematic diagram of the elevation direction finding ambiguity corresponding to a signal received by an 8-element sparse array based on an optimized array manifold dictionary of a sparse array direction of arrival estimation method in an exemplary embodiment of the present application; Figure 7A A schematic diagram of a three-dimensional view of the azimuth-elevation joint optimized array manifold dictionary coherence characteristics based on a full array array in the prior art; Figure 7B A schematic diagram of a side view of the azimuth-elevation joint optimized array manifold dictionary coherence characteristics based on a full array array in the prior art; Figure 7C A schematic diagram of a -3dB contour plot of the azimuth-elevation joint optimized array manifold dictionary coherence characteristics based on a full array array in the prior art; Figure 7D A schematic diagram of a three-dimensional view of the azimuth-elevation joint optimized array manifold dictionary coherence characteristics of an 8-element sparse array based on an optimized array manifold dictionary of a sparse array direction of arrival estimation method in an exemplary embodiment of the present application; Figure 7EA side view diagram illustrating the coherence characteristics of the azimuth-elevation joint optimized array manifold dictionary of the eight-element sparse array in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application; Figure 7F A -3dB contour diagram illustrating the coherence characteristics of the azimuth-elevation joint optimized array manifold dictionary of the eight-element sparse array in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application; Figure 8A A diagram illustrating the azimuth direction finding ambiguity of the full array receiving the sound source signal in the prior art; Figure 8B A diagram illustrating the azimuth direction finding ambiguity of the eight-element sparse array receiving the sound source signal in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application; Figure 8C A diagram illustrating the elevation direction finding ambiguity of the full array receiving the sound source signal in the prior art; Figure 8D A diagram illustrating the elevation direction finding ambiguity of the eight-element sparse array receiving the sound source signal in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application; Figure 9A A comparison curve of the accuracy rate of the full array and the eight-element sparse array in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application; Figure 9B A comparison curve of the root mean square error of using different DOA estimation algorithms on the eight-element sparse array in the sparse array DOA estimation method based on the optimized array manifold dictionary according to an example embodiment of the present application. DETAILED DESCRIPTION

[0015] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these examples are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example implementations to those skilled in the art. The described features, structures, or characteristics can be combined in one or more implementations.

[0016] In addition, the accompanying drawings are included to provide a thorough understanding of embodiments of the application and are not intended to be exhaustive or to limit the application to the precise drawings. Identical reference numerals denote corresponding combined figures throughout the specific description. Like reference numerals indicate like elements in the drawings, and thus continuous numbering in each drawing indicates those components that are the same across all drawings. Some of the figures can be simplified for the sake of clarity; thus, the drawings can not portray all possible implementations in which the principles of the present application can be employed. However, the functions and details of the drawings are adequate for understanding the disclosure. Although each of the drawings can not be to scale, aspects of some embodiments can be employed across different drawings to produce scale with respect to one another.

[0017] In view of the shortcomings and deficiencies of the prior art, the present example embodiment provides a sparse array direction of arrival estimation method based on an optimized array manifold dictionary. Referring to Figure 1 As shown, the method can specifically include the following steps: Step S10, combining a genetic algorithm to optimize the layout of an initial uniform planar array to obtain an eight-element sparse array layout structure; Step S12, using the eight-element sparse array layout structure to obtain an observation signal; wherein the observation signal is a multi-channel time-domain signal collected when the eight-element sparse array receives a sound source signal; In step S12, the sound source signal can be a signal actually radiated by a target in water, or a water acoustic signal used for algorithm simulation. For example, a linear frequency modulation pulse signal is used to simulate the signal radiated by a target in water, with parameters set to a center frequency of 30 kHz, a bandwidth of 10 kHz, and a pulse width of 10 ms, which is suitable for underwater medium and short distance target detection. On this basis, a noise environment is used to simulate an actual underwater complex scene, with a signal-to-noise ratio set to -5 dB, and a Gaussian white noise and multipath reflection interference superimposed.

[0018] Step S14, performing energy screening on the observation signal to obtain a dominant frequency band; and combining the dominant frequency band and the eight-element sparse array layout structure to construct a target array manifold dictionary; Step S16, based on an orthogonal matching pursuit algorithm, sparsely reconstructing the target array manifold dictionary to obtain a sparse vector corresponding to the sound source signal; and performing peak value searching on the sparse vector to obtain a direction of arrival estimation of the sound source signal.

[0019] Based on steps S10 to S16, the initial uniform array is first optimized using a genetic algorithm to construct an eight-element sparse array structure, significantly reducing the number of array elements while maintaining a radiation pattern close to that of a full array. Then, energy filtering is performed on the multi-channel observation signals acquired by the sparse array to extract the dominant frequency bands and improve the effective signal-to-noise ratio. Next, an optimized array manifold dictionary is constructed based on phase compensation and amplitude weighting mechanisms, reducing internal correlations by minimizing the off-diagonal elements of the dictionary Gram matrix. On this basis, the orthogonal matching pursuit algorithm is used to sparsely reconstruct the optimized dictionary, thereby obtaining the direction-of-arrival (DOA) estimation results for the sound source signal. This invention not only effectively suppresses the inherent direction-finding ambiguity and grating lobes of sparse arrays, improving direction estimation accuracy, but also maintains high reliability and robustness despite a significant reduction in the number of array elements.

[0020] The following will describe in more detail each step of a sparse array wave direction-of-arrival estimation method based on an optimized array manifold dictionary in this exemplary embodiment, with reference to the accompanying drawings and embodiments.

[0021] For example, in step S10, the optimization layout of the initial uniform array using a genetic algorithm to obtain an octal sparse array layout structure includes: Step S101: Based on the shape constraints of the array mounting platform, the array elements in the initial uniform area array are initially screened to obtain the initial array layout structure. For details, please refer to Figure 2 As shown. Based on Figure 2 The 7×7 rectangular uniform planar array shown illustrates the receiving array design in the cooperative target sonar system to which this invention applies. Considering the limitations of the real-time data processing hardware in this system, it can currently only process data from a maximum of 8 channels simultaneously. Furthermore, in real-world applications, the array is often mounted on a specific carrier platform. Moreover, the shape and size of the platform itself impose restrictions on the structure and size of the array. Therefore, in this invention, the array elements in the initial uniform planar array are initially screened, i.e., the constraint is that there are 12 array element positions at the four vertices of the 7×7 rectangular uniform planar array (the positions shown by the red solid circles in the figure) where it is impossible to install. Based on this, in order to ensure the performance of the system under the constraints of limited hardware resources and platform structure, the optimized design of the receiving array is particularly necessary. Currently, common measures to improve array performance include: (1) effectively increasing the array aperture by increasing the number of array elements or expanding the array element spacing. (2) applying array optimization algorithms to optimize the array configuration.

[0022] Furthermore, this invention combines the above two array optimization strategies to meet the constraints of limited hardware resources and platform structure, and finally selects array element positions that satisfy the constraints. Figure 2It is represented by a solid blue circle.

[0023] Step S102: Based on the initial array layout structure, binary gene encoding is performed on the positions of candidate array elements, and an initial population for the genetic algorithm is constructed according to the Gaussian distribution and the constraint of the number of array elements; wherein, the initial population contains multiple candidate sparse array individuals; Step S103: Calculate the fitness value corresponding to each candidate sparse array individual based on the fitness function; wherein, the fitness function is constructed based on the main lobe width and side lobe level of the candidate sparse array individual; Step S104: The initial population is iteratively updated using genetic operations such as tournament selection, single-point crossover, and probabilistic mutation until a preset termination condition is met, thus obtaining the target sparse array individuals; and the octal sparse array layout structure is determined based on the gene encoding corresponding to the target sparse array individuals.

[0024] For details, please refer to Figure 3 As shown in the figure. The specific implementation process of optimizing the layout of the initial uniform array using a genetic algorithm to obtain the octal sparse array layout structure is shown below: (1) The candidate array element positions are encoded using binary genes, and the initial population for the genetic algorithm is constructed based on the Gaussian distribution and the constraint of the number of array elements: The number of individuals selected is And the dimension of each individual is . An initial population is constructed using a binary value parameter vector. In this initial population, each individual is represented as shown in the following equation: (1) in, This is the individual's identification number within the corresponding population. For the number of generations of a heredity, The total number of individuals in the population.

[0025] To optimize the search and set a starting point, initial encoding of the individuals in the population is required. Assume the number of elements after sparse processing is... And assume that all randomly initialized populations follow a Gaussian distribution. And satisfy... Figure 2 If the gene value of 12 elements in the array is 0, then the initial parameter values ​​of the individual can be expressed as follows: (2) (3) In equation (2), Indicates the interval Random numerical values ​​generated internally and conforming to a Gaussian distribution; Using equation (3), we can Figure 2 The gene values ​​of the 12 array elements are set to 0. Simultaneously, the individual with the highest gene value is selected. One gene is set to 1, and the rest are set to 0. Finally, each gene of length [missing information] is [missing information]. individual Convert to The matrix is ​​shown in the following formula: (4) In the operation of a genetic algorithm, the individuals that perform genetic operations such as selection, crossover, and mutation are called intermediate individuals. The fitness calculation is used to evaluate the individual. .

[0026] (2) Fitness calculation Fitness functions are used to evaluate the quality of individuals in the population, thereby guiding the search direction of the algorithm. For array optimization problems, constructing a suitable fitness function is crucial. Its core objective is to suppress grating lobes while ensuring the optimized array maintains a narrow main lobe width and minimizes the maximum sidelobe level of the planar array to reduce background noise interference. Therefore, array optimization simultaneously considers reducing the main lobe width and suppressing the sidelobe level to achieve an optimal balance between these two factors.

[0027] (3) Genetic manipulation After calculating individual fitness, if the termination condition is met, the optimal solution is output; otherwise, selection, crossover, and mutation operations are performed, ensuring the new population meets the constraints. Subsequently, the termination check is repeated for the offspring population until the conditions are met. The final selected array element positions that satisfy the constraints are located at... Figure 2 It is represented by a solid blue circle.

[0028] For example, in step S103, calculating the fitness value corresponding to each candidate sparse matrix individual based on the fitness function includes: Step S201: Based on the element position coordinates of the candidate sparse array individuals, and the angle scan of the direction response in the first preset azimuth angle domain and the first preset pitch angle domain, obtain the three-dimensional orientation map of the candidate sparse array individuals. Step S202: Determine the maximum amplitude value in the three-dimensional radiation pattern as the main lobe peak value; Step S203: Determine the angular region corresponding to when the main lobe peak value drops to a preset amplitude threshold as the main lobe width; Step S204: Search for sidelobe peaks in the remaining angular domain; and obtain the sidelobe level based on the difference between the sidelobe peak and the main lobe peak. Step S205: Weight the main lobe width and side lobe level to obtain the fitness value.

[0029] Specifically, the constructed fitness function is shown in the following equation: (5) Among them, the width of the main lobe Represented as a three-dimensional radiation pattern The main lobe gain in the middle drops to its maximum value. The range of angles corresponding to the time; Side lobe level The value is defined as the difference between the highest sidelobe value and the main lobe value. This difference is generally negative and is measured in dB. and This is represented by the weights of each indicator. By selecting appropriate weights, the relationship between main lobe width and side lobe suppression can be balanced.

[0030] Furthermore, in initializing the population size Individual genetic dimension Number of sparse array elements Genetic generations Crossover probability Probability of mutation Under simulation conditions, an optimized 8-element sparse array is obtained. (Reference) Figures 4A-6A As shown, three different array configurations and their direction-finding ambiguities are illustrated. Figures 4A-4C It is a full array (7×7 uniform planar array element array). Figures 5A-5C For 8-element sparse arrays (increasing element spacing) and Figures 6A-6C An 8-element optimal sparse array is used. In this process, test signals are received using a full array, an 8-element sparse array, and an 8-element optimal sparse array, respectively, to obtain the received signals. The received signals are then used to determine the direction-finding ambiguity map.

[0031] Specifically, as shown in the figure, the number, layout, and optimization level of array elements directly affect the direction-finding ambiguity map. A full array, employing a dense array layout, can provide high direction estimation accuracy. However, as the number of array elements increases, the computational complexity also rises, especially in large-scale arrays or real-time processing scenarios, which may lead to a significant increase in computational burden. In contrast, an 8-element sparse array effectively reduces computational complexity by reducing the number of array elements and narrows the main lobe width by increasing the spacing between some array elements. Simultaneously, this sparse configuration also introduces grating lobes, leading to an expansion of the direction-finding ambiguity region and causing a multi-peak phenomenon in azimuth estimation. The optimal 8-element sparse array, through optimized design, minimizes direction-finding ambiguity while reducing the number of array elements. Specifically, compared to an 8-element sparse array, the optimal 8-element sparse array has a smaller direction-finding ambiguity region; compared to a full array, although the optimal 8-element sparse array has a higher sidelobe level, their main lobe widths are similar, so the increase in the direction-finding ambiguity region is smaller. Therefore, the optimal 8-element sparse array can achieve a balance between computational complexity and direction estimation accuracy to a certain extent.

[0032] For example, in step S14, the energy filtering of the observed signal to obtain the dominant frequency band includes: Step S141: The observed signal is segmented according to the preset time window length to obtain segmented signals; and the discrete Fourier transform is performed on the segmented signals to obtain the complex spectrum. Step S142: Calculate the initial energy at each frequency point based on the complex spectrum; Step S143: Filter the initial energy of each frequency point according to the preset energy threshold to obtain the target energy that is greater than or equal to the preset energy threshold; Step S144: Determine the frequency point corresponding to the target energy as the dominant frequency band.

[0033] Specifically, firstly, the observed signal is framed, and the complex spectrum corresponding to each segment is obtained using the Discrete Fourier Transform (DFT). For example, the observed signal is framed in 10ms time windows, resulting in 10 segments, each containing 200 sampling points. A 256-point, 200kHz DFT is then performed on each segment to generate the frequency domain component matrix, as shown in the following equation: (6) in, For the first The array element in the first Duan Di The spectrum of a frequency point.

[0034] Through this step, the original time-domain observation signal is mapped to the frequency domain.

[0035] For example, a linear frequency modulated pulse signal can be used to simulate the radiation emitted by a target in water. The parameters are set to a center frequency of 30 kHz, a bandwidth of 10 kHz, and a pulse width of 10 ms, making it suitable for detecting short- to medium-range underwater targets. The observed signal corresponding to this simulated sound source signal, after signal segmentation and DFT, is displayed in the 25 kHz to 35 kHz frequency band, covering the main energy distribution area of ​​the signal. The following examples all use this sound source signal as an example, and will not be elaborated further.

[0036] Next, frequency point energy screening and dominant frequency band extraction are performed. The initial energy of each frequency point, i.e., the total energy, is calculated, and the frequency band with the highest energy is selected as the dominant frequency band. The calculation process for the initial energy is shown in the following formula: (7) Where K is the number of segments, which is 10 segments in the example; M is the number of array elements, M = 8.

[0037] Finally, select those that meet the requirements. frequency band set As the dominant frequency band, among which, This is the energy threshold. For example, set it to 90%. This step eliminates noise-dominant frequency bands by using the energy threshold, thus improving the signal-to-noise ratio.

[0038] For example, after calculating the initial energy, the 32kHz and 34kHz frequencies have significantly higher energy than other frequency bands and are selected as the focus frequencies for subsequent processing.

[0039] For example, in step S14, constructing the target array manifold dictionary by combining the dominant frequency band and the octal sparse array layout structure includes: Step S145: Determine the frequency parameters of the steering vector based on the dominant frequency band; wherein, the steering vector is an atom of the initial array manifold dictionary; In step S145, the aforementioned guide vector Used to describe The phase response of the incident sound source signal in the direction on an octet sparse array.

[0040] Step S146: Determine the element position coordinates of the guiding vector based on the octal sparse array layout structure; Step S147: Combine the second preset azimuth angle domain and the second preset pitch angle domain to obtain the angle set of the guide vector; and determine the guide vector based on the frequency parameter, the array element position coordinates and the angle set. Specifically, based on the array geometry, a coverage azimuth range is constructed. Pitch angle range The guide vector The step size can be set to 1°. For each set of angles in the angle set... Based on frequency parameters Array element positions Unit vector of angle and direction Calculate the corresponding steering vector, as shown in the following formula: (8) in, For frequency parameters; For the first The position coordinates of each array element; =1500m / s is the speed of sound; It is a unit vector for direction.

[0041] Step S148: Arrange the steering vectors based on the frequency parameters and angle set to obtain the initial array manifold dictionary; Specifically, the guide vectors obtained based on all angle combinations are arranged column-wise to form an initial array manifold dictionary, as shown in the following equation: (9) wherein, is the number of angle combinations.

[0042] The initial array manifold dictionary fully characterizes the response mode of the sparse array to signals of different incident directions under the dominant frequency band.

[0043] Further, for each frequency value in the dominant frequency band, a corresponding steering vector is constructed according to the element position and the angle set, and all steering vectors corresponding to the frequency points are collectively composed into the initial array manifold dictionary.

[0044] For example, the dominant frequency band contains two frequency points, denoted as and . Therefore, the frequency parameter of the steering vector is two frequency values. For each frequency value in the dominant frequency band, a corresponding steering vector is constructed according to the element position and the angle set, and the steering vector matrices of the two frequencies are spliced to construct a multi-frequency initial array manifold dictionary. As shown in the following formula: (10) wherein, is the steering vector matrix constructed by all angle combinations corresponding to the frequency ; is the steering vector matrix constructed by all angle combinations corresponding to the frequency .

[0045] Step S149, phase compensation and amplitude weighting are performed on the initial array manifold dictionary to obtain a target array manifold dictionary.

[0046] Specifically, phase compensation and amplitude weighting operations are performed on the initial array manifold dictionary to correct the phase offset and amplitude imbalance caused by the non-uniform geometry of the sparse array, thereby reducing the correlation within the dictionary. Phase compensation is used to correct the phase mismatch in the steering vector caused by the asymmetric layout of the elements; amplitude weighting is used to suppress the sidelobe uplift phenomenon caused by geometric sparseness. After optimization, the target array manifold dictionary is obtained, which is used for subsequent orthogonal matching pursuit sparse reconstruction to improve the DOA estimation accuracy.

[0047] Illustratively, in step S149, the phase compensation and amplitude weighting of the initial array manifold dictionary to obtain the target array manifold dictionary includes: Step S301, determining the array reference center based on the eight-element sparse array layout structure; and compensating the phase term of the steering vector using the position deviation between each element and the array reference center to obtain the compensated phase of the steering vector; Step S302, a weight is added to the amplitude term of the steering vector by using a weighting function to obtain a compensated amplitude of the steering vector; Specifically, the element position compensation refers to correcting the position deviation of each array element relative to the array reference center when calculating the phase of the steering vector, so as to compensate for the phase mismatch caused by the non-uniform arrangement; the amplitude weighting is to apply appropriate weights to each array element, for example, based on a weighting window function, to reduce the contribution of edge array elements to the beam sidelobes when constructing the dictionary, so as to weaken the possible grating lobe amplitude.

[0048] Step S303, calculating the off-diagonal elements of the dictionary Gram matrix based on the compensated steering vector; Step S304, iteratively adjusting the phase term and the amplitude term of the compensated steering vector by using a gradient descent algorithm until the off-diagonal elements of the dictionary Gram matrix meet a preset termination condition, to obtain a target array manifold dictionary.

[0049] Specifically, the steering vectors after phase compensation and amplitude weighting are combined to form a compensated dictionary , and the off-diagonal elements of the Gram matrix thereof are calculated based on the following formula: (11) wherein, is the conjugate transpose of the compensated dictionary; is the unit matrix; is the Frobenius norm.

[0050] Further, a cost function is constructed with the optimization objective of minimizing the off-diagonal elements of the Gram matrix of the compensated dictionary, aiming to make the Gram matrix closer to the unit matrix, i.e., the steering vectors are as independent as possible. The cost function is shown in the following formula: (12) Specifically, to minimize the cost function, the gradient descent algorithm is used to iteratively update the phase term and the amplitude term of the compensated steering vector. During the optimization process, it is continuously monitored whether the off-diagonal elements of the Gram matrix have decreased to a preset threshold or whether the number of gradient descent iterations has reached a preset upper limit. When the termination condition is met, the final optimized dictionary, i.e., the target array manifold dictionary, is output.

[0051] Further, referring to Figures 7A-7F , after the above phase compensation and amplitude weighting processing, the coherence between the atoms of the optimized dictionary is greatly reduced, and the maximum correlation coefficient is reduced to about 0.2, which is close to the performance of the full-array optimized dictionary, significantly improving the orthogonality of the sparse array dictionary. Specifically, FIG. 7 compares the coherence characteristics distribution of the full-array and the optimized eight-element array after the array manifold dictionary optimization. Among them Figures 7A-7CThe atom correlation distribution based on the full-matrix array manifold dictionary is shown in a three-dimensional graph, a side view and a -3dB contour graph, and it can be seen that the correlation between the dictionary atoms is extremely low under the full-matrix condition, and the non-diagonal elements of the Gram matrix are close to 0. Figures 7D-7F The coherence distribution of the optimized eight-element array after the dictionary optimization strategy is given. The results show that the maximum correlation coefficient between the dictionary atoms of the optimized eight-element array is reduced to about 0.2, which is comparable to the full-matrix case, verifying the effectiveness of the proposed dictionary optimization method.

[0052] Exemplarily, in step S16, the target array manifold dictionary is sparsely reconstructed based on the orthogonal matching pursuit algorithm to obtain a sparse vector corresponding to the sound source signal, including: In step S161, the residual is initialized as the observation signal, and the support set is initialized as an empty set; In step S162, the column vector with the largest correlation with the current residual is searched in the target array manifold dictionary, and the index corresponding to the current column vector is added to the support set; In step S163, the observation signal is least square fitted based on the updated support set to obtain a reconstruction coefficient vector; In step S164, the reconstruction signal is calculated according to the reconstruction coefficient vector, and the residual is updated; In step S165, the iteration is terminated when the norm of the current residual is less than a preset threshold or the number of iterations reaches a preset number, and the sparse vector is obtained.

[0053] Specifically, the specific process of sparsely reconstructing the target array manifold dictionary based on the orthogonal matching pursuit (OMP) algorithm to obtain a sparse vector corresponding to the sound source signal is as follows: First, the residual is initialized as the observation signal, and the support set is initialized as an empty set, as shown in the following formula: , (13) Wherein, is the initial residual, y is the observation signal; the support set is used to record the position number of the currently selected dictionary atom (direction vector).

[0054] Then, the dictionary atom with the largest correlation with the residual is selected step by step, and the support set and the sparse vector are updated.

[0055] In the process of selecting the dictionary atom with the largest correlation with the residual, which direction is most similar to the residual indicates that the sound source signal may come from this direction. The process is shown in the following formula: (14) Wherein, The first column of the target dictionary, i.e., the first dictionary atom; the inner product The first column of the target dictionary, i.e., the first dictionary atom; the inner product The inner product is the relevance of the residual to the dictionary atom.

[0056] The process of updating the support set is shown as follows: (15) The support set stores all direction indexes that are considered to be possible DOAs.

[0057] The reconstruction coefficient vector is updated by least square as shown in the following formula: (16) The process is to take all columns in the current support set from the target array manifold dictionary, and use least square to find the optimal linear combination to reconstruct the signal, so that the reconstructed signal is as close as possible to the observed signal , and obtain the reconstruction coefficient vector .

[0058] Then, the residual is updated according to the reconstructed sparse vector. This process can make the reconstructed signal continuously approach the observed signal, and the specific process is shown in the following formula: (17) Finally, the iteration is terminated when the norm of the current residual is less than a preset threshold, or the number of iterations reaches a preset number, and the sparse vector is obtained. For example, the iteration is terminated when the residual norm is lower than the threshold 0.05 or the number of iterations reaches 10.

[0059] Specifically, in the above process, the maximum sparsity (K=1, i.e., only one non-zero coefficient needs to be reconstructed) is set to preferentially capture the single target angle corresponding to the main lobe energy. In the OMP sparse reconstruction process, a dynamic weight adjustment and residual threshold control mechanism are introduced to filter out noise interference. Among them, the dynamic weight adjustment is reflected in that the algorithm will adaptively select the atom most related to the residual (i.e., the largest correlation coefficient) in each iteration, which is equivalent to giving the main signal component a higher "weight" and suppressing the influence of noise components; and the setting of the residual norm threshold ensures that the iteration is stopped in time when the residual energy is small enough, avoiding overfitting of noise as signal components. Through the combination of dynamic correlation selection and threshold truncation strategy, the pseudo-peak caused by noise can be effectively filtered out, and only the peak value corresponding to the true target is retained in the coefficient vector obtained by sparse reconstruction.

[0060] For example, in step S16, the sparse vector is searched for a peak value to obtain a direction of arrival estimation of the sound source signal, including: ​Step S166: Calculate the magnitude of each element of the sparse vector to obtain the sparse reconstruction magnitude. Step S167: Select the peak amplitude from the sparse reconstructed amplitude. When the peak amplitude is greater than the preset amplitude threshold, obtain the angle index corresponding to the peak amplitude. Step S168: Map the angle index to the target array manifold dictionary to obtain the corresponding azimuth and elevation angles; Step S169: Determine the azimuth and elevation angles as the direction of arrival estimates of the sound source signal.

[0061] Specifically, OMP sparse reconstruction outputs a sparse vector. , of which each The corresponding target array manifold dictionary The first guiding vector (i.e., the first...) (Several candidate directions). Since it is a complex number, its magnitude needs to be taken. This yields a set of "sparse reconstruction amplitudes". Find the maximum amplitude. When the peak amplitude is greater than the preset amplitude threshold, the angle index corresponding to the peak amplitude is obtained; the angle index is mapped to a dictionary to obtain the azimuth and pitch angles.

[0062] Further, refer to Figures 8A-8D As shown. Figures 8A-8D The direction-finding ambiguity regions in the azimuth and elevation directions were compared between a full array and an optimized octagonal array. Figure 8A and Figure 8C These correspond to the ambiguous regions in the azimuth and elevation directions of the full array, respectively, with a range of approximately ±1.5°. Figure 8B and Figure 8D To optimize the octal array, it can be seen that the unoptimized sparse array originally had a relatively large ambiguity region of about ±8°. However, after the array layout optimization and dictionary coherence suppression of this invention, the ambiguity region of the octal array was significantly compressed to about ±2°, which is very close to the full array level. This shows that the present invention can effectively overcome the original direction-finding ambiguity problem of sparse arrays.

[0063] Furthermore, referring to Table 1, regarding real-time performance, Table 1 presents a comparison of the computational load of the optimized octal sparse array of this invention and the traditional 49-element full array under different processing steps. It can be seen that in key processing steps such as covariance matrix operations, eigenvalue decomposition, and sparse reconstruction, the optimized sparse array significantly reduces the computational burden. Moreover, the total computational load of the optimized octal array is less than 12% of that of the full array, resulting in a substantial reduction in hardware resource consumption.

[0064]

[0065] At the same time, refer to Figure 9AAs shown, the DOA estimation accuracy rate comparison curves of the full array and the optimized eight-element array under low signal-to-noise ratio conditions are shown. It can be seen that the estimation accuracy rate curves of the two arrays are almost coincident within the entire measurement condition range, which means that the eight-element sparse array adopting the method of the application has the same direction finding accuracy as the full array, and even under the conditions of extremely low signal-to-noise ratio and few snapshots, the eight-element sparse array can still maintain the high accuracy comparable to the full array.

[0066] and reference Figure 9B As shown. Figure 9B The performance differences of different DOA estimation algorithms under the same eight-element array configuration are shown, and the root mean square error (RMSE) is used as an evaluation index for comparison. Among them, the popular dictionary of the optimized array is the application, the popular dictionary of the array is the method of not performing phase amplitude weighting and dictionary gram matrix optimization, and the MUSIC is the traditional DOA estimation algorithm, multiple signal classification (MUSIC) method. The curve comparison shows that the traditional subspace algorithm and the sparse reconstruction method without dictionary optimization all show high estimation error under low signal-to-noise ratio. In contrast, the optimized array manifold dictionary and the OMP algorithm proposed in the application can significantly reduce the estimation error, and the smallest RMSE is achieved within the entire test range. This result proves the superiority and robustness of the application for sparse array DOA estimation.

[0067] The beneficial effects of the application are as follows: (1) A dynamic frequency point energy screening mechanism is proposed In view of the problem of uneven energy distribution of wideband signals, the main energy frequency band is adaptively selected to participate in direction finding, avoiding the precision decline caused by the fixed frequency band containing noise in the traditional method; (2) An array manifold dictionary optimization method is designed, which greatly reduces the correlation between dictionary atoms through gradient descent, overcoming the problem of low reconstruction efficiency caused by high dictionary redundancy and strong atom coherence in previous sparse representation methods; (3) Combined with the eight-element sparse array layout optimized by the genetic algorithm, the number of array elements is significantly reduced while effectively suppressing the inherent direction finding ambiguity and grating lobe of the sparse array, improving the direction finding accuracy of the sparse array. In previous schemes, random selection of array elements in the sparse array can cause serious performance degradation, while the optimized array selection in the application makes the performance of the sparse array approach that of the full array. (4) An improved sparse reconstruction algorithm (OMP combined with a dynamic threshold) is used to robustly extract the direction of the target signal under low signal-to-noise ratio and few snapshots, avoiding the failure of traditional subspace algorithms under such conditions. In summary, the application scheme realizes low-cost, high-precision and high-robustness DOA estimation performance in complex environments, which is a beneficial improvement over the prior art.

[0068] It is noted that the above figures are only a schematic representation of the processes comprised in the method according to the exemplary embodiments of the present application and are not intended to limit the present application. It is readily understood that the processes shown in the above figures do not indicate or limit the chronological order of these processes. In addition, it is readily understood that these processes can be executed, for example, synchronously or asynchronously in a plurality of modules.

[0069] It is noted that, although in the above detailed description several modules or units of the device for action execution are mentioned, such a division is not mandatory. Indeed, according to embodiments of the present application, the features and functionalities of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functionalities of one module or unit described above can be further divided into embodied by a plurality of modules or units.

[0070] It is noted that the storage medium shown in the embodiments of the present application can be a computer readable signal medium or a computer readable storage medium or any combination of the above. The computer readable storage medium may, for example, be but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, the computer readable storage medium can be any tangible medium that contains or stores a program used or used in conjunction with an instruction execution system, device or apparatus. In the present application, the computer readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer readable program code. Such a propagated data signal can take on many forms, including but not limited to an electromagnetic signal, an optical signal or any suitable combination of the above. The computer readable signal medium can also be any storage medium that can send, propagate or transmit a program for use by or in connection with an instruction execution system, device or apparatus. The program code contained in the storage medium can be transmitted in any suitable medium, including but not limited to wireless, wired, or the like, or any suitable combination of the above.

[0071] The units described in the embodiments of the present application can be implemented by software, or by hardware, or by a combination of software and hardware. The units described can be located in one place, or distributed over several places. The names of the units in some cases do not limit the units themselves.

[0072] It should be noted that, as another aspect, the present application also provides a storage medium, which can be included in an electronic device, or can exist separately without being assembled into the electronic device. The storage medium carries one or more programs, which, when executed by an electronic device, cause the electronic device to implement the method described in the embodiments below. For example, the electronic device can implement each step of the method as shown in Figure 1

[0073] In one embodiment, the present application provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps in the method embodiments described above.

[0074] In addition, the above-described figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present application, and are not intended to be limiting. It is easy to understand that the processes shown in the above-described figures do not indicate or limit the time sequence of the processes. In addition, it is also easy to understand that the processes can be executed synchronously or asynchronously, for example, in multiple modules.

[0075] Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. It is intended that the present application cover any and all variations of the application that come within the scope of the claims and their equivalents. It is intended that the specification and examples be considered exemplary only, with the true scope and spirit of the application being indicated by the following claims.

[0076] It should be understood that the present application is not limited to the precise construction that has been described above and illustrated in the accompanying drawings, and that various modifications and changes can be made by those skilled in the art without departing from the scope of the present application. The scope of the present application is limited only by the appended claims.​

Claims

1. A sparse array wave direction-of-arrival estimation method based on an optimized array manifold dictionary, characterized in that, The method includes: By combining genetic algorithms to optimize the layout of the initial uniform array, an octet sparse array layout structure is obtained. The observation signal is obtained by using an octet sparse array layout structure; wherein, the observation signal is a multi-channel time-domain signal collected when receiving sound source signals using an octet sparse array. Energy filtering of the observed signals yields the dominant frequency band; and, combining the dominant frequency band and the octet sparse array layout structure, a target array manifold dictionary is constructed. The target array manifold dictionary is sparsely reconstructed based on the orthogonal matching pursuit algorithm to obtain the sparse vector corresponding to the sound source signal; and peak search is performed on the sparse vector to obtain the direction of arrival estimation of the sound source signal.

2. The method according to claim 1, characterized in that, The optimization layout of the initial uniform array using a genetic algorithm yields an octet sparse array layout structure, including: Based on the shape constraints of the array mounting platform, the array elements in the initial uniform area array are initially screened to obtain the initial array layout structure. Based on the initial array layout structure, the positions of candidate array elements are encoded using binary genes, and an initial population for the genetic algorithm is constructed according to the Gaussian distribution and the constraint of the number of array elements; wherein, the initial population contains multiple candidate sparse array individuals; The fitness value corresponding to each candidate sparse matrix individual is calculated based on the fitness function; wherein, the fitness function is constructed based on the main lobe width and side lobe level of the candidate sparse matrix individual; Genetic operations such as tournament selection, single-point crossover, and probabilistic mutation are used to iteratively update the initial population until a preset termination condition is met to obtain the target sparse array individuals; and the octal sparse array layout structure is determined based on the gene encoding corresponding to the target sparse array individuals.

3. The method according to claim 2, characterized in that, The calculation of the fitness value corresponding to each candidate sparse matrix individual based on the fitness function includes: Based on the element position coordinates of the candidate sparse array individuals, and by performing an angle scan on the directional response within the first preset azimuth angle domain and the first preset pitch angle domain, a three-dimensional orientation map of the candidate sparse array individuals is obtained. The maximum amplitude value in the three-dimensional radiation pattern is determined as the main lobe peak value; The angular region corresponding to when the main lobe peak value drops to a preset amplitude threshold is determined as the main lobe width. The sidelobe peaks are searched in the remaining angular domain; and the sidelobe level is obtained based on the difference between the sidelobe peaks and the main lobe peaks. The fitness value is obtained by weighting the main lobe width and the side lobe level.

4. The method according to claim 1, characterized in that, The process of energy filtering of the observed signal to obtain the dominant frequency band includes: The observed signal is segmented according to a preset time window length to obtain segmented signals; and a discrete Fourier transform is performed on the segmented signals to obtain the complex spectrum. Calculate the initial energy at each frequency point based on the complex spectrum; The initial energy at each frequency point is filtered according to the preset energy threshold to obtain the target energy that is greater than or equal to the preset energy threshold; The frequency point corresponding to the target energy is determined as the dominant frequency band.

5. The method according to claim 3, characterized in that, The method of constructing a target array manifold dictionary by combining the dominant frequency band and the octal sparse array layout structure includes: The frequency parameters of the steering vector are determined based on the dominant frequency band; where the steering vector is an atom of the initial array manifold dictionary; The element position coordinates of the guiding vector are determined based on the octet sparse array layout structure. By combining the second preset azimuth angle domain and the second preset elevation angle domain, the angle set of the guide vector is obtained; and the guide vector is determined based on the frequency parameter, the array element position coordinates and the angle set. The initial array manifold dictionary is obtained by arranging the steering vectors based on the frequency parameters and angle set; Phase compensation and amplitude weighting are applied to the initial array manifold dictionary to obtain the target array manifold dictionary.

6. The method according to claim 5, characterized in that, The step of performing phase compensation and amplitude weighting on the initial array manifold dictionary to obtain the target array manifold dictionary includes: The array reference center is determined based on the octet sparse array layout structure; and the phase term of the steering vector is compensated by utilizing the positional deviation between each array element and the array reference center to obtain the compensated phase of the steering vector. By adding weights to the magnitude term of the steering vector using a weighting function, the compensation magnitude of the steering vector can be obtained. Calculate the off-diagonal elements of the dictionary Gram matrix based on the compensated guide vector; The phase and magnitude terms of the compensated steering vector are iteratively adjusted using the gradient descent algorithm until the off-diagonal elements of the dictionary Gram matrix satisfy the preset termination condition, thus obtaining the dictionary of the target array manifold.

7. The method according to claim 1, characterized in that, The sparse reconstruction of the target array manifold dictionary based on the orthogonal matching pursuit algorithm to obtain the sparse vector corresponding to the sound source signal includes: Initialize the residuals to the observed signals and the support set to an empty set; Search the target array manifold dictionary for the column vector that is most correlated with the current residual, and add the index corresponding to the current column vector to the support set; The observed signal is fitted with least squares based on the updated support set to obtain the reconstructed coefficient vector. The reconstructed signal is calculated based on the reconstructed coefficient vector, and the residuals are updated. The iteration is terminated when the norm of the current residual is less than a preset threshold or the number of iterations reaches a preset number, resulting in a sparse vector.

8. The method according to claim 1, characterized in that, The peak search of the sparse vector to obtain the direction-of-arrival (DOA) estimate of the sound source signal includes: The magnitude of each element of the sparse vector is calculated to obtain the sparse reconstruction magnitude. Select the peak amplitude from the sparse reconstructed amplitude. When the peak amplitude is greater than the preset amplitude threshold, obtain the angle index corresponding to the peak amplitude. Map the angle index to the target array manifold dictionary to obtain the corresponding azimuth and elevation angles; The azimuth and elevation angles are used to estimate the direction of arrival of the sound source signal.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 8.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 8.

Images