A method and system for MUSIC spectrum estimation based on improved jacobi matrix decomposition
Patent Information
- Application Number
- CN202611084235.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-21
- Publication Date
- 2026-10-16
AI Technical Summary
若相关功能单元之间的接口组织和连接关系不清晰，则不利于形成稳定的硬件处理链路
本采用本发明所提供的方法，利用了稀疏矩阵的特性，充分发挥了矩阵乘法左行右列的特性，在Jacobi矩阵分解方面显著降低了编程复杂度和复用性，并且结构更加简单巧妙，易于理解。其次，通过将阵列信号采集、协方差矩阵构建、Jacobi并行特征分解、特征值排序和谱峰搜索输出集成为一体化硬件结构，便于在FPGA器件中形成连续的测角处理链路；通过设置实扩维协方差矩阵处理结构、并行Jacobi处理结构以及粗搜索和细搜索结合的谱峰搜索结构，便于各处理阶段之间进行接口衔接和并行配合。并且算法全流程采用硬件电路实现，无软件算法的串行执行缺陷，运算速度相比传统串行方案提升显著，硬件资源利用率高，实时性强，适配雷达、无线通信等实时空间谱估计场景，实用性强。
Smart Images
Figure CN122925796A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of array signal processing technology, and more specifically, to a MUSIC spectrum estimation method and system based on improved Jacobi matrix decomposition. Background Technology
[0002] The MUSIC algorithm is a commonly used high-resolution array direction finding method that can estimate the direction of arrival of a target signal by performing covariance analysis, eigenvalue decomposition, and spectral peak search on the received array signal. Existing MUSIC algorithms are mostly implemented in general-purpose processors or software platforms, which leads to problems such as long processing links, insufficient parallelism, and poor hardware adaptability when dealing with real-time processing of multi-channel array signals.
[0003] When deploying MUSIC algorithms onto FPGA devices, it is typically necessary to sequentially complete processes such as array signal acquisition, covariance matrix construction, eigenvalue decomposition, eigenvalue sorting, and two-dimensional angle search. If the interface organization and connection relationships between related functional units are unclear, it will hinder the formation of a stable hardware processing chain.
[0004] Furthermore, amplitude and phase errors may exist between the channels of the array. Without corresponding calibration structures, these errors will affect subsequent covariance matrix construction and spectral peak search results. Therefore, it is necessary to provide a parallel MUSIC spectrum estimation system for FPGA implementation to enable hardware-based angle measurement processing of the array signals. Summary of the Invention
[0005] The purpose of this invention is to provide a MUSIC spectrum estimation method and system based on improved Jacobi matrix decomposition to solve the problems existing in the prior art.
[0006] This invention is achieved through the following technical solution: A MUSIC spectrum estimation method based on improved Jacobi matrix decomposition includes: It receives complex array signal inputs corresponding to multiple array elements, sets the number of snapshots, performs complex covariance matrix accumulation calculation based on the number of snapshots, and converts the obtained complex covariance matrix into a real extended dimension covariance matrix. The extended covariance matrix is subjected to parallel Jacobi rotation iteration to output eigenvalues and eigenvectors. The eigenvalues and their corresponding eigenvectors are sorted in descending order to obtain the noise subspace eigenvectors. The search vector is constructed and calibrated to obtain the target search vector. The spectral value is calculated based on the target search vector and the noise subspace feature vector, and the target parameters are output.
[0007] Preferably, the step of converting the obtained complex covariance matrix into a real extended-dimensional covariance matrix includes:
[0008] In the formula, For the number of snapshots, It is a complex covariance matrix; It is a matrix with real extended dimensions; To obtain the real part of the matrix, To obtain the imaginary part of the matrix, This is the data vector received by a single snapshot of the array. This is the conjugate transpose.
[0009] Preferably, the target search vector obtained after calibration includes: Based on the acquisition of pitch and azimuth angles from complex array signals, the channel with the highest power is selected as the reference channel from all channels, and the first complex coefficient of each channel relative to the reference channel is calculated. Obtain the steering vector of the current true direction of arrival, normalize the steering vector and complex coefficients to the reference channel, and obtain the second complex coefficients that need to be corrected for each channel in the current region based on the steering vector and complex coefficients normalized to the reference channel. Construct a search vector to be calibrated, calibrate the search vector based on the second complex coefficient to obtain the calibrated search vector, and convert the calibrated search vector to a real number to obtain the target search vector.
[0010] Preferably, calculating the first complex coefficient of each channel relative to the reference channel includes:
[0011] In the formula, The first complex coefficient, For the first The first channel Data from each sampling point It is the reference channel number Data from each sampling point It is the conjugate of the reference channel. This represents the total number of sampling points.
[0012] Preferably, the normalization of the guide vector and complex coefficients to the reference channel includes:
[0013]
[0014] In the formula, These are the normalized complex coefficients. For the complex coefficients of the reference channel, As the guide vector, The normalized guide vector, The pitch angle, It is the azimuth angle. The guide vector for the reference channel; The second complex coefficients that need to be corrected for each channel in the current region, obtained based on the steering vector normalized to the reference channel and the complex coefficients, include:
[0015] In the formula, It is the second complex coefficient.
[0016] Preferably, constructing the search vector to be calibrated includes:
[0017] The calibration of the search vector obtained by calibrating the search vector to be calibrated based on the second complex coefficients includes:
[0018] The target search vector is obtained by converting the calibrated search vector to a real number:
[0019] In the formula, Let be the matrix to be decomposed. The imaginary unit, The wavelength of the signal carrier frequency, The pitch angle, It is the azimuth angle. The element spacing along the x-axis of the two-dimensional array structure plane. The element spacing in the y-axis direction of the two-dimensional array structure plane is denoted as ... by denoted as denoted as denoted by denoted as denoted by denoted as denoted by denoted as denoted by To obtain the real part, To obtain the imaginary part, The target search vector.
[0020] Preferably, the spectral value calculation based on the target search vector and the noise subspace feature vector includes:
[0021] In the formula, For spectral values, The eigenvector of the noise subspace is the th A noise intermediate vector.
[0022] Preferably, the output target parameters include: The two-dimensional search area is divided into several regions according to the pitch and azimuth angles. In each region, the first angle value is used for the search, and the azimuth and pitch angles with the minimum spectral values are used as the target points. The target point is searched using the second angle value to find the azimuth and elevation angles corresponding to the minimum spectral value in the current area, which are then output as target parameters.
[0023] Preferred options also include: The scanning angle is mapped to a preset angle range, and the corresponding value is read from the sine lookup table and cosine lookup table according to the mapped angle. Calculate the direction cosine parameter corresponding to the scanning angle, calculate the phase of each array element based on the array coordinates, direction cosine parameter and operating frequency, and output a data format that matches the feature vector of the noise subspace.
[0024] Secondly, the present invention also provides a MUSIC spectrum estimation system based on improved Jacobi matrix decomposition, for performing the above-described MUSIC spectrum estimation method based on improved Jacobi matrix decomposition, comprising: The array signal acquisition module is used to receive complex array signal inputs corresponding to multiple array elements; The covariance matrix construction module is used to set the number of snapshots, perform cumulative calculation of the complex covariance matrix based on the number of snapshots, and convert the obtained complex covariance matrix into a real extended dimension covariance matrix. The Jacobi parallel matrix decomposition module is used to perform parallel Jacobi rotation iteration on the extended dimension covariance matrix and output eigenvalues and eigenvectors. The eigenvalue sorting module is used to sort the eigenvalues and their corresponding eigenvectors in descending order to obtain the noise subspace eigenvectors. The peak search output module is used to construct the search vector and obtain the target search vector after calibration. Based on the target search vector and the noise subspace feature vector, the spectral value is calculated and the target parameter is output.
[0025] The technical solution of the present invention has at least the following advantages and beneficial effects: This invention employs the method provided by the present invention, utilizing the characteristics of sparse matrices and fully leveraging the left-row, right-column property of matrix multiplication. This significantly reduces programming complexity and reusability in Jacobi matrix decomposition, resulting in a simpler, more ingenious, and easier-to-understand structure. Secondly, by integrating array signal acquisition, covariance matrix construction, parallel Jacobi eigenvalue decomposition, eigenvalue sorting, and spectral peak search output into a unified hardware structure, it facilitates the formation of a continuous angle measurement processing chain within FPGA devices. Furthermore, by setting up a real-dimensional extended-dimensional covariance matrix processing structure, a parallel Jacobi processing structure, and a spectral peak search structure combining coarse and fine searches, it facilitates interface connection and parallel cooperation between various processing stages. Moreover, the entire algorithm is implemented using hardware circuitry, eliminating the serial execution defects of software algorithms. The computational speed is significantly improved compared to traditional serial schemes, with high hardware resource utilization, strong real-time performance, and adaptability to real-time spatial spectrum estimation scenarios such as radar and wireless communication, making it highly practical. Attached Figure Description
[0026] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 A schematic diagram of the architecture of the MUSIC spectrum estimation method based on improved Jacobi matrix decomposition provided by this invention; Figure 2 This is a schematic diagram of the process for constructing a real extended dimension covariance matrix provided by the present invention; Figure 3 A schematic diagram of the process for the improved Jacobi matrix decomposition provided by the present invention; Figure 4 This is a schematic diagram of the peak search process provided by the present invention.
[0028] Figure 5 This is a comparison chart of the FPGA and MATLAB results of this invention. Detailed Implementation
[0029] 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 with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0030] The module division in this application is a logical division. In actual application, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not executed.
[0031] Furthermore, the connection, coupling, or communication in this application can be a direct connection, coupling, or communication between related objects, or an indirect connection, coupling, or communication through other devices. Moreover, the connection, coupling, or communication between objects can be electrical or other similar forms, and this application does not impose any limitations on these. Independently described modules or sub-modules may or may not be physically separated; they may be implemented in software or hardware, and some modules or sub-modules may be implemented in software, with the processor calling the software to implement the function of these modules or sub-modules, while other modules or sub-modules may be implemented in hardware, such as through hardware circuits. Furthermore, some or all of the modules can be selected to achieve the purpose of this application's solution according to actual needs.
[0032] Please refer to Figure 1 A method for estimating the MUSIC spectrum based on improved Jacobi matrix decomposition, comprising: S100: Receives complex array signal inputs corresponding to multiple array elements, sets the number of snapshots, performs complex covariance matrix accumulation calculation based on the number of snapshots, and converts the obtained complex covariance matrix into a real extended dimension covariance matrix. The signal from the direction of arrival of the array elements is acquired and digital signals are obtained. The obtained digital signals are then subjected to complex number operations, accumulated according to the number of snapshots, and the matrix values are automatically scaled to construct the covariance matrix, followed by real matrix expansion.
[0033] S102: Perform parallel Jacobi rotation iteration on the extended covariance matrix to output eigenvalues and eigenvectors. Sort the eigenvalues and their corresponding eigenvectors in descending order to obtain the noise subspace eigenvectors. For a matrix with expanded real dimension, matrix decomposition is performed first. Using the current index pair group, a Givens rotated sparse matrix G is generated. Then, the sparse matrix is right-multiplied and then left-multiplied by the decomposed matrix A. Simultaneously, the eigenvectors V also need to be multiplied by the rotated matrix G. This process is repeated until the matrix converges. The diagonal elements of the matrix are then sorted in descending order using bubble sort, and the eigenvectors corresponding to the eigenvalues are also sorted accordingly. S103: Construct the search vector and calibrate it to obtain the target search vector. Calculate the spectral value based on the target search vector and the noise subspace feature vector, and output the target parameters.
[0034] The feature vector is extracted to extract the noise subspace according to the number of signals, and a search angle is generated. The corresponding search vector is generated according to the search angle. Then, a 5-region calibration is performed to obtain an accurate search vector. This vector is multiplied with the corresponding noise subspace, and the absolute values are summed to obtain the spectral value. Then, the pitch and azimuth angles corresponding to the smallest spectral value are found and output.
[0035] This invention employs the method provided by the present invention, utilizing the characteristics of sparse matrices and fully leveraging the left-row, right-column property of matrix multiplication. This significantly reduces programming complexity and reusability in Jacobi matrix decomposition, resulting in a simpler, more ingenious, and easier-to-understand structure. Secondly, by integrating array signal acquisition, covariance matrix construction, parallel Jacobi eigenvalue decomposition, eigenvalue sorting, and spectral peak search output into a unified hardware structure, it facilitates the formation of a continuous angle measurement processing chain within FPGA devices. Furthermore, by setting up a real-dimensional extended-dimensional covariance matrix processing structure, a parallel Jacobi processing structure, and a spectral peak search structure combining coarse and fine searches, it facilitates interface connection and parallel cooperation between various processing stages. Moreover, the entire algorithm is implemented using hardware circuitry, eliminating the serial execution defects of software algorithms. The computational speed is significantly improved compared to traditional serial schemes, with high hardware resource utilization, strong real-time performance, and adaptability to real-time spatial spectrum estimation scenarios such as radar and wireless communication, making it highly practical.
[0036] In one exemplary embodiment of the present invention, the process for constructing the real extended-dimensional covariance matrix is as follows: Figure 2 As shown, the signal is acquired, and autocorrelation matrix operations are performed on the acquired signal, namely complex number operations, snapshot accumulation, adaptive scaling, and real dimension expansion. Specifically:
[0037]
[0038] In the formula, For the number of snapshots, It is a complex covariance matrix; It is a matrix with real extended dimensions; To obtain the real part of the matrix, To obtain the imaginary part of the matrix, This is the data vector received by a single snapshot of the array. It is the conjugate transpose. This is the data vector received by a single snapshot of the array. It is a matrix with real extended dimensions; To obtain the real part of the matrix, To obtain the imaginary part of the matrix, For the first channel The conjugate of each snapshot data For the first channel Each snapshot data point.
[0039] An exemplary embodiment of the present invention, specifically the Jacobi parallel matrix decomposition process, is as follows: Figure 3 As shown, the input matrix is rotated on p and q according to the round-robin scheduling sequence 7. Seven different rotation numbers are generated for each of the seven different p and q pairs. Then, row and column rotations are performed. Finally, the L1 norm convergence segment is calculated for the off-diagonal elements of the matrix. When this value is less than a threshold, the matrix convergence is complete, and the diagonal eigenvalues and eigenvectors are output. If the threshold is not met, rotation sweeping continues until the maximum number of sweeps N is reached without convergence. At this point, the result of the current operation is discarded, and the next operation begins. Detailed steps are as follows: Generate a pq schedule, where p is the index row of the matrix and q is the index column of the matrix.
[0040]
[0041] in, For the row of the matrix index pair, This refers to the columns of matrix index pairs. We take p and q from the top and bottom respectively, meaning each index has 7 pairs. After taking all the pairs, the sequence is shifted once in the direction of the arrow during the next rotation. The resulting 14x14 matrix parallel Jacobi grouping is as follows: s1: {(13,14)(1,12)(2,11)(3,10)(4,9)(5,8)(6,7)}; s2: {(12,14)(11,13)(1,10)(2,9)(3,8)(4,7)(5,6)}; s3: {(11,14)(10,12)(9,13)(1,8)(2,7)(3,6)(4,5)}; s5: {(9,14)(8,10)(7,11)(6,12)(5,13)(1,4)(2,3)} s6: {(8,14)(7,9)(6,10)(5,11)(4,12)(3,13)(1,2)}; s7: {(7,14)(6,8)(5,9)(4,10)(3,11)(2,12)(1,13)}; s8: {(6,14)(5,7)(4,8)(3,9)(2,10)(1,11)(12,13)}; s9: {(5,14)(4,6)(3,7)(2,8)(1,9)(10,13)(11,12)}; s10: {(4,14)(3,5)(2,6)(1,7)(8,13)(9,12)(10,11)}; s11: {(3,14)(2,4)(1,5)(6,13)(7,12)(8,11)(9,10)}; s12: {(2,14)(1,3)(4,13)(5,12)(6,11)(7,10)(8,9)}; s13: {(1,14)(2,13)(3,12)(4,11)(5,10)(6,9)(7,8)}; Then according to Calculate the rotation factor for each group, where, The sweep angle for the current index pair. For real extended dimension matrix The Line number The data in the column, for The Line number The data in the column, for The Line 1 Column data, then take cos and sin Examples of forming a rotated sparse matrix G include:
[0042]
[0043] The elements written in the middle are all 0, and then the selected matrix R_real is rotated in both rows and columns. , For a matrix with real-dimensional extended dimensions due to row and column rotations, performing the same operation for each rotation yields:
[0044] in, For the first A rotational sparse matrix, For the first matrix 1 eigenvalue, This is to extract the data from the diagonal of the matrix.
[0045] Simultaneously, the feature vectors of the noise subspace also need to be rotated accordingly. , Let be the feature vector of the noise subspace, where It is the first A number of rotated sparse matrices are used to complete the matrix decomposition and the calculation of the noise subspace.
[0046] An exemplary embodiment of the present invention, the specific peak search process of this embodiment is as follows: Figure 4 As shown, the target search vector obtained after calibration includes: Based on the acquisition of pitch and azimuth angles from complex array signals, the channel with the highest power is selected as the reference channel from all channels, and the first complex coefficient of each channel relative to the reference channel is calculated. Obtain the steering vector of the current true direction of arrival, normalize the steering vector and complex coefficients to the reference channel, and obtain the second complex coefficients that need to be corrected for each channel in the current region based on the steering vector and complex coefficients normalized to the reference channel. Construct a search vector to be calibrated, calibrate the search vector based on the second complex coefficient to obtain the calibrated search vector, and convert the calibrated search vector to a real number to obtain the target search vector.
[0047] Calculating the first complex coefficient of each channel relative to the reference channel includes:
[0048] In the formula, The first complex coefficient, For the first The first channel Data from each sampling point It is the reference channel number Data from each sampling point It is the conjugate of the reference channel. This represents the total number of sampling points.
[0049] Normalizing the steering vector and complex coefficients to the reference channel includes:
[0050]
[0051] In the formula, These are the normalized complex coefficients. For the complex coefficients of the reference channel, As the guide vector, The normalized guide vector, The pitch angle, It is the azimuth angle. The guide vector for the reference channel; The second complex coefficients that need to be corrected for each channel in the current region are obtained based on the steering vector normalized to the reference channel and the complex coefficients.
[0052] In the formula, It is the second complex coefficient.
[0053] Specifically, the spectral values are calculated based on the eigenvector (i.e., the noise subspace input spectrum) obtained by sorting the eigenvalues in descending order. Then, a search vector is generated. This search vector needs to be multiplied by different coefficients for different regions to reduce errors caused by physical array elements and the RF front-end. The two-dimensional search region is divided according to elevation and azimuth angles: region 1 (elevation 0-20°), region 2 (azimuth 0-90°), region 3 (azimuth 90-180°), region 4 (azimuth 180°-270°), and region 5 (azimuth 270-360°). A coarse search is then performed based on the search vector, starting with 3° increments to calculate the spectral values. The azimuth and elevation angles with the smallest spectral values are obtained. Then, a fine search is performed within a 12° radius around this point, in increments of 0.3°. Finally, the azimuth and elevation angles corresponding to the smallest spectral values in this region are output as the estimated direction of arrival (DOA).
[0054] In one exemplary embodiment of the present invention, constructing the search vector to be calibrated includes:
[0055] The calibration of the search vector obtained by calibrating the search vector to be calibrated based on the second complex coefficients includes:
[0056] The target search vector is obtained by converting the calibrated search vector to a real number:
[0057] In the formula, For the target search vector, The calibrated target search vector, The imaginary unit, The wavelength of the signal carrier frequency, The pitch angle, It is the azimuth angle. The element spacing along the x-axis of the two-dimensional array structure plane. The element spacing in the y-axis direction of the two-dimensional array structure plane is denoted as ... by denoted as denoted as denoted by denoted as denoted by denoted as denoted by denoted as denoted by To obtain the real part, To obtain the imaginary part, This is the target search vector after being converted to a real number.
[0058] An exemplary embodiment of the present invention includes calculating spectral values based on the target search vector and the noise subspace feature vector, comprising:
[0059] In the formula, For spectral values, The eigenvector of the noise subspace is the th A noise intermediate vector.
[0060] Preferably, the output target parameters include: The two-dimensional search area is divided into several regions according to the pitch and azimuth angles. In each region, the first angle value is used for the search, and the azimuth and pitch angles with the minimum spectral values are used as the target points. The target point is searched using the second angle value to find the azimuth and elevation angles corresponding to the minimum spectral value in the current area, which are then output as target parameters.
[0061] Preferred options also include: The scanning angle is mapped to a preset angle range, and the corresponding value is read from the sine lookup table and cosine lookup table according to the mapped angle. Calculate the direction cosine parameter corresponding to the scanning angle, calculate the phase of each array element based on the array coordinates, direction cosine parameter and operating frequency, and output a data format that matches the feature vector of the noise subspace.
[0062] like Figure 5 As shown, this invention provides a result comparison of one embodiment. The left side shows the MUSIC spectrum estimation result of the algorithm of this invention on the FPGA, and the right side shows the MUSIC spectrum estimation result after matrix decomposition using the MATLAB library function eig. It can be seen that both have high accuracy. This invention has great convenience for Verilog programming and module reusability. Secondly, the present invention also provides a MUSIC spectrum estimation system based on improved Jacobi matrix decomposition, for performing the above-described MUSIC spectrum estimation method based on improved Jacobi matrix decomposition, comprising: The array signal acquisition module is used to receive complex array signal inputs corresponding to multiple array elements; The covariance matrix construction module is used to set the number of snapshots, perform cumulative calculation of the complex covariance matrix based on the number of snapshots, and convert the obtained complex covariance matrix into a real extended dimension covariance matrix. The Jacobi parallel matrix decomposition module is used to perform parallel Jacobi rotation iteration on the extended dimension covariance matrix and output eigenvalues and eigenvectors. The eigenvalue sorting module is used to sort the eigenvalues and their corresponding eigenvectors in descending order to obtain the noise subspace eigenvectors. The peak search output module is used to construct the search vector and obtain the target search vector after calibration. Based on the target search vector and the noise subspace feature vector, the spectral value is calculated and the target parameter is output.
[0063] Specifically, the array signal acquisition module is used to receive the in-phase and quadrature components of the seven array elements, and the covariance matrix construction module converts the 7th order complex covariance matrix into a 14th order real extended dimension covariance matrix.
[0064] The covariance matrix construction module includes a complex multiply-accumulate unit array, a snapshot counting unit, an adaptive scaling unit, and a matrix output unit; The complex multiply-accumulate array is used to perform conjugate multiplication and accumulation on the signals of each array element; The snapshot counting unit is used to trigger the output of the covariance matrix when a preset number of snapshots is reached; The adaptive scaling unit is used to determine the truncation scaling parameters based on the maximum value of the diagonal elements of the covariance matrix; The matrix output unit is used for output, that is, to generate a 14-dimensional real expanded covariance matrix according to the formula.
[0065] The Jacobi parallel eigendecomposition module includes a round-robin scheduling unit, a Givens rotation parameter calculation unit, and multiple parallel Jacobi processing units. The round-robin scheduling unit is used to generate multiple non-conflicting matrix element index pairs in each round of cleaning; its specific sequence scheduling is as follows: The Givens rotation parameter calculation unit is used to calculate the rotation cosine and rotation sine in parallel based on the matrix elements of the corresponding index pairs; Multiple parallel Jacobi processing units are used to perform matrix column updates, matrix row updates, and eigenvector updates in parallel.
[0066] The Jacobi parallel eigenvalue decomposition module processes 7 pairs of matrix elements in parallel in each iteration and completes the diagonalization of the 14th-order real extended dimension covariance matrix through multiple rounds of sweeping.
[0067] The eigenvalue sorting module is used to sort the 14 eigenvalues in descending order and simultaneously rearrange the eigenvectors corresponding to the eigenvalues to form a set of noise subspace eigenvectors for spectral peak search.
[0068] The spectral peak search module includes a search vector generation unit, a coarse search angle control unit, a fine search angle control unit, a spectral value calculation unit, and a peak value determination unit; The search vector generation unit is used to generate search vectors based on the scanning angle, array coordinates, and operating frequency. The coarse search angle control unit is used to perform grid scanning of the pitch and azimuth angles according to the first step. The fine search angle control unit is used to perform a local fine scan in the vicinity of the candidate peak obtained by the coarse search according to the second step; The spectral value calculation unit is used for calculating the spectral values at angle points; The peak determination unit is used to determine the pitch and azimuth angles corresponding to the final peak from the coarse and fine search results.
[0069] The search vector generation unit includes an angle mapping unit, a trigonometric function lookup unit, a direction cosine calculation unit, an array phase calculation unit, and a search vector packing unit. The angle mapping unit is used to map the scanning angle to a preset angle range; The trigonometric function lookup unit is used to read the corresponding value from the sine lookup table and the cosine lookup table according to the mapped angle; The direction cosine calculation unit is used to calculate the direction cosine parameters corresponding to the scanning angle; The array phase calculation unit is used to calculate the phase of each array element based on the array coordinates, direction cosine parameters, and operating frequency. The search vector packing unit is used to output a data format that matches the feature vectors of the noise subspace.
[0070] The spectral value calculation unit searches in 3° increments during the coarse search phase, with elevation ranging from 0 to 90° and azimuth from 0 to 360°. During the fine search phase, it expands the search range by 6° around the maximum spectral value from the coarse search.
[0071] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0072] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0073] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A MUSIC spectrum estimation method based on improved Jacobi matrix decomposition, characterized in that, include: It receives complex array signal inputs corresponding to multiple array elements, sets the number of snapshots, performs complex covariance matrix accumulation calculation based on the number of snapshots, and converts the obtained complex covariance matrix into a real extended dimension covariance matrix. The extended covariance matrix is subjected to parallel Jacobi rotation iteration to output eigenvalues and eigenvectors. The eigenvalues and their corresponding eigenvectors are sorted in descending order to obtain the noise subspace eigenvectors. The search vector is constructed and calibrated to obtain the target search vector. The spectral value is calculated based on the target search vector and the noise subspace feature vector, and the target parameters are output.
2. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 1, characterized in that, The step of converting the obtained complex covariance matrix into a real extended-dimensional covariance matrix includes: In the formula, For the number of snapshots, It is a complex covariance matrix; It is a matrix with real extended dimensions; To obtain the real part of the matrix, To obtain the imaginary part of the matrix, This is the data vector received by a single snapshot of the array. This is the conjugate transpose.
3. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 2, characterized in that, After calibration, the target search vector obtained includes: Based on the acquisition of elevation and azimuth angles from complex array signals, the channel with the highest power is selected as the reference channel from all channels, and the first complex coefficient of each channel relative to the reference channel is calculated. Obtain the steering vector of the current true direction of arrival, normalize the steering vector and complex coefficients to the reference channel, and obtain the second complex coefficients that need to be corrected for each channel in the current region based on the steering vector and complex coefficients normalized to the reference channel. Construct a search vector to be calibrated, calibrate the search vector based on the second complex coefficient to obtain the calibrated search vector, and convert the calibrated search vector to a real number to obtain the target search vector.
4. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 3, characterized in that, The calculation of the first complex coefficient of each channel relative to the reference channel includes: In the formula, The first complex coefficient, For the first The first channel Data from each sampling point It is the reference channel number Data from each sampling point It is the conjugate of the reference channel. This represents the total number of sampling points.
5. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 4, characterized in that, The normalization of the steering vector and complex coefficients to the reference channel includes: In the formula, These are the normalized complex coefficients. For the first complex coefficient of the reference channel, As the guide vector, The normalized guide vector, The pitch angle, It is the azimuth angle. The guide vector for the reference channel; The second complex coefficients that need to be corrected for each channel in the current region, obtained based on the steering vector normalized to the reference channel and the complex coefficients, include: In the formula, It is the second complex coefficient.
6. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 5, characterized in that, The construction of the search vector to be calibrated includes: The calibration of the search vector obtained by calibrating the search vector to be calibrated based on the second complex coefficients includes: The target search vector is obtained by converting the calibrated search vector to a real number: In the formula, The matrix to be decomposed, The imaginary unit, The wavelength of the signal carrier frequency, The pitch angle, It is the azimuth angle. The element spacing along the x-axis of the two-dimensional array structure plane. The element spacing in the y-axis direction of the two-dimensional array structure plane is denoted as ... by denoted as denoted as denoted by denoted as denoted by denoted as denoted by denoted as denoted by To obtain the real part, To obtain the imaginary part, The target search vector.
7. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 6, characterized in that, The spectral value calculation based on the target search vector and the noise subspace feature vector includes: In the formula, For spectral values, The eigenvector of the noise subspace is the th A noise intermediate vector.
8. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 7, characterized in that, The output target parameters include: The two-dimensional search area is divided into several regions according to the pitch and azimuth angles. In each region, the first angle value is used for the search, and the azimuth and pitch angles with the minimum spectral values are used as the target points. The target point is searched using the second angle value to find the azimuth and elevation angles corresponding to the minimum spectral value in the current area, which are then output as target parameters.
9. The MUSIC spectrum estimation method based on improved Jacobi matrix decomposition according to claim 7, characterized in that, Also includes: The scanning angle is mapped to a preset angle range, and the corresponding value is read from the sine lookup table and cosine lookup table according to the mapped angle. Calculate the direction cosine parameter corresponding to the scanning angle, calculate the phase of each array element based on the array coordinates, direction cosine parameter and operating frequency, and output a data format that matches the feature vector of the noise subspace.
10. A MUSIC spectrum estimation system based on improved Jacobi matrix decomposition, characterized in that, The method for performing the MUSIC spectrum estimation method based on improved Jacobi matrix decomposition as described in any one of claims 1-9 includes: The array signal acquisition module is used to receive complex array signal inputs corresponding to multiple array elements; The covariance matrix construction module is used to set the number of snapshots, perform cumulative calculation of the complex covariance matrix based on the number of snapshots, and convert the obtained complex covariance matrix into a real extended dimension covariance matrix. The Jacobi parallel matrix decomposition module is used to perform parallel Jacobi rotation iteration on the extended dimension covariance matrix and output eigenvalues and eigenvectors. The eigenvalue sorting module is used to sort the eigenvalues and their corresponding eigenvectors in descending order to obtain the noise subspace eigenvectors. The peak search output module is used to construct the search vector and obtain the target search vector after calibration. Based on the target search vector and the noise subspace feature vector, the spectral value is calculated and the target parameter is output.