An improved DOA estimation algorithm based on the OMP algorithm and its application

Abstract

This invention discloses a DOA estimation algorithm based on an improved OMP algorithm. The algorithm references the DBF algorithm during atom selection, and the method includes: initializing residuals, index sets, iteration counts, and atom selection parameters; performing spectral peak search based on the correlation between the current residuals and the measurement matrix, and batch selecting multiple potential atoms to add to a temporary index set; performing threshold screening on the atoms in the temporary index set according to preset atom selection parameters, removing erroneous atoms, retaining correct atoms, and updating them to the main index set; reconstructing the signal and updating the residuals based on the updated main index set using least squares estimation; determining whether the iteration stopping condition is met; if so, outputting the DOA estimation result; otherwise, returning to step S2 to continue iteration. This invention achieves a combination of efficient DOA estimation and flexible system deployment, significantly improving the overall perception performance in complex dynamic scenarios.

Description

Technical Field

[0001] This invention belongs to the field of array signal processing, specifically relating to a DOA estimation algorithm based on an improved OMP algorithm and its applications. Background Technology

[0002] Direction of arrival (DOA) estimation, a core problem in array signal processing, has wide applications in radar, wireless communication, seismic exploration, and many other fields. Currently, DOA estimation algorithms proposed by researchers both domestically and internationally can be mainly divided into two categories:

[0003] The first category is beamforming methods represented by the Capon algorithm. Its basic idea is to obtain the spectral distribution of the signal in the spatial domain by suppressing interference and noise and enhancing the target signal.

[0004] The second category is subspace methods, represented by the multiple signal classification (MUSIC) algorithm and the rotation-invariant subspace (ESPRIT) algorithm. These algorithms separate the signal subspace and the noise subspace by performing eigenvalue decomposition on the covariance matrix of the array received data, and use the orthogonality between the two to construct a spatial spectrum function to achieve high-resolution DOA estimation.

[0005] However, both of these traditional methods have their limitations: beamforming algorithms, while breaking through the limitations of physical array aperture to some extent, have low angular resolution and are more sensitive to noise; while subspace algorithms rely on a large amount of independent and identically distributed snapshot data to ensure that the covariance matrix has good statistical properties, resulting in high computational complexity and making it difficult to meet the real-time processing requirements of modern systems.

[0006] In recent years, compressed sensing (CS) theory has attracted widespread attention due to its ability to achieve high-precision signal reconstruction at rates far below the Nyquist sampling rate by utilizing the sparse priors of signals. In DOA estimation scenarios, since the number of target sources in the actual space is usually much smaller than the dimension of the entire angle search grid, the signal is naturally sparsity in the spatial domain, making it very suitable for modeling and solving using the compressed sensing framework.

[0007] Nevertheless, traditional compressed sensing-based DOA estimation algorithms (such as Orthogonal Matching Pursuit OMP) still have shortcomings: OMP selects only one most relevant atom to add to the support set in each iteration, resulting in a large number of iterations required and a slow convergence speed, which affects the estimation efficiency and practicality.

[0008] Furthermore, in complex real-world application scenarios (such as intelligent transportation hubs, security of large venues, and monitoring of flexible production lines), the traditional fixed radar array deployment method reveals its lack of flexibility; fixed array positions and viewing angles may lead to perception blind spots, making it impossible to cope with dynamically changing key areas of the task, and the performance of a single array is limited when facing widely distributed or densely angled targets; therefore, how to combine efficient DOA estimation algorithms with flexible and reconfigurable radar hardware systems to build an intelligent perception network that can adapt to changes in the environment and tasks has become the key to improving the overall system performance;

[0009] Therefore, to address the aforementioned shortcomings, this invention proposes an improved DOA estimation algorithm based on the OMP algorithm and its application. Summary of the Invention

[0010] To overcome the shortcomings of existing technologies, this invention proposes an improved DOA estimation algorithm based on the OMP algorithm and its application. It aims to solve the problems of the traditional OMP algorithm having a large number of iterations and slow convergence speed in the direction of arrival estimation process. The efficient algorithm is applied to a dynamically configurable multi-sensor system to improve the overall perception capability and adaptability of the system.

[0011] The present invention discloses an improved DOA estimation algorithm based on the OMP algorithm, comprising the following steps:

[0012] S1. Initialize residuals, index set, iteration count, and atomic sieving parameters;

[0013] S2. Based on the correlation between the current residual and the measurement matrix, perform spectral peak search and select multiple potential atoms in batches to add to the temporary index set; S3. According to the preset atom filtering parameters, perform threshold filtering on the atoms in the temporary index set, remove erroneous atoms, retain correct atoms, and update the main index set; S4. Based on the updated master index set, reconstruct the signal and update the residuals using least squares estimation; S5. Determine whether the iteration stopping condition is met. If it is met, output the DOA estimation result; otherwise, return to step S2 to continue the iteration.

[0014] Input an M×N dimensional measurement matrix A, M dimensional observations X, sparsity K, and atom sieving parameters λ.

[0015] Output an N×1 dimensional sparse signal S.

[0016] Preferably, in S1, the initialization specifically involves setting the initial residual. The observed signal vector Initial primary index set The correct number of atoms found Number of iterations The input parameters include the measurement matrix A, the observed signal X, the signal sparsity K, and the atom sieving parameter λ, where... .

[0017] Preferably, in S2, the specific method for batch selecting multiple potential atoms is: calculating the current residual. The inner product modulus of each column of the measurement matrix A is used to obtain an initial correlation vector; a spectral peak search is performed on the initial correlation vector, and the atomic indices corresponding to the largest Ki spectral peaks are selected to form a temporary index set Ti, where... , This represents the number of correct atoms found after the (i-1)th iteration.

[0018] Preferably, in S3, the specific method for threshold filtering is as follows: calculate the maximum value of all spectral peaks in the temporary index set Ti. Set the filtering threshold Traverse Ti and retain only the atom indices whose spectral peaks are greater than η, forming a finely selected set of atom indices. .

[0019] Preferably, the atomic screening parameters The range of values ​​is .

[0020] Preferably, in S3, the specific method for updating the main index set is as follows: The refined atomic index set... Merge into the primary index set, i.e. And update the number of the correct atoms found. ,in Represents a set The number of elements; the specific method of S4 is: based on the current primary index set Extract the corresponding columns from measurement matrix A to form a submatrix. Solving for sparse signals using least squares estimation: Update residuals: This step is consistent with the traditional OMP algorithm, aiming to determine the origin of the found matrix A. i And the current residual X=r i-1 The sparse signal S is calculated using the least squares method, where S is the spatial incident signal vector described on page 8 of the theoretical introduction. The residuals are then updated based on the calculated S for the next iteration. Here, i represents the iteration number. To find matrix A i The left-hand reverse.

[0021] In S5, the iteration stopping condition is: the number of correct atoms Mi found is equal to the preset signal sparsity K; in S5, the specific method for outputting the DOA estimation result is: mapping the atom indices in the final master index set to the pre-divided spatial angle grid to obtain the corresponding direction of arrival angle estimation value.

[0022] An application of a DOA estimation algorithm based on an improved OMP algorithm is disclosed. This application is set up in a perception system, which includes a central processing unit, at least two movable millimeter-wave radar nodes, and a communication network. The millimeter-wave radar nodes are used to collect echo data and execute the DOA estimation algorithm to obtain local target angle estimation information, and then send this information along with their own position and attitude information to the central processing unit. The central processing unit is used to fuse information from each node to generate a global situational awareness, and to generate movement control commands and / or operating parameter adjustment commands for specific radar nodes based on the global situational awareness and task requirements. The communication network connects the central processing unit with each millimeter-wave radar node to achieve data and command transmission. The millimeter-wave radar nodes are mounted on a mobile platform, which may be a tracked robot, a wheeled robot, a drone, or an electric gimbal. The central processing unit is also used to automatically calculate and issue at least one of the following commands based on the global target distribution, task priority, and environmental constraints: a target movement position command for the radar node, an array baseline adjustment command, a radar transmit waveform adjustment command, and a receive gain adjustment command.

[0023] The beneficial effects of this invention are as follows:

[0024] 1. This invention, by selecting and filtering multiple atoms in batches during each iteration, enables DOA estimation of all signal sources with fewer iterations. This directly reduces the number of core operations such as least squares calculation during algorithm execution, significantly reducing computational complexity and processing latency, and making it more suitable for real-time engineering implementation;

[0025] 2. This invention introduces a batch pre-selection mechanism for peak search based on a reference DBF, combined with an atomic screening strategy based on adaptive parameters. This effectively accelerates the algorithm's convergence process while maintaining estimation accuracy. This enables faster and more accurate DOA estimation results in highly dynamic scenarios or applications with high real-time requirements. Attached Figure Description

[0026] Figure 1 The flowchart shows a proposed DOA estimation algorithm based on an improved OMP algorithm.

[0027] Figure 2 The graph shows the estimation results for simulation experiment 1;

[0028] Figure 3The graph shows the estimation results for simulation experiment 2;

[0029] Figure 4 The graph shows the estimation results for simulation experiment 3;

[0030] Figure 5 The graph shows the estimation results for simulation experiment 4;

[0031] Figure 6 This is a scene diagram from the actual test experiment;

[0032] Figure 7 The graph shows the estimated results of Experiment 1.

[0033] Figure 8 The graph shows the estimated results of Experiment 2. Detailed Implementation

[0034] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0035] like Figure 1-8 As shown, the DOA estimation algorithm based on the improved OMP algorithm of the present invention includes the following steps:

[0036] S1. Initialize residuals, index set, iteration count, and atomic sieving parameters;

[0037] S2. Based on the correlation between the current residual and the measurement matrix, perform spectral peak search and select multiple potential atoms in batches to add to the temporary index set; S3. According to the preset atom filtering parameters, perform threshold filtering on the atoms in the temporary index set, remove erroneous atoms, retain correct atoms, and update the main index set; S4. Based on the updated master index set, reconstruct the signal and update the residuals using least squares estimation; S5. Determine whether the iteration stopping condition is met. If it is met, output the DOA estimation result; otherwise, return to step S2 to continue the iteration.

[0038] Input an M×N dimensional measurement matrix A, M dimensional observations X, sparsity K, and atom sieving parameters λ.

[0039] Output an N×1 dimensional sparse signal S.

[0040] In one embodiment of the present invention, in S1, the initialization specifically involves: setting an initial residual. The observed signal vector Initial primary index set The correct number of atoms found Number of iterations The input parameters include the measurement matrix A, the observed signal X, the signal sparsity K, and the atom sieving parameter λ, where... .

[0041] In one embodiment of the present invention, in S2, the specific method for batch selecting multiple potential atoms is as follows: calculating the current residual. The inner product modulus of each column of the measurement matrix A is used to obtain an initial correlation vector; a spectral peak search is performed on the initial correlation vector, and the atomic indices corresponding to the largest Ki spectral peaks are selected to form a temporary index set Ti, where... , This represents the number of correct atoms found after the (i-1)th iteration.

[0042] In one embodiment of the present invention, in S3, the specific method for threshold filtering is as follows: calculating the maximum value of all spectral peaks in the temporary index set Ti. Set the filtering threshold Traverse Ti and retain only the atom indices whose spectral peaks are greater than η, forming a finely selected set of atom indices. .

[0043] As one embodiment of the present invention, the atomic screening parameters The range of values ​​is .

[0044] Preferably, in S3, the specific method for updating the main index set is as follows: The refined atomic index set... Merge into the primary index set, i.e. And update the number of the correct atoms found. ,in Represents a set The number of elements; the specific method of S4 is: based on the current primary index set Extract the corresponding columns from measurement matrix A to form a submatrix. Solving for sparse signals using least squares estimation: Update residuals: This step is consistent with the traditional OMP algorithm, aiming to determine the origin of the found matrix A. i And the current residual X=r i-1 The sparse signal S is calculated using the least squares method, where S is the spatial incident signal vector described on page 8 of the theoretical introduction. The residuals are then updated based on the calculated S for the next iteration. Here, i represents the iteration number. To find matrix A i The left-hand reverse.

[0045] In S5, the iteration stopping condition is: the number of correct atoms Mi found is equal to the preset signal sparsity K; in S5, the specific method for outputting the DOA estimation result is: mapping the atom indices in the final master index set to the pre-divided spatial angle grid to obtain the corresponding direction of arrival angle estimation value.

[0046] An application of a DOA estimation algorithm based on an improved OMP algorithm is disclosed. This application is set up in a perception system, which includes a central processing unit, at least two movable millimeter-wave radar nodes, and a communication network. The millimeter-wave radar nodes are used to collect echo data and execute the DOA estimation algorithm to obtain local target angle estimation information, and then send this information along with their own position and attitude information to the central processing unit. The central processing unit is used to fuse information from each node to generate a global situational awareness, and to generate movement control commands and / or operating parameter adjustment commands for specific radar nodes based on the global situational awareness and task requirements. The communication network connects the central processing unit with each millimeter-wave radar node to achieve data and command transmission. The millimeter-wave radar nodes are mounted on a mobile platform, which may be a tracked robot, a wheeled robot, a drone, or an electric gimbal. The central processing unit is also used to automatically calculate and issue at least one of the following commands based on the global target distribution, task priority, and environmental constraints: a target movement position command for the radar node, an array baseline adjustment command, a radar transmit waveform adjustment command, and a receive gain adjustment command.

[0047] This invention addresses the problem that OMP only selects the most matching atom to add to the index set in each iteration, resulting in excessive computational complexity, slow convergence speed, and difficulty in meeting the implementation requirements of practical systems. It provides a DOA estimation method based on an improved OMP algorithm, which can effectively reduce the number of algorithm iterations and improve the convergence speed.

[0048] For an M-element uniform linear array, the element spacing is d, and it is assumed that all elements are isotropic; in the far field of the array, with the normal to the linear array axis as a reference... There are K narrowband point sources at (k = 1, 2, ..., K) that are incident with plane waves (wavelength λ). The spatial signal is sparsified by gridding, and the entire spatial range is uniformly divided into N parts at equal angles. Then, all possible directions can be represented as... (k = 1, 2, ..., N); Sparsify the spatial signal so that there are only K actual target signal sources in the directions, where N is much larger than the number of actual signal sources K; then all the directions in space correspond to a steering vector, which is: The data received by the array is represented as follows:

[0049]

[0050]

[0051]

[0052] In the formula: Let K be the spatial incident signal vector, and let K be the sparsity of K. N), There are only K non-zero elements in the array, and the remaining NK elements are zero elements. The position of the non-zero element is the target direction of the real signal source. It is Gaussian white noise with a mean of 0 and a variance of σ²; This is the array received signal vector;

[0053] Sparse representation-based DOA estimation refers to the estimation of the sparse coefficient vector given the observed signal vector X and the array manifold matrix A of the receiving array. Locate the positions corresponding to all non-zero elements; these non-zero positions correspond to the actual incident angles of the signal source. (where k = 1, 2, ..., N);

[0054] To locate sparse signal vectors To determine the location of non-zero elements, the corresponding valid columns (i.e., steering vectors related to the actual signal source direction) in the array manifold matrix A must be identified. The OMP algorithm employs a greedy strategy to achieve this: in each iteration, the algorithm selects the column in A that is most relevant to the current residual; subsequently, the component corresponding to that column is removed from the received signal, the residual is updated, and the index of that column is recorded; after K iterations (K being the preset number of signal sources), the algorithm obtains a set of selected column indices, which correspond to the columns in A involved in the linear representation. The effective steering vector; ultimately, by mapping these indices back to the angle set. (k = 1, 2, ..., N) can be used to determine the direction of arrival of the actual signal source;

[0055] The DOA estimation method based on the OMP algorithm is time-consuming. In order to improve the convergence speed of the algorithm, the atomic selection strategy of the algorithm is optimized.

[0056] Traditional beamforming (DBF) algorithms are the simplest and least computationally complex method for DOA estimation; however, their performance is limited by the physical aperture of the actual antenna array, i.e., constrained by the Rayleigh limit, making it difficult to effectively distinguish multiple signal sources located within the same beamwidth; for a uniform linear array (ULA), its theoretical angular resolution can be given by the following formula:

[0057] ;

[0058] Where N represents the number of array elements, d is the spacing between adjacent elements in the ULA, and θ is the incident angle of the target signal; the following formula gives the theoretical angular resolution of the radar in the normal direction:

[0059] ;

[0060] The following formula gives the distance between adjacent array elements in the radar normal direction. Under these conditions, the theoretical angular resolution that ULA can achieve is:

[0061] ;

[0062] Specifically, this algorithm references the DBF algorithm during the atom selection process, effectively improving the convergence speed, as follows:

[0063] During reconstruction, the improved OMP algorithm no longer selects only one atom to add to the index set in each iteration. Instead, it references the DBF algorithm during iteration, performs a spectral peak search on the calculation results after calculating the inner product of the residual and each column of the measurement matrix, and selects the largest K. i The atoms corresponding to each spectral peak are added to the index set, K i This represents the number of targets not yet found, but this could easily introduce incorrect atoms, so K was then... i Each atom is screened to retain the correct M. i One atom, eliminating erroneous atoms, when K i =M i The iteration stops when the time is right; compared to the OMP algorithm, the improved algorithm's search and selection method can improve the convergence speed and reduce computational complexity.

[0064] The embodiments of this application include two parts: simulation verification and measured data verification;

[0065] The feasibility of the proposed algorithm was verified by MATLAB simulation in the simulation experiment. In the simulation experiment, a uniform linear array with 8 array elements was used. The noise was Gaussian white noise with a signal-to-noise ratio of 5dB. The array element spacing of the uniform linear array was half a wavelength. The spatial signal was divided into 361 equal parts with an angle of 0.5° and the parameter λ was 0.3.

[0066] Simulation Experiment 1: Two signal sources are incident on a uniform linear array at incident angles of 0° and 30°. The proposed algorithm is used to estimate the DOA (Depth of Ability). The results are as follows: Figure 2 As shown;

[0067] Simulation Experiment 2: Two signal sources are incident on a uniform linear array at incident angles of 0° and 10°. The proposed algorithm is used to estimate the DOA (Depth of Ability). The results are as follows: Figure 3 As shown;

[0068] Simulation Experiment 3: Three signal sources are incident on a uniform linear array at incident angles of -60°, -30°, 0°, 10°, and 60°. The proposed algorithm is used to estimate the DOA (Depth of Ability). The results are as follows: Figure 4 As shown;

[0069] Simulation Experiment 4: Five signal sources are incident on a uniform linear array at incident angles of -60°, -30°, 0°, 30°, and 60°. The proposed algorithm is used for DOA estimation, and the results are as follows. Figure 5 As shown;

[0070] In simulation experiment 1, when the difference in DOA between the two incident signals is greater than the Rayleigh limit, the improved OMP algorithm finds the two target angles in just one iteration.

[0071] In simulation experiment 2, when the difference in DOA between the two incident signals is less than the Rayleigh limit, the improved OMP algorithm still requires two iterations to find the two target angles.

[0072] In simulation experiment 3, two of the DOA differences of the five incident signals were less than the Rayleigh limit. The improved OMP algorithm found four target angles with DOA differences greater than the Rayleigh limit in the first iteration and found the remaining target angle in the second iteration. A total of two iterations were required.

[0073] In simulation experiment 4, the DOA differences of the five incident signals were all greater than the Rayleigh limit. The improved OMP algorithm only needed one iteration to find the angles of the five targets.

[0074] It is evident that when the estimated target angle exceeds the Rayleigh limit, the improved OMP algorithm can effectively reduce the number of iterations and increase the number of convergences. Furthermore, the more targets that exceed the Rayleigh limit, the better the results.

[0075] In the field data verification, TI's IWR1443BOOST evaluation board and DCA1000EVM were used to collect and analyze data.

[0076] In the experiment, the radar was configured with 2 transmitters and 4 receivers, and TDMA mode was used to simulate 8 arrays;

[0077] The spatial signal is still divided into 361 equal-angled parts, each with an angle of 0.5°, and the parameter λ is 0.3, according to the equal-angle division method.

[0078] To minimize interference from other objects on the experimental results, this experiment was conducted in an open playground, such as... Figure 6 As shown;

[0079] In Experiment 1, two angle reflectors of equal height were placed at a distance of about 3 meters, with a distance of about 0.53 meters between them. The horizontal azimuth angle between one of the angle reflectors and the radar was kept as close as possible to 0°. The horizontal incident angle of the other angle reflector was calculated to be about 10°.

[0080] In Experiment 2, two angle reflectors of equal height were placed at a distance of about 3 meters, with a distance of about 1.09 meters between them. The horizontal azimuth angle between one of the angle reflectors and the radar was kept as close as possible to 0°. The horizontal incident angle of the other angle reflector was calculated to be about 20°.

[0081] The results of Experiment 1 show that the improved OMP algorithm requires two iterations to find the two target angles, while in Experiment 2, since the difference in DOA between the two incident signals is greater than the Rayleigh limit, the two target angles are found in just one iteration, which is consistent with the simulation results above and proves the effectiveness of the algorithm in practical applications.

[0082] This invention provides a specific scheme for applying the aforementioned improved OMP-DOA estimation algorithm to a sensing system;

[0083] The system includes a central processing unit, three (expandable) mobile millimeter-wave radar nodes, and a wireless / wired communication network connecting them; each millimeter-wave radar node integrates an antenna array, a radio frequency front-end, the signal processing module of this invention (for running the improved OMP-DOA algorithm), and a mobile platform (e.g., a robot chassis that can glide along a fixed track); the central processing unit has powerful computing capabilities and runs data fusion algorithms and decision control software.

[0084] The specific steps of the system's workflow are as follows:

[0085] Initial deployment and startup: Based on the default or previous task scenario, the system deploys three radar nodes at predetermined locations (e.g., the three vertices of a triangular region). Each node starts up and scans its assigned sector.

[0086] Distributed DOA estimation: Each radar node independently collects echo data and uses the improved OMP algorithm described in this invention to quickly calculate the target angle list within its field of view. Because this algorithm converges quickly and has high accuracy, it can rapidly provide reliable local perception results for upper-level decision-making. Simultaneously, the node reports its precise position obtained from its GPS / odometry and the gimbal attitude angle.

[0087] Central Fusion and Situation Assessment: The central processing unit receives all data. It first transforms the local angle information of each node, combined with the node's own pose, into a unified global coordinate system, generating a global target distribution map covering the entire monitoring area. Subsequently, the central unit assesses the current system performance, for example: Does area A have any blind spots due to building obstructions? Are there two targets in area B with angles very close together, exceeding the Rayleigh resolution limit of a single radar at that distance?

[0088] Intelligent Decision-Making and Command Generation: Based on the evaluation results and preset task priorities (e.g., ensuring the distinguishability of targets in area B is higher than scanning area A), the central unit makes a decision; it may calculate that moving Radar2 20 meters towards area B and adjusting its antenna beamwidth can significantly improve the angular resolution of area B. Simultaneously, it instructs Radar3 to change its scanning mode, focusing on covering area A; these decisions are translated into specific control commands and radar parameter settings.

[0089] Execution and Closed-Loop Feedback: After receiving instructions via the communication network, Radar2 and Radar3 automatically control the mobile platform to reach the new location and adjust the radar signal parameters. Upon completion, the system enters a new round of perception-estimation-fusion-decision loop.

[0090] Through the aforementioned closed loop, the system achieves adaptive adjustment. For example, in traffic monitoring, when a sudden gathering of vehicles (dense target) is detected at an intersection, the system can automatically dispatch nearby radars to adjust their positions and beams to monitor the intersection with more precise angular resolution and avoid misjudgment. In security scenarios, when a radar detects an abnormal moving target, the system can instruct adjacent radars to coordinate their turn to cross-locate and continuously track the target.

[0091] The improved OMP-DOA estimation algorithm provided by this invention is a key enabling technology for building such intelligent adaptive radar networks due to its high efficiency and reliability. Only when the underlying perception algorithm is fast and accurate enough can the upper-layer dynamic deployment and resource allocation strategies play a timely and effective role.

Claims

1. A DOA estimation algorithm based on an improved OMP algorithm, characterized in that, Includes the following steps: S1. Initialize residuals, index set, iteration count, and atomic sieving parameters; S2. Based on the correlation between the current residual and the measurement matrix, perform spectral peak search and select multiple potential atoms in batches to add to the temporary index set; S3. According to the preset atom filtering parameters, perform threshold filtering on the atoms in the temporary index set, remove erroneous atoms, retain correct atoms, and update the main index set; S4. Based on the updated master index set, reconstruct the signal and update the residuals using least squares estimation; S5. Determine if the iteration stopping condition is met. If it is met, output the DOA estimation result; otherwise, return to step S2 to continue the iteration.

2. The DOA estimation algorithm based on the improved OMP algorithm according to claim 1, characterized in that, In S1, the initialization specifically involves setting the initial residual. The observed signal vector Initial primary index set The correct number of atoms found Number of iterations The input parameters include the measurement matrix A, the observed signal X, the signal sparsity K, and the atom sieving parameter λ, where... .

3. The DOA estimation algorithm based on the improved OMP algorithm according to claim 2, characterized in that, In S2, the specific method for batch selecting multiple potential atoms is as follows: calculate the current residual. The inner product modulus of each column of the measurement matrix A is used to obtain an initial correlation vector; a spectral peak search is performed on the initial correlation vector, and the atomic indices corresponding to the largest Ki spectral peaks are selected to form a temporary index set Ti, where... , This represents the number of correct atoms found after the (i-1)th iteration.

4. The DOA estimation algorithm based on the improved OMP algorithm according to claim 1, characterized in that, In S3, the specific method for threshold filtering is as follows: calculate the maximum value of all spectral peaks in the temporary index set Ti. Set the filtering threshold ; Traverse Ti and retain only the atom indices whose spectral peaks are greater than η to form a finely selected set of atom indices. .

5. The DOA estimation algorithm based on the improved OMP algorithm according to claim 4, characterized in that, The atomic screening parameters The range of values ​​is .

6. The DOA estimation algorithm based on the improved OMP algorithm according to claim 4, characterized in that, In S3, the specific method for updating the primary index set is as follows: The refined atomic index set... Merge into the primary index set, i.e. And update the number of the correct atoms found. ,in Represents a set The number of elements; the specific method of S4 is: based on the current primary index set Extract the corresponding columns from measurement matrix A to form a submatrix. Solving for sparse signals using least squares estimation: Update residuals: In S5, the iteration stopping condition is: the number of correct atoms Mi found is equal to the preset signal sparsity K; in S5, the specific method for outputting the DOA estimation result is: mapping the atom indices in the final master index set to the pre-divided spatial angle grid to obtain the corresponding direction of arrival angle estimation value.

7. An application of a DOA estimation algorithm based on an improved OMP algorithm, applicable to the DOA estimation algorithm based on an improved OMP algorithm as described in any one of claims 1-6, characterized in that, The application is set in a perception system, which includes a central processing unit, at least two movable millimeter-wave radar nodes, and a communication network. The millimeter-wave radar nodes are used to collect echo data and execute the DOA estimation algorithm to obtain local target angle estimation information, and send the information and their own position and attitude information to the central processing unit. The central processing unit is used to fuse information from each node to generate a global situation, and to generate movement control commands and / or operating parameter adjustment commands for specific radar nodes based on the global situation and mission requirements. The communication network is used to connect the central processing unit with each of the millimeter-wave radar nodes to realize data and command transmission; the millimeter-wave radar nodes are installed on a mobile platform, which is a tracked robot, a wheeled robot, a drone, or an electric gimbal; the central processing unit is also used to automatically calculate and issue at least one of the following commands based on the global target distribution, task priority, and environmental constraints: target movement position command for radar nodes, array baseline adjustment command, radar transmission waveform adjustment command, and receiver gain adjustment command.

Images