A DOA estimation method based on K-Means clustering with minimized cross entropy
Through K-Means clustering and minimizing cross entropy function selection, and the support set is constructed in combination with the orthogonal least squares criterion, the existing DOA estimation method has solved the problem of low accuracy and insufficient noise robustness under low signal-to-noise ratio. , achieving higher DOA estimation accuracy and robustness.
Patent Information
- Application Number
- CN202311472200.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-11-07
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing DOA estimation method has low accuracy under low signal-to-noise ratio and is not very robust to noise, making it difficult to avoid problems in the local optimal solution.
The minimized cross-entropy DOA estimation method based on K-Means clustering is adopted to select atoms through the cross-entropy function to reduce the noise impact, and to construct a support set using the orthogonal least squares criterion to avoid local optimal solutions.
It improves the accuracy and robustness of DOA estimation, and more accurately estimates the incoming wave signal angle at low signal-to-noise ratio, reduces the impact of noise on atomic selection, and improves the accuracy and convergence accuracy of support set construction.
Smart Images

Figure CN117518081B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of direction of arrival estimation, and in particular to a cross entropy minimization DOA estimation method based on K-Means clustering. Background Art
[0002] As one of the important array signal processing technologies, DOA estimation is widely used in radio communications, radar navigation, electronic countermeasures and other fields. DOA estimation mainly estimates the direction of arrival of the incident signal based on the phase difference relationship between the received data of each array element in the array in space. Traditional multiple signal classification algorithms based on subspace decomposition and signal parameter rotation invariance algorithms require multiple snapshots of data to ensure the accuracy of the estimation. However, the estimation performance of coherent signals by traditional methods is severely reduced and it is difficult to meet the demand.
[0003] Due to the sparsity of the spatial target itself, sparse recovery technology is currently also used to construct the DOA estimation model. Compressed sensing, as a newly proposed sparse sampling theory, can effectively restore sparse signals under undersampling conditions. Among them, the greedy algorithm represented by the Orthogonal Matching Pursuit (OMP) algorithm has attracted widespread attention due to its low complexity and relatively excellent reconstruction performance. This method finds atoms that provide large energy in an overcomplete dictionary through multiple iterations, and then estimates the direction of the outgoing wave signal based on the position of the atom in the spatial grid division. Although the greedy algorithm can effectively estimate the outgoing wave direction, its estimated reconstruction rate is limited by the Restricted Isometry Property (RIP). Therefore, when the number of array elements in the array decreases, the accuracy of DOA estimation also decreases. In addition, the greedy algorithm is greatly affected by noise, so the error in estimating the incoming wave direction is large under low signal-to-noise ratio conditions. At present, machine learning methods have been introduced into greedy algorithms, which makes them not restricted by RIP conditions. However, this type of method has the characteristics of fast convergence, so it is easy to fall into the local optimal solution under low signal-to-noise ratio conditions, resulting in low DOA estimation accuracy. Summary of the invention
[0004] The purpose of the present invention is to solve the problem of low accuracy of existing DOA estimation, and propose a DOA estimation method based on K-Means clustering and minimizing cross entropy.
[0005] A method for minimizing cross entropy DOA estimation based on K-Means clustering. The specific process is as follows:
[0006] Step 1: Obtain the signal Y received by the array element in the uniform linear array;
[0007] Step 2: Set the number of repetitions i and initialize i to i=1;
[0008] Step 3: Use the received signal Y obtained in step 1 to obtain the i-th repeated minimized cross entropy atom selection range position set
[0009] Step 4: Construct the orthogonal least squares support set position set for the i-th repetition according to the orthogonal least squares criterion
[0010] Step 5: Set the number of iterations k and initialize k to k=1;
[0011] Step 6: Select the range of positions of atoms based on the minimum cross entropy obtained in step 3 and the orthogonal least squares support set position set obtained in step 4 for the i-th repetition Construct a candidate support set position set, and use the candidate support set position set to obtain the excellent support set position set for the i-th repetition k-th iteration
[0012] Where m = 1, 2, ..., N f , N f is the number of excellent support sets, is the set of positions of the mth excellent support set in the kth iteration;
[0013] Step 7: Determine whether the current number of iterations k reaches the preset maximum number of iterations k max , if k = k max Then use the i-th repetition of the k-th max The probability distribution of dictionary atoms at iteration Get the optimal support set position set Then execute step 8, if k < k max Then use the obtained value in step 6 Get the probability distribution of atoms in the overcomplete dictionary at the i-th repetition k+1 iteration And set k=k+1, and then return to step 6;
[0014] Step 8: Determine whether to stop repeating according to the current number of repetitions i. If so, execute step 9; otherwise, set i=i+1 and return to step 3.
[0015] Step 9: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal And estimate the angle of the incoming signal
[0016] Furthermore, the step 1 of obtaining the signal Y received by the array element in the uniform linear array is specifically:
[0017] Y=Ax+n,
[0018]
[0019]
[0020] Among them, Y is the array element receiving signal, A is the overcomplete dictionary, For noise, is the incoming wave signal, L is the number of incoming wave signals, is the nth element in the overcomplete dictionary, represents the nth division angle of the uniform space. Each division angle in the uniform space represents an atom. λ is the wavelength of the received signal. [·] T represents transpose, j is the imaginary unit, M is the number of array elements, and d is the element spacing.
[0021] Furthermore, the step 1 of obtaining the signal Y received by the array element in the uniform linear array is specifically:
[0022] Y=Ax+n,
[0023]
[0024]
[0025] Among them, Y is the array element receiving signal, A is the overcomplete dictionary, For noise, is the incoming wave signal, L is the number of incoming wave signals, is the nth element in the overcomplete dictionary, represents the nth division angle of the uniform space. Each division angle in the uniform space represents an atom. λ is the wavelength of the received signal. [·] T represents transpose, j is the imaginary unit, M is the number of array elements, and d is the element spacing.
[0026] Furthermore, the received signal Y obtained in step 1 is used in step 3 to obtain the i-th repeated minimized cross entropy atom selection range position set Specifically:
[0027] Step 3.1. Obtain the correlation coefficient γ of atoms in the complete dictionary A i (n), specifically:
[0028]
[0029] Among them, γ i (n) is the correlation coefficient of the nth atom in the overcomplete dictionary at the i-th repetition, A(:,n) is the nth atom in the overcomplete dictionary A, (·)H represents the conjugate transpose, and ||·||2 represents norm, is the residual of the i-1th repetition, i is the number of repetitions, and when i=1, the residual is the residual error at the first repetition;
[0030] Step 3.2: Use the K-Means clustering method to use the atomic correlation coefficient γ obtained in step 3.1 i (n) Divide the N atoms in the overcomplete dictionary A into P clusters, and obtain the optimal cluster among the P clusters, thereby obtaining the set of atomic correlation coefficients in the optimal cluster Specifically:
[0031]
[0032] in, represents the set of atomic correlation coefficients in the optimal cluster at the i-th repetition, p = 1, 2, ..., P, is the pth cluster divided, ∑γ i (n) represents the sum of the atomic correlation coefficients in the pth cluster, is the number of atoms in the pth cluster, N is the number of atoms in the overcomplete dictionary A, and P is the total number of partitioned clusters;
[0033] Step 33: Use the optimal cluster atomic correlation coefficient set from step 32 Get the set of positions for the atom selection range that minimizes cross entropy
[0034] Furthermore, the optimal cluster atomic correlation coefficient set in step 32 is used in step 33 Get the set of positions for the atom selection range that minimizes cross entropy Specifically:
[0035]
[0036] in, The set of positions for the range of atom selection that minimizes cross entropy, is the set of atomic positions of K-Means clustering at the i-th repetition, and Respectively The set of atomic positions adjacent to the left and right of the atomic position in . union(·) represents the union operation.
[0037] Furthermore, the orthogonal least squares support set position set for the i-th repetition is constructed according to the orthogonal least squares criterion in step 4. Specifically:
[0038]
[0039]
[0040] in, is the i-1th orthogonal least squares support set position set, Indicates that the i-th repetition is The positions of atoms are added to , Ψ represents the set of positions of all N atoms, Indicates that it is in Ψ but not in The atomic positions of Represents the set of positions in the orthogonal least squares support set Add the position of the nth atom to is an overcomplete dictionary A The support set of the position set, yes The orthogonal projection of , I is the unit matrix, express The pseudo-inverse.
[0041] Furthermore, the minimum cross entropy atom selection range position set Π obtained in step 3 in step 6 f and the orthogonal least squares support set position set obtained in step 4 for the i-th repetition Construct a candidate support set position set, and use the candidate support set position set to obtain the excellent support set position set for the i-th repetition k-th iteration Specifically:
[0042] Step 6.1: Based on the probability distribution of atoms in the overcomplete dictionary at the i-th repetition of the k-th iteration and minimize cross entropy atom selection range position set Generate N according to Bernoulli distribution c -1 candidate support set location set Specifically:
[0043] First, set the number of loops t and initialize t to t=1;
[0044] Then, the tth cycle is in The position set of the atoms selected in is used as the temporary position set π t , determine the temporary position support set π obtained in the tth cycle t Is the number of atoms in the card less than the number of incoming signals L? t )≤L, then the current temporary position support set is taken as the candidate support set, that is, If Card(π t )>L, then calculate the temporary position support set π for the t+1th cyclet+1 , until Card(π t )≤L to obtain the candidate support set;
[0045] Among them, π t+1 is the temporary position support set for the t+1th cycle, π t is the temporary position support set of the tth cycle, j′ is the number of the candidate support set, L is the number of incoming signals, is the j'th candidate support set position set at the kth iteration, Card(π t ) is π t The number of atoms in
[0046] Finally, repeat the above process N c -1 times, get N c -1 candidate support set location set
[0047] Step 62: Will As a candidate support set, add it to the N obtained in step 6. c -1 candidate support set location set In the c Candidate support set positions
[0048] Step 63: Get N c Candidate support set positions Corresponding error:
[0049]
[0050] in, is the corresponding error of the j'th candidate support set in the kth iteration, is the j'th candidate support set position set The corresponding support set is express of pseudo-rebellion;
[0051] Step 64: In N c The candidate support set position set corresponding error is obtained by f The candidate support set position set with the smallest error is taken as the excellent support set position set
[0052] Where m = 1, 2, ..., N f , N f is the number of excellent support sets, is the mth excellent support set position set in the kth iteration.
[0053] Furthermore, the temporary position support set π for the t+1th cycle t+1 , specifically:
[0054]
[0055] in, To satisfy the Bernoulli probability distribution The {0,1} value parameter, 0 means the current atom will not be selected, 1 means the current atom is a candidate atom.
[0056] Furthermore, the step 7 determines whether the current number of iterations k reaches the preset maximum number of iterations k. max , if k = k max Then use the i-th repetition of the k-th max The probability distribution of dictionary atoms at iteration Get the optimal support set position set Then execute step 8, if k < k max Then use the obtained value in step 6 Get the probability distribution of atoms in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6, specifically:
[0057] Step 71. According to N f Excellent support set location set Build N f A vector of {0,1} values
[0058]
[0059] Among them, β m represents the mth {0,1} value vector, β m,n Represents β m The value of the nth element in;
[0060] Step 72: Determine whether the current number of iterations k is equal to k max , if k = k max Then execute step 74; otherwise, execute step 73;
[0061] Among them, k max is the preset maximum number of iterations;
[0062] Step 7.3: Obtain the atomic probability distribution in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6:
[0063]
[0064] in, is the probability of the nth atom in the overcomplete dictionary at the k+1th iteration;
[0065] Step 74: Repeat the kth step according to the i-th step max The probability distribution of dictionary atoms at iteration Select the L atomic position sets with the largest probability to form the optimal support set position set
[0066] Furthermore, in step eight, it is determined whether to stop the repetition according to the current number of repetitions i. If the repetition is to be stopped, step nine is executed. Otherwise, i=i+1 is set and the process returns to step three, which is specifically:
[0067] like Or i=L, then stop repeating and execute step 9, otherwise set i=i+1 and return to step 3;
[0068] in, To calculate the residual norm, and ε is the set stopping repetition standard.
[0069] Furthermore, the optimal support set position set in step nine is used Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal And estimate the angle of the incoming signal Specifically:
[0070] Step 91: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal
[0071]
[0072] in, Represents the reconstructed incoming signal The optimal support set position set in Elements in
[0073] Step 92: Use the optimal support set position set Estimate the angle of the incoming signal and obtain the estimated angle of the incoming signal
[0074]
[0075] in, To estimate the angle of the incoming signal The angle of the lth incoming wave signal, l=1,2,…,L, is the optimal support set position set The corresponding position of the lth atom in the overcomplete dictionary.
[0076] The beneficial effects of the present invention are:
[0077] The present invention incorporates the idea of machine learning into atom selection, and uses the cross entropy function as a standard for atom selection. The support set is constructed by minimizing the cross entropy, and the influence of noise on atom selection is reduced. At the same time, the present invention can overcome the problem that the existing DOA estimation method using compressed sensing reconstruction is not robust to noise, so that the present invention can more accurately estimate the angle of the incoming wave signal under low signal-to-noise ratio conditions, thereby improving the accuracy of DOA estimation. The present invention uses K-Means clustering to pre-process the atomic selection range for minimizing cross entropy, and by using the correlation between adjacent angles of the incoming wave signal, the selection range of subsequent atoms is greatly reduced, avoiding falling into the local optimal solution when minimizing the cross entropy, and improving the correctness of the support set construction. The present invention uses the support set constructed by the orthogonal square minimum criterion as one of the candidate support sets in each repetition, which can accelerate the convergence speed of the cross entropy function while improving the convergence accuracy, avoiding the cross entropy function from falling into the local optimal solution, so that the present invention has a higher DOA estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0078] Figure 1 It is a flow chart of the present invention;
[0079] Figure 2 Root mean square error diagram of DOA estimation under different signal-to-noise ratios of the present invention;
[0080] Figure 3 This is a diagram showing the success rate of DOA estimation under different signal-to-noise ratios according to the present invention. DETAILED DESCRIPTION
[0081] The present invention incorporates the idea of machine learning into linear array DOA estimation, classifies atoms through the K-Means clustering algorithm, thereby reducing the selection range of atoms and improving the estimation accuracy of the algorithm. At the same time, the cross entropy function is used as the atom selection criterion, and the distribution probability of atoms in the dictionary is updated by minimizing the cross entropy function through multiple iterations, and then atoms with higher angle accuracy with the incoming wave signal are selected to join the support set, effectively reducing the impact of noise on the accuracy of atom selection. In addition, the invention can achieve faster convergence of the cross entropy function by adding the support set constructed by the least squares criterion to the candidate support set, while avoiding falling into the local optimal solution when minimizing the cross entropy function. The present invention is described below in conjunction with specific embodiments.
[0082] Specific implementation method 1: Figure 1 As shown, the specific process of the method for minimizing cross entropy DOA estimation based on K-Means clustering in this embodiment is as follows:
[0083] Step 1: Obtain the signal Y received by the array element in a uniform linear array with M array elements and d array element spacing:
[0084] Y=Ax+n,
[0085]
[0086]
[0087] in, is the signal received by the array element, A is an overcomplete dictionary, For noise, There are L far-field narrowband incoming wave signals with values not equal to 0, L is the number of incoming wave signals, is the nth element in the overcomplete dictionary, represents the nth division angle of the uniform space. Each division angle in the uniform space represents an atom. λ is the wavelength of the received signal. [·] T In the present invention, Y, L, and A are known, and j is an imaginary unit.
[0088] Step 2: Set the number of repetitions i and initialize i to i=1;
[0089] Step 3: Obtain the position set of the atomic selection range that minimizes the cross entropy for the i-th repetition through the K-Means clustering algorithm The following steps are involved:
[0090] Step 3.1. Obtain the correlation coefficient γ of atoms in the overcomplete dictionary i (n):
[0091]
[0092] Among them, γ i (n) is the correlation coefficient of the nth atom in the overcomplete dictionary at the i-th repetition, A(:,n) is the nth atom in the overcomplete dictionary A, (·) H represents the conjugate transpose, and ||·||2 represents norm, is the residual of the i-1th repetition, i is the number of repetitions, and when i=1, the residual in is the residual error at the first repetition, and Y is the signal received by the array element.
[0093] Step 3.2: Correlation coefficient γ of the atom obtained according to step 3.1 i (n) Use K-Means clustering method to divide N atoms in the overcomplete dictionary into P clusters, and obtain the best cluster among P clusters, so as to obtain the set of atomic correlation coefficients in the best cluster.
[0094]
[0095] in, represents the set of atomic correlation coefficients in the optimal cluster at the i-th repetition, p = 1, 2, ..., P, is the pth cluster divided, ∑γ i (n) represents the sum of the atomic correlation coefficients in the pth cluster, is the number of atoms in the pth cluster, N is the number of atoms in the overcomplete dictionary A, and P is the number of clusters partitioned by K-Means clustering;
[0096] Step 33: Use the optimal cluster atomic correlation coefficient set from step 32 Get the set of positions for the atom selection range that minimizes cross entropy for:
[0097]
[0098]
[0099] in, The set of positions for the range of atom selection that minimizes cross entropy, is the set of atomic positions of K-Means clustering at the i-th repetition, which is composed of the set of atomic correlation coefficients in the optimal cluster The atomic positions in the and Respectively The set of atomic positions adjacent to the left and right of the atomic position in . union(·) represents the union operation.
[0100] Step 4: Construct the orthogonal least squares support set position set for the i-th repetition according to the orthogonal least squares criterion Specifically:
[0101]
[0102]
[0103] in, is the i-1th orthogonal least squares support set position set, Indicates that the i-th repetition is Add the position of the atom in is the orthogonal least squares support set position set for the i-th repetition, Ψ represents the position set of all N atoms, Indicates that it is in Ψ but not in The atomic positions of Represents the set of positions in the orthogonal least squares support set Add the position of the nth atom to It is A The support set of the position set, yes The orthogonal projection of , I is the unit matrix, express is the pseudo-inverse of , and Y is the signal received by the array element.
[0104] Step 5: Set the number of iterations k and initialize k to k=1.
[0105] Step 6: Select the range of positions of atoms based on the minimum cross entropy obtained in step 3 and the orthogonal least squares support set position set of the i-th repetition obtained in step 4 Construct a set of candidate support set positions, and use the set of candidate support set positions to obtain an excellent support set position set
[0106] Step 6.1: Based on the probability distribution of atoms in the overcomplete dictionary at the i-th repetition of the k-th iteration and minimize cross entropy atom selection range position set Generate N according to Bernoulli distribution c -1 candidate support set location set
[0107] First, set the number of loops t and initialize t to t=1;
[0108] Then, the tth cycle is in The position set of the atoms selected in is used as the temporary position set π t , determine the temporary position support set π obtained in the tth cycle t Is the number of atoms in the card less than the number of incoming signals L? t )≤L, the current temporary position support set is taken as the candidate support set, that is, If Card(π t )>L, then calculate the temporary position support set π for the t+1th cycle t+1 Until Card(π t )≤L to obtain the candidate support set:
[0109]
[0110] Among them, π t+1 is the temporary position support set for the t+1th cycle, π t is the temporary position support set for the tth cycle, To satisfy the Bernoulli probability distribution The {0,1} value parameter, 0 means the current atom will not be selected, 1 means the current atom may be selected, k is the number of iterations, N c is the total number of candidate support sets, j′ is the index of the candidate support set, L is the number of incoming signals, is the j'th candidate support set position set at the kth iteration;
[0111] Finally, repeat the above generation process to obtain N c -1 candidate support set location set
[0112] Step 62: Will As a candidate support set, add it to the N obtained in step 6. c -1 candidate support set location set In the c Candidate support set positions
[0113] Step 63: Get N c The corresponding error of the candidate support set position set is:
[0114]
[0115] in, is the corresponding error of the j'th candidate support set in the kth iteration, is the j'th candidate support set position set The corresponding support set is express The pseudo-inverse.
[0116] Step 64: In N c The candidate support set position set corresponding error is obtained by f The candidate support set position set with the smallest error is taken as the excellent support set position set
[0117] Where m = 1, 2, ..., N f , N f is the number of excellent support sets, is the mth excellent support set position set in the kth iteration.
[0118] Step 7: Determine whether the current number of iterations k reaches the maximum number of iterations k max , if k = k max Then use the i-th repetition of the k-th max The probability distribution of dictionary atoms at iteration Get the optimal support position set Then execute step 8, if k < k maxThen use the obtained value in step 6 Get the probability distribution of atoms in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6, including the following steps:
[0119] Step 71. According to N f Excellent support set location set Build N f A vector of {0,1} values
[0120]
[0121] Among them, β m represents the mth {0,1} value vector, β m,n Represents β m The value of the nth element in the set when the position of the nth atom is in the excellent support set position set China Times Beta m,n is 1 if the value is true, otherwise it is 0.
[0122] Step 72: Determine whether the current number of iterations k is equal to k max , if k = k max Then execute step 74; otherwise, execute step 73;
[0123] Among them, k max is the preset maximum number of iterations;
[0124] Step 7.3: Obtain the atomic probability distribution in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6:
[0125]
[0126] in, is the probability of the nth atom in the overcomplete dictionary at the k+1th iteration.
[0127] Step 74: Repeat the kth step according to the i-th step max The probability distribution of dictionary atoms at iteration Select the L atomic position sets with the highest probability as the optimal support set position set
[0128] When the number of iterations k = k max The i-th iteration ends at , according to the probability distribution of atoms in the overcomplete dictionary at this time Select the atomic position sets corresponding to the L probabilities with the largest probabilities to form the optimal support set position set
[0129] Step 8: Determine whether to stop repeating according to the current number of repetitions i. If so, execute step 9; otherwise, set i=i+1 and return to step 2.
[0130] like Or i=L, then stop repeating and execute step 9, otherwise set i=i+1 and return to step 3;
[0131] in, To calculate the residual norm, and ε is the set stopping repetition standard.
[0132] Step 9: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal And estimate the angle of the incoming signal The following steps are involved:
[0133] Step 91: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal
[0134]
[0135] in, Represents the reconstructed incoming signal The optimal support set position set in The elements in .
[0136] Step 92: Use the optimal support set position set Estimate the angle of the incoming signal and obtain the estimated angle of the incoming signal
[0137]
[0138] in, To estimate the angle of the incoming signal The angle of the lth incoming wave signal, l=1,2,…,L, is the optimal support set position set The corresponding position of the lth atom in the overcomplete dictionary.
[0139] Embodiment: In order to verify the beneficial effects of the present invention, the following simulation experiments were carried out:
[0140] The DOA estimation root mean square error and DOA estimation success rate of the DOA estimation method based on K-Means clustering and minimizing cross entropy of the present invention are experimentally simulated, and the experimental parameters are selected as follows:
[0141] The root mean square error is defined as:
[0142]
[0143] Among them, N T represents the Monte Carlo order, For nth T The present invention estimates the lth source angle through a Monte Carlo experiment.
[0144] The DOA estimation success rate is defined as:
[0145]
[0146] Among them, F(θ l ) indicates N T The number of times the lth source angle is estimated correctly within the allowable error range in the Monte Carlo method.
[0147] The experimental simulation setting allows an error range of 1°. The receiving uniform linear array has an element spacing of d = λ / 2, the number of elements is set to M = 16, the number of sources is L = 3, and the incident angle is θ = [-30°, 20°, 40°]. The (-90°, 90°) space is evenly divided into N = 181 parts, each with a spacing of 1°. The iteration parameter is set to k max =5,N c =2N f =30, the stopping criterion is ε=10 -3 .
[0148] Experiment 1: The root mean square error of DOA estimation under different signal-to-noise ratios of the present invention is tested. The signal-to-noise ratio range is set to -10dB to 4dB, and the step size is 2dB. The simulation results are as follows: Figure 2 As shown. When the signal-to-noise ratio is less than 0 dB, the present invention can maintain a low root mean square error, indicating that the present invention can estimate DOA under low signal-to-noise ratio conditions. At the same time, the mean square error of the present invention is less than 1° under each signal-to-noise ratio, indicating that the present invention has strong robustness and can maintain excellent estimation performance under different noise intensities.
[0149] Experiment 2: The DOA estimation success rate of the present invention under different signal-to-noise ratios is tested. The signal-to-noise ratio range is set to -20dB to 20dB, with a step size of 5dB. The simulation results are shown in Figure 3 As shown. When the signal-to-noise ratio is higher than 10dB, the present invention can fully estimate the angle of the incoming signal, which reflects the ability of the present invention to accurately estimate DOA under high signal-to-noise ratio. In particular, when the signal-to-noise ratio is less than 0dB, the estimation success rate of the present invention is close to 90%. This shows that the present invention has a higher estimation accuracy under low signal-to-noise ratio. At the same time, the DOA estimation success rate of the present invention is always maintained at around 90% under low signal-to-noise ratio, reflecting the strong robustness of the present invention.
Claims
1. A method for DOA estimation based on K-Means clustering with minimized cross entropy, characterized in that The specific process of the method is: Step 1: Obtain the signal Y received by the array element in the uniform linear array; Step 2: Set the number of repetitions i and initialize i to i=1; Step 3: Use the received signal Y obtained in step 1 to obtain the i-th repeated minimized cross entropy atom selection range position set Step 4: Construct the orthogonal least squares support set position set for the i-th repetition according to the orthogonal least squares criterion Step 5: Set the number of iterations k and initialize k to k=1; Step 6: Select the range of positions of atoms based on the minimum cross entropy obtained in step 3 and the orthogonal least squares support set position set obtained in step 4 for the i-th repetition Construct a candidate support set position set, and use the candidate support set position set to obtain the excellent support set position set for the i-th repetition k-th iteration Where m = 1, 2, ..., N f , N f is the number of excellent support sets, is the set of positions of the mth excellent support set in the kth iteration; Step 7: Determine whether the current number of iterations k reaches the preset maximum number of iterations k max , if k = k max Then use the i-th repetition of the k-th max The probability distribution of dictionary atoms at iteration Get the optimal support set position set Then execute step 8, if k < k max Then use the obtained value in step 6 Get the probability distribution of atoms in the overcomplete dictionary at the i-th repetition k+1 iteration And set k=k+1, and then return to step 6; Step 8: Determine whether to stop repeating according to the current number of repetitions i. If so, execute step 9; otherwise, set i=i+1 and return to step 3. Step 9: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal And estimate the angle of the incoming signal 2. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 1, characterized in that: The step 1 of obtaining the signal Y received by the array element in the uniform linear array is specifically as follows: Y=Ax+n, Among them, Y is the array element receiving signal, A is the overcomplete dictionary, For noise, is the incoming wave signal, L is the number of incoming wave signals, is the nth element in the overcomplete dictionary, represents the nth division angle of the uniform space. Each division angle in the uniform space represents an atom. λ is the wavelength of the received signal. [·] T represents transpose, j is the imaginary unit, M is the number of array elements, and d is the element spacing.
3. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 2, characterized in that: In step 3, the received signal Y obtained in step 1 is used to obtain the i-th repeated minimized cross entropy atom selection range position set Specifically: Step 31: Get the correlation coefficient γ of atoms in the complete dictionary A i (n), specifically: Among them, γ i (n) is the correlation coefficient of the nth atom in the overcomplete dictionary at the i-th repetition, A(:,n) is the nth atom in the overcomplete dictionary A, (·) H represents the conjugate transpose, and ||·||2 represents norm, is the residual of the i-1th repetition, i is the number of repetitions, and when i=1, the residual is the residual error at the first repetition; Step 3.2: Use the K-Means clustering method to use the atomic correlation coefficient γ obtained in step 3.1 i (n) Divide the N atoms in the overcomplete dictionary A into P clusters, and obtain the optimal cluster among the P clusters, thereby obtaining the set of atomic correlation coefficients in the optimal cluster Specifically: in, represents the set of atomic correlation coefficients in the optimal cluster at the i-th repetition, p = 1, 2, ..., P, is the pth cluster divided, ∑γ i (n) represents the sum of the atomic correlation coefficients in the pth cluster, is the number of atoms in the pth cluster, N is the number of atoms in the overcomplete dictionary A, and P is the total number of partitioned clusters; Step 33: Use the optimal cluster atomic correlation coefficient set from step 32 Get the set of positions for the atom selection range that minimizes cross entropy 4. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 3, characterized in that: The optimal cluster atomic correlation coefficient set in step 32 is used in step 33 Get the set of positions for the atom selection range that minimizes cross entropy Specifically: in, The set of positions for the range of atom selection that minimizes cross entropy, is the set of atomic positions of K-Means clustering at the i-th repetition, and Respectively The set of atomic positions adjacent to the left and right of the atomic position in . union(·) represents the union operation.
5. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 4, characterized in that: The orthogonal least squares support set position set for the i-th repetition is constructed according to the orthogonal least squares criterion in step 4. Specifically: in, is the i-1th orthogonal least squares support set position set, Indicates that the i-th repetition is The positions of atoms are added to , Ψ represents the set of positions of all N atoms, Indicates that it is in Ψ but not in The atomic positions of Represents the set of positions in the orthogonal least squares support set Add the position of the nth atom to is an overcomplete dictionary A The support set of the position set, yes The orthogonal projection of , I is the unit matrix, express The pseudo-inverse.
6. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 5, characterized in that: The minimum cross entropy atom selection range position set Π obtained in step 3 in step 6 f and the orthogonal least squares support set position set obtained in step 4 for the i-th repetition Construct a candidate support set position set, and use the candidate support set position set to obtain the excellent support set position set for the i-th repetition k-th iteration Specifically: Step 6.1: Based on the probability distribution of atoms in the overcomplete dictionary at the i-th repetition of the k-th iteration and minimize cross entropy atom selection range position set Generate N according to Bernoulli distribution c -1 candidate support set location set Specifically: First, set the number of loops t and initialize t to t=1; Then, the tth cycle is in The position set of the atoms selected in is used as the temporary position set π t , determine the temporary position support set π obtained in the tth cycle t Is the number of atoms in the card less than the number of incoming signals L? t )≤L, then the current temporary position support set is taken as the candidate support set, that is, If Card(π t )>L, then calculate the temporary position support set π for the t+1th cycle t+1 , until Card(π t )≤L to obtain the candidate support set; Among them, π t+1 is the temporary position support set for the t+1th cycle, π t is the temporary position support set of the tth cycle, j′ is the number of the candidate support set, L is the number of incoming signals, is the j'th candidate support set position set at the kth iteration, Card(π t ) is π t The number of atoms in Finally, repeat the above process N c -1 times, get N c -1 candidate support set location set Step 62: Will As a candidate support set, add it to the N obtained in step 6. c -1 candidate support set location set In the c Candidate support set positions Step 63: Get N c Candidate support set positions Corresponding error: in, is the corresponding error of the j'th candidate support set in the kth iteration, is the j'th candidate support set position set The corresponding support set is express of pseudo-rebellion; Step 64: In N c The candidate support set position set corresponding error is obtained by f The candidate support set position set with the smallest error is taken as the excellent support set position set Where m = 1, 2, ..., N f , N f is the number of excellent support sets, is the mth excellent support set position set in the kth iteration.
7. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 6, characterized in that: The temporary position support set π for the t+1th cycle t+1 , specifically: in, To satisfy the Bernoulli probability distribution The {0,1} value parameter, 0 means the current atom will not be selected, 1 means the current atom is a candidate atom.
8. The method for DOA estimation based on K-Means clustering by minimizing cross entropy according to claim 7, characterized in that: The step 7 of determining whether the current number of iterations k reaches the preset maximum number of iterations k max , if k = k max Then use the i-th repetition of the k-th max The probability distribution of dictionary atoms at iteration Get the optimal support set position set Then execute step 8, if k < k max Then use the obtained value in step 6 Get the probability distribution of atoms in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6, specifically: Step 71. According to N f Excellent support set location set Build N f A vector of {0,1} values Among them, β m represents the mth {0,1} value vector, β m,n Represents β m The value of the nth element in; Step 72: Determine whether the current number of iterations k is equal to k max , if k = k max Then execute step 74; otherwise, execute step 73; Among them, k max is the preset maximum number of iterations; Step 7.3: Obtain the atomic probability distribution in the overcomplete dictionary at the i-th repetition k+1 iteration And let k=k+1, then return to step 6: in, is the probability of the nth atom in the overcomplete dictionary at the k+1th iteration; Step 74: Repeat the kth time according to the i-th time max The probability distribution of dictionary atoms at iteration Select the L atomic position sets with the largest probability to form the optimal support set position set 9. The method for minimizing cross entropy DOA estimation based on K-Means clustering according to claim 8, characterized in that: In step 8, it is determined whether to stop repeating according to the current number of repetitions i. If the repetition is to be stopped, step 9 is executed. Otherwise, i=i+1 is set and step 3 is returned, which is specifically: like Or i=L, then stop repeating and execute step 9, otherwise set i=i+1 and return to step 3; in, To calculate the residual norm, ε is the set stopping repetition standard.
10. The method for DOA estimation based on K-Means clustering and minimizing cross entropy according to claim 9, characterized in that: The optimal support set position set in step nine is used Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal And estimate the angle of the incoming signal Specifically: Step 91: Use the optimal support set position set Reconstruct the incoming wave signal to obtain the reconstructed incoming wave signal in, Represents the reconstructed incoming signal The optimal support set position set in Elements in Step 92: Use the optimal support set position set Estimate the angle of the incoming signal and obtain the estimated angle of the incoming signal in, To estimate the angle of the incoming signal The angle of the lth incoming wave signal, l=1,2,…,L, is the optimal support set position set The corresponding position of the lth atom in the overcomplete dictionary.
Citation Information
Patent Citations
Direction of arrival angle optimization method combining MMP algorithm and combined tree pruning
CN114254525A
Broadband signal DOA estimation method based on Dirichlet process prior
CN115343673A
