Source arrival estimation method and system based on minimum variance distortion-free response

Abstract

The invention belongs to the field of array signal processing, and relates to a method and system for estimating a source arrival based on a minimum variance distortion-free response. The scheme comprises: establishing a source arrival estimation model, and judging a source arrival direction by calculating the output power of a sensor antenna array at each scanning angle; converting the source arrival estimation model into a constrained minimization problem according to a minimum variance distortion-free response criterion, and the constrained minimization problem is specifically as follows: adjusting the sensor antenna weight to maximize the output power gain of the source arrival angle and suppressing the output power gain of other interference angles; in solving the constrained minimization problem, using a recursive least squares method as a framework, and calculating the product of an inverse matrix of a snapshot covariance matrix in the output power of the antenna array at each angle and a scanning steering vector by a bisection coordinate descent method, and obtaining the output power of the antenna array of the arrival signal at each angle, and estimating the source arrival direction by using the output power spectrum peak value, with low complexity and good estimation performance.

Description

Technical Field

[0001] The invention belongs to the technical field of array signal processing, and in particular relates to a source arrival estimation method and system based on minimum variance distortion-free response. Background Art

[0002] Direction of Arrival (DOA) estimation is an important branch of array signal processing and is widely used in radar, sonar, detection, and medical engineering. The steps to determine the direction of arrival are as follows: the receiving antenna array performs beam scanning within a certain range of angles, and then calculates the signal power received at each angle. Finally, the angle corresponding to the maximum signal power is determined as the direction of arrival of the source. Among the many adaptive direction of arrival estimation algorithms, the Minimum Variance Distortionless Response (MVDR) algorithm is the most representative one. It has good performance and is easy to implement. However, the standard MVDR algorithm requires a matrix inversion operation, and the direct matrix inversion operation requires O(M 3 ) operations, where M is the number of sensor array antennas. When the number of array antennas is large, the complexity is very high and is not suitable for hardware implementation. Therefore, how to reduce the complexity of the MVDR inversion operation is a challenge.

[0003] In recent years, many scholars have proposed a variety of adaptive direction of arrival estimation algorithms to reduce the complexity of the MVDR algorithm inversion. For example, the least mean square (LMS) algorithm is used to implement MVDR, and its computational complexity is O(M). However, the convergence speed of LMS is very slow. When the recursive least squares (RLS) algorithm is used to implement MVDR, it can provide a good compromise between convergence speed and complexity. Many researchers are committed to converting the MVDR problem into an RLS minimization problem. However, the RLS algorithm decays exponentially quickly when the initial value is regularized. The diagonal loading technique can enhance the robustness of the RLS algorithm by adding a suitable value to the diagonal elements of the matrix, but this additional regularization operation requires O(M) per matrix inversion operation. 3 ) operands, resulting in poor practical operability. Summary of the invention

[0004] To this end, the present invention provides a source arrival estimation method and system based on minimum variance distortion-free response, which has simple operation, low complexity, good estimation performance, and is convenient for application in actual scenarios.

[0005] According to the design scheme provided by the present invention, a source arrival estimation method based on minimum variance distortion-free response includes the following contents:

[0006] A signal source arrival estimation model is established. The signal source arrival estimation model obtains a scanning steering vector by discretely scanning the sensor antenna array in a certain angle range, and calculates the antenna array output power at each angle in combination with the received information to estimate the signal source arrival direction.

[0007] According to the minimum variance distortion-free response criterion, the signal source arrival estimation model is converted into a constrained minimization problem. The constrained minimization problem is as follows: the sensor antenna weights are adjusted to maximize the output power gain of the signal source arrival angle and suppress the output power gain of other interference angles;

[0008] In solving the constrained minimization problem, the recursive least squares method is used as the framework, and the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector in the antenna array output power at each angle is calculated by the bisection coordinate descent method to obtain the antenna array output power of the wave arrival signal at each angle;

[0009] According to the calculated antenna array output power of the wave arrival signal at each angle, the output power spectrum peak is used to estimate the source wave arrival direction.

[0010] As a source arrival estimation method based on minimum variance distortion-free response of the present invention, further, in the source arrival estimation model, a sensor antenna receiving array is used to perform beam scanning within a preset area, the antenna array output power of each scanning angle is calculated, and the angle corresponding to the maximum output power is determined as the source arrival direction.

[0011] As the source arrival estimation method based on minimum variance distortion-free response of the present invention, further, in the beam scanning of the sensor antenna receiving array, assuming that N narrowband signals are received within the observation period of Q snapshots, the steering vector matrix A(θ n ) is used to represent the response of the sensor antenna receiving array to a narrowband signal source with unit energy at the i-th snapshot, where i∈{1,2,3,…,Q}.

[0012] As the source arrival estimation method based on the minimum variance distortion-free response of the present invention, further, the observation signal of the sensor antenna receiving array at the i-th snapshot is expressed as: r(i)=A(θ n )·s(i)+n(i), where, A(θ n ) represents the steering vector matrix, and A(θ n )=[A0(θ1),A0(θ2),...,A0(θ N )],A0(θ N ) represents the Nth source angle θ N, s(i) represents a group of unrelated arrival signals sent by multiple sources at the i-th snapshot moment, and n(i) represents Gaussian random vector noise.

[0013] As the source arrival estimation method based on minimum variance distortion-free response of the present invention, further, for the nth source angle θ n , the constrained minimization problem is expressed as: And satisfy w H (i)A(θ n )=1, where R(i) is the M×M-dimensional snapshot covariance matrix, and w(i) is a complex weight vector of dimension M×1, K is the number of training snapshots, A(θ n ) represents the steering vector matrix, and r(k) represents the observation signal of the antenna receiving array.

[0014] As the source arrival estimation method based on the minimum variance distortion-free response of the present invention, further, in solving the constrained minimization problem, it is assumed that the scanning steering vector A s (θ p ) is stable during the DOA estimation, p = 1, 2, ..., P, P is the number of scanning steps, then each scanning angle θ p The output power is expressed as Among them, v θ (i) = R -1 (i)A s (θ p ).

[0015] As a source arrival estimation method based on minimum variance distortion-free response of the present invention, further, in the process of output power calculation, the snapshot covariance matrix R(i) is regularized using the diagonal matrix μI to obtain a diagonal loaded snapshot covariance matrix, wherein the regularization process is expressed as: The covariance matrix R(i) of the i-th snapshot is expressed as R(i)=ηR(i-1)+r(i)r H (i)+(1-η)μI; Use the bisection coordinate descent method to calculate the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector, and set the forgetting factor η to 1-2 -Ω , the covariance matrix R(i) is updated by adding and shifting the forgetting factor η, where μ is the diagonal loading factor, I is the unit matrix, r(i) is the observation signal of the antenna receiving array, and Ω is a positive integer.

[0016] Furthermore, the present invention also provides a source wave arrival estimation system based on minimum variance distortion-free response, comprising: a model building module, a model conversion module, a model solving module and a signal processing module, wherein:

[0017] A model building module is used to establish a signal source arrival estimation model and determine the signal source arrival direction by calculating the sensor antenna array output power at each scanning angle;

[0018] The model conversion module is used to convert the source arrival estimation model into a constrained minimization problem according to the minimum variance distortion-free response criterion. The constrained minimization problem is specifically: adjusting the sensor antenna weight to maximize the output power gain of the source arrival angle and suppressing the output power gain of other interference angles;

[0019] The model solving module is used to calculate the product of the inverse matrix of the snapshot covariance matrix and the scanning steering vector in the antenna array output power at each angle by using the bisection coordinate descent method in solving the constrained minimum problem, and obtain the antenna array output power of the wave arrival signal at each angle;

[0020] The signal processing module is used to estimate the direction of arrival of the signal source by using the peak value of the output power spectrum according to the calculated output power of the antenna array of the signal arriving at each angle.

[0021] Beneficial effects of the present invention:

[0022] The invention is based on the minimum variance distortion-free response criterion and takes the recursive least square method as the framework to calculate the output power of the arrival signal at each angle received by the antenna array. In the output power calculation, the product of the inverse matrix of the snapshot covariance matrix and the scanning steering vector is calculated by the bisection coordinate descent method. While reducing the computational complexity in the process of direction of arrival estimation, it can provide better DOA estimation resolution, have good estimation performance, simple operation, and is easy to implement, and has good application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The figure is a schematic diagram of the source arrival estimation process based on the minimum variance distortion-free response in the embodiment;

[0024] Figure 2 It is a schematic diagram of the RMSE performance of the comparison algorithm changing with snapshots at different signal-to-noise ratios when the true angle of arrival is 29° in the embodiment;

[0025] Figure 3 Schematic diagram of the RMSE of the angle of arrival obtained by the comparison algorithm with the increase of snapshots under different signal-to-noise ratios in the embodiment, and the actual angle of arrival is 29.9°;

[0026] Figure 4 The RMSE performance of the comparison algorithm in the embodiment varies with the signal-to-noise ratio, and the true angle of arrival is 29° and 29.9°;

[0027] Figure 5The LEDCD-RLS-DL algorithm is used to estimate the angle of arrival at different scanning steps in the embodiment. The actual angle of arrival is 37.5°.

[0028] Figure 6 It is a block diagram of DOA estimation based on LEDCD algorithm in the embodiment;

[0029] Figure 7 The figure is an illustration of the fixed-point DOA estimation performance when the actual angle of arrival is 5° and 25.3° when the signal-to-noise ratio is 10 dB in the embodiment. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solutions and advantages of the present invention clearer and more understandable, the present invention is further described in detail below in conjunction with the accompanying drawings and technical solutions.

[0031] The embodiment of the present invention provides a method for estimating the source wave arrival based on the minimum variance distortion-free response, see Figure 1 As shown, it contains the following contents:

[0032] S101, establishing a signal source arrival estimation model, the model determines the signal source arrival direction by calculating the sensor antenna array output power at each scanning angle;

[0033] S102, according to the minimum variance distortion-free response criterion, converting the source arrival estimation model into a constrained minimization problem, wherein the constrained minimization problem is specifically: adjusting the sensor antenna weight to maximize the output power gain of the source arrival angle and suppressing the output power gain of other interference angles;

[0034] S103, in solving the constrained minimization problem, using the recursive least squares method as a framework, the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector in the antenna array output power at each angle is calculated by the bisection coordinate descent method to obtain the antenna array output power of the wave arrival signal at each angle;

[0035] S104 , estimating the direction of arrival of the signal source using the peak value of the output power spectrum of the antenna array output power of the signal arriving at each angle calculated.

[0036] The Dichotomous Coordinate Descent (DCD) algorithm has low complexity and can maintain numerical stability. This is because the DCD algorithm does not require multiplication and division operations when solving system problems constructed by linear equations, but only addition and shift operations. The RLS problem can be expressed as a set of auxiliary regular equations, and the DCD iteration approximates the solution of this set of equations. In the embodiment of this case, based on the minimum variance distortion-free response criterion and taking the recursive least squares method as the framework, the inverse of the covariance matrix of the output power of the wave arrival signal at each angle received by the antenna array is calculated by the bisection coordinate descent algorithm, and the product of the scanning steering vector is solved, so as to solve the high complexity caused by the large number of array antennas in the existing wave arrival direction estimation, the unsuitability for hardware implementation and the low practical operability.

[0037] As a source arrival estimation method based on minimum variance distortion-free response in an embodiment of the present invention, further, in the source arrival estimation model, a sensor antenna receiving array is used to perform beam scanning within a preset area, the antenna array output power at each scanning angle is calculated, and the angle corresponding to the maximum output power is determined as the source arrival direction.

[0038] Assume there are N far-field signal sources s1,s2,...,s N Send information to a uniform linear array consisting of M sensor antennas, and the direction of arrival is:

[0039] Z=[θ1,θ2,θ3,...,θ N ], (1)

[0040] where θ N represents the angle with the Nth signal source. The signal X received by the sensor antenna array at sampling time t is expressed as

[0041]

[0042] Where s(t) is a set of uncorrelated arrival signals. n ) is the M×1-dimensional steering vector of the nth source angle, which is represented by A0(θ n )=[1,e τ ,e 2τ ,...,e (M-1)τ ], τ = jγd sin(θ), γ = 2π / λ, λ is the wavelength, and d is the distance between the sensor antennas. An M×1-dimensional signal vector X(t) can also be written as:

[0043] X(t)=A(θ n )·s(t), (3)

[0044] Where A(θ n)=[A0(θ1),A0(θ2),...,A0(θ N )] is the M×N dimensional array steering vector matrix. The signal vector plus noise is called a snapshot. Assume that N narrowband signals are received within Q snapshot observation periods. When the i-th (i∈{1,2,3,...,Q}) snapshot is received, the observation signal of the antenna array is yes

[0045] r(i)=A(θ n )·s(i)+n(i), (4)

[0046] where r(i) is an M×1 dimensional vector and n(i) is a vector with variance σ 2 An M×1-dimensional zero-mean complex Gaussian random noise vector.

[0047] According to the MVDR criterion, the sensor antenna receiving array estimates the direction of arrival by calculating the received signal power in all possible directions. Mathematically, the MVDR-DOA estimation method can be expressed as a constrained minimization problem, that is, the sensor antenna array adjusts the antenna weights to maximize the output power gain of the source arrival angle and minimize the output power gain of other interference angles. The weighted signal F(i) received by the sensor antenna array is expressed as:

[0048] F(i)=w H (i)r(i), (5)

[0049] Where w(i) is an M×1 dimensional complex weight vector. For each possible source angle θ n According to the MVDR standard, it is necessary to meet

[0050]

[0051] in

[0052]

[0053] is the M×M-dimensional snapshot covariance matrix, and K is the number of training snapshots. The weight vector w(i) is expressed as:

[0054]

[0055] Where A s (θ p )(p=1,2,...,P) is a scanning steering vector of M×1 dimension, and P is the number of scanning steps.

[0056] Each scanning angle θ p The output power is

[0057]

[0058] The DOA estimate can be obtained by calculating the output power spectrum peak, that is:

[0059]

[0060] Assuming that the steering vector corresponding to the source signal is known, the DOA estimator based on the MVDR criterion has better resolution and more significant interference suppression capability than the traditional DOA estimator that does not rely on steering data. However, the operation of inverting the covariance matrix R is relatively high in terms of computational complexity. In the embodiment of this case, the inverse of the matrix may not be calculated directly, but the high computational burden in the hardware implementation may be avoided by solving the product of the inverse of the covariance matrix and the scanning steering vector. Among them, for the i-th snapshot, the MVDR solution based on the exponentially weighted RLS algorithm can be summarized as follows:

[0061] (1) Update the gain vector

[0062] (2) Update weight coefficient where ε n is a gain factor that strengthens the constraint,

[0063] (3) Through R -1 (i) = η -1 R -1 (i-1)-η -1 g(i)r H (i)R -1 (i-1) Update the inverse matrix.

[0064] In summary, the traditional RLS algorithm applies the matrix inversion lemma and obtains the inverse matrix R of the (i-1)th snapshot. -1 (i-1) can calculate the inverse matrix R -1 (i)

[0065]

[0066] Where η is the forgetting factor (0<η≤1). However, the RLS matrix inversion operation comes at the cost of increasing computational complexity, that is, each snapshot executes O(M 2 ) operands.

[0067] The recursive matrix inversion operation is the source of complexity and instability in most known RLS adaptive filtering algorithms. If the inverse matrix of the complex matrix R is not directly calculated, formula (9) can be expressed as

[0068]

[0069] where v θ (i) = R -1(i)A s (θ p ), assuming A s (θ p ) is stable during the estimation of the direction of arrival.

[0070] Formula R(i)v θ (i) = A s (θ p ) can be expressed as:

[0071] R(i)[v θ (i-1)+Δv θ (i)]=A s (θ p ), (13)

[0072] Formula (13) can also be rewritten as

[0073]

[0074] A s (θ p ; i) is the scan steering vector of the i-th snapshot, Λ sr (i-1) = A s (θ p ;i-1)-R(i-1)v θ (i-1). Increment ΔR(i), ΔA s (θ p ; i), and Δv θ (i) are respectively expressed as

[0075]

[0076] In order to improve the robustness of the algorithm, the diagonal matrix μI is added to the snapshot covariance matrix for regularization to obtain the diagonal loading index snapshot covariance matrix:

[0077]

[0078] Where μ is the diagonal loading factor (μ is a small constant) and I is the identity matrix. Then equation (16) can be modified as

[0079] R(i)=ηR(i-1)+r(i)r H (i)+(1-η)μI, (17)

[0080] Through equations (15), (17) and ΔA s (θ p ; i) = 0, we can get

[0081] ΔR(i)=(η-1)R(i-1)+r(i)rH (i)+(1-η)μI, (18)

[0082] Multiply both sides of formula (18) by Can get

[0083]

[0084] in Applying formula (19) to formula (14), A a (i) can be expressed as:

[0085]

[0086] If R(i) and A are known a (i), the approximate solution Δv can be obtained by equation (14): θ (i). In order to avoid high computational complexity, an iterative descent search method—DCD—can be used to solve equation (14), because DCD algorithms use shift operations instead of multiplication / division operations, which is very suitable for algorithm hardware implementation.

[0087] A DCD-type algorithm that selects update elements by leading indices has been shown to speed up convergence. In this algorithm, the “leading” index q corresponds to the residual vector Λ sr The largest element in the . This type of DCD algorithm is called the Leading Element DCD algorithm LEDCD (Leading Element DCD) algorithm. Table 1 describes the working steps of the LEDCD algorithm. The LEDCD algorithm can effectively solve Δv in step 4 of Table 2 θ (i). Vector Δv θ Each element of B max The parameter H is the initial value of the variable step size ζ. Each time it is updated, the step size ζ is divided by 2. The LEDCD algorithm updates Δv θ and Λ sr The qth element in can only execute U max times. If If the absolute value of is small enough, the algorithm will not be updated. The absolute value of is large enough, and the algorithm updates the solution and vector Λ sr If the comparison operation (step ③) in Table 1 is always satisfied, considering the amount of computation, the update times U can be preset. max (Step ⑥) to stop the algorithm. Another parameter to stop the algorithm is B max , this parameter can control the number of step size updates (step ②). A single update of the LEDCD algorithm requires M shift operations, (M+2) addition operations and 1 comparison operation.

[0088] Table 1: LEDCD algorithm

[0089]

[0090] Table 2: LEDCD-RLS-MVDR DOA estimation algorithm for diagonal loading in this embodiment ( Complex number system)

[0091]

[0092] Furthermore, based on the above method, an embodiment of the present invention also provides a source arrival estimation system based on minimum variance distortion-free response, comprising: a model construction module, a model conversion module, a model solution module and a signal processing module, wherein:

[0093] A model building module is used to establish a signal source arrival estimation model and determine the signal source arrival direction by calculating the sensor antenna array output power at each scanning angle;

[0094] The model conversion module is used to convert the source arrival estimation into a constrained minimization problem according to the minimum variance distortion-free response criterion. The constrained minimization problem is specifically: adjusting the sensor antenna weight to maximize the output power gain of the source arrival angle and suppressing the output power gain of other interference angles;

[0095] The model solving module is used to calculate the product of the inverse matrix of the snapshot covariance matrix and the scanning steering vector in the antenna array output power at each angle by using the bisection coordinate descent method in solving the constrained minimization problem, and obtain the antenna array output power of the wave arrival signal at each angle;

[0096] The signal processing module is used to estimate the direction of arrival of the signal source by using the peak value of the output power spectrum according to the calculated output power of the antenna array of the signal arriving at each angle.

[0097] In order to verify the effectiveness of this solution, the following is a further explanation based on specific data:

[0098] The algorithmic complexity is measured by the number of multiplications and additions required for each snapshot. Note that these methods are used to solve complex equations. Solving a complex linear system of equations in M / 2×M / 2 dimensions can be transformed into solving a real system of equations in M×M dimensions.

[0099]

[0100] Where R1 and R2 are the real and imaginary parts of R, which are M / 2×M / 2 dimensional real number matrices. M / 2×1 dimensional vector Δv θ1 and Δv θ2 is Δvθ The real and imaginary parts of M / 2×1 dimensional vector A a1 and A a2 Yes A a The real and imaginary parts of the transformed real matrix R are symmetric. Therefore, only the upper triangular matrix of the matrix R needs to be calculated. The direct inversion method is used to solve R -1 Requires O(M 3 ) operations (addition and multiplication), while classic RLS requires O(M 2 ) operations. Unlike the RLS algorithm, which solves the inverse matrix recursively, the LEDCD-RLS-MVDR algorithm solves the product R(i)Δv recursively. θ (i). Among them, the parameters μ, η and A s (θ P ) is assumed to be a constant during the iteration process, and (1-η)μ is also a constant. In addition, the forgetting factor η is set to 1-2 -Ω , where Ω is a positive integer, so that the multiplication operation of (1-η)μ can be replaced by addition and shift operations. In hardware implementation, shift operations are much simpler than multiplication operations. Therefore, the matrix R(i) can be updated as follows: R(i) = (1-2 -Ω )R(i-1)+r(i)r H (i)+2 -Ω μI. This operation can reduce the complexity of step 1 in Table 2 from 3M 2 Reduce to M 2 The complexity of step 4 in Table 2 is the computational complexity required to solve the problem using the LEDCD algorithm. max Updates, the complexity of the LEDCD algorithm (Ψ LEDCD ) The upper limit is 2MU max +B max However, in actual implementation, the vector Δv with the same accuracy is obtained θ (i), LEDCD algorithm usually requires less than U max The diagonal loading technique can improve the performance of the algorithm. However, this technique will increase (MP+M) multiplications and M on the basis of the LEDCD-RLS algorithm. 2 +MP additions. The traditional RLS algorithm uses diagonal loading technology and requires matrix inversion for each update, which results in an O(M 3 In summary, the RLS algorithm requires 3.5M to calculate the output power of a beam-arrival angle array. 2 +3.5M multiplications and 2.5M 2 -M additions. The LEDCD-RLS-MVDR algorithm requires M 2 +3M multiplications and 2M2 +6M - 2 + Ψ LEDCD The second addition. The LEDCD - RLS - MVDR - DL algorithm requires M 2 +5M multiplications and 3M 2 +7M - 2 + Ψ LEDCD additions. Through analysis and comparison, the sorting of these algorithm complexities from small to large is: LEDCD - RLS < LEDCD - RLS - DL < RLS < direct inversion.

[0101] The performance of the proposed DOA estimation scheme is evaluated through simulation experiments. The LEDCD - RLS - DL method proposed in the embodiments of this case is compared with other algorithms in terms of Root Mean Square Error (RMSE) and Signal Noise Ratio (SNR). Among them, the direct inversion, RLS, LEDCD - RLS, and LEDCD - RLS - DL methods calculate the estimated angle The RMSE is as follows:

[0102]

[0103] where is the DOA estimated by the algorithm at the i - th snapshot, θ true is the true DOA of the signal source. The vector A s (θ p ) is the scanning steering vector obtained at a scanning interval of 0.1° within the angle range from - 90° to 90°. The value of P is calculated to be 1801.

[0104] 1. Assume that the number of array elements of the uniform linear array is M = 16, the element spacing d is 1 / 2 of the wavelength of the incident signal, Ω = 8, U max = 32, B max = 15, N = 1.

[0105] See Figure 2 As shown, when θ true = 29°, the RMSE performance of the direct inversion, RLS, LEDCD - RLS, and LEDCD - RLS - DL algorithms with the increase of snapshots at SNR = 5dB, SNR = 10dB, and SNR = 15dB respectively. The results show that the estimation accuracy of the LEDCD - RLS algorithm is the worst, and the performance of the direct inversion method is the best. With the increase of the number of snapshots, the RMSE performance of the LEDCD - RLS - DL algorithm is close to that of the direct inversion algorithm.

[0106] See Figure 3 As shown, when θ true=29.9°, the RMSE performance of the four algorithms, direct inversion, RLS, LEDCD-RLS and LEDCD-RLS-DL, as the number of snapshots increases at SNR=5dB, SNR=10dB and SNR=15dB respectively. The results show that when SNR=5dB, as the number of snapshots increases, the RMSE performance of the LEDCD-RLS algorithm is similar to that of the RLS algorithm, and the performance of the LEDCD-RLS-DL algorithm is better than that of RLS and LEDCD-RLS. Since the scanning step size is 0.1°, Figure 2 In comparison, the RMSE performance of these four algorithms at 29.9° angle of arrival estimation is lower than their RMSE performance at 29° angle of arrival estimation.

[0107] 2. Assume that the number of uniform linear array elements is M = 16, the element spacing d is half the wavelength of the incident signal, the SNR range is 0dB ~ 20dB, K = 500, Ω = 8, U max =32, B max =15, N=1.

[0108] See also Figure 4 As shown in the figure, the direct inversion, RLS, LEDCD-RLS and LEDCD-RLS-DL algorithms are true =29° and θ true =29.9°. The results show that the RMSE performance of the direct inversion algorithm, RLS, LEDCD-RLS, and LEDCD-RLS-DL algorithms improves with the increase of SNR. The RMSE performance of the LEDCD-RLS-DL algorithm is better than that of the LEDCD-RLS and RLS algorithms, and is very close to that of the direct inversion algorithm.

[0109] LEDCD-RLS-DL MVDR DOA estimator can first obtain a rough estimate of the source wave arrival with low computational complexity. Then, a small angle range is set with the rough DOA estimate as the center, and the received power of the possible source angles in the set angle range at the antenna array is calculated, and the peak power is selected to obtain the precise estimate of the source DOA. Figure 5 It shows that when θ true =37.5°, the LEDCD-RLS-DL MVDR DOA estimation algorithm uses different scanning step lengths to estimate the wave arrival effect. Step 1: Obtain the scanning steering vector A at intervals of 5° in the range of -90° to 90°. s (θ p), the results show that there is an obvious power amplitude peak at about 40°. Step 2, set the scanning range of 5° to the left and right with 40° as the center, and the scanning step size is 1°. The results show that the power peak is about 38°. Step 3, with 38° as the center angle, the scanning range is 37° to 39°, and the scanning step size is 0.1°. The results show that there is a power amplitude peak at 37.5°. This example shows that the scanning step size will affect the accuracy of DOA estimation. The rough DOA estimation (step 1) of the LEDCD-RLS-DL algorithm requires less computation, and the precise DOA estimation (steps 2 and 3) requires more scanning data to improve the estimation accuracy, but it also increases the computational complexity. Different methods can be used for DOA estimation according to different application requirements.

[0110] like Figure 6 As shown in the figure, the LEDCD algorithm in the LEDCD-RLS-DL MVDR DOA estimator is implemented on the Nexys4-DDR development board using VHSIC Hardware Description Language (VHDL) and synthesized using Vivado with parameters set to M = 16, N = 1, Ω = 8, μ = 0.1, U max =32. Scanning steering vector A s (θ p ) is obtained by scanning in the range of -60° to 60° with an interval of 0.1°. The vector r(i) and A s (θ p ) is represented in 16-bit fixed-point Q15 format. The matrix R(i) and the vector Λ sr (i) and v θ The real and imaginary parts of (i) are represented by 32 bits in Q15 format. When SNR = 10 dB, the LEDCD-RLS-DL algorithm performs true =5° and θ true =25.3° fixed-point DOA estimation performance is as follows Figure 7 As shown in (a) and (b) of Figure 2, the results show that the hardware fixed-point DOA estimation performance is very close to its floating-point performance. 7 No numerical instability was observed during the experiment. Table 3 shows the hardware resources occupied by the LEDCD algorithm with different M values.

[0111] Table 3 LEDCD hardware resources

[0112]

[0113] The data show that as the value of M increases, the hardware occupancy of LEDCD remains basically unchanged, and its power consumption is only 0.239 mW at different M values. Therefore, the MVDR DOA estimator based on LEDCD-RLS-DL is very suitable for low-power and cheap hardware implementation and is easy to implement.

[0114] Unless otherwise specifically stated, the relative steps, numerical expressions and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0115] Based on the above method and / or system, an embodiment of the present invention also provides a server, including: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the above method.

[0116] Based on the above method and / or system, an embodiment of the present invention further provides a computer-readable medium having a computer program stored thereon, wherein the above method is implemented when the program is executed by a processor.

[0117] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0118] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0119] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A method for source arrival estimation based on minimum variance distortion-free response, characterized in that: Contains the following: A signal source arrival estimation model is established, which determines the signal source arrival direction by calculating the sensor antenna array output power at each scanning angle; According to the minimum variance distortion-free response criterion, the source arrival estimation model is converted into a constrained minimization problem. The specific constrained minimization problem is: adjust the sensor antenna weight to maximize the output power gain of the source arrival angle and suppress the output power gain of other interference angles; where, for the nth source angle θ n , the constrained minimization problem is expressed as: And satisfy w H (i)A(θ n )=1, R(i) is the M×M-dimensional snapshot covariance matrix, and w(i) is an M×1 dimensional complex weight vector, K is the number of training snapshots, and A(θ n ) represents the steering vector matrix, r(k) represents the observation signal of the antenna receiving array; In solving the constrained minimization problem, the recursive least squares method is used as the framework, and the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector in the antenna array output power at each angle is calculated by the bisection coordinate descent method to obtain the antenna array output power of the wave arrival signal at each angle; Among them, in solving the constrained minimization problem, it is assumed that the scan steering vector A s (θ p ) is stable during the DOA estimation, p = 1, 2, ..., P, P is the number of scanning steps, then each scanning angle θ p The output power is expressed as v θ (i) = R -1 (i)A s (θ p ) ; During the output power calculation process, the snapshot covariance matrix R(i) is regularized using the diagonal matrix μI to obtain the diagonal loaded snapshot covariance matrix. The regularization process is expressed as: The covariance matrix R(i) of the i-th snapshot is expressed as R(i) = ηR(i-1) + r(i)r H (i)+(1-η)μI; Use the bisection coordinate descent method to calculate the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector, and set the forgetting factor η to 1-2 -Ω , the covariance matrix R(i) is updated by adding and shifting the forgetting factor η, μ is the diagonal loading factor, I is the unit matrix, r(i) is the observation signal of the antenna receiving array, and Ω is a positive integer; According to the calculated antenna array output power of the wave arrival signal at each angle, the output power spectrum peak is used to estimate the source wave arrival direction.

2. The method for source arrival estimation based on minimum variance distortion-free response according to claim 1, characterized in that: In the signal source arrival estimation model, the sensor antenna receiving array is used to perform beam scanning within a preset area, the antenna array output power at each scanning angle is calculated, and the angle corresponding to the maximum output power is determined as the signal source arrival direction.

3. The method for source arrival estimation based on minimum variance distortion-free response according to claim 1 or 2, characterized in that: In the beam scanning of the sensor antenna receiving array, assuming that N narrowband signals are received within the observation period of Q snapshots, the steering vector matrix A(θ n ) is used to represent the response of the sensor antenna receiving array to a narrowband signal source with unit energy at the i-th snapshot, where i∈{1,2,3,...,Q}.

4. The method for source arrival estimation based on minimum variance distortion-free response according to claim 3, characterized in that: The observation signal of the sensor antenna receiving array at the i-th snapshot is expressed as: r(i) = A(θ n )·s(i)+n(i), where, A(θ n ) represents the steering vector matrix, and A(θ n )=[A0(θ1),A0(θ2),...,A0(θ N )],A0(θ N ) represents the Nth source angle θ N , s(i) represents a group of unrelated arrival signals sent by multiple sources at the i-th snapshot moment, and n(i) represents Gaussian random vector noise.

5. A source arrival estimation system based on minimum variance distortion-free response, characterized in that: It includes: model building module, model conversion module, model solving module and signal processing module, among which, A model building module is used to establish a signal source arrival estimation model and determine the signal source arrival direction by calculating the sensor antenna array output power at each scanning angle; The model conversion module is used to convert the source arrival estimation model into a constrained minimization problem according to the minimum variance distortion-free response criterion. The constrained minimization problem is specifically as follows: adjust the sensor antenna weights to maximize the output power gain of the source arrival angle and suppress the output power gain of other interference angles; where, for the nth source angle θ n , the constrained minimization problem is expressed as: And satisfy w H (i)A(θ n )=1, R(i) is the M×M-dimensional snapshot covariance matrix, and w(i) is an M×1 dimensional complex weight vector, K is the number of training snapshots, and A(θ n ) represents the steering vector matrix, r(k) represents the observation signal of the antenna receiving array; The model solving module is used to calculate the product of the inverse matrix of the snapshot covariance matrix and the scanning steering vector in the antenna array output power at each angle by using the bisection coordinate descent method in solving the constrained minimization problem, and obtain the antenna array output power of the wave arrival signal at each angle; wherein, in solving the constrained minimization problem, it is assumed that the scanning steering vector A s (θ p ) is stable during the DOA estimation, p = 1, 2, ..., P, P is the number of scanning steps, then each scanning angle θ p The output power is expressed as v θ (i) = R -1 (i)A s (θ p ); During the output power calculation process, the snapshot covariance matrix R(i) is regularized using the diagonal matrix μI to obtain the diagonal loaded snapshot covariance matrix. The regularization process is expressed as: The covariance matrix R(i) of the i-th snapshot is expressed as R(i) = ηR(i-1) + r(i)r H (i)+(1-η)μI; Use the bisection coordinate descent method to calculate the product of the inverse matrix of the snapshot covariance matrix and the scan steering vector, and set the forgetting factor η to 1-2 -Ω , the covariance matrix R(i) is updated by adding and shifting the forgetting factor η, μ is the diagonal loading factor, I is the unit matrix, r(i) is the observation signal of the antenna receiving array, and Ω is a positive integer; The signal processing module is used to estimate the direction of arrival of the signal source by using the peak value of the output power spectrum according to the calculated output power of the antenna array of the signal arriving at each angle.

6. A computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the method according to any one of claims 1 to 4.

7. A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 4.

Images