阅读说明：本技术 一种基于粒子群算法的无网格波达角估计方法 (Non-grid DOA (angle of arrival) estimation method based on particle swarm optimization ) 是由 段克清 李想 杨兴家 王煜岗 于 2021-08-04 设计创作，主要内容包括：本发明公开了一种基于粒子群算法的无网格波达角估计方法,所述方法包括步骤如下：S1：根据所用天线阵列结构和待估计DOA的维度构建导向矢量,并结合导向矢量与观测数据构建出自变量为信源估计角度的适应度函数；S2：设定粒子群算法的参数,并设置迭代终止条件；将适应度函数作为粒子群算法的优化目标进行迭代优化,从连续的角度域里进行参数搜索求解,直到达到粒子群算法迭代终止条件；S3：粒子群算法迭代完成之后,从中选取使得适应度函数最优的粒子作为最优解输出,其输出的参数即为所估计的DOA。本发明适用于阵列信号DOA估计,具有样本快拍需求少、无需预先角度离散化处理、目标超分辨的特点。(The invention discloses a non-grid DOA estimation method based on a particle swarm algorithm, which comprises the following steps: s1: constructing a guide vector according to the antenna array structure and the dimension of DOA to be estimated, and constructing a fitness function taking an independent variable as an information source estimation angle by combining the guide vector and observation data; s2: setting parameters of a particle swarm algorithm and setting iteration termination conditions; performing iterative optimization by taking the fitness function as an optimization target of the particle swarm algorithm, and performing parameter search solution from a continuous angle domain until an iteration termination condition of the particle swarm algorithm is reached; s3: and after the iteration of the particle swarm optimization is finished, selecting particles which enable the fitness function to be optimal from the particle swarm optimization as an optimal solution to be output, wherein the output parameters are the estimated DOA. The method is suitable for array signal DOA estimation, and has the characteristics of less sample snapshot requirements, no need of angle discretization in advance and target super-resolution.)

1. A grid-free DOA estimation method based on particle swarm optimization is characterized in that: the method comprises the following steps:

s1: constructing a guide vector according to the antenna array structure and the dimension of DOA to be estimated, and constructing a fitness function taking an independent variable as an information source estimation angle by combining the guide vector and observation data;

s2: setting parameters of a particle swarm algorithm and setting iteration termination conditions; performing iterative optimization by taking the fitness function as an optimization target of the particle swarm algorithm, and performing parameter search solution from a continuous angle domain until an iteration termination condition of the particle swarm algorithm is reached;

s3: and after the iteration of the particle swarm optimization is finished, selecting particles which enable the fitness function to be optimal from the particle swarm optimization as an optimal solution to be output, wherein the output parameters are the estimated DOA.

2. The particle swarm algorithm-based grid-free angle of arrival estimation method of claim 1, wherein: the antenna array structure adopts a uniform linear array ULA, so that the constructed steering vector is as follows:

in the formula, lambda represents signal wavelength, and d represents array element interval; n represents the number of array elements owned by the uniform linear array; theta is the angle of incidence,is an imaginary unit, (.)^TIs a matrix device operation.

3. The particle swarm algorithm-based grid-free angle of arrival estimation method of claim 1, wherein: the DOA dimension to be estimated can be one-dimensional or two-dimensional.

4. The particle swarm algorithm-based mesh-free angle of arrival estimation method of claim 2, wherein: and combining the steering vector and the observation data to construct a fitness function f (theta) with the independent variable as the source estimation angle as follows:

D(θ)＝[v(θ₁),v(θ₂),…,v(θ_K)] (2)

f(θ)＝||D(θ)D(θ)⁺X-X||_F (3)

in the formula, K represents the number of information sources (.)⁺Representing the pseudo-inverse of the matrix, | · |. non-woven phosphor_FA Frobenius norm representing a matrix; d (θ) is a matrix formed by arranging K steering vectors, i.e. the source subspace estimator.

5. The particle swarm algorithm-based mesh-free angle-of-arrival estimation method of claim 4, wherein: the parameters of the particle swarm algorithm comprise iteration times T, the number of the population particles, the particle dimension and a self-learning factor c₁Social cognition factor c₂Maximum value of inertia factor w_MAXMinimum value of inertia factor w_MINAnd a convergence factor a.

6. The particle swarm algorithm-based mesh-free angle of arrival estimation method of claim 5, wherein: before iterative optimization, initializing initial positions of all particles randomly, and taking the initial positions as the self-generation optimal particles of each particle; and then, calculating the adaptive value of each particle by using the fitness function, and selecting the particle with the lowest adaptive value as the optimal particle in the population.

7. The particle swarm algorithm-based mesh-free angle of arrival estimation method of claim 6, wherein: the formula of the iterative optimization is as follows:

in the formula, w^tAn inertia factor representing a current iteration;representing the velocity of the ith particle at the t-th iteration;representing the position of the ith particle at the time of the t-th iteration; r is₁And r₂A uniformly distributed random value representing a obedience interval of (0, 1); theta_{pbest_i}Representing the generation-optimal particle of the ith particle; theta_gbestRepresenting the optimal particles in the current population; a denotes a convergence factor, set to a constant less than 1.

8. The particle swarm algorithm-based mesh-free angle of arrival estimation method of claim 7, wherein: and (3) performing parameter search solving on the fitness function by utilizing a particle swarm algorithm: for each particle, comparing the particle with the generation optimal particle, and if the adaptive value is lower, replacing the generation optimal particle with the particle; and simultaneously comparing each particle with the optimal particle in the population, and replacing the optimal particle in the population if the adaptive value is lower.

9. The particle swarm algorithm-based mesh-free angle of arrival estimation method of claim 8, wherein: the iteration termination condition is as follows: if the current iteration time T is more than or equal to T, or the optimal particle fitness f (theta) in the group_gbest)≤σ²Wherein σ is²Representing the receiver noise power, stopping iteration; otherwise, continuing the iterative optimization.

10. A computer system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein: the processor, when executing the computer program, performs the steps of the method according to any of claims 1 to 9.

Technical Field

The invention relates to the technical field of array signal processing, in particular to a non-grid DOA estimation method based on a particle swarm algorithm.

Background

Estimation of the signal angle of arrival (DOA) is an important research topic in array signal processing, and the main task is to arrange a plurality of sensor array elements in a certain structure in a space domain, and receive, process and analyze signals from the space, so as to acquire information such as the DOA of the signals. As one of the key technologies of adaptive arrays, DOA estimation has been widely applied in the fields of wireless communication, radar, sonar, reconnaissance, detection, and the like.

Early array DOA estimation is achieved by beamforming-like algorithms, including beamforming algorithm (DBF) and least-variance signal distortion-free response algorithm (MVDR). The performance of the DBF algorithm is limited by the antenna elements, and multiple targets in the same beam cannot be distinguished, that is, a rayleigh limit exists. Therefore, the DOA resolution of the algorithm is not high. The MVDR algorithm estimation performance is still poor in resolving power for multiple targets and angularly neighboring targets with low signal-to-noise ratio (SNR) and few fast sample beats. With the increasing development and improvement of communication technology, the requirements for precision and resolution in practical application measurement are gradually increased, and the performance of the DOA method cannot meet the requirements.

Currently, there are two main categories of high-resolution and high-precision DOA estimation techniques. The first major category is subspace-based methods, and the most representative algorithms are the multiple signal classification (MUSIC) algorithm and the signal parameter rotation invariant (ESPRIT) algorithm. And the MUSIC algorithm decomposes the eigenvalue of the covariance matrix of the received signal to obtain a noise subspace and a signal subspace, and designs a space spectrum search function by utilizing the orthogonality between the noise subspace and the signal subspace, thereby obtaining a target information source DOA estimation value. The algorithm breaks through the Rayleigh limit and obviously improves the estimation performance of the signal DOA, but the proper search interval is difficult to set so that the real DOA falls on the grid. The ESPRIT algorithm does not need to search a signal space spectrum, so the calculation complexity is relatively low, but the application of the ESPRIT algorithm is limited by an array structure, partial array apertures are sacrificed, and the estimation performance is slightly poor compared with the MUSIC algorithm. In addition, the subspace method utilizes the second-order cumulant of the signal to analyze and solve, so the performance of the subspace method is extremely dependent on the accurate estimation of the signal covariance matrix.

However, in engineering application, due to the influence of various factors of an application scene, enough sample snapshots can not be obtained, so that the estimation of the covariance matrix is difficult to achieve convergence in a statistical sense; in addition, in the source coherent environment, the covariance matrix even has rank loss, and at this time, the performance of the subspace-like algorithm is sharply reduced. The second category is sparse recovery method, which considers that the distribution of the incident signal in the whole discrete angle domain has certain sparsity and represents the received signal as a linear combination of several bases in an over-complete base set. Based on this assumption, DOA estimation can be converted into an underdetermined equation solving problem under constraint conditions, and a general solving framework is provided. A large number of direction finding methods have been developed under the theoretical framework of sparse recovery, and mainly include three types: matching pursuit, norm constraint optimization and sparse Bayesian learning, and realizes super-resolution estimation of the signal DOA under the condition of small samples.

However, the conventional sparse recovery method requires the angular domain to be discretized into a large number of finite grids in advance, and assumes that the incident signal DOA falls on the grid points exactly without offset. In practice, the value of the signal DOA is continuous, which means that the problem of lattice mismatch is difficult to avoid. Increasing the grid density can obtain higher angular resolution, so that the probability that the true DOA falls on the grid points is improved, and meanwhile, the correlation of the steering vectors corresponding to the adjacent grids is increased, and the sparse recovery performance is reduced. More importantly, increasing the grid density also leads to an increase in the computation amount, and is particularly reflected in a planar array, and the computation amount of the algorithm is larger because the planar array divides the space by one dimension than a linear array.

By combining the analysis, the contradiction between the discreteness of the grid and the continuity of the real DOA is the problem of the traditional DOA method of sparse recovery. Therefore, to avoid estimation bias due to lattice mismatch, a sparse recovery DOA estimation algorithm that can be solved in a continuous parameter space is needed.

Disclosure of Invention

The invention provides a non-grid DOA (direction of arrival) estimation method based on a particle swarm algorithm, which aims to solve the problem of grid mismatch of the traditional sparse recovery algorithm, is suitable for DOA (direction of arrival) estimation of array signals, and has the characteristics of less sample snapshot requirements, no need of angle discretization in advance and target super-resolution.

In order to achieve the purpose of the invention, the technical scheme is as follows:

a mesh-free DOA estimation method based on a particle swarm algorithm comprises the following steps:

s1: constructing a guide vector according to the antenna array structure and the dimension of DOA to be estimated, and constructing a fitness function taking an independent variable as an information source estimation angle by combining the guide vector and observation data;

s2: setting parameters of a particle swarm algorithm, and setting iteration termination conditions according to the actual problem scale; performing iterative optimization by taking the fitness function as an optimization target of the particle swarm algorithm, and performing parameter search solution from a continuous angle domain until an iteration termination condition of the particle swarm algorithm is reached;

s3: and after the iteration of the particle swarm optimization is finished, selecting particles which enable the fitness function to be optimal from the particle swarm optimization as an optimal solution to be output, wherein the output parameters are the estimated DOA.

Preferably, the antenna array structure adopts a uniform linear array ULA, so that the constructed steering vector is as follows:

in the formula, lambda represents signal wavelength, and d represents array element interval; n represents the number of array elements owned by the uniform linear array; theta is the angle of incidence,is an imaginary unit, (.)^TIs a matrix device operation.

Preferably, the dimension of the DOA to be estimated may be one-dimensional or two-dimensional.

Further, a fitness function f (theta) with the independent variable as the source estimation angle is constructed by combining the guide vector and the observation data as follows:

D(θ)＝[v(θ₁),v(θ₂),…,v(θ_K)] (2)

f(θ)＝||D(θ)D(θ)⁺X-X||_F (3)

in the formula, K represents the number of information sources (.)⁺Representing the pseudo-inverse of the matrix, | · |. non-woven phosphor_FA Frobenius norm representing a matrix; d (θ) is a matrix formed by arranging K steering vectors, i.e. the source subspace estimator.

Still further, the parameters of the particle swarm algorithm comprise iteration times T, population particle number, particle dimension and self-learning factor c₁Social cognition factor c₂Maximum value of inertia factor w_MAXMinimum value of inertia factor w_MINAnd a convergence factor a.

Further, before iterative optimization, the initial positions of all the particles are initialized randomly, and the initial positions are used as the self-generation optimal particles of each particle; and then, calculating the adaptive value of each particle by using the fitness function, and selecting the particle with the lowest adaptive value as the optimal particle in the population.

Still further, the formula of the iterative optimization is as follows:

in the formula, w^tAn inertia factor representing a current iteration;representing the velocity of the ith particle at the t-th iteration;representing the position of the ith particle at the time of the t-th iteration; r is₁And r₂A uniformly distributed random value representing a obedience interval of (0, 1); theta_{pbest_i}Representing the generation-optimal particle of the ith particle; theta_gbestRepresenting the optimal particles in the current population; a denotes a convergence factor, set to a constant less than 1.

And further, performing parameter search solving on the fitness function by using a particle swarm algorithm: for each particle, comparing the particle with the generation optimal particle, and if the adaptive value is lower, replacing the generation optimal particle with the particle; and simultaneously comparing each particle with the optimal particle in the population, and replacing the optimal particle in the population if the adaptive value is lower.

Still further, the iteration termination condition is: if the current iteration time T is more than or equal to T, or the optimal particle fitness f (theta) in the group_gbest)≤σ²Wherein σ is²Representing the receiver noise power, stopping iteration; otherwise, continuing the iterative optimization.

A computer system comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method when executing the computer program.

The invention has the following beneficial effects:

according to the method, through modeling analysis, grid-free sparse recovery DOA estimation is regarded as a combined optimization problem; then, constructing a guide vector according to the array antenna structure and the DOA to be estimated, and further constructing a reasonable fitness function; and finally, regarding the signal DOA as a parameter to be solved, and performing iterative optimization solution by using the advantages of global random search and independence on function gradient of a particle swarm algorithm, so that the optimal solution with higher precision can be obtained in a continuous domain. The method is suitable for array signal DOA estimation, has the characteristics of less sample snapshot requirement, no need of angle discretization in advance and target super-resolution, and can effectively avoid the problem of grid mismatch of the traditional sparse recovery algorithm.

Drawings

Fig. 1 is a flow chart of the steps of the method according to the present embodiment.

Detailed Description

The invention is described in detail below with reference to the drawings and the detailed description.

Example 1

Assume that the number of sources is 2, and the signal wavelength λ is 0.15 m. The array antenna used is a uniform linear array with N equal to 20 array elements, and the interval d between the array elements is half wavelength of the received signal (i.e. d equal to 0.075 m). Receiver noise power of sigma²＝10^-12W, the received observation data is X.

The dimension of the DOA to be estimated in this embodiment may be one-dimensional or two-dimensional, that is, the method provided in this embodiment is suitable for both one-dimensional DOA estimation and joint two-dimensional DOA estimation of azimuth and pitch. For convenience of understanding, the present embodiment uses one-dimensional DOA as an example to describe the method in detail.

As shown in fig. 1, a method for estimating a grid-free angle of arrival based on a particle swarm algorithm includes the following steps:

s1: constructing a guide vector according to the antenna array structure and the dimension of DOA to be estimated, and constructing a fitness function taking an independent variable as an information source estimation angle by combining the guide vector and observation data;

s2: setting parameters of a particle swarm algorithm, and setting iteration termination conditions according to the actual problem scale; performing iterative optimization by taking the fitness function as an optimization target of the particle swarm algorithm, and performing parameter search solution from a continuous angle domain until an iteration termination condition of the particle swarm algorithm is reached;

s3: and after the iteration of the particle swarm optimization is finished, selecting particles which enable the fitness function to be optimal from the particle swarm optimization as an optimal solution to be output, wherein the output parameters are the estimated DOA.

In a specific embodiment, the antenna array structure employs a uniform linear array ULA, and thus the constructed steering vectors are as follows:

in the formula, lambda represents signal wavelength, and d represents array element interval; n represents the number of array elements owned by the uniform linear array; theta is the angle of incidence,is an imaginary unit, (.)^TIs a matrix device operation.

In a specific embodiment, a fitness function f (theta) with an argument as a source estimation angle is constructed by combining the steering vector and the observation data as follows:

D(θ)＝[v(θ₁),v(θ₂),…,v(θ_K)] (2)

f(θ)＝||D(θ)D(θ)⁺X-X||_F (3)

in the formula, K represents the number of information sources (.)⁺Representing the pseudo-inverse of the matrix, | · |. non-woven phosphor_FA Frobenius norm representing a matrix; d (θ) is a matrix formed by arranging K steering vectors, i.e. the source subspace estimator.

In a specific embodiment, the parameters of the particle swarm algorithm are specifically shown in table 1;

TABLE 1 parameters of particle swarm optimization

Number of iterations T	100
		Number of population particles	50
Dimension of particle	2
		Self learning factor c1	2
Social cognition factor c2	2
		Maximum value w of inertia factorMAX	0.9
Minimum value of inertia factor wMIN	0.4
		Convergence factor a	0.729

In a specific embodiment, before iterative optimization, the initial positions of all particles are initialized randomly, and the initial positions are used as the historical optimal particles of each particle; and then, calculating the adaptive value of each particle by using the fitness function, and selecting the particle with the lowest adaptive value as the optimal particle in the population.

In a specific embodiment, the formula of the iterative optimization is as follows:

in the formula, w^tRepresenting the inertia factor of the current iteration, and adopting a linear attenuation strategy to improve the convergence speed of the algorithm;representing the velocity of the ith particle at the t-th iteration;representing the position of the ith particle at the time of the t-th iteration; r is₁And r₂A uniformly distributed random value representing a obedience interval of (0, 1); theta_{pbest_i}Representing the generation-optimal particle of the ith particle; theta_gbestRepresenting the optimal particles in the current population; a represents a convergence factor, and is generally set to be a constant less than 1, which can ensure the convergence of the algorithm in the optimization process.

In a specific embodiment, a particle swarm algorithm is used for carrying out parameter search solving on the fitness function: for each particle, comparing the particle with the generation optimal particle, and if the adaptive value is lower, replacing the generation optimal particle with the particle; and simultaneously comparing each particle with the optimal particle in the population, and replacing the optimal particle in the population if the adaptive value is lower.

In a specific embodiment, the iteration termination condition is:

condition 1: the current iteration time T is more than or equal to T;

condition 2: optimal particle fitness within a population f (θ)_gbest)≤σ²

When the iteration meets one of the two conditions, stopping the iteration; otherwise, continuing the iterative optimization.

Through the steps, when the particle swarm optimization stops iterative optimization, the particles with the optimal fitness function in the swarm are used as the optimal solution to be output, and the output parameters are the estimated DOA.

The particle swarm optimization in the embodiment is a heuristic search algorithm in nature, the solving process is random combinatorial optimization including directionality, and the method is more suitable for sparse recovery DOA estimation which is actually a combinatorial optimization problem. The method utilizes the characteristic that the particle swarm algorithm can carry out parameter solution in a continuous domain, and is applied to grid-free sparse recovery DOA estimation. And the particle swarm optimization has wide applicability, is independent of the structural characteristics of the array antenna and only needs to construct a proper fitness function for search optimization. Therefore, the implementation mode of the method is simple, and the method can be directly applied to DOA estimation of arrays such as uniform linear arrays, L-shaped arrays, uniform area arrays and uniform circular arrays. Compared with the existing DOA algorithm technology, the method does not need to divide angle grids in advance, is more suitable for high-precision and super-resolution DOA estimation, and can avoid estimation deviation caused by the fact that a real angle does not fall on a grid point.

Example 2

A computer system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to perform the steps of:

s1: constructing a guide vector according to the antenna array structure and the dimension of DOA to be estimated, and constructing a fitness function taking an independent variable as an information source estimation angle by combining the guide vector and observation data;

s2: setting parameters of a particle swarm algorithm and setting iteration termination conditions; performing iterative optimization by taking the fitness function as an optimization target of the particle swarm algorithm, and performing parameter search solution from a continuous angle domain until an iteration termination condition of the particle swarm algorithm is reached;

s3: and after the iteration of the particle swarm optimization is finished, selecting particles which enable the fitness function to be optimal from the particle swarm optimization as an optimal solution to be output, wherein the output parameters are the estimated DOA.

It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.