DOA Estimation Method Based On Non-convex Greedy Matching

Direction of Arrival(DOA)estimation is a research hotspot in the field of array signal processing.Its basic task is to estimate the azimuth angle of the target signal reaching the array antenna in a certain area or multiple areas in space.DOA estimation has extremely important applications in the fields of radio communication,navigation,radar,and medicine.With the development of sparse recovery theory and algorithms in recent years,more and more scholars apply it to the DOA estimation problem.Compared with the traditional estimation method based on feature space,the DOA sparse recovery estimation method has better algorithm performance.Therefore,in order to pursue better DOA estimation results,we design a method that is related to the sparse recovery from the perspective of the basic theoretical research of sparse recovery.The model of our method is equivalent to an alternative model,and a non-convex greedy matching DOA estimation algorithm is proposed,and good experimental results are obtained.The research content of the full text is as follows:1.The basic uniform linear array antenna DOA estimation model is studied,and based on this,the subspace algorithms represented by Mulitiple Signal Classification(MUSIC)and Estimating Signal Parameters via Rotational Invariance Techniques(ESPRIT),and the DOA sparse recovery estimation algorithm represented by l1 Singular Value Decomposition(l1SVD)and Orthogonal Matching Pursuit(OMP)algorithms are studied.Through in-depth exploration of the principles of these algorithms,combined with simulation and comparative experiments,the performance of various algorithms is analyzed.2.A non-convex greedy matching DOA estimation algorithm under the condition of single snapshot is proposed,and the method is applied to the multi-shot DOA estimation problem based on the smoothed covariance vector model.The original model of DOA sparse recovery estimation l0 norm minimization is an NP-HARD problem.At present,the mainstream algorithm is mostly to relax the problem to a convex optimization problem.However,the related algorithm theory requires the observation dictionary to satisfy Restricted Isometry Property(RIP)and other statistical conditions that are difficult to verify.Therefore,this article first designs a non-convex fractional alternative model for the single-shot DOA estimation problem,and proves the equivalence of this model and the original model.Secondly,based on the first-order KKT(Karush Kuhn Tucker)optimization properties of the optimal solution of the alternative model,this paper proposes a fixed point iterative algorithm for non-convex greedy matching and proves its convergence.Finally,this article applies the algorithm to the multi-shot DOA estimation problem.The simulation experiment results show that the method proposed in this paper has better algorithm performance for both single snapshot and multiple snapshot DOA estimation problems.3.Aiming at the problem of DOA estimation of single snapshot in the off-grid situation,an adaptive method for adjusting the grid of atomic updates is proposed.In practical problems,the direction of arrival of the target signal usually does not fall on the divided grid points,that is,grid mismatch or off-grid situation.In order to solve this problem,this paper designs a method to compare the current An off-grid adjustment method in which the reconstruction result is similar to the real solution.First,determine whether replacement is required based on the preliminary reconstruction results.During replacement,the observation dictionary atoms are updated according to a certain replacement strategy.After the update,the proposed fractional minimization algorithm based on non-convex greedy matching is used for sparse recovery.When the function sum of the structure vector is less than the set threshold or reaches the maximum number of update steps,the estimation ends,and the angle information represented by the updated atom corresponding to the final reconstruction vector of the algorithm is the final estimation result.Finally,the performance of the algorithm is verified through simulation experiments.