Research On The Direction Of Arrival Tracking Algorithm Based On Bayesian Theory

Direction of Arrival(DOA)estimation is a method of finding direction.It estimates the arrival angle of the incident signal using the array received data.It plays an important role in many fields including mobile location,sonar system and radar tracking.In many practical scenarios,the signal source is usually moving,so real-time DOA estimation i.e.DOA tracking is desirable.However,the classical high resolution DOA estimation algorithms are based on the assumption that the signal source is motionless during the observation time.The movement of source moves would cause the diffusion of the spatial spectrum.In addition,this kind of algorithms usually require a large number of snapshots or operations of eigenvalue decomposition.Thus,when they are used for dynamic tracking problems directly,the amount of computation is too larger,and the real-time performance will be affected.Therefore,this paper makes a deep research on the high-precision DOA tracking algorithm.Firstly,this paper presents the background of the research issue,and provides the overview of the research status of DOA tracking algorithms.Then,the basic theory of DOA tracking and the classical subspace tracking algorithms are introduced,and the influence of the coherence of the signal source on the DOA tracking algorithm based on subspace tracking is studied.Secondly,a DOA tracking algorithm based on particle filter is studied in this paper.The particle filter is built in the framework of the Bayesian theory,which avoids the problem that subspace algorithms can not deal with coherent signal sources.Then,the scene with unknown initial angle is considered,and the initial angle is corrected.The basic particle filter algorithm needs the known statistical characteristics of noise and the motion state model of the system.However,these information may not be accurately known in practical situations.Therefore,the cost reference particle filter algorithm and the interacting multiple model algorithm are introduced.The cost reference particle filter does not need the statistical characteristics of the system noise,and the interacting multiple model can be applied to various motion state models simultaneously.Then,considering the characteristics of these two methods,an interacting multiple model cost reference particle filter algorithm is proposed.In the simulation analysis,the performance of various algorithms under various conditions is compared,and the effectiveness of the DOA tracking algorithm based on the interacting multiple model cost reference particle filter is verified.Finally,the DOA tracking algorithm based on Bayesian learning is studied,and a DOA tracking algorithm based on dynamic array manifold is proposed.This algorithm can deal with coherent signal sources,and does not need to build state space models,so it is more versatile.In the DOA tracking algorithm based on sparse Bayesian learning,the division of the spatial angle will lead to grid effect,and the correction of the grid offset is needed.Then,two fast algorithms are analyzed to reduce the complexity of the algorithm.Because the sparse model extends the signal to a very high dimension,the computation complexity of the fast algorithms are still large.Considering the time-varying characteristics of the DOA tracking problem,the dynamic manifold matrix is constructed.Without space angle division and signal dimension expansion,DOA tracking is realized directly in low dimension under the framework of Bayesian learning,so that the computational complexity is greatly reduced.And there is no grid effect,so the precision is higher.Then,using the corresponding parameters obtained in the solution process,the DOA tracking in the scenario of variable number of sources is realized.Simulation results show that the proposed algorithm has better tracking performance than PSBL.