Research On Particle Filter For Maneuvering Target DOA Tracking

Maneuvering target tracking is a kind of state estimation problem continuously for uncertainly moving objects, which is widely applied in many fields. The DOA(Direction of Arrival) is often changing as time passing that cannot be described exactly, therefore,researching on DOA tracking has a great important meaning.Firstly, the dissertation establishes a model of maneuvering target DOA tracking based on the array signal processing model and CV(Constant Velocity) model, which is also called the state space model. Several typical DOA tracking algorithms are designed through studying two subspace tracking algorithms and Extended Kalman filtering tracking algorithm combining with the state space model. Simulations proved the effectiveness of tracking, and compared the tracking performance of above algorithms.Then a nonlinear filtering method of PF(Particle Filter) is introduced in detail, and apply it to the DOA tracking domain. Several modified PF DOA tracking algorithms are proposed to improve the tracking performance. The simulation results show that the proposed algorithm is superior to the traditional subspace tracking algorithms and standard particle filter algorithm, in addition, solves the problem of multiple DOAs tracking for crossing trajectories without targets association and has less tracking error, as well as higher resistance to SNR(signal-to-noise ratio) and robustness. Moreover, the tracking performance and convergence properties are compared under different SNR and particle numbers.Finally, on account of vector sensor can collect sound pressure and vibration velocity information simultaneously,a novel DOA tracking algorithm is proposed based on single vector sensor united with PF algorithm. In view of the dynamic tracking model, this paper not only introduces the derivation and calculation of PCRB(Posterior Cramer-Rao Bound),obtaining the lower limit influence factor: SNR and snapshot, but also draw the PCRB and CRB curve. The PCRB and CRB curves of simulation generally align with the PCRB and CRB that acquired from single vector sensor and Capon beam forming tracking algorithms.Furthermore, it also verifies the correctness of the PCRB's derivation and calculation, and obtains the lower limit PCRB than CRB in the framework of dynamic tracing model.