Based On 4 Ts201s High-speed Parallel Direction Finding Algorithm

In the field of electronic surveillance, it is necessary to determine the orientationof the targets through DOA(Direction of Arrival) estimation. In the field of civilcommunications, DOA estimation is used to estimate the location of mobile users fordirectional beamforming. Subspace-based DOA estimation algorithms have a highprecision, high-resolution performance. As the beginning of the subspace-basedalgorithms, high-speed parallel implementation of MUSIC is not only the basis of suchalgorithms, but also the precondition of wideband DOA estimation. However, due tothe complexity of the algorithm, it is still difficult to achieve real-time processing.So, it is important to restudy each module of MUSIC, realize high-speed parallelDOA search. The main contents are as follows:(1) The correlation matrix estimation module is analyzed. The essay uses dataallocation method with the properties of Hermite matrix, and Linkport synchronoustransmission method with interrupt control to realize matrix parallel computing.Through hardware simulation, the parallel algorithm can effectively accelerate thecalculation speed of the correlation matrix.(2) The important EVD(Eigenvalue Decomposition) module is analyzed. Most ofthe traditional EVD algorithm cannot directly decompose the complex Hermite matrix,so the essay uses improved rotation matrix to reformulate the double Jacobi method,single Jacobi method and QR method. The simulations show that the EVD accuracy ofthese algorithms and real symmetric matrix EVD algorithms are equivalent at differentSNR, could meet the high-precision demand.The essay uses threshold control andcalculation optimization to improve double Jacobi method for fast eigen decomposition.The essay illustrates the improved double Jacobi method has an obvious advantage ofcomputing, through comparative analysis of different series/parallel algorithms.(3) Subspace iteration algorithms are studied. PCA signal subspace estimationalgorithm is implemented on a single DSP. This essay includes the comparison of theperformance of PCA estimation and MUSIC-EVD. The results illustrate PCA subspaceestimate method is faster than MUSIC-EVD for more arrays, and the memory resources occupy is significantly less than MUSIC-EVD. Then, the DOA parallelcomputation algorithm based on PCA is implemented on4DSP.(4) The fast extremum solving methods are analyzed. The essay uses variablestep search method to speed the peak search in a single DSP. In order to avoidrepetitive search, the essay proposes to limit the scope of the small-step search. Thesimulation results show that the search veracity of improved variable step search andthe traditional search method are equivalent. When search points greater than1000points, variable step search is superior to traditional peak search. The peak search isimplemented on4DSP, through dividing search range, multi-chip data comparison,and interrupt control.(5) The methods of wideband DOA estimation are analyzed. The Auto-focusmethod without pre-estimate is studied, and the performance of it is analyzed throughmatlab simulation. It illustrates that the performance of this algorithm is subject to thereference frequency choise. So, in the essay proposes an advanced Auto-focus method.Using this method could eliminate the DOA bias caused by the improper referencefrequency selected. Based on the performance of narrowband DOA parallelimplementation, the essay presents an advanced Auto-focus parallel design.