Algorithm Of Direction-of-Arrival Estimation Based On GPU

Direction of arrival is an important research field in array signal processing,and it is widely used in radar,sonar,communication,navigation etc.Determining the direction of arrival of a radiation source is one of the essential functions of an electronic surveillance system and an important prerequisite for the electronic warfare to carry out information warfare.However,the conditional high resolution direction-of-arrival estimation algorithms are computationally expensive,which limit the wide range of applications in practical engineering,including direction finding algorithm based on subspace,direction finding algorithm based on maximum likelihood estimation,MVDR algorithm.With the rapid development of parallel processors,graphic process unit(GPU)has gradually developed into a computing tool with ultra high floating-point opertations,and provide a strong guarantee for solving the big data processing.The parallel processing capability is determined by the large number of steam processors in the chip,which make the GPU have a greater advantage than the CPU when processing large amounts of data.The main researchs are summarized as follows:1.First of all,this paper introduces the basic signal model of array signal processing,and then research the basic theory of direction finding algorithm and analyse algorithm performance and computational complexity by simulation.2.The maximum likelihood estimation algorithm can get optimum direction finding performance in theory.However,it requires a multidimensional nonlinear search,which calls for a very large amount of calculation.Intelligent search algorithm is a random search algorithm that can find the optimal solution quickly and effectively.This paper studies the intelligent search algorithm,and propose to apply the cuckoo search algorithm and particle swarm optimization algorithm to the problem of direction finding.The simulation experiment shows that the proposed method can greatly reduce the computational complexity under the premise of ensuring the performance of direction finding algorithm.Furthermore,the execution time of direction finding can be greatly reduced by utilizing GPU owing to the natural parallelism of particle swarm optimization.3.The direction finding algorithm based on subspace decomposition has less real-time duing to the traditional subspace decomposition algorithms have high computational complexity.This paper proposes a new method that can fast compute signal subspace depend on repeatedly apply householder transformation to reduce the convariance matrix order based on the power method,and has fewer iterations and less amount of calculation.Furthermore,the time of calulating signal subspace can be further reduced by using GPU.4.MVDR algorithm is also a direction finding algorithm.However,it needs to compute the inversion of spatial correlation matrix which is computationally expensive.This paper proposes to decompose the spatial correlation matrix into two triangular matrices,and then compute the inversion of spatial correlation matrix with the inversion of triangular matrix.Furthermore,this paper studies the parallelization of a recursive algorithm for triangular matrix inversion,using the “divide and conquer” paradigm.