Ultrasound DOA Estimation Based On Sparse Reconstruction In The Air And Its Parallelism

Since the unsatisfactory of traditional optical and electromagnetic devices in In view of the limitation of traditional optical and electromagnetic devices under harsh environments,such as the fog,rain,snow and strong radiation.This thesis describes the MEMS ultrasonic sound field characteristics of L-shaped array,then elaborates the DOA(direction of arrivals from)estimation based on the sparse reconstruction,and designs the parallel optimization for DOA estimation algorithm through the NVIDIA CUDA tool.In the practical application,the number of source is usually far less than the number of potential source signals,and only exists a non-zero value of source signal intensity in the corresponding spatial orientation.Due to the sparseness of the source signals sparse reconstruction algorithm can be used to address the DOA estimation,and the sparse model can be converted to l0 norm model according to the zero space conditions.Two methods of l0 norm model are investigated in the first method l0 norm model is further converted to l1 norm model and calculated by CVX toolkit;in the second method l0 norm model is directly calculated through the greedy algorithm by using compressed sensing theory.Two algorithms for calculating the l1 norm model were investigated.For the two-dimensional DOA estimation based on feature vector sparse decomposition,the sparse model is established through the array covariance matrix of the largest eigenvector and finally is calculated by using CVX toolkits.For the two-dimensional DOA estimation based on sparse representation of reduced covariance matrix,the noise elimination for the column of each array covariance matrix is conducted,then the column number of array covariance matrix is reduced,and DOA estimation is realized only by using the covariance matrix of certain columns.The two kinds of algorithm simulation experiment were carried out.Simulation results show that the error of estimation performance of both algorithms will be minimized when the signal-to-noise ratio,the snapshots and the angle interval reaches 5d B,100 and 1° respectively.For DOA estimation based on the compressed sensing,three key components of the compressed sensing were redesigned: signal sparse representation,design of measurement matrix,signal sparse reconstruction algorithm,and the orthogonal matching pursuit(OMP)algorithm and generalized orthogonal matching pursuit(GOMP)algorithm of greed algorithm were used to recover sparse signal respectively.Simulation results show that the error of estimation performance of both algorithms will be minimized when the signal-to-noise ratio,the snapshots and the angle interval reaches-5d B,10 and 1° respectively.Besides,the recovery performance as a function of measured values M(measure matrix dimensions)and sparse level K(the number of source)was simulated.For OMP algorithm,the sparse signals can be completely reconstructed when M reach 70,and K is less than 20.For GOMP algorithm,the sparse signals can be completely reconstructed when M reach 45,K is less than 40.Finally,the characteristics of DOA estimation algorithm based on compressed sensing was analyzed and the parallel optimization was conducted through Eigen and CUDA toolkits.Simulation results show that speedup ratio can reach up to 2 times.