Assistant vehicle locating based on DOA estimation of deep unfolded network

Abstract

In view of the decline in the positioning accuracy of the vehicles with small snapshots, this paper proposes a direction of arrival(DOA) estimation method based on deep unfolded network to assist the global positioning system to achieve high-precision locating. Firstly, the signal emitted from the on board unit installed on the vehicle is received by the array antenna of the road side unit(RSU), then RSU transmits the array data to cloud service platform, in which the traditional fixed point continuation algorithm is improved by using l1 norm instead of l2 norm, and the adaptive outlier point tracking algorithm is used to deal with the problem of symbol flipping caused by noise. Then we unfold the improved fixed point continuous algorithm for obtaining a neural network. In order to improve the DOA estimation performance and convergence speed, this paper uses the black widow optimization algorithm to determine the soft threshold shrinkage parameters. According to the mapping between the signal and the array manifold matrix, the DOA of the target is estimated, and its location can be evaluated from the geometry relation between vehicle and RSU. The simulation results show that the algorithm proposed in this paper has better direction finding performance and improves the vehicle positioning accuracy under small snapshots and high signal-to-noise ratio effectively.