Large Array DOA Estimation Based on Extreme Learning Machine and Random Matrix Theory
Anqi Zhao, Hong Jiang, Qi Zhang · 2020
Estimation of the direction-of-arrival (DOA) in large array systems owns wide prospect in radar applications. However, the traditional subspace DOA estimation algorithm has deteriorated performance when the numbers of antenna elements and samples grow in the same rate. Also, they are poorly adapted to the actual low signal-to-noise ratio (SNR) environment. In this paper, we investigate the large array DOA estimation based on extreme learning machine (ELM) and random matrix theory (RMT). Conventionally, ELM requires the activation function to be infinitely differentiable, which may lead to slow training rate for large arrays. According to RMT, it is proved that ELM with ReLU function as its activation function has asymptotic convergence in large dimension regime. Thus, a ReLU-ELM method for large array DOA estimation is put forward. Numerical simulations show that under low SNR, it has better performance than the traditional algorithms. Compared with the least square support vector machine (LSSVM) method, the ReLU-ELM algorithm can greatly reduce the training and testing time and improve the learning efficiency under the premise of ensuring excellent estimation performance.