Signal separation and super-resolution DOA estimation based on multi-objective joint learning

Houhong Xiang, Meibin Qi, Baixiao Chen, Shuo Zhuang

Abstract: The performance of traditional physics-driven direction-of-arrival (DOA) methods suffers from array imperfection, coherent signals, and computational complexity. In this paper, the multi-objective joint learning (MOJL) model is proposed to mine the latent joint features of multiple coherent signals and separate them into different subnetworks. The data of each signal are reconstructed and used for super-resolution DOA estimation according to the output of the subnetwork. The simulation results show that the proposed method can separate the data of multiple coherent signals with small errors. The statistical results demonstrate that the performance of the proposed method is superior to the traditional physics-driven methods and state-of-the-art data-driven methods in DOA estimation accuracy, and generalization to signal-to-noise ratio (SNR) and array imperfection.