DOA Estimation Based on an Adversarial Learning Network via Small Antenna Arrays

Abstract

As a key technology for radio monitoring and positioning, direction-of-arrival (DOA) estimation has garnered significant attention and has undergone in-depth research. This article proposes a new subspace-based DOA estimation algorithm based on an adversarial learning network. Considering the impact of the number of antennas in the signal-receiving array on the resulting DOA estimation accuracy, the proposed algorithm takes a covariance matrix corresponding to a small antenna array as the input of the adversarial learning network and reconstructs an extended covariance matrix corresponding to a virtual large antenna array. By introducing subspace technology, the multiple signal classification (MUSIC) algorithm can achieve high-resolution DOA estimation. Therefore, the extended covariance matrix corresponding to the virtual large antenna array is combined with the MUSIC to achieve DOA estimation. Simulated and real-world experimental results demonstrate that compared with conventional subspace-based DOA estimation algorithms, the proposed algorithm achieves significantly improved DOA estimation performance.