CAWE-ACNN Algorithm for Coprime Array Adaptive Beamforming

Abstract This paper presents a coprime array robust adaptive beamforming algorithm based on attention convolutional neural network (ACNN), named as CAWE-ACNN algorithm. Firstly, a spatial attention unit and a channel attention unit are used to construct an attention network that can strengthen the features contributing to beamforming weight vector estimation from spatial and channel dimension, so beamformer output signal-to-interference-plus-noise ratio (SINR) is improved, and an ACNN model is established which is suitable for coprime array beamforming. Then, a virtual array of the coprime array is utilized to achieve direction-of-arrival (DOA) estimation, and a least squares method and a quadratic convex optimization problem are used to reconstruct the interference-plus-noise covariance matrix (INCM) and correct mismatched steering vector, respectively. Furthermore, the beamforming weight vector is calculated and it is regarded as the training label of the ACNN model, and the mapping procedure from covariance matrix to beamforming weight vector is completed. Finally, the well-trained ACNN model is utilized to obtain the beamforming weight vector. Simulation results verify the effectiveness of the proposed algorithm.