TCN-Based Virtual Array Element Beamforming for Restricted Aperture Underwater Sensing Arrey

Virtual element beamforming technology is a means to enhance the array resolution of conventional beamforming algorithms. However, the linear predictor-based virtual element beamforming algorithm therein is equivalent to the early autoregressive models in time series prediction tasks, exhibiting limited capability in learning the patterns of data variations, particularly for small-scale arrays. Considering the emergence of deep learning techniques, time series prediction methods based on deep learning have gradually evolved. Therefore, this paper proposes a virtual element beamforming algorithm based on temporal convolutional network (TCN) neural network. This algorithm utilizes TCN neural network to learn the temporal characteristics between real element received signals and virtual element received signals, and enhances computational efficiency through GPU, thereby achieving efficient target detection. Simulation results demonstrate that, under small-scale arrays, compared to the linear predictor-based virtual element beamforming algorithm, the TCN-based virtual element beamforming algorithm not only improves the direction finding accuracy of target azimuth estimation but also enhances angular resolution.