Adaptive Space-time Broadband Beamforming using Machine Learning approach

Adaptive space-time broadband beamforming technology is a significant branch of array signal processing. For the conventional iterative algorithms of the broadband beamforming, the iteration step is always the fixed constant for each approach. In this paper, a variable step gradient algorithm based on machine learning is proposed for the adaptive space-time broadband beamforming, namely adaptive gradient Frost beamformer (AGFB) and adaptive gradient frequency constrained Frost beamformer (AGFCFB). Using the concept of adaptive gradient (AdaGrad) in machine learning (ML), the iterative formulas of adaptive broadband beamformers are ameliorated. The algorithm has the capability of adjusting the step of each approximation for different parameter values independently, making the entire process of iteration more flexible. Meanwhile, the convergence rates of the proposed beamformers are faster than the conventional algorithms. Simulation results highlight the effectiveness of the proposed methods.