Adaptive Beamforming Algorithm Based on Reconstructed Covariance Matrix and Adaptive Projection Gradient Descent

This paper proposes an adaptive beamforming algorithm that updates the covariance matrix based on a recursive method. In beamforming technology, the estimation of the covariance matrix is a crucial aspect of all positioning algorithms. However, traditional covariance matrix estimations rely on static short-frame data, leading to biases in positioning results. Due to the dynamic nature of speech signals, this method replaces old data with new data to obtain a real-time covariance matrix, enhancing the accuracy of positioning results. It introduces a method for updating the old covariance matrix with new observational data. Additionally, mismatches in the steering vector can result in poor resolution of beamforming. To address this, we propose an adaptive projection gradient algorithm for weight estimation, which adaptively adjusts the learning rate and iteratively calculates weights, solving the issue of inaccurate calculation of steering vector weights, improving the resolution of beamforming, and achieving high-resolution imaging of sound source signals. Simulation results show that the proposed method effectively enhances the localization and identification of sound sources, further improving the resolution of beamforming.