Adaptive Beamforming Algorithm Based on Automatic Deep Neural Network Optimization for Multiple Noise Signals

Abstract

As modern communication systems demand increasingly higher speed and accuracy in signal processing, traditional adaptive beamforming algorithms face challenges in real-time response to rapidly changing, multiple-noise signal environments. To address this issue, this paper proposes an Automated Deep Neural Network Adaptive Beamforming (A-DNNABF) algorithm for multiple noise signal environments. A-DNNABF uses the angle of arrival vector as input, employs an attention mechanism, and optimizes network architecture via Differentiable Architecture Search. Simulation results show A-DNNABF outperforms traditional Minimum Variance Distortionless Response (MVDR) and Deep Neural Network Adaptive Beamforming (DNNABF) methods in computational efficiency (10 times faster), prediction accuracy, and robustness to varying interference sources. The algorithm maintains stable performance with changing numbers of interference signals, demonstrating lower angular deviation in estimating both desired and interference signals. A-DNNABF provides an efficient solution for real-time adaptive beamforming in rapidly changing, multiple-noise signal environments.