A neural network for robust LCMP beamforming

Calculating an optimal beamforming weight is a main task of beamforming. Quadratic constraints on the weight vector of an adaptive linearly constrained minimum power (LCMP) beamformer can improve robustness to pointing errors and to random perturbations in sensor parameters. This paper presents a neural network approach to the robust LCMP beamformer with the quadratic constraint. Compared with the existing neural networks for the LCMP beamformer, the proposed neural network converges fast to an optimal weight. Compared with the existing adaptive algorithms for the robust LCMP beamformer, in addition to parallel implementation, the proposed neural network is guaranteed to converge exponentially to an optimal weight. Simulations demonstrate that the proposed neural network has better interference suppression and faster convergence than the existing neural networks and the adaptive algorithms.