A Parallel Beamforming System with Real-Time Implementation

For voice control applications, it is common to employ a microphone array to enhance received signals via beamforming techniques. In designing beamformers, different criteria will lead to different signal performance. It is known that speech recognition accuracy relies heavily on the trade-off between signal distortion and noise reduction. In this paper, we propose a novel beamformer structure which can give a continuous profile in signal distortion and noise reduction. The proposed structure combines two existing optimal beamformers to form the final filter. Moreover, since both optimal beamforming filters can be executed in parallel, a method is proposed to implement the noise reduction algorithm in the frequency domain. By studying the accuracy and efficiency of different modules, a hybrid fixed-floating point arithmetic is proposed within an FPGA hardware architecture to form an embedded system for industrial applications.