Decoupled Beamforming and Noise Cancellation

Abstract

The enhancement of noise-corrupted speech acquired by microphones is indispensable to the functioning of a wide variety of digital signal processing algorithms. Many existing products are equipped with steerable, stand-alone fixed beamformers which provide moderate levels of directivity. Moreover, many applications have long employed the classical adaptive noise canceller configuration with a reference sensor near the noise source to cancel unwanted noise. In this paper, the cascading of stand-alone beamformers with back-end adaptive noise cancellers is studied. A decoupled model for signal enhancement using front-end beamformers and cascaded noise cancellers is presented. The inter-operation of the beamforming and noise canceling units is studied by defining the signal-to-interference ratio (SIR) gain, directivity index, and white noise gain offered by the beamforming and noise cancelling components. The performance of decoupled beamformer-noise canceller structures is evaluated using experimental measurements. An experimental procedure for evaluating output SIR is presented. Results reveal SIR improvements of up to 27 dB, and are compared to those stemming from conventional adaptive beamformers