Dual MVDR Architecture for Adaptive Cancellation of Dynamic Interference

Microphone arrays are widely used for speech enhancement applications. We consider the enhancement of multiple, desired and undesired, sources in a noisy environment, the approximate locations of which are assumed to be known a priori. For the given scenario, the linear constraint minimum variance (LCMV) beamformer is commonly used. Although an LCMV-based beamformer provides the optimal solution for the case of static sources, the problem becomes more challenging when the sources' locations are constantly changing. The LCMV spatial notch pointing at the interference location is usually very sharp, resulting in cancellation degradation in the case of even small movements of the interfering sources. We propose an alternative scheme to the traditional LCMV beamformer that efficiently tracks and cancels the interfering source. The scheme is presented for a dual source scenario. We prove that for the static scenario, the proposed method and the LCMV beamformers are mathematically equivalent. However, for practical uses, we demonstrate that the proposed algorithm outperforms the LCMV in terms of signal-to-interference ratio (SIR) using a simulated room environment, as well as in real recorded data.