Abstract
Fast-moving, close-range interferers with high bearing rates transition through resolution cells rapidly, challenging the minimum variance distortionless response (MVDR) beamformer’s ability to accurately place notches in the interferers’ actual direction. This results in the MVDR notch slightly lagging behind the interferer’s actual location, reducing the beamformer’s ability to effectively suppress the interferers. A more effective approach is to apply a flatter and broader notch near the interferer’s location, accommodating the interferer’s motion. Increasing the notch width improves suppression of the interferer but reduces the array’s white noise gain (WNG). Conversely, decreasing the notch width preserves WNG but increases the risk that moving interferers will not be fully suppressed. Many existing methods address this challenge by employing fixed, wide notches in the direction of the interferer. This conservative approach effectively ensures attenuation but causes inefficient degrees of freedom (DoFs) use and significantly reduces WNG. This dissertation introduces two hybrid beamformers, hybrid double zero beamformer(HDZ) and hybrid MVDR beamformer (HMVDR), designed to address the suppression of moving interferers while providing a high WNG. The proposed hybrid beamformers suppress the moving interferers by creating implicitly wide notches in the beampattern. These notches result from the convolution of smaller adaptive and fixed weight vectors, which, in the spatial domain, corresponds to factoring the beampattern into adaptive and fixed components. The adaptive component implicitly places notches in the interferer direction for suppression and reduces some background noise, while the fixed component improves white noise gain. The HDZ and HMVDR beamformers differ in their approach to defining the adaptive component. The HDZ beamformer generates second-order notches by convolving the adaptive weights with themselves, while the HMVDR beamformer employs a first-order notch for the adaptive component. The factored beampattern enables the hybrid beamformers a flexible allocation of DoFs between the adaptive and fixed components. However, the primary challenge for hybrid beamformers lies in optimizing the allocation of DoFs for the adaptive component, particularly in dynamic environments with an unknown and time-varying number of interferers, to avoid overestimating or underestimating the adaptive notches. To address this challenge, we introduce a universal hybrid beamformer framework that encompasses the HDZ and HMVDR beamformers. The universal beamformer integrates a mixture of experts universal algorithm into the hybrid beamformer, blending different factorizations to dynamically allocate DoFs between the adaptive and fixed components. Simulations and microphone array experiments confirm that the universal hybrid beamformer achieves better array output power and white noise gain than competing beamformers.