Performance for randomly described arrays

The complex relationship between irregular microphone distributions and its spatial gain hinders the efficient use of beamforming in various applications. This paper examines beamformer performance based on statistical descriptions of microphone geometries. In particular, array centroid, dispersion, and entropy based statistics are used to characterize irregular geometries relative to beamformer focal points. Simulations are performed to determine relationships between geometry descriptors and beamformer performance for speech applications. Descriptors are used in a genetic algorithm (GA) for optimizing microphone placements for various acoustic target and noise source distributions. Results show significant SNR enhancements for optimized irregular arrays over comparable regular arrays. GA objective functions using proposed geometry descriptors, result in rapid convergence to superior performing arrays, indicating a strong correlations with array performance.