Statistical Array Geometries for Real-Time Covert Surveillance with Parallel Computer Implementations

Abstract: The proposed work is a continuation of the work started in FY 2007/2008, which seeks to provide law enforcement agencies with enhanced abilities to covertly record and listen to remote conversations of interest (COI) in areas where multiple conversations and interfering noises are ongoing The technology involves a distribution of microphones mounted in non-regular patterns around an area or on moving platforms (i.e. undercover agents, trained animals, or small robotic devices), and using clusters of computers to provide near-real time processing for tracking and intelligibility enhancements of the COI. Year 1 focused on building a set of tools and data sets for assessment of near-field and non-regular arrays, and developing models for relating non-regular geometries to performances of beamforming and sound source location. Year 2 will continue with more aggressive development of beamforming algorithms exploiting both non-regular and near-field properties of the arrays in order the achieve target SNR and intelligibility enhancements with as few microphones as possible, an examination of noise and interference reductions with wavelet technology, and parallel algorithm implementations on low-cost/ low-power computer clusters for near real-time operations in the field.