Automated real-time active sonar decision making by exploiting massive parallelism

Multibeam active sonar systems acquire huge amounts of data on each ping. For example a 24 beam system in the 20 kyd search mode, which was used in the demonstration of the real-time automated system, would have to perform nearly a million FFTs per ping cycle just to do matched filtering. In order to accomplish real-time decision making the algorithms were developed on a massive parallel computer. The actual computer utilized was MASPAR´s MP-2 with 4096 parallel elements. System development utilized Fortran-90 and MP-2 library routines. The paper discusses the real-time decision making requirement for a 24 beam active sonar system, describes the sequential detection algorithm and the sequential classifier algorithm, the parallel array mapping algorithm that was the key to achieving real-time throughput. In the search mode of an active sonar system all possible range and bearing locations are processed for targets for each ping. A decision making algorithm then decides if the target is of interest. It was demonstrated that a sequential detection algorithm and a sequential classifier algorithm could process data in real-time that automatically detects and classifies targets by integrating over several pings. However on each ping decisions can also be made. Both the sequential detector and classifier can be designed to look for low SNR targets or high SNR targets. These design settings affect the performance of the algorithm. The purpose of this paper is to describe the automated algorithms and the parallel array mapping algorithm for a 24 beam active sonar system operating in a 20 kyd range mode