Application of parallel processors to real-time sensor array processing

Historically, most radar sensor array processing has been implemented using dedicated and specialized processing systems. This approach was necessary because the algorithm computation requirements were several orders of magnitude higher than any commercial supercomputer could provide in an acceptable size, weight and power. Most recently, with the interest by the military to employ commercial-off-the-shelf (COTS) technology, and the rapid advances in computation throughput by COTS supercomputers, we are able to meet the present algorithm requirements with COTS massively parallel processors (MPPs). However, there are still many issues to be overcame to get the most out of a parallel processor for our application. Some of the important issues are the balance between high computation throughput and total exchange communication over the processor network, message passing or shared memory access with high communication throughput for small message sizes, and the scalability of these machines to more demanding algorithm and application requirements. In this paper we present the application requirements, review the important computational kernels, and conclude with our experiences in evaluating representative MPPs.