Scalable GMTI tracker

This paper describes the design and preliminary implementation of a scalable multiple hypothesis tracker on a distributed memory multiprocessor computer. The throughput requirements of the tracker strongly depend on the desired performance and can amount to several giga floating point operations per second. We show that individual track processing tasks constitute more than 80% of the overall processing, and the tasks for any one track can be performed independently of other tracks. This inherent parallelism allows the tracker to be partitioned into data parallel tasks that can then be processed on different processors. Distributed memory multiprocessor computers allow a user to increase the number of processors, the amount of input/output bandwidth, memory, and inter-processor bandwidth to support increasing processing needs of an application. We combine the inherent parallelism in the tracker and the flexibility of a distributed memory computer to construct a scalable tracker. The user can scale either the tracker´s performance or the number of targets tracked by adding more hardware (processor boards, I/O boards) and without making any changes to the tracker software.