Parallel implementation of the tracking Kalman filter using a network of transputers

In many radar tracking applications it is required to estimate the position, velocity and acceleration of a manoeuvring target using noisy measurements. Linear filters have been shown to be relatively inaccurate due to the dependence of the optimal filter gains on the target position. On the other hand the use of non-linear filters, such as the extended Kalman filter (EKF), can be computationally expensive leading to difficulties in real-time implementation. The aim of the paper is to describe techniques used in exploiting the concurrency within the tracking algorithms and to report on their implementation on a transputer array. The transputer implementations are all written using the Occam2 language and run using a motherboard with 4 TRAMs, each containing one T800-G20 transputer and 1 Mbyte of RAM. The relevant links were connected, in software, using the C004 programmable crossbar switch