Transputer implementation of tracking Kalman filters

The development of efficient parallel implementations of the tracking Kalman filter on networks of transputers is described. The approach is based on using the structure and sparsity of the system matrices to partition the filter equations. Both the more numerically robust square-root filter and the regular covariance filter are considered and, while the regular form requires twice the number of bits for the same numerical accuracy, it is shown to be much less computational expensive. The performances of the transputer based filters are found to compare favourably with results reported elsewhere in the literature. >