Development and evaluation of an adaptive algorithm for predicting tank motion

This paper discusses the development and evaluation of an adaptive filter for predicting tank motion during the time-of-flight of a projectile. Tank accelerations are assumed to be the output of stationary Markov processes. The parameters of these models are determined by a combination of spectral analysis and maximum likelihood identification using tank tracks obtained under tactical conditions. The determination of model parameters and structure provides a case study of several complimentary features of different types of identification procedures. The various motion models corresponding to different tanks and tests were examined for their similarities and a reduced set of four models was chosen. These four models were used in a parallel bank of extended Kalman filters as an adaptive tracking filter. The filter with the greatest likelihood function at the time of firing was assumed to have the best motion model and thus its state vector was used to determine the lead offset of the gun. A Monte Carlo evaluation of hit probability was made for the adaptive filter and for conventional first-order prediction. The results demonstrate the superiority of the adaptive filter. The final phase of this effort involves the implementation of the adaptive filter on a microprocessor.