Hybrid Computer Synthesis of Optimal Discrete Nonlinear Filters

Abstract : The total number of digital operations required to generate an estimate using a discrete optimal nonlinear filter is extremely large. The paper delineates the computer task and introduces an algorithm for mechanizing an optimal 1-step predictor using a hybrid computer to reduce the computation time. The advantages and limitations of the hybrid mechanization are described. Hybrid computer predictions (as simulated by a continuous system simulation language) are compared to those generated by an all digital mechanization. Timing estimates, accuracy estimates and computer equipment requirements are given. Extensions of the algorithm to multidimensional systems and observations are discussed. Use of a parallel digital multiprocessor to perform the algorithm is demonstrated. The use of interactive computer graphics for debugging and control of the algorithm is discussed. (Author)