Comparison of three nonlinear filtering schemes (Q2266356)

This paper introduces three numerical realization schemes for the nonlinear filtering algorithm, namely the Chebychev, Laguerre-Gauss, and the Hermite-Gauss quadrature schemes. Their efficiencies are compared between themselves, but not to the well known ''Extended Kalman Filter''. The Chebychev quadrature scheme (CQS) employs the oriented floating grid with finite domain of integration; the Laguerre-Gauss and Hermite-Gauss quadrature schemes (LGQS and HGQS) have the advantage of a simple floating grid with infinite domain of integration. A simple example of second order is presented. The more robust LGQS and HGQS algorithms required typically between 12 and 13 seconds of computing time per stage on the UNIVAC 1106 computer. The CQS algorithm required about 2.5 second per stage but was sensitive to the initial estimate.