Fast Kalman Filtering for ARMA (Autoregressive Moving Average) Processes: Fixed Point Implementation.

摘要

Kalman predictors and filters are implemented in fixed point arithmetic on a 16-bit INTEL 8086 microprocessor. Results from this implementation are compared with corresponding results for a 16-bit floating point implementation on an 8-bit 8080 microprocessor. Both implementations are carried out using an INTEL MDS 230 development system. Floating point code is written in FORTRAN and fixed point code is written in Assembly language. The Kalman filters and predictors are realized in a fast form that uses the so-called fast Kalman gain algorithm. This algorithm for the gain is inherently fixed point. Scaling rules for Kalman filters and predictors are derived, and expressions are derived for rounding error variances. The numerical results show that low order fixed point realizations perform very close to the floating point realizations. (Author)