An FPGA-based fully hardware Kalman filter has been designed and prese
nted and a reconfigurable Kalman filter-based coprocessor in FPGAs has
been proposed. High-speed arithmetic function implementations and pip
elining have been used and a substantial improvement in performance ha
s been gained. The cycle time (one iteration) for computing Kalman fil
ter is reduced from 1.8274 mu s in our previous design to 0.4013 mu s.
The performance gained in our approach includes two to four orders of
magnitude higher speed than other implementations. The high-speed, re
congifuration and easy-to-develop characteristics of the FPGA-based Ka
lman filter will largely broaden the real-time application area of Kal
man filter. (C) 1997 Elsevier Science B.V.