A parallel decoupled Kalman filtering algorithm and systolic architecture

In this paper, a new parallel decoupled Kalman filtering algorithm and U-D factorized implementation are proposed. The algorithm is based on decoupling the time update and measurement update equations of the conventional Kalman filter by forcing the measurement update to lag the time update by one time step so that these computations can be done simultaneously on separate processors of a parallel computer. In order to get high numerical stability and efficiency, the U-D covariance factorization technique is introduced in the parallel decoupled Kalman filtering algorithm. In addition, the systolic decoupled Kalman filter architecture is developed based on mapping the Kalman filter recursions directly onto a linear systolic array. The simulated computation on a PD-100 parallel computer simulator is presented. The results show that the U-D factorized parallel decoupled Kalman filtering algorithm can not only remain the precision of conventional Kalman filtering, but also achieve a speedup of 1.8 over the conventional Kalman filtering, with a corresponding efficiency of 91%