Parallel computation of the modified extended kalman filter

In this paper, we describe certain techniques for mapping the modified extended Kalman filter (MEKF) onto systolic array processors. First, we introduce a square-root algorithm based on the singular value decomposition (SVD) for the Kalman filter. Then, we develop a VLSI architecture of the systolic array type for its implementation. Compared with other existing square-root Kalman filtering algorithms, our new design is numerically more stable and has nicer parallel and pipelining characteristics when it is applied to the MEKF. Moreover, it achieves higher efficiency. For n-dimensional state vector estimations, the proposed architecture consists of O(3/2n 2) processing elements and completes an iteration in time O((s + 8)n), in contrast to the time complexity of O((s + 3)n 3) for a sequential implementation, where s ≈ log n.