Optimized UD filtering algorithm for floating-point hardware execution

The Kalman filter is an effective tool for fusing signals from multiple sources. The UD filtering is a well-known, numerically-stable formulation of the Kalman filter, owing to G.J. Bierman and C. Thornton. The most popular version of this filter is oriented to be executed in a traditional, sequential microprocessor. In this paper a new algorithm for the UD filtering is presented, specially designed for execution in hardware. It is based upon operations involving matrices and vectors, which is a more suitable approach for hardware optimization. To the best of the authors' knowledge, this is the first reported work about a fully UD filtering implementation in hardware with floating-point arithmetics. Since no previous works were found, the sequential UD filtering is synthesized as a benchmark. When compared with this sequential version, the UD filtering for hardware provides a speed-up of ~10x and presents a performance-vs.-area improvement by ~2x.