Processing performance of two Kalman filter algorithms with a DSP32C by using assembly and C languages

摘要

Two Kalman filter algorithms are implemented with a DSP32C processor. These two Kalman filters use conventional matrix operation and U-D factorization algorithms, respectively. The real-time processing performance of each algorithm is evaluated in terms of throughput, program and data memory sizes. Both DSP32C assembly and high-level C language programs of these two algorithms are developed (a total of four programs) for evaluating the coding efficiency. It is observed that both algorithms can be more efficiently programmed by using assembly language, a matrix-based algorithm enjoys its simple and regular operations so that less program memory is required in both assembly and in C languages, the U-D factorization algorithm involves fewer multiply-accumulate operations and provides a fast throughput in C language only, and the advantage of less multiply-accumulate operations in U-D factorization algorithm no longer exists in assembly language when the number of states of a Kalman filter is large.