The efficiency of computational procedures for recursive digital filters

The paper contrasts a number of approaches for computing recursive digital filters, with a particular emphasis upon high sample rate designs such as are typical for real-time digital control applications. Firstly, a useful method is presented by which the precision of computation can be determined. This is then applied to first- and second-order filter examples. A discussion follows of suitable data types for a high level language approach, and computation times are determined as compared with the opposite extreme of a purely Assembly language approach. Two additional intermediate approaches are assessed: a totally high level approach but using a special purpose multiplication procedure, and a partially high level approach which employs an Assembly language multiplication routine. The comparison is broadened by giving computation times for the four approaches when implemented on three different types of processor system. Important conclusions regarding the choice between high and low level programming are drawn from the assessment of computation times.