Analysis of digital filtering algorithms implemented with signal processors

Reports the results in microanalysis of DSP digital filter programs. Some popular DSPs have been considered. The results may be very inviting from the designers point of view. The major filter speed differences refer to the structures, where dedicated instructions are used for a given DSP (e.g. transversal structure realized by TMS32010 is four times slower as compared with the looped TMS32020 realization). Furthermore the number of multiplications in the filter structure is not a critical parameter of the structure. More important is the number of transfers and processor registers. Filters, for which the programs are optimized for low consumption of program memory are slower from direct realizations, if there is no hardware loop available. The original point of the paper is the systematic and impartial comparison, allowing the reader to estimate the processor throughput for a given filter structure and DSP