Parallelizing small finite state machines, with application to pulsed signal analysis
Lee Barford · 2012
Due to the increasing dominance of multicore processors, performing measurement analysis on larger data sets using such parallel processors is of interest. The particular measurement analysis problem studied here is that of recognizing patterns that require further processing within a stream of measured data. That is, it is desired to search through a signal looking for a particular sequence of values, states, or events. Previous work has shown one particular case where such a pattern recognition (identifying transitions in pulsed waveforms) could be done using parallel scans. However, the previous work was based on intuitive arguments and also left open the question of how in general to derive the required associative operators given a pattern of measured data-derived symbols to be recognized. The novel contribution of this work is (1) to provide a theoretical explanation for those previous results in parallelizing pulsed waveform analysis that obtained correct results but had only an intuitive but not a theoretical explanation, and (2) to demonstrate that the explanation leads to a general method for parallelizing pattern finding with FSMs. This method is tested on actual measured data.