A fast algorithm for the on-line estimation of polarity correlation using binary patterns

Abstract

An algorithm which utilises the binary patterns of the polarity-sampled input signals for polarity correlation is presented. When programmed on a Z-80 microprocessor operating at 4 MHz, a double precision correlator which estimates 128 correlation coefficients with a maximum allowable sampling rate of about 2.5 kHz is realised. All the operations are performed on-line, except for an interval of about 15 ms of off-line processing to estimate the correlation coefficients from the binary patterns. This interval is constant, irrespective of the signal interval used. The efficiency of this algorithm is derived from the ability of microprocessors to handle 8 bits of data simultaneously, and from the fact when the sampling rate: input-signal bandwidth ratio is high (> 15, for the random signal tested) then only a limited set of input signal binary patterns are obtained. The resulting correlator was tested on real natural signals (the Electromyographic signal (EMG)).