ANALYSIS OF ERRORS IN INTEGER COEFFICIENT DIGITAL FILTERS.

An analysis is presented of integer coefficient digital filters to identify sources of errors. Microprocessor implementation of the filters usually limits the word length to a finite 8 or 16-bit size. This could result in truncation or round-off errors. Fortunately, allowing filter coefficients in the integer form only eliminates the error due to numeric representation that can occur when more general floating-point numbers are allowed. Still, due to repeated additions and multiplications, there is a danger of an overflow. Overflow results in errors in the form of oscillations that persist for several samples. The authors present some examples of integer coefficient filters used to process noisy electrocardiogram recordings.