On the correction of the differential non-linearity arising from discrete values for a digital signal processing system

Abstract The differential non-linearity arising from digital signal operations has been examined. Serious degradation of the differential linearity is observed when a series of discrete integer data is processed by a digital system. Two practical correction methods have been proposed: one is to utilize random numbers for converting the discrete integer data into floating point data. The other method is to consider each integer number with its width in order to cover the number line until the neighboring discrete value. Experimental results show great improvements of the differential linearity for both of the two methods, however, the latter method gives a better result.