A fast algorithm for linear convolution of discrete time signals

A new, computationally efficient, algorithm for linear convolution is proposed. This algorithm uses an N point instead of the usual 2N-1 point circular convolution to produce a linear convolution of two N point discrete time sequences. To achieve this, a scaling factor is introduced which enables the extraction of the term representing linear convolution from any algorithm that computes circular convolution. The proposed algorithm is perfectly accurate provided that the chosen circular convolution algorithm does not introduce round-off errors. The analysis is supported by simulation examples for several typical application cases.