Error analysis and complexity optimization for the multiplier-less FFT-like transformation (ML-FFT)

Abstract

This paper studies the effect of the signal round-off errors on the accuracies of the multiplier-less Fast Fourier Transform-like transformation (ML-FFT). The idea of the ML-FFT is to parameterize the twiddle factors in the conventional FFT algorithm as certain rotation-like matrices and approximate the associated parameters inside these matrices by the sum-of-power-of-two (SOPOT) or canonical signed digits (CSD) representations. The error due to the SOPOT approximation is called the coefficient round-off error. Apart from this error, signal round-off error also occurs because of insufficient wordlengths. Using a recursive noise model of these errors, the minimum hardware to realize the ML-FFT subject to the prescribed output bit accuracy can be obtained using a random search algorithm. A design example is given to demonstrate the effectiveness of the proposed approach.