Optimizing a Fast Fourier Transform Algorithm Implemented with Integer Arithmetic
Matheus Martins Rodrigues, Aldário C. Bordonalli · 2023
Digital signal processing fixed-point arithmetic is widely adopted because of its lower execution cost than that of floating-point arithmetic. Since defining the number of bits ($NBW$) of fixed-points variables and its integer part ($NBI$) representation can be time consuming, a simulation-based tool for optimizing$NBW$and$NBI$is proposed. Its performance is evaluated for the 256-point Radix-2 Fast Fourier Transform algorithm. The results suggest that the evolutionary optimization step for$NBW$can be skipped when dealing with linear transformation algorithms, reducing the number of simulations required and, therefore, speeding up the tool.