Improved Unified Architecture for 3, 5, and 7-point Winograd Fourier Transform Algorithm

This paper proposes an improved unified architecture for 3, 5, and 7-point Winograd Fourier Transform Algorithm (WFTA). We define the mapping rules of inputs and outputs and modify the matrices in WFTA by matrix transformation. This design not only reduces hardware resource consumption by reducing multiplexers from 7 to 3, but also simplifies the control flow by performing data input and ports mapping through certain rules.