Memory-Based Self-Ordering FFT for Efficient I/O Scheduling

Abstract

A complex-valued self-ordering radix-2 memory-based Fast Fourier Transform (FFT) architecture suitable for low end Field Programmable Gate Arrays (FPGA) is presented. Employing a self-ordering algorithm within the data flow, both input and output data are kept in normal sequential order, not in digit-reversed-order. This way, with an appropriate scheduling, last stage of the FFT and I/O operations are performed in parallel with no wait states. Self-ordering FFT algorithms are generally designed for software implementations. We designed and implemented one on FPGA (hardware), showing that considerable number of clock cycle savings can be obtained compared to unordered FFT counterparts. The approach is implemented on various FPGAs. The results are compared with similar radix-2 architectures in terms of required clock cycles and resource usage, confirming the advantage of the approach.

A complex-valued self-ordering radix-2 memory-based Fast Fourier Transform (FFT) architecture suitable for low end Field Programmable Gate Arrays (FPGA) is presented. Employing a self-ordering algorithm within the data flow, both input and output data are kept in normal sequential order, not in digit-reversed-order. This way, with an appropriate scheduling, last stage of the FFT and I/O operations are performed in parallel with no wait states. Self-ordering FFT algorithms are generally designed for software implementations. We designed and implemented one on FPGA (hardware), showing that considerable number of clock cycle savings can be obtained compared to unordered FFT counterparts. The approach is implemented on various FPGAs. The results are compared with similar radix-2 architectures in terms of required clock cycles and resource usage, confirming the advantage of the approach.