Novel Access Patterns Based on Overlapping Loading and Processing Times to Reduce Latency and Increase Throughput in Memory-based FFTs

Abstract

This paper presents novel access patterns for P-parallel N-point radix-2 memory-based fast Fourier transform (FFT) architectures. This work aims to reduce the latency and increase the throughput by changing the data order and/or choosing different places of the architectures to input/output data. In this way, we can eliminate the loading time and/or the time to collect the output data. This results in a reduction in latency and an increase in throughput. Likewise, the architectures use the same permutation circuits for each iteration, which simplifies the circuit. In addition to improvements in latency and throughput with different access patterns for memory-based FFTs, this work also offers bit-reversed or natural input/output alternatives.