Processor-efficient FFT Implementation Scheme for Active Noise Control Applications

TLDR: An alternative method to compute the DFT is proposed, which distributes the computations over a span of several sampling instants, and is applied to the reduced delay-less frequency-domain block filtered-x least-mean-square algorithm.

Abstract: Most of the frequency-domain (FD)-based active noise control (ANC) applications involve the computation of several discrete Fourier transforms (DFTs). Conventionally, an N-point DFT of a sequentially arriving data is computed only after the arrival of the N th sample. For applications involving ANC, such an approach will overload the processor. In this paper, an alternative method to compute the DFT is proposed, which distributes the computations over a span of several sampling instants. As an example to prove the efficiency of the proposed algorithm, it is applied to the reduced delay-less frequency-domain block filtered-x least-mean-square (RD-FBFXLMS) algorithm, wherein about 24% (for a block length of 1024 samples) of the multiplications and about 29% of additions (which were supposed to have been done at the last sampling instant of each block) are shifted to earlier sampling instants during which the processor is idle. The percentage of computational redistribution will be higher for multi channel non-linear systems.