Constraining error—A sliding discrete Fourier transform investigation

The sliding discrete Fourier transform provides an alternative to the FFT, permitting a custom choice of frequency decomposition which outputs an update after each input sample. The technique relies on the application of the Fourier shift property, and is recursive by nature. This work investigates the error performance of alternative techniques (SDFT; gSDFT; mSDFT, rSDFT; and Douglas and Soh algorithms) under both floating point and fixed point arithmetic constraints. The results highlight that the sliding discrete Fourier transform with error correction provides consistent error performance over a range of test cases, and indicates the limitations applicable to all techniques.