Accelerated frequency domain adaptive filter using modified DFT pair and its application to noise canceling

The frequency domain adaptive filter (FDAF) converges faster than the time domain adaptive filter (TDAF) even when an input signal is colored since the input autocorrelation matrix is approximately diagonalized by using the discrete Fourier transform (DFT). The FDAF using the modified DFT pair (MDFTP) has been proposed to enable DFT operation in nonblock processing. This paper presents the introduction of the spectral error method into the FDAF using the MDFTP to accelerate its convergence speed. The effectiveness is examined through computer simulation in the adaptive line enhancer (ALE). Furthermore, it is applied to a noise canceling system for speech signals. The system based on the ALE requires one microphone; therefore, it is suitable for miniaturizing portable electronic equipment. The ability of noise canceling is confirmed through computer simulation using a speech signal.