Quality improvement of signals corrupted by additive white noise using Extended Kalman filter with quantile based noise variance estimation

Perceived quality of signal is degraded by the presence of additive noises. Hence we regard removal of these noises as quality improvement of the signal. There are many works in literature addressing this issue using adaptive filters. To this end Kalman filter and Extended Kalman filter are applied to the signals successfully in earlier works. But in these works the estimate of the corrupting noise, knowledge of which is a pre-requisite for the application of filter, is calculated from known regions of signal where only noise is present. Thus this method prevents real time application of the filter for signal quality improvement. We propose incorporating the quantile based estimation of noise variance to the extended Kalman filter so that it can be adapted for real time applications. For demonstration of filter effectiveness with this method, noise degraded audio signal is used as the input signal.