A Bayesian Filtering Approach with Time-Frequency Representation for Corrupted Dual Tone Multi Frequency Identification

This paper addresses the identification of the Dual Tone Multiple Frequency (DTMF) signal from real environments where the signal was corrupted during acquisition and transmission processes by using the sequential Bayesian filtering along with the time-frequency representation of time-series. The mathematical and statistical models were employed to estimate the frequency content embedded in the real DTMF signal. A sequential state-space framework, by treating each frequency component as a target to be tracked, that is developed in this work for the extraction of time-frequency information from time slice spectrograms provides excellent results, stemming from an efficient representation of the DTMF signals in the frequency domain. The paper also illustrates the accuracy of the estimates by displaying the probability density functions (PDFs) of the frequencies obtained from the filter. The performance of the proposed approach was compared to those of the conventional method. The comparison demonstrates a significant benefit of our method for DTMF signal identification under various noisy environments.