Dual-tone Multifrequency Signal Detection using Support Vector Machines

The need for efficient detection of Dual-tone Multifrequency (DTMF) tones for developing telecommunication equipment is justifiable. This paper presents an artificial intelligence based approach for efficient detection of DTMF tones under the influence of White Gaussian Noise (WGN) and frequency variation, using Support Vector Machines (SVM). Additive WGN in the DTMF input samples is removed by filtering out unwanted frequencies. Detection of DTMF carrier frequencies from input samples employs a traditional software based approach using the power spectrum analysis of the Discrete Fourier Transform (DFT) signals. The Goertzel's Algorithm is used to estimate the seven fundamental DTMF carrier frequencies. A SVM classifier is trained using the estimated fundamental DTMF carrier frequencies, and is validated using the input samples for classification of low and high DTMF frequency groups. The tone detection scheme employs decision logic using a rule-base expert system for classification of low and high DTMF frequency groups, corresponding to valid DTMF frequency groups. Comparison of this hybrid DTMF tone detection model with existing DTMF detection techniques proves the merits of this proposed scheme. This hybrid DTMF tone detection scheme is simulated in a MATLAB environment and results from performance tests are given in this paper.