High-order covariance analysis for the classification of facsimile signalling tones

A method is presented for detecting and classsifying facsimile signalling waveforms on the general switched telephone network (GSTN). Linear prediction is employed to discriminate the signalling tones from speech and other tones on the network. A novel feature of the system allows for the detection of both the `fall-back´ rate and the full-rate transmission. This could enable the demodulation of the facsimile data, leading to a major saving in network bandwidth. A high-order predictor is required for fast detection and robust operation under noisy conditions. It was found that with high-order covariance analysis it is possible to detect V.27ter and V.29 signals rapidly and at signal-to-noise ratios worse than 15 dB