Identification of voiceband data signal constellations using a divisive cluster algorithm

A divisive cluster algorithm has been applied to identify the type of signal constellation used by a voice band modem. Tests with simulated signals showed that this cluster technique is capable to classify a variety of voiceband signals in use (DPSK2A/B, DPSK4A/B, V.29, V.29Fallback (7.2 kbit/s) and V.32Fallback (9.6kbit/s)). Compared to former proposed algorithms to classify linear digital modulated signal constellations, this technique is able to classify both higher rated PSK and QAM signals at high to moderate signal-to-noise ratios (SNR). With respect to signal impairments such as phase jitter, residual carrier or incomplete equalisation the classifier has proven to be reasonably robust. This paper describes the applied cluster algorithm and provides examples of its performance.