A novel voiceband QAM constellation discrimination technique

We describe a novel technique for discriminating between two or more voiceband QAM signals that have identical carrier frequencies and symbol rates but different constellations. The characteristics of the expected signal classes are stored in the form of histograms that record the frequencies of all possible amplitude changes and phase changes of subsequent symbols. The histogram of a received signal of unknown class is cross-correlated with all the stored histograms. The resulting scores are combined using a discriminant function to yield a classification decision. The new QAM discrimination method may offer several advantages over existing methods due to the fact that the classification decisions are based on a large number of symbol to symbol changes. Specifically, the relative nature of the symbol to symbol measurements and the finite bin sizes used in the histograms make the proposed method substantially immune to receiver errors related to carrier recovery, symbol timing recovery, channel equalization, and channel noise.