Letter: Neural classifiers as optimal decision devices for warped M‐QAM signal formats

The paper is focused on the application of perceptron-based neural networks to optimal decision for M-QAM signals transmitted over nonlinear channels. The decision is performed by two cascaded blocks: the first denotes one of the 4 quadrants and the second selects one of the M/4 symbols. A specific initialization procedure has been adopted in order to avoid local minima. The classifier performance, for the specific application, has been evaluated by the CNR (Carrier to Noise Ratio) degradation due to nonlinearity (ΔC/N) for a target error rate P e = 10 -3 . The results, obtained by MonteCarlo simulations, denote optimal matching with respect to upper bounds obtained with some minor simplifying hypothesis, even if the overall method's effectiveness can be adequate only for mild nonlinearity conditions.