Modulation classification in fading multipath channel

A system for modulation classification in a dispersive communication channel is developed. The channel equalizer and symbol rate are blindly estimated from data. The features used for recognition include cumulants and moments computed from a polar coordinate representation of the IQ-plane. The actual classification is performed using a neural MLP classifier. Simulation results are presented using the WSSUS (wide sense stationary uniform scattering) channel model in typical urban and hilly terrain scenarios. Highly accurate recognition is achieved even at low signal-to-noise ratios. The importance of the equalization stage is clearly demonstrated.