Digital Modulation Scheme Recognition Technique Using Minimal Radial Basis Function Neural Networks for ISI Channels

Modulation recognition is extremely important in communication intelligence applications for several reasons. At the moment, the most attractive application area is radio and other re-configurable communication systems. This work describes an attempt to classify six digital modulation schemes using a gaussian radial basis function neural network with smaller complexity using efficient identifier and a localized generalization error model to improve the generalization capability of the network.. In this technique, higher order cumulants and sample kurtosis in addition to spectral features are utilized as the effective features. Tests and simulations using an additive white guassian noise and Raleigh fading channel show that the classifier has a success rate of 95.6% for signals with signal to noise ratio (SNR) equal to 5 dB with the computational units in the network equal to six. Also the evolved network with an additive white guassian noise consisted of four neurons with a classification efficiency of 99.6% for SNR of 10 dB and 99.3% for SNR of 5dB. Tests using mixed SNR ranging from 5dB to 15dB and Raleigh fading channel show that the classifier has a success rate of 91.2% with eight computational units. Simulation results show that the proposed minimal network has high performance for identification of the considered digital signal types even at very low SNRs and exhibits a great degree of generalization.