Neural networks applied to the classification of spectral features for automatic modulation recognition

Abstract

The use of back-error propagation neural networks for the automatic modulation recognition (AMR) of an intercepted signal is demonstrated. In all, ten modulation types are considered and a variety of spectral preprocessors are investigated for feature extraction. For the given training and test sets, the Welch periodogram is found to give the best results. For classification, experimental results show that neural networks match and even outdo the performance of the conventional k-nearest neighbor (k-NN) classifier for this preprocessor. Moreover, optimization of selected neural networks is demonstrated using the optimal brain damage (OBD) pruning technique.<>