Using Kullback Leibler Divergence to Perform Modulation Recognition on a TorchSig Dataset

This paper discusses a novel approach to modulation recognition (classification) of RF signals based on Kullback Leibler divergence calculations. The classification systems use small amounts of known (labeled) signals from each source to help determine the class of an unknown (test) signal. This method performs significantly better than a previous method that utilizes mutual information and may be suitable in scenarios where labeled data is unavailable in significant quantities. Testing is performed on clean signals as well as signals that are impaired with white Gaussian noise and wireless channel effects.