Investigation Of Compression Technique applied for Automatic Modulation Classification Deep Neuron Network learning

The aim of this paper is to investigate efficiency of usage information theory methodes to train DNN for Automatic modulation Classification. Compression mechanism, based on Information Bottleneck method (IB), applied for training dataset, lets to retreive minimum sufficient information from input data and improve performance of the model. In the field of deep learning-based wireless communication, dataset RML2016 have been used to train model. Current research applies information theory principe, in the field of Automatic modulation Classification and investigates impact for resulting model accuracy and time for learning and prediction process. CNN model was taken for research.