Automatic Modulation Classification for Cognitive Radio Systems using CNN with Probabilistic Attention Mechanism

Abstract

This paper studies automatic modulation classification (AMC) for cognitive radio systems. We propose a deep learning neural network approach enhanced with an intelligent attention mechanism to correctly classify, detect, and segment spatially distributed modulation data. AMC is achieved by training the neural network to focus only on specific significant regions learnt using the attention mechanism. The proposed approach is tested for detection efficiency and accuracy to distinguish different modulation data using the publicly available RML2016.10a dataset. The outcome shows the accuracy of the proposed scheme is comparable to other state-of-the-art deep learning algorithms with a reduced complexity. The real-time assessment of the temporal states is achieved based on the spectral characteristics of modulation constellation images at various signal-to-noise ratio (SNR) values. The model performance is evaluated using mean average precision (mAP), F1 score, and speed-accuracy trade-off.