Automatic Modulation Classification in Wireless Communications Using Multi-Aspect Graph Attention Network with Avg-TopK Pooling and Osprey Optimization

AMC stands for automatic modulation classification, a significant tool in wireless communication for identification of the signal modulations of immense relevance to dynamic spectrum management, interference identification and cognitive radio networks. Many of the previous AMC approaches include feature extraction and manual classification which are not particularly portable and accurate in other contexts of communications. Recently, with the development of deep leaning, attention based methods have brought obviously enhanced performance in perceiving comprehensive signal characteristics; but there are still problems in focusing to multi-aspect features and training at a large scale signal data with high computational complexity. In order to address this problem, this research introduces a Avg-TopK pooling layer along with a Multi-Aspect Graph Attention Network (MAGAN) for the purpose of the AMC while using the Osprey Optimization Algorithm for the optimization of the proposed model. The Avg-TopK pooling only keeps K most important feature representations and decreases noise levels for better signal modulation classification. MAGAN uses multi-aspect attention mechanism allowing the emphasize various properties of the signal while perceiving mutual connections between the features of the signal. Additionally, the proposed Osprey Optimization Algorithm improves the training time of the model through optimization of hyperparameters. The presented results obtained from experimental analysis of standard AMC performance metrics show that the proposed model provides better AMC classification capability and insensitivity to the noise and existence of multiple types of modulation modes in contrast to other AMC models. In light of this, this work aims at solving some of the major challenges facing AMC through presenting a highly adaptive, efficient and accurate solution for the ever evolving wireless communication systems. The introduced approach attains higher accuracy as 99%.