Deep Learning Empowered MAC Protocol Identification With Squeeze-and-Excitation Networks

Medium access control (MAC) protocol identification in the context of cognitive radio is a challenging issue, which is usually solved by the support vector machine (SVM) method. To avoid manual feature extraction in SVM and realize accurate identification, a graphical scheme is adopted in this work. Firstly, the identification problem is modeled to analyze the distinctions of four MAC protocols by exploiting time and power features of the received signal. Then, we propose a convolutional neural network (CNN)-based MAC protocol identification (C-MPI) method, which combines CNN with the time-frequency image that contains time, frequency and energy three dimensional information. Moreover, in order to further improve the identification performance, a modified CNN with a squeeze-and-excitation mechanism-based MAC protocol identification (CSE-MPI) approach is developed, which plugs the SE block, a channel-wise attention mechanism, into classical CNN. Numerical simulations are provided to demonstrate the effectiveness of the proposed approaches.