Convolutional Neural Network Architectures for Modulation Scheme Classification in RF Signals within Cognitive Radio Systems

Dynamic Spectrum Access (DSA) plays a crucial role in cognitive radio (CR) by effectively utilizing unused spectrum and meeting the escalating traffic demands of advanced cellular networks like 5G and beyond. A significant focus in recent years has been on employing Deep Learning (DL) architectures for Radio Frequency (RF) signal classification in CR-based applications. This paper presents the development of a DL-based framework using Convolutional Neural Network (CNN) architecture to classify various modulation schemes, including Continuous-phase frequency-shift keying (CPFSK), Gaussian Frequency Shift Keying(GFSK), 64 quadrature amplitude modulation (QAM), Binary Phase-shift keying (BPSK), 16QAM, Gaussian Minimum Shift Keying (GMSK), QPSK of a signal received from the base stations in Global System for Mobile communication (GSM) network. The performance evaluation of the proposed CNN architecture will be conducted using real-time GSM signals obtained from nearby base stations. The performance analysis of each network architecture is presented and analyzed.