Generation and Deep Learning-Based Classification of RF Signals Represented as I/Q Time Series

Generation and Deep Learning-Based Classification of RF Signals Represented as I/Q Time Series

연구 분야: Infrastructure

학회: 2025 IEEE International Mediterranean Conference on Communications and Networking (MeditCom)

In today's interconnected world, the radio frequency (RF) spectrum is an essential yet limited resource, requiring precise monitoring and efficient management. Although deep learning has demonstrated significant potential in RF signal classification and analysis, major challenges remain, including the limited availability and variability of labeled data, the need for large datasets, and difficulties in model generalization across different environments. To address this limitation, this paper presents a method for generating radio signals that comply with standard communication protocols, replicating the characteristics of real-world transmissions. Although this approach can be applied to generate signals compliant with any protocol, this study specifically focuses on data generated in accordance with the IEEE 802.11 standard. Furthermore, two CNN-LSTM neural network architectures are evaluated for classifying the generated signals based on multiple criteria: (1) identification of IEEE 802.11 standard version, (2) classification of modulation and coding schemes (MCSs), and (3) assessment of propagation conditions in varying wireless channel environments. The results indicate strong classification performance, validating the effectiveness and robustness of the proposed approach across various wireless communication scenarios.