Juhyoung Sung, Kiwon Kwon, Kyoungwon Park, Journal of Internet Computing and Services, Vol. 26, No. 4, pp. 31-39, Aug. 2025
10.7472/jksii.2025.26.4.31, Full Text:  HTML
Keywords: Artificial intelligence, CNN, Data Preprocessing, modulation type classification, Radar signal, STFT, WVD

Abstract

The ability to accurately classify radar signal modulation types has become increasingly critical in the fields of electronic warfare, and spectrum management. As traditional signal analysis methods face significant limitations in adapting to the increasing complexity and diversity of modern radar signals, prompting artificial intelligence (AI)-based automatic modulation recognition (AMR) techniques have been attracted. This study aims to improve the performance of AI-based modulation classification models by analyzing the impact of different data preprocessing techniques on radar signals sampled at the IQ (In-phase and Quadrature) level. Two approaches are compared: one using a conventional model that directly flattens and inputs raw IQ data without preprocessing, and another utilizing a convolutional neural network (CNN) model trained on two-dimensional images generated by time-frequency analysis techniques—specifically, Wigner-Ville Distribution (WVD) and Short-Time Fourier Transform (STFT). Experimental results demonstrate that the CNN models fed with time-frequency transformed images significantly outperform the baseline model using raw flattened IQ data. Additionally, we compare the properties of the images generated by WVD and STFT-based preprocessing methods and the classification performance, thereby verifying how the choice of time-frequency transformation method can influence the performance of AI models. This study underscores the potential of time-frequency image-based preprocessing to overcome the limitations of traditional domain-specific approaches and substantially improve modulation classification accuracy. The proposed framework is expected to serve as a foundational technique for building scalable and robust radar signal recognition systems across diverse electromagnetic environments.

Statistics
Cite this article