Authors: Assistant Professor CH.Sravan Kumar, G.Niharika
Abstract: Automatic identification of wireless communication signals is a fundamental capability of intelligent radio systems, enabling efficient spectrum utilization, interference management, network monitoring, and electronic surveillance. Conventional automatic signal identification (ASI) techniques, including likelihood-based and feature-based methods, often experience limitations associated with high computational complexity, sensitivity to channel variations, and reduced robustness under practical deployment conditions. Recent advances in machine learning have demonstrated significant potential for improving signal identification accuracy while reducing computational overhead. This study presents an Extreme Learning Machine (ELM)-based framework for the automatic identification of cellular signals using real Power Spectral Density (PSD) measurements collected over the air. The proposed methodology processes PSD measurements belonging to Global System for Mobile Communications (GSM), Universal Mobile Telecommunications System (UMTS), and Long-Term Evolution (LTE) networks by transforming PSD measurements into two-dimensional binary images through a PSD image-mapping process. Following preprocessing and image standardization, the generated signal images are supplied to an ELM classifier that utilizes randomly initialized hidden-layer parameters and analytically determined output weights to achieve rapid model training. The proposed model is evaluated using two independent datasets, DS1 and DS2, where DS1 is employed for hyperparameter optimization and DS2 is used to evaluate robustness and generalization capability. Experimental results demonstrate that the proposed ELM framework achieves high identification accuracy while significantly reducing training complexity compared with existing neural network approaches. These findings indicate that the proposed model provides an efficient and reliable solution for intelligent wireless communication systems requiring fast and accurate cellular signal identification.
