Entropy-Based Wireless Technology Recognition

Abstract

Wireless Technology Recognition (WTR) enables identifying wireless technologies from received radio signals and is important for spectrum monitoring and management systems. Existing approaches either rely on handcrafted signal features with limited robustness or computationally intensive Deep Learning (DL) models operating on high-dimensional In-phase and Quadrature (IQ) samples or spectrograms. In this paper, we investigate a different hypothesis: wireless technologies can be distinguished through the temporal evolution of their local time-frequency complexity. To capture this behavior, we propose a lightweight WTR framework based on two-dimensional dispersion entropy. Entropy values are extracted from successive spectrogram regions to characterize the evolution of local spectrogram complexity throughout a transmission and summarized into a compact feature representation for classification using a Multilayer Perceptron (MLP). The framework is evaluated on seven wireless technologies, namely IEEE 802.11ah, IEEE 802.15.4 SUN-FSK, IEEE 802.15.4 SUN-OFDM, LoRa, IEEE 802.11n, LTE, and 5G NR, at Signal-to-Noise Ratios (SNRs) ranging from -20 dB to 40 dB. Experimental results show that the proposed approach consistently outperforms IQ-based and Short-Time Fourier Transform (STFT)-based Convolutional Neural Network (CNN) baselines, particularly at low and moderate SNRs. Averaged across all evaluated SNR levels, it achieves 82.73% accuracy, corresponding to improvements of 23.8% and 14.6% over the IQ-based and STFT-based CNN approaches, respectively. Furthermore, the approach maintains its recognition advantage with data-augmented training while requiring substantially shorter training times. The results demonstrate that temporal evolution of local spectrogram complexity provides a robust technology-specific fingerprint and an effective, computationally efficient representation for WTR.