RF Jamming Detection in Vehicular Wireless Networks Using Deep Learning Approaches: A Comparative Study
TL;DR
Findings suggest that whilst CNN-LSTM architecture offers amazing robustness to velocity changes in comparison with other models, higher relative speed produce more unique noise patterns.
Abstract
This study presents a comparative study with regard to RF jamming detection in Vehicular Ad-hoc Networks (VANETs) using both deep learning (DL) and machine learning (ML) approaches. Provide a framework for classifying various jamming situations by exanimate conventional techniques (Gradient Boosting, Random Forest (RF), and Decision Trees (DT)) and DL architectures (CNN-LSTM, 1D-CNN). The method depends on a unique set of features comprising variations in PDR SNR, relative velocity (VRS), and RSSI. Two datasets with varying relative speed thresholds (15 m/s and 25 m/s) each with 3000 balanced samples exhibiting interactive attack, interference operation, and continuous attack were used to test the performance. Clearly surpassing traditional approaches, CNN-LSTM model shows remarkable performance (98% accuracy at 25 m/s, 92.67% at 15 m/s). These findings suggest that whilst CNN-LSTM architecture offers amazing robustness to velocity changes in comparison with other models, higher relative speed produce more unique noise patterns.