Path Loss Estimation and Jamming Detection in Hybrid RF-VLC Vehicular Networks: A Machine-Learning Framework

Abstract

Emerging vehicle-to-everything (V2X) networks employ machine-learning (ML) techniques to provide adaptive, reliable, secure, and low-latency communication. Previously proposed fitting-based path loss and rule-based jamming detection schemes in vehicular networks lack accuracy due to the highly mobile and complex vehicular environment. Standalone ML models are utilized for different regression and classification problems in vehicular networks; however, these models still provide limited accuracy in the case of limited datasets. This article proposes a hybrid learning framework for jamming detection and path loss predictions based on the successive usage of multiple deep neural network (DNN) blocks, in which each block extends the feature set of the succeeding block for efficient and fast learning. The proposed hybrid DNN model for jamming detection comprises three DNN blocks, that is, a multilayer perceptron (MLP) block and two sequential learning blocks with bidirectional LSTM and gated recurrent unit (GRU) layers. Similarly, for path estimation, the proposed hybrid model includes one MLP block followed by two bagged tree-based learning blocks. The proposed models for jamming and path loss prediction are trained and tested on a real-world dataset. In the IEEE 802.11p link, the proposed hybrid model has been demonstrated to classify injected jamming signals with a significantly improved accuracy of 90.4% compared to existing benchmarks, including the random forest (RaFo) classifier and the deep convolutional neural network (DCNN). Similarly, the proposed hybrid DNN model for path loss estimation outperforms existing RaFo and fitting-based models, with a mean absolute error (MAE) reduction of 4.5 and 0.9 dB in the case of vehicular visible light communication (V-VLC) and IEEE 802.11p links, respectively.