Deep learning-based selective spectrum sensing and allocation in cognitive vehicular radio networks

Abstract

The main challenge with Vehicular Ad-Hoc Networks (VANETs) for assisting Intelligent Transportation Services (ITSs) is ensuring effective data delivery under various network circumstances despite the scarcity of radio frequency spectrum channels. Meanwhile, Dynamic Spectrum Access (DSA) utilizing Cognitive Radio (CR) technology shows its potential to solve the channel shortage issue. Reliable Spectrum Sensing (SS) and Opportunistic Spectrum Allocation (OSA) are the two most important and challenging aspects of deciding the success of the deployment of CR-enabled VANETs (CR-VANETs). In this paper, we propose three relevant issues of CR-VANETs in a single framework, i.e., reliable Cooperative Spectrum Sensing (CSS), channel indexing for selective SS, and best channel allocation to the CR users. In CSS, we use the local SS decision with more critical attributes like the geographical position of sensing signal acquisition and timestamp to obtain the global CSS session using the Deep Reinforcement Learning (DRL) technique. Selective channel-based spectrum sensing is essential to minimize the sensing overload by CR users. The present work uses the time series analysis through the deep learning-based Long short-term memory (LSTM) model for indexing the primary user channels for selective SS. Finally, for the channel allocation to the CR-VANETs, we model the complex environment as a Partial Observable Markov Decision Process (POMDP) framework and solve using a value iteration-based algorithm. The simulation results show the proposed work's efficacy over the existing works.