Learning-Based Resource Allocation for Backscatter-Aided Vehicular Networks

Wali Ullah Khan, University of Luxembourg
Tu N. Nguyen, Kennesaw State University
Furqan Jameel, Aalto University
Muhammad Ali Jamshed, University of Glasgow
Haris Pervaiz, School of Computing and Communications, Lancaster University
Muhammad Awais Javed, COMSATS University Islamabad
Riku Jantti, Aalto University


Heterogeneous backscatter networks are emerging as a promising solution to address the proliferating coverage and capacity demands of next-generation vehicular networks. However, despite its rapid evolution and significance, the optimization aspect of such networks has been overlooked due to their complexity and scale. Motivated by this discrepancy in the literature, this work sheds light on a novel learning-based optimization framework for heterogeneous backscatter vehicular networks. More specifically, the article presents a resource allocation and user association scheme for large-scale heterogeneous backscatter vehicular networks by considering a collaboration centric spectrum sharing mechanism. In the considered network setup, multiple network service providers (NSPs) own the resources to serve several legacy and backscatter vehicular users in the network. For each NSP, the legacy vehicle user operates under the macro cell, whereas, the backscatter vehicle user operates under small private cells using leased spectrum resources. A joint power allocation, user association, and spectrum sharing problem has been formulated with an objective to maximize the utility of NSPs. In order to overcome challenges of high dimensionality and non-convexity, the problem is divided into two subproblems. Subsequently, a reinforcement learning and a supervised deep learning approach have been used to solve both subproblems in an efficient and effective manner. To evaluate the benefits of the proposed scheme, extensive simulation studies are conducted and a comparison is provided with benchmark techniques. The performance evaluation demonstrates the utility of the presented system architecture and learning-based optimization framework.