Toward Intelligent and Sustainable IoT: A DRL Approach for Backscatter-Based Communication Under QoS Constraints

With the rapid proliferation of vehicular communications and the need to support intelligent Internet-of-Things (IoT) devices in dynamic transportation environments, enhancing the functionality of low-power sensors is essential to realize the vision of sustainable and connected road networks. In this paper, we investigate a deep reinforcement learning (DRL)-based framework to optimize the long-term energy efficiency (EE) for an energy-harvesting (EH) backscatter node (BN) operating in the vicinity of mobile vehicles transmitting data to a roadside unit (RSU). These vehicular users (VUs), serving as primary devices, generate ambient radio-frequency (RF) signals that are opportunistically exploited by the BN for simultaneous data transmission and energy harvesting via a quality-of-service (QoS)-aware non-orthogonal multiple access (NOMA) scheme. We formulate a joint optimization problem that captures the BN's reflection and time-sharing decisions under dynamic vehicular channels, while guaranteeing the QoS of the primary vehicular users. The problem is addressed in two phases: (i) a convex optimization-based derivation of optimal reflection and time-sharing coefficients and (ii) a DRL-assisted solution using the deep deterministic policy gradient (DDPG) algorithm to adaptively manage uplink transmissions under varying channel and energy conditions. Simulation results show that the proposed DRL-driven scheme significantly enhances the throughput performance of the BN in highly dynamic vehicular environments, outperforming conventional baselines. Simulation results show that the proposed DRL-driven scheme improves the backscatter node's long-term throughput by up to 45% and EE by 35% compared to conventional strategies, highlighting its effectiveness for sustainable vehicular IoT communications.