DRL-Driven Optimization of a Wireless Powered Symbiotic Radio With Nonlinear EH Model

• Published in: IEEE open journal of the Communications Society 2024, Vol.5, p.5232-5247
• Main Authors: Ullah, Syed Asad, Mahmood, Aamir, Nasir, Ali Arshad, Gidlund, Mikael, Hassan, Syed Ali
• Format: Article
• Language: English
• Subjects: Algorithms; Circuits; Cognitive radio; cognitive radio-inspired non-orthogonal multiple access (CR-NOMA); Convexity; deep deterministic policy gradient (DDPG); Energy efficiency; energy efficiency (EE); Energy harvesting; Integrated circuit modeling; NOMA; Nonlinearity; Nonorthogonal multiple access; Optimization; Parameter modification; Power management; Radio frequency; RF EH; Signal processing algorithms; Symbiosis; Symbiotic radio; Wireless communication
• ISSN: 2644-125X; 2644-125X
• DOI: 10.1109/OJCOMS.2024.3447152

Given the rising demand for low-power sensing, integrating additional devices into an existing wireless infrastructure calls for innovative energy- and spectrum-efficient wireless connectivity strategies. In this respect, wireless-powered or energy-harvesting symbiotic radio (EHSR) is gaining attention for establishing the secondary relationship with the primary wireless systems in terms of RF EH and opportunistically sharing the spectrum or schedule. In this paper, assuming the commensalistic relationship with the primary system, we consider the energy-efficient optimization of such an EHSR by intelligently making EH and transmission decisions under the inherent nonlinearity of the EH circuitry and dynamics of pre-scheduled primary devices. We present a state-of-the-art deep reinforcement learning (DRL)-engineered, energy-efficient transmission strategy, which intelligently orchestrates EHSR's uplink transmissions, leveraging the cognitive radio-inspired non-orthogonal multiple access (CR-NOMA) scheme. We first formulate the energy efficiency (EE) optimization metric for EHSR considering the nonlinear EH model, and then we decompose the inherently complex, non-convex problem into two optimization layers. The strategy first derives the optimal transmit power and time-sharing coefficient parameters, using convex optimization. Subsequently, these inferred parameters are substituted in the subsequent layer, where the optimization problem with continuous action space is addressed via a DRL framework, named modified deep deterministic policy gradient (MDDPG). Simulation results reveal that, compared to the baseline DDPG algorithm, our proposed solution provides a 6% EE gain with the linear EH model and approximately a 7% EE gain with the non-linear EH model.