Abstract: In wireless powered communication networks (WPCNs), battery-free energy harvesting (EH) devices harvest energy from radio frequency (RF) signals and transmit information using the harvest-then-transmit scheme, leading to infrequent, irregular transmissions. Consequently, conventional medium access protocols, designed for devices with a stable power supply, are not suitable for EH devices as they impose stringent constraints and operational require- ments that cannot be met by EH devices. Therefore, it is imperative to design efficient, distributed, low-overhead medium access control protocols for WPCNs that account for irregular transmissions from EH devices.
Our work explores the application of adaptive energy beam steering to control both charging rate and access to the wireless medium in WPCNs under the slotted ALOHA. In particular, we examine the dynamics of the slotted ALOHA protocol in conjunction with the charging process of EH devices to develop a framework for maximising throughput through beam steering, despite partial observability constraints, such as unknown device locations,channel conditions and current charge levels. We devise deep reinforcement learning approaches, starting with a Deep Q-Network and then moving to a Deep recurrent Q-Network, which leverages past observations in its action decision process. We also design a global knowledge-based oracle policy for benchmarking purposes. Numerical results show that our proposed adaptive scheme outperforms the non-learning baselines by up to 68% in terms of throughput.Publications:
Journals:
S. Adhikari, N. Awasthi, S. Sarma, “Deep Recurrent Q-Learning Based Beam Steering Strategy for Throughput Maximization in WPCNs ”, [Under revision].
Conferences: