code of paper: AoI-aware Efficient Energy Harvesting for D2D IoT Network via Multi-task MARL Mechanism

Abstract—A novel approach is proposed in this article to boost the energy efficiency (EE) of an AoI-aware IoT network. In particular, we propose a new approach that is based a combination of simultaneous wireless information and power

transfer (SWIPT) as well as energy harvesting (EH). Time switching (TS) is used by device-to-device (D2D) users to harvest energy from the surrounding environment, while power splitting (PS) is used by Internet of Things (IoT) users to compile energy from base stations (BS). It is our objective to investigate the EE optimization problem that takes into account the feasibility conditions for transmitting power between D2D and IoT users,

the minimum rate requirements for D2D and IoT users, as well as the joint frequency sharing and time allocations for D2D links. Intractable mixed-integer nonlinear problem (MINLP). The global optimal solution to the problem is highly nonconvex. The original problem can be reduced to three subproblems: 1) joint subchannel allocation and PS; 2) power control; and 3) time allocation. Due to the difficulty in finding an exact state model approach in a dynamic environment with a large state space, we propose an approach based on reinforced learning (RL) called modified federated learning. In this mechanism, not only do we transmit the weight of the neural networks but also the central server acts as a central trainer to foster the collaboration. The simulations show that the proposed algorithm, in addition to showing the superiority of the proposed algorithm over other methods in the literature, also demonstrates impressive EE gains due to the sharing of spectrum and energy harvesting that is possible via IoT and D2D devices.