Empowering Energy-Sustainable IoT Devices With Harvest Energy-Optimized Deep Neural Networks

There is a growing demand for low-power network devices; therefore, enabling technologies for the Internet of Things (IoT) is significantly important. This paper proposed resource allocation by maximizing the harvested energy to substantially improve Energy Efficiency (EE) and regulate transmission power for the scheduled IoT devices. Energy Harvesting (EH) is a viable technology that enables long-term and self-sustainable operations for IoT devices. The Simultaneous Wireless Information and Power Transfer (SWIPT) has been proposed as a promising solution for maximizing EE while ensuring the quality of service of all IoT devices, where the ultra-low power devices harvest energy in Power Splitting (PS) mode. This paper applied the proposed Optimal Transmit Power and PS Ratio (OTPR) algorithm to maximize the EE for SWIPT based on the partial derivative of Lagrange dual decomposition methods. The algorithm jointly optimized the allocation of the channel, PS, and power control to solve the distributed non-convex and NP-hardness caused by co-channel interference. A novel training was proposed for Deep Neural Network (DNN) algorithms chain rules to minimize the loss function based on updating the parameters of the weights hidden layer and convergence training to achieve near-optimal performance and minimize unneeded label data. The simulation results showed that the DNN training for the chain rule provided a near-optimal performance EE with the shortest training time. This observation indicated that decreasing the loss function at every training optimizes the co-channel conditions for IoT devices by assigning the EH requirement to meet the minimum harvesting need.