A Deep Reinforcement Learning Framework for Data Compression in Uplink NOMA-SWIPT Systems

Abstract

<comment< Non-orthogonal multiple access (NOMA) shall play an important role in the current and foreseeable design of 5G and beyond networks. NOMA allows multiple users to share the same time-frequency resources, thus allowing higher spectrum efficiency. In addition, massive deployment of Internet of Things (IoT) nodes for different applications calls for exploiting state-of-art protocols to allow for the increasing demand of spectrum. Therefore, NOMA is one key protocol for IoT applications. Those IoT nodes usually have limited access to energy source and are clustered around a cluster head (CH), which collects data from multiple nodes and transfers it to a fusion or data center. Data compression at the IoT nodes reduces the number of transmitted samples, which translates into reduction in energy consumption. Energy is typically a scarce resource in IoT nodes. Consequently, data compression at the IoT nodes could be essential in maximizing its lifetime. Another key technology is simultaneous wireless information and power transfer (SWIPT), which allows for the transfer of information and energy concurrently. We propose a framework that enables the CH to harvest energy from uplink transmission by IoT nodes employing data compression under NOMA protocol. Our framework enables the CH to maximize the harvested energy while meeting constraints on outage probability, consumed energies by the transmitting IoT nodes and compression and distortion ratios. We provide necessary analysis for our framework and derive closed form expressions for the outage probability and average harvested energy under NOMA protocol. We formulate an optimization problem with NOMA factors, SWIPT factor and NOMA user distances as optimization parameters. We first solve the optimization problem and find the optimized values using a grid-based search. Then we exploit deep reinforcement learning algorithm to solve the optimization problem more efficiently. Throughout this work, we prove the feasibility of such framework and deliver key observation that will help the CH scheduling different IoT nodes such that the harvested energy is maximized while the constraints are met. </comment< We propose a framework that enables the cluster head (CH) to harvest energy from uplink transmission by Internet of Things (IoT) nodes employing data compression under non-orthogonal multiple access (NOMA) scheme. Our framework enables the CH to maximize the harvested energy while meeting constraints on outage probability, consumed energies by the transmitting IoT nodes and compression and distortion ratios. We provide necessary analysis for our framework and derive an expression for the outage probability and average harvested energy under NOMA scheme. We formulate an optimization problem with NOMA factors, simultaneous wireless information and power transfer (SWIPT) factors, and NOMA user distances as optimization parameters. We first solve the optimization problem and find the optimized values using a grid-based search. Then, we exploit deep reinforcement learning algorithm to solve the optimization problem more efficiently. Throughout this work, we prove the feasibility of such framework and deliver key observation that will help the CH scheduling different IoT nodes such that the harvested energy is maximized while the constraints are met. IEEE