Deep Reinforcement Learning Algorithm for Smart Data Compression under NOMA-Uplink Protocol

One of the highly promising radio access strategies for enhancing performance in the next generation cellular communications is non-orthogonal multiple access (NOMA). NOMA offers a number of advantages including better spectrum efficiency. This paper focuses primarily on proposing an energy efficient system for transmitting medical data, such as electroencephalogram (EEG), collected from patients for the sake of continuous monitoring. The framework proposes the use of deep reinforcement learning (DRL) to provide smart data compression in uplink-NOMA protocol. DRL enforces the data compression ratios for the nodes in order to avoid outage constraints at any sensor node. Jointly, it optimizes the power consumption of these sensor nodes. The data compression for such sensor network is vital in order to minimize the power every sensor consumes to maximize its service lifetime. We minimize the expected distortion under practical channel realization and outage probability constraints using NOMA-uplink protocol. Meanwhile, we optimize the power efficiency of the user node in order to increase the battery lifetime.