Device-Level Learning-Based Distributed Power Control for Uplink NOMA in IoT Networks

Non-orthogonal multiple access (NOMA) addresses the massive connectivity issues in Internet of Things (IoT) networks operating under limited radio resources. This paper proposes a statistical learning-based novel power control mechanism for uplink data transmission in dense IoT networks where energy-constrained devices employ the power domain NOMA. The proposed mechanism has two stages and follows a distributed approach. The first stage uses an orthogonal multiple access (OMA) scheme, enabling devices to predict the number of active devices (network load) and device activation probability. In the second stage, devices utilize their network load prediction to determine the optimal number of transmission power levels for NOMA-based data transmission that meets the required latency-reliability constraint. Moreover, by using the probability of exception in network load prediction, devices are enabled to compute the optimal duration of the first phase. Results show that the average battery life in the proposed network load-aware adaptive NOMA scheme outperforms the conventional power-domain NOMA, where devices are unaware of the current network load and randomly select power levels. Moreover, when device activation probability is increased, the number of optimal power levels increases, consequently average battery life decreases. Furthermore, for the given number of installed devices, normalized mean square error (NMSE) in the device-level estimation of the current network load decreases as device activation probability increases.