Unsupervised Learning Based Emission-Aware Uplink Resource Allocation Scheme for Non-Orthogonal Multiple Access Systems

Abstract

The densification of wireless infrastructure to meet ever-increasing quality of service (QoS) demands, and the evergrowing number of wireless devices may lead to higher levels of electromagnetic field (EMF) exposure in the environment, in the 5G era. The possible long term health effects related to the EMF radiation are still an open debate and requires attention. Therefore, in this paper, we propose a novel EMFaware resource allocation scheme based on the power domain non-orthogonal multiple access (PD-NOMA) and machine learning (ML) technologies for reducing the EMF exposure in the uplink of cellular systems. More specifically, we use the K-means approach (an unsupervised ML approach) to create clusters of users to be allocated together and to then strategically group and assign them on the subcarriers, based on their associated channel properties. Finding the best number of clusters in the PD-NOMA environment is a key challenge, and in this paper, we have used the elbow method in conjunction with the F-test method to effectively control the maximum number of users to be allocated at the same time per subcarrier. We have also derived an EMF-aware power allocation by formulating and solving a convex optimization problem. Based on the simulation results, our proposed ML-based strategy effectively reduces the EMF exposure, in comparison with the state-of-the-art techniques.

ML-Based Radio Resource Management in 5G and Beyond Networks: A Survey

“…In [91], the authors propose an integrated scheme for resource management in NOMA environments. The first stage of the algorithm refers to the users' grouping and subcarrier allocation, while the latter one to the power control.…”

“…The second criterion is the usage of these algorithms in RRM-related KPI prediction task in 5G/B5G networks, according to the presented literature in the previous sections (i.e. [79], [80], [81], [89], [90], [91]). More specifically, using the Lumos-5G dataset [112], the problem of throughput prediction is investigated.…”

“…In other words, accuracy is the sum of True Positive (TP) and True Negative (TN) predictions, divided by the number of the total predictions (TP + TN + False Positive (FP) + False Negative (FN)). Then, F1-score is given by the following formula: [83], [84], [85], [86], [87] Unsupervised Learning RN or BS selection unsupervised k-NN, k-Means clustering variations [88] Unsupervised Learning user grouping, clustering, handover management k-NN, k-Means, Agg-GNN, f-test [89], [90], [91], [92] Reinforcement Learning subcarrier allocation, power control, frequency selection MDP, DRL, Water-Filling, WMMSE [93]. [94], [95], [96], [97] Reinforcement Learning minimization of difference between requested and active KPIs (throughput, SNIR, CSI)…”

Enhancing URLLC in Integrated Aerial Terrestrial Networks: Design Insights and Performance Trade-offs

“…Considering delay-sensitive applications, NOMA-assisted aerial terrestrial network in conjunction with a finite blocklength regime is also considered a key enabler for URLLC. Authors in [8] exploit the intrinsic attributes of power domain NOMA (PD-NOMA) to accommodate maximum devices without compromising the weighted sum-rate (WSR). Driven by real-time benefits, the concept of hybrid NOMA-OMA is also taken into consideration to support aerial-based communication.…”

“…Then, bisection search method is applied to search the range until an optimal values of blocklengths are computed. Secondly, we solve the problem (8) with fixed values of (θ (i+1) , B (i+1) ) to find next optimal value of P (i+1) .…”