Unmanned aerial vehicle (UAV)-assisted communication is becoming a streamlined technology in providing improved coverage to the internet-of-things (IoT) based devices. Rapid deployment, portability, and flexibility are some of the fundamental characteristics of UAVs, which make them ideal for effectively managing emergency-based IoT applications. This paper studies a UAV-assisted wireless IoT network relying on non-orthogonal multiple access (NOMA) to facilitate uplink connectivity for devices spread over a disaster region. The UAV setup is capable of relaying the information to the cellular base station (BS) using decode and forward relay protocol. By jointly utilizing the concepts of unsupervised machine learning (ML) and solving the resulting non-convex problem, we can maximize the total energy efficiency (EE) of IoT devices spread over a disaster region. Our proposed approach uses a combination of k-medoids and Silhouette analysis to perform resource allocation, whereas, power optimization is performed using iterative methods. In comparison to the exhaustive search method, our proposed scheme solves the EE maximization problem with much lower complexity and at the same time improves the overall energy consumption of the IoT devices. Moreover, in comparison to a modified version of greedy algorithm, our proposed approach improves the total EE of the system by 19% for a fixed 50k target number of bits.

翻译：无人机辅助通信正在成为提升物联网设备覆盖范围的一项流线化技术。无人机具备快速部署、便携性和灵活性等基本特征，使其成为有效管理基于应急场景的物联网应用的理想选择。本文研究了一种依赖非正交多址接入技术的无人机辅助无线物联网网络，旨在为分布在灾区的设备提供上行链路连接。无人机设备能够采用解码转发中继协议将信息中继传输到蜂窝基站。通过联合利用无监督机器学习概念并解决由此产生的非凸问题，我们可以最大化分布在灾区的物联网设备的总能量效率。我们提出的方法采用k-medoids聚类与轮廓分析相结合的方式进行资源分配，而功率优化则使用迭代方法进行。与穷举搜索方法相比，我们提出的方案以更低的复杂度解决了能效最大化问题，同时改善了物联网设备的整体能耗。此外，与改进的贪婪算法版本相比，在固定目标比特数为50k的情况下，我们提出的方法使系统的总能效提升了19%。