Optimizing Computation Offloading in Satellite-UAV-Served 6G IoT: A Deep Learning Approach

Abstract

Satellite networks can provide Internet of Things (IoT) devices in remote areas with seamless coverage and downlink multicast transmissions. However, the large transmission latency, serious path loss, as well as the energy and resource constraints of IoT terminals challenge the stringent service requirements for throughput and latency in the 6G era. To address these problems, technologies including space-air-ground integrated networks (SAGINs), machine learning, edge computing, and energy harvesting are highly expected in 6G IoT. In this article, we consider the unmanned aerial vehicles (UAVs) and satellites to offer wireless-powered IoT devices edge computing and cloud computing services, respectively. To accelerate the communications, Terahertz frequency bands are utilized for communications between UAVs and IoT devices. Since the tasks generated by terrestrial IoT devices can be conducted locally, offloaded to the UAV-based edge servers or remote cloud servers through satellites, we focus on the computation offloading problem and consider deep learning techniques to optimize the task success rate considering the energy dynamics and channel conditions. A deep-learning-based offloading policy optimization strategy is given where the long short-term memory model is considered to address the dynamics of energy harvesting performance. Through the theoretical explanation and performance analysis, we discover the importance of emerging technologies including SAGIN, energy harvesting, and artificial intelligence techniques for 6G IoT.