Machine learning for large-scale optimization in 6G wireless networks
作者：白琳 时间：2023年08月01日 22:10 点击数：
语言：English
发表刊物：IEEE Communications Surveys and Tutorials
发表日期：2023年8月1日
摘要：
The sixth generation (6G) wireless systems have recently attracted considerable attention from both industry and academia, whose visions are towards ubiquitous 3D coverage (space-air-ground-sea integrated network) [1], the intelligent and green networks [2], Internet of Everything (IoE) [3], etc. Compared with the previous generations, 6G can provide services with more stringent requirements, such as higher throughput, lower latency, higher reliability, denser connection, higher energy efficiency, as well as connected intelligence with machine learning capability [3]. Driven by the new industrial and technological revolution, 6G can also support new services/applications beyond 5G such as immersive cloud extended reality (XR), holographic communications, sensory interconnection, digital twins, metaverse, and so on [4], [5], which may demand new performance metrics to facilitate diversified and personalized user services. The requirements of 6G system have made the fine-grained optimization of radio resources and effective learning of network-related information urgent necessities. Due to the large-scale, high density, heterogeneous qualities of services, and integrated multi-functional cross-layer design, the optimization problems in 6G can be extremely time-sensitive and complex, which pose great challenges for efficient optimization algorithm design. Machine learning (ML) has been recently leveraged as a disruptive technology to solve challenging optimization problems in 6G, as well as support ubiquitous artificial intelligence (AI) services and IoE applications [4], [6], [7] including synaesthesia Internet, digital twins, smart industry, smart agriculture, super traffic, precision medicine, and blockchain economy. In this section, we first discuss the properties of optimization problems in 6G wireless networks and summarize the advantages and disadvantages of classic optimization-based methods. Then we introduce the motivation for ML-based optimization frameworks and summarize the existing design paradigms to solve different classes of optimization problems. Table I summarizes the main notations used throughout this paper.
原文链接
