Digital Twin for Intelligent Network: Data Lifecycle, Digital Replication, and AI-Based Optimizations

Mobile networks are becoming increasingly complex as generations evolve, thus making manual organization and management no longer a viable option for network operators. To achieve intelligent deployment, management, operation, and optimization of networks, AI/ML-based approaches are gaining a tremendous amount of interest in both academia and industry for accomplishing the full potential of 5G, Beyond 5G (B5G), and the upcoming 6G. However, AI/ML in networks is in its infancy for many practical challenges, such as the lack of AI/ML training and running frame-works/platforms, and the potential risks of service degradation when training AI models involving trial-and-error in real networks. Digital Twin, which creates virtual surrogates of real networks, thereby deriving insights into their use, is a promising approach to address the challenges. Using digitally replicated twins, one can test a newly designed algorithm without affecting user experience, predict the network status, analyze what-if scenarios, and train/validate network operating AI/ML models. This article presents the practical challenges and considerations of making use of Network Digital Twin for network management and optimization in terms of the network data lifecycle, digital replication, and finally effective and real-world applicable AI/ML-based network optimizations. As a case study, the cell on/off energy-saving problem using AI and digital twin is presented along with the results of its trial on the real network.