Hierarchical Transfer Learning for UE Channel-Adaptive AI Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing demand for high data throughput and diverse communication services in wireless communication systems, coupled with limited resources, necessitates a method to efficiently manage communication with high-density nodes and user equipment (UEs), reduce latency, and support various service requirements using artificial intelligence.
Innovation Solution
Implementing transfer learning (TL) by mapping UEs to different levels and groups based on channel characteristics, training neural networks to enhance communication efficiency and reduce latency, and transferring learning results among UEs within these groups.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the number of UEs and data throughput demand increase, then communication capacity requirements improve, but available radio resources become insufficient
Solution Approach 1:
The patent segments UEs into different groups based on channel characteristics and service requirements. By dividing the UE population into distinct segments, the system can allocate radio resources more efficiently to each group, allowing high-throughput services to receive priority resources while maintaining overall system capacity despite increasing UE density.
2Productivity
If AI training is applied to enhance communication efficiency, then service performance improves, but training time and latency increase
Solution Approach 1:
The patent performs preliminary AI training by pre-processing channel characteristics and organizing training data before actual communication operations. By preparing training models in advance based on historical channel data, the system reduces real-time training latency while maintaining improved communication efficiency from the trained AI models.
3Adaptability or versatility
If transfer learning is implemented across diverse UE groups, then adaptation to different channel characteristics improves, but system complexity increases
Solution Approach 1:
The patent applies local quality by transferring learning results specifically between UEs with similar channel characteristics within the same group, rather than across all UEs universally. This targeted approach maintains adaptability to different channel conditions while reducing system complexity by limiting transfer learning operations to relevant, homogeneous groups.
Data Source
AI summary
The communication device can perform hierarchical transfer learning (TL). The communication device may comprise: receiving information about channel characteristics from a plurality of user equipments including a first user equipment; mapping the plurality of user equipments to a plurality of TL levels and a plurality of groups for each of the plurality of TL levels on the basis of the information about the channel characteristics; and training a neural network of the first user equipment on the basis of the TL levels and the groups to which the plurality of user equipments is mapped. Training of the neural network of the first user equipment can comprise training a training portion for a first TL level from among TL levels to which the first user equipment is mapped.