NG-RAN ML-Split Signaling for Fast UE Reconfiguration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The distribution of machine learning (ML)-assisted functionalities between next generation random access networks (NG-RAN) and user equipment (UE) is complicated due to varying computational capabilities and coordination challenges, especially in radio resource management (RRM) functions, necessitating improved techniques for distributing ML-split setups.
Innovation Solution
A distribution framework for ML-assistance is introduced, where a core network entity, such as an ML-split orchestrator (MSO), provides ML-split setups to network nodes and UEs based on their capabilities, using hierarchical RRC/MAC protocols for low-overhead and fast reconfiguration, including RRC, MAC CE, and DCI signaling to manage ML-split configurations and activations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If ML-processing is distributed between network nodes and UEs, then system gains in spectral efficiency and RRM function optimization are improved, but coordination complexity and signaling overhead increase
Solution Approach 1:
The patent segments the ML processing task into distinct components: network nodes handle ML model training and parameter estimation, while UEs execute simplified inference operations. This segmentation allows each component to operate at its optimal complexity level, improving spectral efficiency while managing coordination complexity through clear functional division.
Solution Approach 2:
The patent introduces ML-split setup data as an intermediary mechanism that carries configuration information between network nodes and UEs. This intermediary structure simplifies the coordination process by pre-configuring ML parameters and processing instructions, reducing real-time signaling overhead while maintaining optimization benefits.
2Adaptability or versatility
If multiple ML-split setups are provided for different UEs, then adaptability to varying computational capabilities is improved, but signaling data transmission complexity increases
Solution Approach 1:
The patent applies local quality by tailoring ML-split setups to individual UE capabilities rather than using a uniform approach. Each UE receives customized configuration data matching its computational power, enabling optimal ML processing adaptation while managing signaling complexity through targeted, capability-based configuration.
Solution Approach 2:
The patent utilizes parameter changes by adjusting ML model complexity and processing parameters based on UE capabilities. The system dynamically modifies operational parameters (such as model depth, computation intensity) to match individual UE strengths, achieving adaptability without proportionally increasing signaling overhead.
3Speed
If fast reconfiguration is implemented using hierarchical RRC/MAC protocols, then response time is improved, but protocol complexity increases
Solution Approach 1:
The patent implements preliminary action by pre-configuring ML-split setups and preparing reconfiguration parameters in advance through higher-layer RRC protocols. This allows faster execution at lower MAC layers during actual reconfiguration events, achieving high speed while managing protocol complexity through hierarchical preparation.
Solution Approach 2:
The patent leverages the hierarchical dimension of RRC/MAC protocols to separate reconfiguration tasks across different layers. Higher RRC layers handle setup and configuration, while lower MAC layers execute fast reconfiguration decisions. This dimensional separation enables fast response time while distributing protocol complexity across multiple hierarchical levels.
Data Source
AI summary
An improved NR-U framework allows a gNB to repeat RLM-RS with the same index over each sub-band of an unlicensed spectrum. For example, a RLM-RS with the same indices are repeated by the gNB over the sub-bands of a wideband channel in an unlicensed spectrum. By monitoring the different sub-bands, a UE may find one or more sub-bands in which the RLM-RS having a particular index is successfully transmitted at a subsequent time.