AI-ML Model Identification via Unique IDs in 5G Wireless Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In the 3GPP 5G new radio (NR) network, identifying and managing AI-ML models across wireless network entities is challenging due to data volume and complexity, hindering common understanding and efficient model selection, activation, and fallback between user equipment (UE) and network entities.
Innovation Solution
The implementation of AI-ML model identification using model IDs, where UE receives AI-ML models and related information from an AI server, and the RAN node maps and configures these models based on unique IDs, facilitating model management and control through procedures like RRC and NAS, ensuring consistent model identification across the network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI-ML models are delivered with complete model information from AI server to UE, then model identification accuracy is improved, but data transmission volume and network overhead increase
Solution Approach 1:
The patent extracts only the essential model identification information (model ID, model type, version) from the complete AI-ML model data, separating this identification metadata from the full model parameters. This allows the network to transmit minimal identification data rather than complete model information, resolving the contradiction between identification accuracy and data volume.
Solution Approach 2:
The patent segments the AI-ML model information into distinct components: model identification information (ID, type, version) and model parameter data. The identification components are transmitted separately for model selection, while detailed parameters are handled separately during model activation, reducing overall transmission overhead while maintaining identification precision.
2Adaptability or versatility
If detailed model information is transmitted for each AI-ML model, then model selection capability is improved, but network signaling overhead increases
Solution Approach 1:
The patent implements partial information transmission by sending only the necessary model identification fields (model ID, model type, version) without transmitting complete model parameters during the selection phase. This partial action approach provides sufficient information for model selection while avoiding the excessive signaling overhead that would result from transmitting full model details.
3Reliability
If UE performs comprehensive model identification procedures, then model compatibility is improved, but processing complexity increases
Solution Approach 1:
The patent performs model identification information extraction and validation as a preliminary step before full model deployment. The UE first processes lightweight identification data (model ID, type, version) to determine compatibility, and only if compatibility is confirmed does it proceed to receive and process the complete model parameters. This preliminary action reduces processing complexity by avoiding unnecessary detailed model analysis for incompatible models.
Data Source
AI summary
Apparatus and methods are provided for AI-ML model/functionality identification. In one novel aspect, model ID is used to identify the AI-ML model. In one embodiment, the UE receives AI-ML model from an AI server, obtains related model information of the AI-ML model and provides the related model information to the wireless network. In one embodiment, the AI-ML model and the related model information are delivered together. The UE receives AI model and related information through NF. In one embodiment, the UE receives AI model and related information by UP traffic. In one embodiment, the UE provides the related model information to the RAN node through a RRC procedure, or to a core network (CN) node through a NAS procedure. In one novel aspect, the radio access network (RAN) node and CN of the wireless identifies the AI-ML model by model ID. In one embodiment, the RAN node assigns a model index for the AI-ML model mapping to at least one of the first model ID and the second model ID.