Compressed RAN State Representation for ML Configuration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Integrating Machine Learning (ML) models into Radio Access Networks (RAN) operations poses challenges such as signaling complexities, resource management, and standardization delays, particularly in defining features for ML model execution across network nodes.
Innovation Solution
A method involving a first and second node in a communication network where the second node generates a compressed representation of parameter values describing a physical or radio environment, which is then transmitted to the first node to facilitate RAN operation configuration, using Reinforcement Learning to optimize configuration actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If ML models are signalled to UEs for execution, then resource saving at radio access nodes is achieved, but signaling cost and complexity increase
Solution Approach 1:
The patent extracts the ML model execution from the radio access node and places it in the UE, thereby removing the computational burden from the network side while retaining the benefits of intelligent resource management. The model is extracted and transferred to the UE for local execution.
Solution Approach 2:
The patent implements preliminary actions by pre-configuring the UE with ML models and their required input parameters before actual execution is needed. The UE is prepared in advance with the necessary computational capabilities and model parameters to execute ML functions without requiring real-time configuration.
2Adaptability or versatility
If frequent ML model signalling is performed, then model updates are achieved, but signaling overhead increases
Solution Approach 1:
The patent implements dynamics by enabling flexible, on-demand model updates triggered by specific events such as handover decisions or network conditions changes. The system dynamically adjusts model parameters based on actual operational needs rather than following fixed signaling schedules.
Solution Approach 2:
The UE performs self-service by autonomously selecting and executing appropriate ML models based on local conditions without requiring continuous network direction. The UE independently manages model execution and can trigger updates when local conditions change, reducing dependency on frequent network signaling.
3Measurement precision
If detailed parameter values are transmitted for ML model execution, then model accuracy is improved, but data volume increases
Solution Approach 1:
The patent applies local quality by transmitting only the specific parameters that are actually needed for each ML model's execution, rather than all available measurement data. Each parameter is selectively transmitted based on its relevance to the particular model being executed.
Solution Approach 2:
The patent segments the parameter transmission by dividing the complete measurement set into discrete, model-specific parameter groups. Each ML model receives only its required parameters as separate, identifiable units, enabling efficient data selection and transmission.
Data Source
AI summary
A method is disclosed for managing a Radio Access Network (RAN) operation performed by a first node in a communication network that comprises a RAN. The method is performed by the first node and comprises receiving a representation of a state of a second node with respect to the RAN operation, wherein the state of the second node comprises a compressed representation of parameter values that describe at least one of a physical state, a radio environment or a physical environment experienced by the second node or experienced by at least one node that is connected to the communication network via the second node. The method further comprises using the received state representation to generate a configuration action for the RAN operation and initiating configuration of the RAN operation in accordance with the generated configuration action.