Cross-Node AI/ML Session Management in RAN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current wireless communication systems face challenges in efficiently managing cross-node AI/ML operations in radio access networks (RAN), leading to computational resource bottlenecks and increased costs for RAN operators.
Innovation Solution
The implementation of signaling to manage cross-node AI/ML sessions between user equipment (UE) and a RAN controller in a cloud-based RAN architecture, allowing for efficient AI/ML processing and offloading of computations to specialized AI/ML computing devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cross-node AI/ML operations are implemented in RAN, then wireless communication performance is improved, but computational resource bottlenecks and costs increase
Solution Approach 1:
The patent segments AI/ML computational tasks across multiple nodes including UE, base station, and cloud-based RAN controller. Each node performs specific AI/ML operations locally while exchanging results with other nodes, distributing the computational burden and avoiding concentration of complexity in a single device.
Solution Approach 2:
The patent introduces a cloud-based RAN controller as an intermediary that coordinates cross-node AI/ML sessions, manages computational resource allocation, and facilitates information exchange between UE and base station. This intermediary handles complex session management and resource orchestration, reducing the computational burden on edge devices.
2Speed
If AI/ML processing is performed at base station, then processing speed is improved, but computational burden on base station increases
Solution Approach 1:
The patent divides AI/ML processing into segments performed at different locations: UE performs sensing and initial processing, base station performs rapid processing for time-critical functions, and cloud RAN controller handles complex model training and management. This segmentation allows base station to maintain high processing speed for urgent tasks while offloading heavy computational burden to the cloud.
Solution Approach 2:
The patent adds a temporal dimension to AI/ML processing by performing different types of processing at different times: urgent processing occurs at base station for immediate response, while less time-critical model training and optimization occur in the cloud over longer periods. This resolves the contradiction between speed and computational burden by separating time-sensitive and non-time-sensitive operations.
3Loss of time
If cross-node AI/ML sessions are established, then processing latency is reduced, but system complexity increases
Solution Approach 1:
The cloud-based RAN controller acts as an intermediary that manages cross-node session establishment, configuration, and coordination. It handles the complex signaling and resource allocation required for cross-node AI/ML sessions, reducing processing latency by pre-configuring sessions and maintaining persistent connections while abstracting system complexity from individual nodes.
Solution Approach 2:
The patent implements preliminary setup of AI/ML sessions during idle periods or low-load conditions, pre-configuring processing pipelines, data flows, and resource allocations. This preliminary action reduces processing latency during actual AI/ML operations by having everything ready in advance, while the complexity of session management is handled separately during setup rather than during time-critical processing.
Data Source
AI summary
Certain aspects of the present disclosure provide techniques for managing cross-node artificial intelligence (AI) and/or machine learning (ML) operations in a radio access network (RAN). An example method of wireless communication by a first network entity includes obtaining machine learning input data associated with a user equipment (UE); providing, to a second network entity, an indication of machine learning output data generated using the machine learning input data; and providing, to the second network entity, control signaling for a cross-node machine learning session between the UE and the first network entity based at least in part on one or more performance indicators associated with the cross-node machine learning session.