NWDAF AI/ML Processing Requests for Low-Latency 5G Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing 5G communication systems face challenges in reducing delays and ensuring end-to-end (E2E) processing of AI/ML operations, particularly in real-time transmission of high-definition video and increasing AI/machine learning applications, without a mechanism for handling AI/ML processing by network data analytics functions (NWDAF).
Innovation Solution
An information processing device that requests AI/ML model processing to a device with network data analysis function (NWDAF) via a network exposure function, allowing for distributed processing between cloud servers and edge application servers, and controlling processing based on wireless communication capabilities and quality to reduce delays and ensure synchronization across multiple PDU sessions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If AI/ML processing is performed via cloud servers, then processing capability is improved, but communication delay increases
Solution Approach 1:
The patent segments AI/ML processing across multiple locations: cloud servers for model training and management, edge application servers for real-time inference, and user equipment for local processing. This segmentation allows workloads to be distributed according to latency requirements, with time-critical operations performed at the edge or device level rather than remotely in the cloud.
Solution Approach 2:
The patent introduces edge application servers as intermediary components between cloud servers and user equipment. These edge servers receive requests from UE, perform AI/ML inference locally, and communicate with cloud servers for model updates. This intermediary architecture reduces round-trip communication delay while maintaining access to cloud-based model resources.
2Productivity
If distributed processing is implemented between cloud and edge servers, then processing efficiency is improved, but system complexity increases
Solution Approach 1:
The patent designs edge application servers with multi-functional capabilities: they can perform AI/ML inference, manage local model versions, handle multiple PDU sessions simultaneously, and communicate with both cloud servers and user equipment using standardized protocols. This universality reduces the need for specialized components and simplifies the overall system architecture.
Solution Approach 2:
The patent dynamically adjusts processing parameters such as model complexity, batch size, and inference frequency based on network conditions, device capabilities, and service requirements. This parameter adaptation allows the distributed system to optimize performance without requiring complex reconfiguration, simplifying deployment and management.
3Reliability
If end-to-end processing is ensured for AI/ML operations, then reliability is improved, but latency increases
Solution Approach 1:
The patent performs preliminary actions by pre-loading AI/ML models to edge application servers before they are needed, and by pre-establishing communication paths and resource allocations. This allows the system to guarantee end-to-end processing reliability without incurring latency during actual inference operations, as resources are already prepared and configured.
Solution Approach 2:
The patent implements dynamic resource allocation and path selection that adapts to changing network conditions. When end-to-end reliability is required, the system dynamically activates redundant paths and reserves resources. When conditions permit, it relaxes these guarantees to minimize latency, allowing flexible trade-offs based on real-time requirements.
Data Source
AI summary
An information processing device executes an application. The information processing device includes a control unit. The control unit makes a request regarding processing of a learning model used in the application to a device having a network data analysis function of a mobile network via a network exposure function and/or directly.