5G AI/ML Traffic Differentiation for Slice and QoS Handling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The 5G system lacks support for AI/ML communications, failing to differentiate AI/ML traffic from non-AI/ML traffic, which impacts performance and decision-making in AI/ML operations between end devices and cloud servers.
Innovation Solution
Enhancements to network functions such as AMF, SMF, and UPF enable AI/ML traffic differentiation at network slice, PDU session, and QoS flow levels, along with AI/ML triggering rules, policies, and communication interfaces to support UE-AF interactions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the 5G system treats all traffic uniformly without differentiation, then the network operation is simple, but AI/ML operations performance deteriorates due to inability to optimize for specific traffic types
Solution Approach 1:
The patent segments traffic into distinct categories by introducing AI/ML traffic differentiation indicators at multiple network levels (PDU session level, QoS flow level). This allows the network to identify and handle AI/ML traffic separately from other traffic types, enabling optimized resource allocation and charging without fundamentally restructuring the entire network architecture.
Solution Approach 2:
The patent applies local quality by implementing AI/ML-specific handling at specific network levels where it is most beneficial. Rather than changing the entire network uniformly, the differentiation is applied locally at PDU session establishment and QoS flow levels, allowing AI/ML traffic to receive specialized treatment (prioritization, charging, resource allocation) only where needed while leaving other traffic flows unchanged.
2Adaptability or versatility
If the 5G system implements comprehensive AI/ML traffic differentiation and support mechanisms, then AI/ML communications are optimized, but the system complexity increases significantly
Solution Approach 1:
The patent achieves versatility through multi-functionality by designing network functions that can handle both traditional traffic and AI/ML traffic using the same infrastructure. The AMF, SMF, and UPF are enhanced to recognize AI/ML traffic indicators and apply appropriate handling, allowing a single network function to serve multiple purposes (general traffic management plus AI/ML optimization) without requiring separate dedicated systems.
Solution Approach 2:
The patent applies preliminary action by establishing AI/ML traffic differentiation at the PDU session establishment phase, before actual data transmission begins. The network function receives an indicator during session setup that pre-configures the session for AI/ML traffic, enabling subsequent optimized handling without requiring real-time detection or complex runtime decision-making.
3Productivity
If AI/ML traffic is differentiated at network slice level, then AI/ML traffic optimization is achieved, but the charging and resource allocation complexity increases
Solution Approach 1:
The patent adds another dimension to traffic differentiation by implementing AI/ML indicators at multiple hierarchical levels (PDU session level and QoS flow level). This multi-dimensional approach allows the network to differentiate AI/ML traffic without being constrained to a single level, enabling flexible resource allocation and charging strategies that can operate at the appropriate granularity for each scenario.
Data Source
AI summary
Methods, systems, and devices may assist in artificial intelligence (AI) or machine learning (ML) communications in 5G system, AI or ML traffic differentiation, AI or ML slice type, AI or ML triggering rules, AI or ML policy, AI or ML operations, or user equipment communication interface exposure.