UPF Selection via Predicted Load Contributions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mechanisms for User Plane Function (UPF) selection in 5G networks are reactive and may lead to sub-optimal resource utilization and congestion, impacting Quality of Service (QoS), as they do not effectively account for predicted load variations among users.
Innovation Solution
Implementing a data-driven, proactive approach using Network Data Analytics Function (NWDAF) to predict load contributions based on historical usage data and session-related parameters, allowing for the selection of optimal UPF instances for packet data sessions by considering both current and predicted load information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If reactive UPF selection mechanisms are used, then current load status is considered, but predicted load variations are not accounted for leading to sub-optimal resource utilization
Solution Approach 1:
The patent applies preliminary action by predicting future load contributions of UEs before they actually consume network resources. The SMF uses historical usage data and machine learning models to estimate upcoming data usage patterns, allowing the network to proactively allocate UPF resources and avoid congestion before it occurs, rather than reacting to current load status alone
Solution Approach 2:
The patent implements feedback mechanisms where the SMF continuously monitors actual data usage of UEs and compares it against predicted load contributions. This feedback loop allows the system to refine its predictions and adjust UPF allocations dynamically, improving both resource utilization and QoS reliability over time through iterative optimization
2Productivity
If all end users are treated equally during resource allocation, then allocation simplicity is maintained, but sub-optimal utilization of network resources results
Solution Approach 1:
The patent applies local quality by differentiating resource allocation based on individual UE characteristics and historical usage patterns. Instead of uniform treatment, the system analyzes each UE's data consumption behavior and assigns them to UPFs based on their specific predicted load contributions, optimizing local resource distribution while maintaining overall system efficiency
Solution Approach 2:
The patent changes the allocation parameter from simple current load status to a composite parameter that includes predicted load contributions derived from historical usage data. This parameter transformation enables more sophisticated resource allocation that accounts for future demand patterns, improving utilization without requiring fundamentally complex allocation mechanisms
3Reliability
If multiple data-greedy users are allocated to the same UPF, then allocation simplicity is maintained, but congestion occurs impacting QoS
Solution Approach 1:
The patent prevents congestion through preliminary action by identifying data-greedy users before they cause network overload. The SMF predicts which UEs will consume large amounts of data and proactively distributes them across different UPFs based on current and predicted load conditions, avoiding the formation of congested UPFs before congestion occurs
Solution Approach 2:
The patent introduces an intermediary intelligence layer (the SMF with machine learning capabilities) that mediates between simple allocation requests and complex UPF selection decisions. This intermediary analyzes predicted load contributions and makes informed UPF selection decisions, balancing the need for allocation simplicity with the requirement to prevent congestion and maintain QoS
Data Source
AI summary
When a packet data session is established for a user equipment (UE), a comparative assessment of load information factors from different sets of load information factors associated with a plurality of user plane function (UPF) instances may be performed. Each set of load information factors of a UPF instance may include predicted load information factors indicative of a predicted load at the UPF instance. A UPF instance may be selected for the packet data session of the UE based on the comparative assessment. The comparative assessment may additionally consider a predicted load contribution of the packet data session to be established for the UE. A data analytics function may utilize a model (e.g. a multiple linear regression model) to calculate predicted load contribution factors, where the model is derived based on historical usage data from previous sessions of one or more UEs, for example, data from charging data records (CDRs).