ML-Based Handover Parameter Tuning for Mobility Robustness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wireless communication networks face challenges in maintaining seamless connectivity due to improperly set mobility parameters like Cell Individual Offset (CIO) and Time-To-Trigger (TTT), leading to Radio Link Failures (RLFs) and Ping-Pong Handovers (PP HOs), which are exacerbated by time-varying traffic patterns.
Innovation Solution
A network entity employs Machine-Learning (ML) models, specifically Double Deep Q-Networks (DDQN), to predict and adjust CIO and TTT parameters based on historical HO data, reducing RLFs and PP HOs by aggregating data vectors and using reward-based training to enhance mobility experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If mobility parameters (CIO and TTT) are manually configured, then network complexity is reduced, but mobility robustness deteriorates due to inability to adapt to time-varying traffic patterns
Solution Approach 1:
The system enables self-service by implementing automated ML-based parameter tuning where the network entity independently collects HO data, trains models, predicts optimal CIO and TTT values, and applies parameter adjustments without manual intervention, allowing the system to adapt to changing traffic patterns autonomously
Solution Approach 2:
The invention applies parameter changes by using ML models to dynamically adjust mobility parameters (CIO and TTT) based on predicted HO problem trends, transforming static manually-configured parameters into adaptive variables that respond to real-time network conditions and traffic patterns
2Reliability
If ML-based automated parameter tuning is implemented, then mobility robustness improves through adaptive parameter adjustment, but device complexity increases due to additional processing requirements
Solution Approach 1:
The system segments the parameter tuning process into distinct ML model components: a first ML model predicts TTT parameter adjustments and a second ML model predicts CIO parameter adjustments, allowing complex processing to be divided into manageable, specialized modules that can be executed efficiently
Solution Approach 2:
The invention applies preliminary action by collecting and storing HO data in data vectors during normal operation, pre-processing this data into training datasets, and training ML models in advance so that when parameter tuning is needed, the system can quickly predict optimal values without extensive real-time computation
3Adaptability or versatility
If traditional parameter tuning methods are used, then processing complexity is low, but adaptability to changing traffic patterns deteriorates
Solution Approach 1:
The system implements feedback by continuously monitoring HO problems, comparing actual performance against predicted outcomes, and using this feedback to retrain and refine ML models, enabling the system to adapt to changing traffic patterns through iterative learning and improvement
Solution Approach 2:
The invention applies dynamics by transforming static mobility parameters into dynamic, adaptive values that change in response to real-time network conditions, using ML models to continuously adjust CIO and TTT parameters based on current traffic patterns and HO problem trends
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A technical solution is proposed, which allows mobility experience to be improved by properly tuning Cell Individual Offset (CIO) and/or Time-To-Trigger (TTT) parameters in a wireless communication network. More specifically, information about a number of Handover (HO) attempts made between a source network node and each of its neighboring network nodes for a predefined period of time and a number of occurrences of each type of HO problems during the HO attempts are collected as a data vector. Then, for each "source network node - neighboring network node" pair, two data sub-vectors are obtained by aggregating the numbers from the data vector at a carrier level and an inter-node level, respectively. After that, the two data sub-vectors are used by an ensemble of two Machine-Learning models to predict whether the TTT and/or CIO parameters need to be changed.