Mixed-Numerology Configuration Selection via ML QoS Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for mixed-numerology configuration in wireless networks either reduce the search space without identifying an optimal configuration, add significant network overhead, or are not scalable for dense networks due to high computational complexity.
Innovation Solution
A base station apparatus and method that extracts statistical features related to service requirements, traffic, and wireless channel conditions, using a Machine Learning model array to estimate Quality of Service (QoS) for mixed-numerology configurations and select an optimal configuration for data transmission and reception.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a greedy algorithm is used to compute network flexibility for all possible mixed-numerology configurations, then the optimal configuration can be identified, but the computational complexity increases exponentially with the number of UEs making it not scalable for dense wireless networks
Solution Approach 1:
The patent segments the complex optimization problem into two distinct phases: (1) an offline training phase where Machine Learning models are pre-trained using brute-force computation on historical data to learn the relationship between network conditions and optimal configurations, and (2) an online inference phase where the pre-trained models rapidly predict QoS for current conditions without exhaustive computation. This segmentation allows the system to benefit from accurate brute-force training while achieving scalable real-time performance through efficient model inference.
2Reliability
If mobility information is periodically transmitted from each UE to the gNB to mitigate Doppler spreading effects, then the adverse effects of Doppler spreading are reduced, but significant overhead is added to the network
Solution Approach 1:
The patent enables the system to self-determine optimal mixed-numerology configurations by utilizing existing network measurements and pre-trained ML models at the gNB side, eliminating the need for UEs to periodically transmit mobility information. The gNB autonomously predicts QoS for different configurations based on current network conditions and selects the optimal configuration without requiring additional feedback loops or periodic reporting from UEs, thus achieving Doppler spread mitigation without adding network overhead.
3Device complexity
If the search space of possible mixed-numerology configurations is reduced, then the method becomes more feasible and less complex, but the ability to identify an optimal mixed-numerology configuration within the feasible set is not provided
Solution Approach 1:
The patent performs preliminary action by pre-training ML models offline using exhaustive brute-force computation across the entire configuration space. During this offline phase, the system learns the optimal configurations for various network conditions and stores this knowledge in the model parameters. When deployed online, the pre-trained model can rapidly predict QoS for current conditions without needing to search or reduce the configuration space, thus maintaining full optimization capability while achieving computational efficiency.
Data Source
AI summary
The requirement feature extractor extracts statistical features related to service requirements of UEs. The traffic feature extractor extracts statistical features related to incoming and outgoing traffic. The channel feature extractor extracts statistical features related to a wireless channel condition and a wireless channel configuration. The context unit generate a context vector based on the statistical features. The ML model array estimates QoS for mixed-numerology configurations based on the context vector. The decision unit selects a mixed-numerology configuration used for data transmission and data reception based on the estimated QoS.