ML-Based Path Selection for 5G Multi-Connectivity Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In dual or multi-connectivity wireless telecommunication systems, existing methods for estimating transmission delays between 5G and 4G network nodes are inaccurate, leading to potential congestion and reordering of data packets, which can increase latency and reduce data rates due to dynamic load conditions and congestions on interfaces.
Innovation Solution
A machine learning model, such as a neural network or recurrent neural network, is used to estimate transmission delays by collecting data sets including data in flight and throughput, and updating the model based on measured delays to improve accuracy and adapt to changing conditions, thereby selecting the optimal path for packet transmission between 5G and 4G nodes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning model is used to estimate transmission delay, then delay estimation accuracy is improved, but device complexity increases
Solution Approach 1:
A machine learning model acts as an intermediary between raw network parameters (data in flight, throughput) and delay estimation. The model processes these parameters through trained algorithms to produce accurate delay predictions without requiring complex direct measurement systems. This intermediary approach resolves the contradiction by providing high measurement precision through the ML model while keeping the underlying implementation manageable through standardized interfaces and training procedures.
2Adaptability or versatility
If machine learning model is updated with measured delays, then adaptability to changing conditions is improved, but loss of time for model training increases
Solution Approach 1:
The machine learning model is pre-trained offline with extensive network data before deployment. This preliminary action prepares the model to handle various network conditions without requiring extensive real-time training. When deployed, the model can quickly adapt to changing conditions using pre-learned patterns, thus achieving high adaptability while minimizing real-time training time loss.
Solution Approach 2:
The model undergoes periodic retraining with newly collected measured delay data to maintain and improve its adaptability. Instead of continuous training that would cause significant time loss, the system periodically updates the model with accumulated data, balancing adaptability improvement with acceptable training time overhead.
3Loss of time
If optimal path selection based on delay estimation is implemented, then latency is reduced, but device complexity increases
Solution Approach 1:
The system implements feedback loops where actual transmission delays are measured and used to update the machine learning model. This feedback mechanism enables the system to learn from real performance data and continuously improve path selection accuracy. The feedback approach reduces latency by making increasingly accurate predictions while managing complexity through iterative learning rather than complex deterministic algorithms.
Solution Approach 2:
The machine learning model performs self-service by automatically selecting optimal transmission paths based on its delay estimates without requiring complex external routing control. The model autonomously processes network parameters, generates delay predictions, and determines optimal paths, thereby reducing latency while keeping the control architecture relatively simple through self-contained decision-making capabilities.
Data Source
AI summary
An apparatus comprises means for performing, at a CU of a master node: receiving an individual PDU from one or more incoming PDUs, wherein each PDU is to be transmitted to a UE via one amongst respective paths towards at least one DU of the master node and at least one secondary node, the UE being in DC or MC with the master node and the at least one secondary node; collecting a respective data set for each of the paths; receiving, from a ML model inputting each data set, a respective estimated delay of transmission for each of the paths; selecting, from each of the paths, a path based on the estimated delay of transmission; and transmitting the PDU to the UE via the selected path.