Mobile Terminal Positioning Through Federated Learning for Lower Data Overhead
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mobile terminal positioning methods face challenges in training data transmission overhead and training efficiency due to local data collection at positioning components and the presence of multiple terminal categories, leading to redundant and inefficient machine learning model training.
Innovation Solution
A federated learning approach is employed to train positioning machine learning models using multiple positioning components as clients, reducing data transmission overhead and enabling efficient training of category-specific models by aggregating and filtering local data, and utilizing a base model as a starting point for category-specific enhancements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If training data is collected locally at positioning components for machine learning model training, then positioning accuracy can be improved, but training data transmission overhead increases
Solution Approach 1:
A federated learning server acts as an intermediary between positioning components and the central system. The server coordinates model training across multiple positioning components without requiring centralized data collection, thus maintaining positioning accuracy while eliminating the need for large-scale training data transmission.
Solution Approach 2:
Positioning components perform local model training using their own collected training data. Each component independently trains positioning models locally and uploads only the trained models or model parameters to the federated learning server, enabling self-service training that avoids transmitting raw training data.
2Measurement precision
If machine learning models are trained for multiple categories of mobile terminals, then positioning accuracy for different terminal types improves, but training efficiency decreases due to redundant training
Solution Approach 1:
The system segments terminal categories into different groups and trains category-specific positioning models for each segment. The federated learning server manages multiple specialized models (e.g., for indoor terminals, outdoor terminals, high-speed terminals) rather than training a single general model, improving accuracy for each category while organizing training efficiently.
Solution Approach 2:
The system performs preliminary categorization of mobile terminals and pre-trains category-specific models using federated learning. By anticipating different terminal categories and preparing specialized models in advance, the system avoids redundant training when actual positioning is needed and improves overall training efficiency.
3Loss of information
If federated learning is used to train positioning models across multiple positioning components, then training overhead is reduced, but system complexity increases
Solution Approach 1:
The federated learning server implements a universal training framework that handles multiple terminal categories, multiple positioning components, and various model types through a single coordinated system. This multi-functional approach reduces overall system complexity by providing a unified solution rather than requiring separate training mechanisms for each scenario.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
According to one embodiment, a method for mobile terminal positioning is described comprising training a positioning machine learning model to output positioning information in response to mobile terminal status information by federated learning using a plurality of positioning components as federated learning clients and determining a position of a plurality of mobile terminals using the trained positioning machine learning model.