Mobile Geolocation Machine Learning with Polar Coordinate Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cellular network geolocation methods are resource-intensive and computationally complex, particularly when using machine learning algorithms to predict latitude and longitude coordinates, due to the need for extensive memory and processing resources for each cell in large metropolitan areas.
Innovation Solution
A two-stage approach is employed, transforming the geolocation problem into a polar coordinate system, using a Random Forest algorithm to determine azimuth and range independently, with a Machine Learning Timing Advance (MLTA) correction model followed by a Geo prediction model, reducing computational complexity and adapting to changing cell site attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used to predict latitude and longitude coordinates for geolocation, then geolocation accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent segments the geolocation prediction task into two independent components: azimuth prediction and range prediction. The Random Forest model separately predicts the azimuth angle and the distance range, then combines these predictions to determine the final geolocation. This segmentation reduces the computational complexity of each individual prediction task while maintaining overall geolocation accuracy.
2Productivity
If machine learning algorithms are deployed for real-time geolocation in large metropolitan areas, then geolocation capability is improved, but memory and processing resources required increase
Solution Approach 1:
By dividing the geolocation problem into azimuth and range components, the patent reduces the memory footprint and processing requirements for each model instance. The segmented approach allows the system to handle larger numbers of cell sites and users with limited resources.
Solution Approach 2:
The patent transforms the traditional Cartesian coordinate prediction (latitude, longitude) into polar coordinate prediction (azimuth, range). This parameter transformation simplifies the prediction task and reduces the computational resources needed, as the Random Forest model can more efficiently predict angular and distance parameters independently.
3Loss of energy
If traditional geolocation methods are used, then resource consumption is reduced, but adaptability to changing cell site attributes decreases
Solution Approach 1:
The patent implements a dynamic machine learning-based geolocation system that can adapt to changing cell site attributes, radio patterns, and environmental conditions. The Random Forest model is trained on historical data and can be retrained as conditions change, providing continuous adaptation without requiring complete system redesign. This dynamic approach maintains low resource consumption while improving adaptability compared to static traditional methods.
Data Source
AI summary
Disclosed herein is a method to determine a geolocation that includes receiving, by a processor, from a base station (BS), radio predictors, a user equipment (UE) location history, and a geolocation of a first UE for which a minimization of drive test (MDT) mode is activated, radio predictors and a UE location history of a second UE for which the MDT mode is not activated, and cell physical parameters. The method includes training, by the processor, a machine learning (ML) model at least based on the radio predictors, the UE location history, and the geolocation of the first UE, and the cell physical parameters. The method includes executing, by the processor, the ML model to determine the azimuth of the second UE and providing, by the processor, to a downstream application, a geolocation of the second UE at least based on the azimuth and the TA of the second UE.