ML-Based Wireless Device Density Estimation for Real-Time Coverage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing population distribution estimation methods fail to provide real-time data due to reliance on non-real-time network activity data with delays and limited participation from wireless devices, making accurate and precise estimation of population distribution challenging, especially in dynamic environments.
Innovation Solution
Training a machine learning model using non-real-time network activity data with location information to predict real-time population distribution by associating network activity patterns with geographic subregions, adjusting for partial device participation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If non-real-time network activity data is used for population distribution estimation, then data collection coverage is improved, but real-time accuracy deteriorates
Solution Approach 1:
The system pre-processes non-real-time network activity data to create training datasets that capture historical population distribution patterns. By preparing this data in advance and using it to train machine learning models, the system can then make real-time predictions without waiting for complete non-real-time data collection, thus resolving the contradiction between broad data coverage and real-time accuracy
Solution Approach 2:
The patent introduces machine learning models as an intermediary between non-real-time network activity data and real-time population distribution estimates. The ML models learn from historical data and serve as a bridge that translates delayed data into accurate real-time predictions, allowing the system to maintain both broad data collection and real-time precision
2Reliability
If diagnostic metrics data is collected from opted-in users only, then user privacy is protected, but data representativeness deteriorates
Solution Approach 1:
The system changes the parameters of the ML model training process to account for sampling bias. By adjusting for the fact that data comes from opted-in users only, the model learns to generalize predictions to the broader population, thus maintaining privacy protection while improving data representativeness through parameter optimization
3Productivity
If real-time network activity data is processed without location information, then processing speed is improved, but location accuracy deteriorates
Solution Approach 1:
The system pre-trains ML models using datasets that include location information from non-real-time data. This preliminary training equips the models with the ability to infer location patterns from network activity alone, enabling real-time processing to be both fast and accurate without requiring location data during real-time operation
Data Source
AI summary
A computer-implemented method generates an estimate of a real-time distribution density of wireless devices in a geographic area. The method includes receiving real-time network activity data collected at cell sites. The method includes processing the real-time network activity data with an ML model. The ML model is trained based on non-real-time network activity data indicating network activities as well as location data of wireless devices served by the cell sites. The ML model is configured to predict a real-time distribution of wireless devices in the geographic area at different points in time. The method includes predicting a real-time distribution of the wireless devices based on a pattern of network activities and locations of wireless devices. The method also includes generating an estimate of a distribution density of the wireless devices in the geographic area based on the predicted real-time distribution and adjusted by the real-time count of network activities.