Deep Learning Path Loss Prediction Without Model Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing network planning technologies face low prediction efficiency due to the need for calibration of propagation models based on actual scenario data, limiting the accuracy and scope of coverage prediction in radio network planning.
Innovation Solution
A path loss prediction method utilizing deep learning models constructed from feature images generated by rasterizing characteristic parameters such as building height, ground object type, and transmitter parameters, eliminating the need for model calibration by directly mapping these parameters to path loss predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If propagation models are calibrated based on actual scenario data, then prediction accuracy is improved, but prediction efficiency deteriorates due to the time-consuming calibration process
Solution Approach 1:
The patent pre-calibrates multiple propagation models for different scenario types before actual prediction tasks. Sample data is collected and used to calibrate models in advance, creating a library of pre-configured models that can be directly applied during prediction without time-consuming on-the-spot calibration
Solution Approach 2:
The patent changes the approach from calibrating model parameters during prediction to clustering scenarios based on characteristic parameters beforehand. By categorizing scenarios into types based on their parameters, the system selects pre-calibrated models matching each scenario type, eliminating the need for real-time parameter calibration
2Adaptability or versatility
If multiple propagation models are calibrated for different scenarios, then adaptability to various planning scenarios is improved, but device complexity increases due to multiple model calibrations
Solution Approach 1:
The patent transforms the complex process of calibrating multiple propagation models into a simpler scenario clustering process based on characteristic parameters. By defining scenario types through parameter thresholds and selecting pre-calibrated models based on scenario type, the system maintains high adaptability while significantly reducing operational complexity
Solution Approach 2:
The patent creates a universal framework where a single path loss prediction system can handle multiple different scenarios by selecting from pre-calibrated models. The scenario clustering mechanism provides a unified approach that adapts to various planning scenarios without requiring separate calibration processes for each
Data Source
Figure 1~2
Figure 3
Figure 3a
AI summary
Embodiments of this application disclose a path loss prediction method and apparatus, to resolve an existing problem of low prediction efficiency caused by propagation correction during prediction of a propagation path loss. The method includes: obtaining sample data, where the sample data includes a characteristic parameter that affects radio propagation between a transmitter and a cell, a parameter value of the characteristic parameter, and a measured path loss between each receiver in the cell and the transmitter; rasterizing the characteristic parameter to generate a feature image; constructing a deep learning model based on the feature image and the measured path loss, where an input variable of the deep learning model is the feature image, and an output variable of the deep learning model is used to indicate the path loss between each receiver in the cell and the transmitter; and predicting the path loss by using the constructed deep learning model. The method provided in the embodiments of this application is used to predict a path loss between a transmitter and a receiver.