Ionospheric TEC Prediction Using LSTM-GAN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In areas with limited ground transceivers, the prediction of total electron content (TEC) in the ionosphere is inaccurate due to sparse data collection, leading to errors in signal delay and location determination, which affects satellite communications.
Innovation Solution
A machine learning model, specifically a combination of a long short-term memory (LSTM) neural network and a generative adversarial network (GAN), is used to predict TEC values by training on historical data and improving predictions in regions with minimal transceivers, thereby enhancing the accuracy of signal delays and location determination.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If ground transceivers are deployed sparsely in certain areas, then deployment cost is reduced, but TEC measurement accuracy deteriorates
Solution Approach 1:
The patent introduces machine learning models as an intermediary between sparse ground transceiver data and TEC prediction requirements. The models process limited measurements from few transceivers and generate accurate TEC predictions for regions without direct measurements, effectively mediating the gap between sparse data and comprehensive coverage needs
Solution Approach 2:
The patent creates virtual copies of TEC data through machine learning predictions. Instead of physically deploying transceivers everywhere, the system generates synthetic TEC values for locations without direct measurements by learning patterns from areas with transceiver coverage, effectively copying reliable data patterns to data-sparse regions
2Measurement precision
If more ground transceivers are deployed to improve TEC measurement accuracy, then TEC measurement accuracy is improved, but deployment and maintenance cost increases
Solution Approach 1:
The patent transitions from a spatial dimension solution (adding more physical transceivers across the geography) to a temporal and computational dimension solution (using machine learning models that process data over time and generate predictions). This dimensional shift allows accurate TEC prediction without linearly increasing the number of physical devices
3Loss of information
If data is interpolated in areas with minimal transceiver coverage, then TEC values are estimated, but data accuracy deteriorates
Solution Approach 1:
The patent changes the fundamental parameter of how TEC values are derived in data-sparse regions. Instead of using simple interpolation methods that assume linear variations, the system employs machine learning models that learn complex nonlinear relationships from training data, fundamentally changing the parameter of data derivation from geometric interpolation to pattern-based prediction
Solution Approach 2:
The patent performs preliminary training of machine learning models using data from regions with good transceiver coverage before applying these models to predict TEC in data-sparse regions. This preliminary action of model training on high-quality data prepares the system to accurately predict values where direct measurements are unavailable
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The model effectively predicts TEC values with high accuracy, reducing errors and improving satellite communication reliability by providing better knowledge of atmospheric signal delays, even in areas with limited transceiver coverage.
Implementation Method 1
Doppler shifts and other signal delays caused by the ionosphere (e.g., which reflects radio waves directed into the sky back toward the Earth)
Data Source
AI summary
A system may be configured to predict total electron content (TEC) in an ionosphere. Some embodiments may: provide a machine learning (ML) model; obtain a dataset; input the dataset into the ML model; predict, for a predetermined number of days, the TEC using the ML model; and observe a performance improvement over the obtained dataset based on the prediction, the prediction being made for a region having a number of ground transmitters satisfying a sparseness criterion.