Complete Large-Scale Parameter Maps Using Transformer Prediction

The patent uses machine learning models to create a virtual copy of the physical channel characteristics. Instead of directly measuring and processing all frequency channels physically, the system trains an ML model on limited measurement data to predict channel behavior across all locations and frequencies, replacing complex physical measurements with a computational model that captures channel characteristics without requiring exhaustive measurement procedures

The patent transforms the channel estimation problem from a physical measurement problem to a data-driven parameter prediction problem. By changing the approach from direct physical channel analysis to predicting channel parameters (LSPs) using machine learning models trained on spatial and channel state information, the system reduces the complexity of analyzing each frequency channel individually while maintaining estimation accuracy

The patent performs preliminary action by training machine learning models in advance using a subset of measured LSP data and spatial information. The model learns the relationship between spatial coordinates, channel state information, and large scale parameters during training, so that when actual channel estimation is needed, the pre-trained model can quickly predict LSPs at unmeasured locations without requiring time-consuming on-site measurements at every location

The system creates a virtual replica of the channel characteristics through machine learning. Instead of physically measuring LSPs at every location, the patent uses an ML model that has learned from limited measurements to generate a complete virtual map of channel characteristics, copying the essential patterns and relationships without requiring exhaustive physical measurement

The patent introduces machine learning models as an intermediary between the limited LSP measurements and the complete channel parameter maps. The ML model acts as a mediator that takes a subset of measured data and spatial information as input, processes it through learned patterns and relationships, and outputs complete LSP maps for all locations, thereby bridging the gap between limited measurements and comprehensive modeling without requiring measurements at every location

The system changes the measurement approach from collecting all possible LSP parameters at all locations to collecting a representative subset of measurements and using machine learning to infer the remaining parameters. By transforming the problem from complete data collection to intelligent inference, the system reduces measurement overhead while maintaining or improving modeling accuracy through the ML model's ability to capture complex spatial and channel relationships