"NA-TPRF: Neural Heteroscedastic Noise Modeling for Robust TDOA UAV Localization with Two-Pass Kalman Smoothing"

"NA-TPRF is a MATLAB repository for UAV TDOA trajectory post-processing, benchmarking, and reproducible evaluation. The project provides a complete workflow that starts from raw flight CSV data, performs preprocessing and coordinate conversion, and then runs multiple localization pipelines under a unified setup. Implemented methods include raw TDOA results, fixed-noise Kalman\/RTS smoothing, CRLB-guided adaptive noise modeling, learned noise prediction with MLP-R, and a two-pass filtering strategy for suppressing clustered outliers. The codebase also includes training\/validation\/test split handling, metric reporting (e.g., RMSE, MAE, median error, and error-threshold statistics), ablation scripts, and automated figure generation for publication-ready comparisons. It is designed to support fast replication of experiments, method comparison, and extension to new trajectory refinement ideas"