Comparative study demonstrates superior accuracy of neural networks for ionospheric delay prediction using NavIC signals, highlighting improved positioning precision.
Key Points
To forecast Vertical Total Electron Content (VTEC) using machine learning algorithms and compare their predictive accuracy against standard International Reference Ionosphere (IRI) empirical models.
Trained and evaluated DeepAR, DLinear, and Feedforward Neural Network (FFNN) algorithms using L5- and S-band pseudorange measurements from an operational NavIC receiver in Bangalore, India.
Quantified predictive performance using Normalized Root Mean Square Error (NRMSE), Mean Absolute Scaled Error (MASE), and Symmetric Mean Absolute Percentage Error (SMAPE) against IRI-2016 and IRI-2020 benchmarks.
FFNN achieved the highest prediction accuracy with an NRMSE of 0.136, SMAPE of 9.51%, and MASE of 0.839.
DLinear obtained an NRMSE of 0.157, MASE of 0.961, and SMAPE of 11.57%, while DeepAR yielded an NRMSE of 0.225, MASE of 1.468, and SMAPE of 18%.
FFNN outperformed all evaluated approaches, including DeepAR, DLinear, IRI-2016, and IRI-2020 models.