NEURAL INFERENCE OF OBJECT LOCALIZATION FROM DISCONTINUOUS TIME-OF-ARRIVAL SEQUENCES
The article addresses the problem of recovering the coordinates of a moving object in conditions of temporary signal loss or degradation (blackout), typical for complex radio environments. A method based on a Long Short-Term Memory (LSTM) recurrent neural network is proposed for interpolating and predicting positions using sequences of Time Difference of Arrival (TDoA) measurements from three fixed anchors. The developed model was trained on synthetic trajectories simulating object movement and demonstrates high robustness to missing data, with a median error of less than 30 meters even under significant signal degradation. A comparative analysis with alternative methods, including GRU, TCN, and Kalman Filter, confirms the superior performance of the LSTM architecture in unstable environments with limited measurements. The results indicate the proposed approach is promising for real-time applications in autonomous navigation and positioning systems.
Minina A.V., Nikulin R.R., Sidorenko I.A. Neural Inference of Object Localization from Discontinuous Time-of-arrival Sequences // Research result. Information technologies. – T.10, №3, 2025. – P. 45-54. DOI: 10.18413/2518-1092-2025-10-3-0-4
While nobody left any comments to this publication.
You can be first.