Timing Advance Estimation for LEO-based NTN with Regenerative Payloads

Non-terrestrial networks (NTNs) utilizing low Earth orbit (LEO) satellites have gained significant interest due to their potential to provide seamless global coverage and low latency. However, LEO satellites require accurate timing advance (TA) estimation considering the variation of propagation delay due to their high speed. In particular, random access channel-less (RACH-less) handovers require precise initial TA estimation within the cyclic prefix (CP) duration, as timing adjustments during the random access procedure cannot be performed. Therefore, this paper proposes an TA prediction framework based on a long short-term memory (LSTM) neural network, which compensates for time-series errors originating from the simplified general perturbations 4 (SGP4) orbital model. Since the SGP4 cannot achieve the required margin of error within the CP duration, the proposed LSTM-based TA estimation model is introduced to address this limitation. Simulation results demonstrate that the proposed method estimates initial TA values, maintaining positional errors within the CP duration for all numerologies and significantly outperforming conventional methods in terms of mean error and variance. These results demonstrate that the proposed method reliably achieves uplink synchronization, even in scenarios requiring high accuracy of the initial TA.