An LSTM-Assisted Calibration Approach for Self-Time Transfer Performance Optimization in Multi-UAV Networks

Multiuncrewed aerial vehicle (UAV) networks have already been deployed for complex tasks in both public and military domains. Central to the effectiveness of multi-UAV networks is self-time-synchronization (STS), which ensures that all UAVs operate based on a unified time reference where external time references are unavailable or signal strength is weak. The foundation of STS depends on the accurate acquisition of time-of-arrival (TOA) measurements. However, in wide-area sparse deployment scenarios, the increasing distance between UAVs reduces the signal-to-noise ratio (SNR), posing significant challenges to improving TOA estimation accuracy in multi-UAV networks. In this article, we propose a calibration-assisted TOA estimation scheme that enables accurate self-time transfer. By incorporating modulated signals directly into the loop-based estimation process, the proposed scheme eliminates the reliance on predefined pilot structures, thereby enhancing adaptability to diverse modulation waveforms and significantly improving TOA estimation accuracy under dynamic signal conditions. To improve TOA estimation accuracy under low SNR conditions, we introduce a calibration scheme based on long short-term memory networks to refine current estimations. Numerical results demonstrate that the proposed algorithm outperforms existing approaches in terms of both accuracy and reliability, particularly under low SNR conditions. Compared to the existing method, our scheme achieves a 17.3\% improvement in the cumulative distribution function at the time error of 3 ns when SNR = -10 dB, indicating higher estimation accuracy.