LeTera: Stochastic beam control through ESN learning in terahertz-band wireless UAV networks

Terahertz (THz)-band communication is a key technology to achieve ultra-high-data-rate wireless links in beyond-5G wireless networks. A main challenge with this frequency band is that the wireless links can be easily disconnected because of beam misalignment in mobile environments. This paper focuses on beam control in THz-band wireless Unmanned Aerial Vehicle (UAV) networks, where perfect beam alignment is hard to achieve in the presence of multi-scale mobility uncertainties of the flying UAVs. We propose a learning-based stochastic beam control scheme called LeTera to reduce the outage probability of the THz-band wireless links based on Echo State Networks (ESN). The scheme dynamically predicts through echo state learning the best beam width based on statistical information of the UAV mobility pattern. The scheme is evaluated using mobility traces collected through a series of UAV flight field experiments in different weather. Results show that LeTera can predict the optimal beam width with 99% accuracy and nearly optimal link capacity can be achieved in the presence of beam alignment latency.