Transformer Neural Network for Throughput Improvement in Non-Terrestrial Networks

The integration of Non-Terrestrial Networks (NTNs) within the 6G framework provides full connectivity worldwide but some system procedures can be ineffective and should be modified for example in the timer thresholds, which may expire in NTN. Moreover, some procedures may also be affected by the different timeline of terrestrial networks and NTN. For example, the aging of the Channel State Information (CSI) collected by the User Equipment (UE) and sent to the gNode B (gNB) for the scheduler activities may affect its behavior reducing the user throughput, due to the high propagation latency in NTN and outdated data. In this paper we propose a strategy to overcome this problem by using a Transformer Neural Network (TNN) to predict the experienced Signal-to-Interference plus Noise Ratio (SINR) of a generic UE to the transmitting gNB. We assume a period in which the gNB receives frequently CSI/SINR reports by a given UE in order to train the TNN model. Next, the gNB uses data provided by the trained TNN to predict the UE channel behavior and then to properly select the best Modulation and Coding Scheme (MCS) for that UE, avoiding to transmit any CSI data. Results show that the throughput obtained with the TNN trained model is close to the highest CSI update rate (1 ms), while it is higher than the throughput obtained with a CSI update rate of 200 ms of about 5 Mbit/s in Sub-Urban environment for both always LOS and variable LOS/NLOS channel conditions, which corresponds to about a gain of 10 %.