Bandwidth Usage Reduction by Traffic Prediction Using Transfer Learning in Satellite Communication Systems

Recently, Internet traffic has surged due to the widespread demand for teleworking and flat-rate video distribution services. It is expected that such enormous traffic with diverse patterns will be managed by employing beyond fifth-generation backbone networks, such as satellite networks. Satellite communication resources are scarcer than those of terrestrial communication. Therefore, this study has focused on ensuring efficient satellite network resource operations. Traffic forecasting is a promising approach to facilitate optimal resource allocation. While there are several methods for traffic prediction in terrestrial communication using machine learning, an insufficient number of studies have been conducted regarding satellite communications. Existing traffic prediction approaches consume high bandwidths, which can be a problem for the bandwidth used by users over the limited bandwidth of satellite networks. Therefore, this study proposes a lightweight machine-learning- based traffic prediction method using transfer learning to reduce bandwidth consumption. Furthermore, we demonstrated the effectiveness of the proposed method via simulations by comparing its accuracy with conventional approaches.