Throughput Prediction by Radio Environment Correlation Recognition Using Crowd Sensing and Federated Learning
Satoshi Nakaniida, Takeo Fujii · 2022
We propose an approach using federated learning for predicting Wi-Fi and LTE transmission control protocol (TCP) throughput to reduce the delay between the output of prediction results and the problem of security risks by sharing the datasets, which is a problem with conventional machine learning methods. The proposed method collects measurement datasets such as received signal strength index from distributed edge devices. Then, a shared learning model is created using the measured dataset on the server. The created model is retrieved by edge devices at any time and used to predict TCP throughput. To evaluate the effectiveness of the proposed method, we perform the emulation evaluation using measured datasets obtained in a real environment. The emulation results reveal that the proposed method can skillfully predict the TCP throughput in the realistic communications. Additionally, the prediction accuracy of the TCP throughput can be improved by creating a learning model for each network area.