A Method for Predicting / Controlling Wireless Network Quality Preserving User Privacy by Federated Learning

DOI

Bibliographic Information

Other Title
  • Federated Learningによるプライバシー保護を考慮した無線品質予測・制御手法

Abstract

The switching network algorithm of smartphones between WLAN and cellular is independent of the communication channel quality, which severely affects the user's experience when connecting to low-quality networks. Predicting WLAN quality before switching is, therefore, imperative. While parameters like BSSID can help predict WLAN rate, their collection from a privacy protection standpoint is challenging. Additionally, inferring Wi-Fi quality on smartphones requires lightweight models. To achieve large-scale training while protecting users' privacy in the training data, we employ Federated Learning and FedHLCR [1]. We present a lightweight algorithm that accurately predicts WLAN quality, achieving training and inference times within a second. Our proposed method reduces disruption time by up to 85% when selecting networks.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top