A Differentially Private Federated Learning Model for Fingerprinting Indoor Localization in Edge Computing

With the increasing demand for location-based services, fingerprinting indoor localization based on received signal strength (RSS) has attracted widespread attention due to its well-established infrastructures and easy implementation. Deep learning (DL) has become a very attractive solution for RSS-based fingerprinting indoor localization because of its powerful capabilities of feature extraction and automatic classification. However, these solutions require repeated training of DL model with a large amount of RSS fingerprinting data via cloud computing. Since the RSS data contain users’ personal sensitive information, it may cause serious users’ privacy violations and data transmission delays when sending these RSS data directly to the untrusted cloud for processing. To address these challenges, a differentially private federated learning model for fingerprinting indoor localization in edge computing (DP-FLocEC) is proposed in this paper, which builds an edge computing enabled federated learning protocol and a convolutional neural network (CNN) based lightweight indoor localization model. The DP-FLocEC does no longer need to upload a large amount of RSS data to the cloud for model training, which improves localization accuracy while reducing data transmission delay. Then, we employ differential privacy technology to solve the problem of user privacy leakage in the offline training phase and the online localization phase of indoor localization. Security analysis and experimental results on multiple real datasets show that, compared with the centralized model based on cloud architecture, the DP-FLocEC obtains higher localization accuracy and reduces communication loss while providing provable privacy-preserving; compared with the distributed model based on the federated learning architecture, our DP-FLocEC provides more comprehensive privacy-preserving for users with almost the same localization accuracy and resource overhead.