A Three-Level Federated Learning Framework for CSI Fingerprint Based Indoor Localization in Multiple Servers Environment

This letter develops a federated learning (FL) framework for indoor localization using channel state information (CSI) fingerprint, aiming to (a) design a FL framework for multiple servers which can protect data privacy, improve data transmission quality and enlarge the coverage area of global server (b) maximize localization performance under the given measurement data. To this end, we first propose a three-level FL framework where the FL is processed between client and intermediate server, intermediate server and intermediate server, intermediate server and global server, respectively. Then, the training accuracy based model aggregation is proposed for first and second level of FL. In order to accelerate the convergence, the inner production based aggregation rule is used for the last level FL processing. After the model aggregation, in order to protect data privacy, the obtained model parameters are sent to the given local client connected with the intermediate server. Experimental results illustrated that the proposed scheme has better localization performance than some existing approaches under two different scenarios.