MobiFi: Fast Deep-Learning Based Localization Using Mobile WiFi

In most indoor localization systems deployed on commodity WiFi infrastructure, channel state information (CSI) data is usually transmitted over multiple subcarriers of different frequencies. An observation is that there exists a certain subcarrier that can best estimate the location of the target. Based on it, we propose MobiFi to leverage deep learning to automatically select the best subcarrier. MobiFi mainly consists of two steps: First, a lightweight end-to-end Convolution Neural Network (CNN) is taken as the backbone network to extract features and do classification while avoiding serious overfitting. After selecting the best subcarrier by the first two steps, MobiFi calculates the AoA estimation and corresponding location estimation in the same way as SpotFi. Since the backbone network is lightweight, MobiFi can realize near real-time on mobile devices with guaranteed localization performance. Extensive experiments show that MobiFi is comparable to SpotFi; both methods achieve a median AoA estimation error of 8.6° and median location estimation error of 1. 5m in an indoor office scenario. At the same time, MobiFi which consumes less than 0. 21s and 1. 7s on Personal Computer (PC) and mobile devices respectively is 5 times faster than SpotFi. Particularly, because MobiFi enables real-time localization on mobile devices, it provides an economical solution for some cases where a central server is replaced by a mobile device.