Enabling wide deployment of GSM localization over heterogeneous phones

Wide deployment of GSM based location determination systems is a critical step towards moving existing systems to the real world. The main barrier towards this critical step is the heterogeneity of existing types of cell phones which results in different readings of received signal strength. Specially, in the context of fingerprinting localization where offline phases are needed for system training and different types of phones may be used in the offline and the online phases. Therefore, a mapping function, that maps the RSSI values between different types of cell phones, is inevitably needed. A trivial solution is to build a radio map for each type of phone. Obviously, this solution can neither scale in terms of number of phone types nor fingerprint size. In this paper, we address this problem by proposing the following two-way approach: A mathematical approach that maps RSSI values of different types of phones using linear transformation with regression, or logging ratios of readings instead of absolute values. We have empirically evaluated the proposed approach on Android-based phones. Our experimental results show that applying our approach can improve location accuracy with at least 127.84% in multiple cell tower configuration and at least 22.11% in the single cell tower configuration compared to the state-of-the-art GSM localization systems.