SiFi: Siamese Networks Based CSI Fingerprint Indoor Localization with WiFi

Deep neural networks (DNN) based Channel state information (CSI) fingerprint indoor localization schemes have been widely investigated. However, existing DNN based schemes usually require to collect massive amount of training data samples, which is labor intensive and time consuming. In order to reduce the labor effort and time consumption, in this paper, we propose a Siamese network based CSI fingerprint indoor localization scheme with WiFi (SiFi), which only requires small number of training data samples to achieve a higher localization accuracy. Specifically, in our experiments, with only limited number of training data samples, for localization error less than 0.4m, the proposed SiFi scheme has the probability 82% to fall into this range, while the CNN based scheme only has the probability 67%. The experiment results demonstrate the proposed SiFi scheme has distinct advantages when only small number of training data samples are available.