A 3-D RSS distribution model based on statistical properties for indoor localization systems

In the field of indoor localization based on Wi-Fi fingerprints, if the training data of received signal strength (RSS) in the sampling positions are not sufficient during the training phase, it is necessary to estimate the missing RSS training data in the non-sampling positions. Most of the previous works simply estimate the average values of RSS based on the methods of surface fitting or interpolation. However, these methods ignore the statistical properties of RSS. Because of the complex indoor environment, the value of RSS at a fixed position follows a specific distribution, which also manifests the diversity among different positions. Thus we need to consider the probability distributions of different RSS values instead of estimating a single average RSS at each position only. To do this, we propose a new 3-D RSS distribution model based on statistical properties. We consider the unstable nature of RSS, and then model the probability distribution of RSS in the whole indoor space. Based on the fitting function provided by our model, the method of fitting can be used to estimate the RSS distribution at the non-sampling positions. Thus our model can be used to reduce the necessary amount of RSS training data for 2-D/3-D indoor localization systems. The results of experiments also confirm the benefit of our model.