Analysis and design of neural network-based WLAN indoor location system

Abstract

As IEEE 802.11 WLAN networks are widely deployed, WLAN-based indoor location systems have obtained a good chance to develop, because they can be easily achieved on the existing WLAN infrastructure and by the RSSI parameter provided by the standard. Due to the complexity of the indoor RF propagation environment, the traditional triangulation method becomes impractical for indoor location systems and the propagation pattern matching method becomes the major method instead. RSSI values from a set of APs form the unique signatures or the propagation map for locations. The neural network method is one of the major location estimation methods for propagation map matching. Compared with the probabilistic method and the nearest neighbor method, the NN method has advantages in saving computing resources. However, the performance of a NN-based WLAN-based location system is determined by its training process and the resolution of the propagation map. By analyzing the properties of the process of RSSI creation, the underlying relationship between the WLAN RSSI parameter and space or RSSI resolution over space is obtained by applying the log-distance path loss model. Propagation map creation rules or rules for creation of neural network training data set are proposed based on the analysis and simulation results. And finally, guidelines about the design and implementation of an NN-based WLAN location system are proposed.