Error Corrected Location Determination in An Outdoor Wireless Environment by Using Estimation, Filtering, Prediction and Fusion Techniques: A WIFI Application by Using Terrain Based Knowledge

Abstract

Location estimation of wireless nodes has been a very popular research area for the past few years. The research in location estimation is not limited to satellite communication, but also in WLAN, MANET , WSN and cellular communication. Because of the growth and advancement in cellular communication architecture, the usage of handheld devices has increased rapidly, therefore mobile users originating calls are also increasing. It is estimated that more than 50% emergency calls are originated by mobile phones. Researchers have used different location estimation techniques, such as satellite based, geometrical, statistical and mapping techniques. In order to achieve accuracy, researchers have combined two or more techniques. However the terrain based location estimation is an area which is not considered by researchers extensively. Due to the fact that radio waves behave differently in different atmospheres , the calculation of few parameters is not sufficient to achieve accuracy in differen terrains , especially when it is totally based on RSS which is carrying impairments. This research is focusing on the localization of wireless nodes by using geometrical and statistical techniques with the consideration of impairment/attenuation of terrains. The proposed model is consisting of four steps, which are estimation, filtering, prediction and fusion. A prototype has been built using the WiFi IEEE 802.11x standard. In step one , by using signal to noise ratio ,the peninsular Malaysia geographical area is categorized into 13 different terrains/clutters . In step two , point-to-point data points are recorded by using available signal strength and receive signal strength with the consideration of different terrains. Estimation of the location is done in step three by using the triangulation method. The result of estimated locations are further filtered in step four by using average and mean of means. For error correction, filtering of the location is also done by using k – nearest neighbor rule. Prediction is done in step five by using combined variance which predicts the region of interest. Region of interest helps to eliminate locations outside of the selected area. In step six filtering results are fused with prediction in order to achieve accuracy. Result show that the current research is capable of reducing errors from 18 m to 6 m in highly attenuated terrains and form 3.5 m to 0.5 m in low attenuated terrains.