A STATISTICAL MODELLING BASED LOCATION DETERMINATION METHOD USING FUSION TECHNIQUE IN WLAN Reetu Singh, Luca Macchi and Carlo. S. Regazzoni * D.I.B.E University of Genova Opera Pia 11a, Genova, Italy Kostas.N. Plataniotis Edward S. Rogers department University of Toronto Ontario, Canada, MSS3G4 ABSTRACT Location information is of paramount importance in con- text aware Ambient Intelligence (AmI), Smart Space, traffic monitoring, surveillance network and cooperative commu- nications services. This paper describes a Positioning de- termination solution based on wireless local area network (WLAN) signals. Position determination is based on the statistical modeling of the received signal at any position. This paper presents a probabilistic based statistical model- ing approach for location estimation which incorporates fu- sion strategy in final step to combine efficiently the location individually reported by each WLAN transmitter. The sys- tem builds a radio map of the environment. The presented system is easier to implement and provide sufficiently good performance under all conditions. The accuracy with the 90% probability is reported to be 1.85 meters where as av- erage error is reported to be 2.1 meters. 1. INTRODUCTION Advancement in mobile world is catapulted by incorporat- ing the mobile users position information. The position of a hand held device is an important information needed to enhance the communication efficiency. Basically, the posi- tioning has been indigenously put apart into two parts, In- doors and Outdoors [1]. Since there are many ways to cat- egorize the positioning systems, in this paper we will just refer to indoors positioning systems based on WLAN. This paper presents a probabilistic based statistical modeling ap- proach for location estimation which incorporates fusion strategy in final step to combine efficiently the location in- dividually reported by each WLAN transmitter. The system builds a radio map of the environment. The concept of radio map is based on collection of signal strength over a set of strategically selected coordinates in a known referenced en- vironment. The set of signal strengths collected over known positions is treated as its signature. The issues related to * This work was performed under co-financing of the MIUR within the project FIRB-VICOM. duration of time in which the signal strength should be col- lected is still needed to be cleared. However on average, in the current state of art, typical time spent on collecting sur- vey data set is 5 minutes with a sampling period of 1 second . Most of the indoor positioning systems are based on the radio map [1-12]. The presented system is easier to implement and provide sufficiently good performance. The system is efficient and works well. The accuracy with the 90% probability is re- ported to be 1.85 meters where as average error is reported to be 2.1 meters. The paper is divided into the following parts: section II explains the formalization of the problem and the maximum likelihood approach utilized. Section III talks about the experimental setup used for algorithm’s per- formance verification. The experimental results are shown in section IV followed by the conclusion in the end. 2. STATISTICALMETHOD FOR LOCATION ESTIMATION Statistical methods are generally based on knowledge of sta- tistical distribution of the propagation model. Using statis- tical parameters of the propagation model, location can be estimated by utilizing a probabilistic approach. A model is said to be statistic if it gives a probability distribution of a signal in certain conditions. The statistical modeling of the environment should be the good choice where deterministic methods fails to imitate the environment. However, propa- gation parameters used for modeling should be derived from the environment itself. The best way to know these parame- ter is doing actual measurements. The parameters are com- puted by incorporating real measurements and the technique applied herein is maximum likelihood approach. Having es- timated the propagation parameters, the location estimation simply reduces to an inference problem. The bayesian ap- proach has been considered in the presented problem. It utilizes a priori knowledge of signal distribution at known locations. The location with the highest likelihood is chosen to be the estimate of user’s position from individual WLAN