Improving the Performance of a RSS-based Location Estimation System, Study and Evaluation Geoffrey Ottoy, Anneleen Van Nieuwenhuyse, Kevin D’hoe, Jean-Pierre Goemaere and Lieven De Strycker DraMCo research group, Katholieke Hogeschool Sint-Lieven Gent, Belgium dramco@kahosl.be Abstract— Development of RSS-based zero-configuration indoor localization systems is a hot-topic nowadays. However many problems arise due to unpredictable radio propagation effects. Several algorithms and commercial applications use the log- distance path loss model as basis for distance calculations. However, this model assumes the absence of fast-fading which is often not the case in reality, when no precautions are taken. In this paper we propose some measures which increase the accuracy of the localization techniques and we evaluate the influence of our changes in comparison to the more commonly used systems. I. INTRODUCTION VER the past decade, wireless networking has emerged in various applications and in a broad range of approaches. From ubiquitous network access via Wireless LAN, over simple wireless data exchange between electronic equipment via Bluetooth, IR or ZigBee to the seemingly infinite number of applications of widespread wireless sensor networks... In these last networks, the physical location of a sensor is one of the most important context parameters. The capability of automatically capturing this context information in wireless-enabled computing devices is therefore highly desirable. One way to determine the position of a so-called blindfolded node is to use the attenuation of the signals between this node and some fixed reference nodes with a known location. The Received Signal Strength (RSS) characterizes the attenuation. Two different approaches exist on how to use the RSS for location estimation. The static approach – often called fingerprinting [1] – uses radio maps to search for the most likely position of the blindfolded node. The survey of these radio maps however, is time-consuming and when the setup has to be changed, the radio maps have to be re-determined. On the other hand, results obtained with fingerprinting are still superior to many other techniques. The dynamic approach tries to bypass these flaws by using some relationship between the RSS and the distance to calculate a position. This way, no exhaustive deployment is needed. In an indoor environment however, conditions are continuously changing and several propagation effects disturb the theoretical relationship between the RSS and the distance. This paper is organized as follows. First we examine two algorithms, suitable for dynamic location estimation (Section II). RSS Self-calibration [2] and Maximum Likelihood Estimation (MLE) [3] both use the same propagation model i.e. the log-distance path loss model or shadowing model, but they start from a different point of view. This makes it interesting to compare both techniques. Next (Section III), indoor radio propagation in terms of RSS is studied to get a better insight into the effects which disturb the expected relationship between the RSS and the distance. Several commercial applications do not cope with these phenomena. Some measures to reduce the influence of the thwarting effects are proposed. Both algorithms are tested and compared in different situations (Section IV). The measurements are performed in a typical indoor environment using a ZigBee network operating at 2.4 GHz. The results should make it possible to state some conclusions (Section V) for further development of RSS- based “zero-configuration” localization systems. II. ALGORITHMS FOR LOCALISATION A. RSS Self-Calibrartion The RSS Self-calibration algorithm proposed in [2] uses the RSS measurements and known distances between the reference nodes to determine the parameters of the log- distance model. This model in terms of RSS can be written as: 0 0 10   d log n RSS     (1) Where: RSS = the Received Signal Strength [dBm] Π 0 = the RSS in dBm at a distance Δ 0 n = the path loss exponent [ ] O