A Sugeno-Mamdani Fuzzy System Based Soft Computing Approach Towards Sensor Node Localization with Optimization

Abstract

Localization implies determining or tracking the position of sensor nodes accurately within the deployment area. Most of the localization approaches involve use of some deployed nodes whose position coordinates are already known to us (using Geographical Positioning System (GPS) or some other method) called landmarks or anchors. As fuzzy systems are apt at handling imprecise and uncertain values, this study attempts to leverage the imprecision and uncertainty handling ability of soft computing techniques such as Fuzzy Logic. An aggregated Mamdani- Sugeno Fuzzy Inference System based localization approach has been proposed using triangular membership functions. One input Received Signal Strength Indicator (RSSI), 5 rules and one output (weight) model has been implemented. The output weight indicated the proximity of a particular anchor to an unknown node. The weight was then used in weighted centroid to compute the estimated position of the unknown sensor node. The solution was optimized using Gauss Newton method. The accuracy of the proposed scheme was 50% to 90% better than centroid, weighted centroid and some other works done using soft computing techniques. A coverage of almost 97–98% can be achieved in least computational time. The average processing time of the proposed technique was approximately 1.5 s. The number of anchors required to localize the nodes is also extremely less. Furthermore, being computationally simple, the algorithm does not require any extra hardware and can be implemented in pure decentralized manner. This study also offers new insights into how optimization techniques such as Gauss Newton method can be used to significantly improve the localization accuracy and to solve the problem of localization in large scale sensor networks where number of sensor nodes are in range of thousands.