Refinement Approach for Node Localization Using Dynamic Cell Structures Neural Networks

Abstract

Wireless Sensor Network (WSN) localization is still an open topic for improvement and continues to prove its necessity in our daily life in parallel with the technological evolution and our needs in various fields. The issue of energy efficiency and network lifetime extension, localization, secure data communication, latency reduction, efficient quality of service, data communication assurance, high scalability also remain topics for improvement.In a context of indoor localization based on learning a database of collected RSSI values, we detail how to proceed with a refinement approach in order to gain in memory, execution time, and localization accuracy. We took advantage of our previous studies in which dynamic cell structure neural networks (DCSNN) have proven superior performance to apply our approach.The results obtained show a memory gain of 82% relative to the data to be extracted from the training database and to be learned, and consequently a gain in execution time and energy consumed. The precision of the coordinates is about 0.031m.Note that the training database is presented as a matrix of RSSI (Received Signal Strength Indicator) values collected by the anchor nodes for each reference point.