Comparison of Neural Network Training Functions for RSSI Based Indoor Localization Problem in WSN

Abstract

AbstractIndoor target localization using received signal strength indicator (RSSI) is an important research areas in wireless sensor network (WSN) domain. The environmental issues such as reflections, multi path fading pose a major challenge in front of localization systems to achieve high localization accuracy. The important reason behind this is the noise uncertainty in RSSI measurements. The application of artificial neural network (ANN) does not necessitates the prior knowledge of noise distribution and therefore they can be successfully applied in RSSI based target localization applications in indoor environment. However, choosing an appropriate training function for training the ANN is a very crucial task. The objective of this chapter is to compare the performance of 11 different types of training functions used to train the proposed Feed Forward Neural Network (FFNT) for the RSSI based indoor target localization in WSN. The comparison of these training functions is made with respect to Average Localization Error by varying the Noise Variance in the RSSI measurements from 0 dBm to 5 dBm in the steps of 1 dBm. The simulation results conclude that out of all the proposed training functions as well as trilateration-based technique, Levenberg-Marquardt (LM) based FFNT implementation shows higher Average Localization Error and is more consistent in providing better location estimates.KeywordsReceived signal strength indicator (RSSI)Wireless sensor network (WSN)Artificial neural network (ANN)Feed Forward Neural Network (FFNT)Levenberg-Marquardt (LM)