Improved Generalized Regression Neural Network for Target Localization

Knowledge of location is of utmost importance in many indoor Location-Based Services (LBS). Although traditional technique such as trilateration involving the use of received signal strengths (RSS’s) is quite popular and simple to use for wireless sensor network (WSN) based target localization, the location estimates obtained using it are not accurate and reliable. The reason behind this is the highly fluctuating nature of RSS’s due to dynamic RF environment and non-linear system dynamics. If the dataset is sparse, the concept of centroid is very useful to estimate fairly closer approximation to the underlying relationship in the given dataset. The GRNN architecture is well known for mapping any nonlinear relationship between input and output. To address the problems with the RSS based target localization and tracking (L&T) using WSN for indoor environment, a novel range free Centroid Generalized Regression Neural Network (C-GRNN) algorithm is presented in this paper. The proposed C-GRNN algorithm is formed by combining the advantages of both centroid and GRNN. In order to realize the dynamicity in given RF environment, the variance in the RSSI measurements is varied from 3 to 6 dBm. During simulation experiments, although the variance in the RSSI measurements is doubled, the average RMSE and average localization error are increased by only approximately 28.31%, and 22.28% respectively. This rise in localization errors with the proposed C-GRNN architecture is very less as compared to the trilateration as well as GRNN based technique.