Machine Learning-Based Path Loss Estimation Model for a 2.4 GHz ZigBee Network

Abstract

Wireless sensor networks (WSNs) and Internet of Things (IoT) have received remarkable attention from the past few years in various applications. Modeling of path loss (PL) for the deployment of a developed WSN system is a crucial task owing to the time-consuming and elegant operation. However, radiofrequency (RF) engineers adopted either deterministic or stochastic empirical models to estimate the PL. In general, empirical models utilize predefined influenced parameters including path loss (dB), path loss exponent ($\Upsilon$), and other significant parameters. Although, empirical models differ significantly from original measurement due to consideration of different terrains. In this study, an endeavor has been made to develop a machine learning-based model to estimate the path loss for a standard ZigBee communication network operating on a 2.4 GHz carrier frequency deployed in an urban area. An experimental setup was designed and tested in line-of-sight (LOS) and non-line-of-sight (NLOS) conditions to collect the influence parameters such as received signal strength indicator (RSSI), frequency, distance, and transmitter antenna gain. Besides that, environmental parameters such as temperature and humidity are also included. In this context, a three-layer, feed-forward back-propagation multilayer perception neural network (BPNN) machine learning and Log-Distance empirical models were employed to estimate the PL. The obtained results reveal that the BPNN model noticeably enhanced the coefficient of determination (R²) and reduced root mean square error (RMSE) compared with the empirical model. The R² and RMSE metrics were obtained as 0.97220 and 0.03630 in the NLOS scenario as well as 0.99820 and 0.00773 in the LOS environment, respectively.