Abstract

Wireless Sensor Networks (WSNs) are applied in various fields under the umbrella of Internet of Things (IoT). These sensors are used for remote monitoring, target tracking, efficient transportation, industrial monitoring, and patient observation through physiological sensors, smart agriculture, smart homes, disaster monitoring and efficient management and many other useful applications. Estimation of location of wireless sensor nodes is known as Localization. Localization finds useful applications both indoor and outdoor. Different methods can be used to estimate the location, these include both range based and range free localization techniques. We apply supervised machine learning algorithms to estimate the performance of localization using received signal power (RSSI) as a parameter for wireless sensor networks. IEEECTW Challenge 2019 localization dataset for 1.25 GHz is used for training purpose. It consists of 16 element OFDM sounding data.  The data set is collected through massive MIMO channel sounder. Four algorithms are used for the purpose. The results show  Mean Square Error (MSE)  of Feed Forward Back propagation network (FFBPN) 0.41165 at 63 epoch, Cascaded feed forward neural network (CFFNN) gives 0.3953 at epoch 80, Elman back propagation (EBN) gives 0.38148 at epoch 20 and Nonlinear Autoregressive Network with Exogenous Inputs (NARX) gives 0.0042304 in 9 epochs. Hence, NARX performs the best among other algorithms.

Statistics