Blind Transmitter Localization in Wireless Sensor Networks: A Deep Learning Approach

This paper describes a blind transmitter localization technique based on the deep neural network (DNN) framework. Blind localization assumes no previous knowledge on the transmit signal. It is shown that DNN based location approaches the maximum likelihood solution with reduced computational complexity. Moreover, the maximum likelihood, least squares and radio environment map localization estimators are presented in order to compare the design and performance of the proposed DNN algorithm. The system model is built based on a wireless sensor network that collects received signal strength measurements assuming disturbances of distance dependent correlated shadowing noise. Performance evaluation using numerical simulations shows that the proposed DNN scheme achieves location accuracy similar to the optimum maximum likelihood estimator while presenting computational complexity reduction of more than 90%.