A Deep Learning Framework for Modeling Wireless Propagation Channel

Traditionally, the development of wireless channel propagation models has been based on mathematical analysis, measurements of received signal levels, or a mixture of both, which make possible to define propagation models for planning wireless networks. Additionally, the development of traditional Machine Learning and Deep Learning techniques in conjunction with the measurements of reception levels is an effective alternative to continue developing this topic. The aim of this article is to present a Deep Learning framework that allows to predict the value of the power received as a distance function over a wireless channel in the 5.8 GHz band with WiMAX technology. Our approach considers performing a nonlinear regression of data by using the autoencoder to determine signal levels as a distance function. We work with a database, which contains information related to transmitting power measurements. Then, we proposed the use of Data Augmentation technique based on Bootstrap method. Finally, these data are inserted into a non-linear regression system by considering an autoencoder, and we compare their results with other proposals. The obtained result shows an improvement of 14.5% related to the mean square error compared with other models.