Modeling and inferring the attenuation induced by vegetation barriers at 2G/3G/4G cellular bands using Artificial Neural Networks

Abstract

Modeling vegetation is a recurrent problem for wireless communications industry. The raising number of available frequency bands increases this issue, since most of the existing methods nowadays rely on measurement campaigns. The presence of vegetation in urban areas (such as parks or gardens) is bothersome for radio planners, which have to deal with an in-excess attenuation difficult to predict due to the large number of different cases (i.e. vegetation species, topologies of vegetation volumes, frequencies…). Usually, these vegetation formations appear in the form of forests or barriers, emphasizing the problem, since their impact in the transmitted power is not negligible. This paper proposes the use of Artificial Neural Networks as powerful tools to model and infer the excess attenuation induced by vegetation formations. The study is held at cellular frequency bands (2G/3G/4G) for different vegetation species and barrier configurations, where a multilayer perceptron has been trained over existing experimental data at 2G/3G frequencies. We demonstrate the efficiency of the model to predict accurately the attenuation in the frequencies for which it has been trained for, and to infer and extend the model obtained to new frequencies, e.g. 4G, while maintaining an overall low median error. The proposed framework, which is sought to be a powerful tool for radio planners to predict attenuation due to a vegetation formation, has been validated against measurements conducted in controlled environments at several mobile radio frequencies, but it could be easily extended to other radio frequencies, such as WiFi, WiMax or 5G frequency bands, as long as a proper training is performed, to include different propagation effects at such bands.