| Popis: |
During the current decade, wireless data traffic has been increasing very rapidly, a trend which is expected to accelerate over the next decade driven by the widespread use of video streaming and the rise of the Internet-of-Things (IoT). In this framework, cellular technology is rapidly moving towards its 5th generation (5G) that will employ millimeter wave (mmWave) frequencies in the attempt to exploit more spectrum and offer multi-Gigabit-per-second (Gbps) data rates to mobile devices. Various propagation phenomena affect adversely mmWave communications, rain fading being the most severe one. The existing ITU-R prediction model for rain induced attenuation over terrestrial line-of-sight (LOS) links does not perform accurately on a global level. This weakness constitutes the main motivation to formulate enhanced models which, by employing appropriate attributes, apply more satisfactorily to specific locations or climatic zones. ITU-R databank includes experimental data of real LOS links operating in various locations that can be used to facilitate supervised machine learning (SML) to formulate methods towards accurate prediction of rain attenuation. Based on a set of past examples or instances, SML aims at exploring/identifying the relationship between a set of descriptive features (inputs) and a target feature (output). After been appropriately trained with past data, SML can be used to make predictions about new instances. This paper proposes a new prediction method which uncovers the latent dependence of rain attenuation on predictors such as path length, operation frequency, wave polarization, rain rate distribution, etc. ensuring high prediction accuracy without necessitating complex mathematical expressions. |
