Machine Learning Projection in Performance Evaluation of Cloud Attenuation Prediction Models for Satellite Transmission Quality Improvement

Abstract

Artificial satellite applications to information transmission remain of great importance now and in the foreseeable future. While machine learning is breaking research achievement records for good, the increase of political influence on scientific potentials needs to be managed cohesively by all for sustainability. The reliability of social and business interactions on communication infrastructure determines the technological advancement of every nation – developed or still underdeveloped. In the disclaimer notices of most financial institutions' transaction forms and mandatory customer business agreements, they declared that they are not liable for communication channel malfunction that may lead to transaction interruption, transmission blackout, and subsequent delay in their services. These prescribe effective hydrometeors attenuation margins determination periodically, from more accurate models – such as machine trained ones, to guarantee an increase in reliability of signal transmissions for every geographic location globally. Earlier research works established that required increases in transmission frequency for better efficiency are directly proportional to consequent hydrometeor attenuation on the signal, and that satellite communication unavailability in most tropical regions is above the allowed 1% outage percentage, significantly due to cloud attenuation contribution at satellite bands - which have been increasing consistently. The existence of clouds in tropical climates is almost perpetual, making cloud models all the more fundamental in tropical regions – which include Africa and not less than half of the rest of the world. The published new tropical cloud attenuation algorithm and its resulting new tropical cloud attenuation model (NTM) - derived from it, are hereby further analysed with respect to a wider frequency range. In the primary research of this work, data were collected from a spectrum analyzer, weather-link, and radiosonde equipment. The data were used to calculate values of cloud attenuation by each major existing cloud model in the signal propagation range of 12 to 50 GHz. The predicted cloud attenuation values were spectrally processed and analysed, resulting in the observation that the NTM’s predictions generally average the characteristics prediction values of existing models as shown by presented graphical outputs, though its differences in values relative to each of the other models are substantial in most cases, as either an increase or a reduction. Also, the predicted attenuation values by each of the cloud models converge increasingly direction-wise with frequency. The stated periodicity requirement above in these regards needs a machine learning approach to at least increase the periodicity of the result’s integrity and reliability by several tens of years, for every geographic location globally.