ML-Based Channel Parameters Estimation For Sensing Applications in Industrial IoT Scenarios

This paper analyses an Industrial IoT scenario with sensing capabilities where a set of base stations (BS) is deployed to provide connectivity to devices located over industry assets. The objective of this study is to propose a machine learning (ML)-based channel propagation model which can be applied to sub-6 GHz, mmWaves, and THz frequencies for the purpose of sensing the surrounding environment. The dataset, generated with the 3DScat ray tracing tool and made publicly available, is initially processed by a ML classifier to identify line-of-sight (LoS) and non-LoS (NLoS) links, achieving an accuracy of over 99.1%. Subsequently, the best fit line for both types is derived through regression and it is employed to construct the channel model by extrapolating channel parameters such as the path loss exponent and the standard deviation of shadowing. Finally, we investigate the impact of the industrial layout on channel propagation and network performance, to determine the optimal BS height.