Understanding sub-GHz Signal Behaviour in Deep-Indoor Scenarios

Critical Internet of Things (IoT) services require seamless connectivity, which is not always simple to provide and particularly in deep-indoor scenarios, it can be even impossible in some cases. The existing outdoor-to-indoor path loss models lack the accuracy in the underground situations, thus IoT coverage planning in such areas cannot rely on robust tools and becomes a process of trial and error. In this work, we derive and analyse various environmental features that can be useful in understanding sub-GHz deep-indoor signal propagation. Based on a large-scale field trial in an underground tunnel system, we formulate several parameters related to TX-RX distance and tunnel geometry. Through feature relevance studies in linear (Ordinary Least Squares (OLS) regression) and non-linear (Gaussian Process Regression) realms we show that 2D indoor distance and the distances to the tunnel walls may be useful in sub-GHz signal strength prediction in deep-indoor situations. We construct a linear and a Gaussian Process model for indoor path-loss prediction that outperform the 3rd Generation Partnership Project (3GPP) model by 1.8 dB and 4.1 dB, respectively.