A Contextual Bandit Approach to the Interface Selection Problem

Increasing public trust in the Industrial Internet of Things (IIoT) requires establishing a strong reliability of the technology. This does not only apply to the participants in the wireless network, but also to the backbone that connects the network to the Internet. A particular challenge is present in the railway domain where a gateway on a train is forwarding data from sensors and/or cameras to the cloud using mobile uplink transmissions. Trains often travel large distances between areas with good coverage and high congestion (train stations, urban environments) and areas with very poor coverage with low congestion (rural areas between cities). One way to achieve reliable data delivery is to utilize the advantages of heterogeneous wireless networks by implementing an intelligent interface selection methodology. In this context we present an approach based on an enhanced version of the linear upper confidence bound (LinUCB) algorithm. The algorithm is able to translate channel quality parameters into an estimated data rate and perform interface selection such that not only throughput, but also reliability are maximized. We present two main contributions in this paper: First, to the best of our knowledge, there has not been any other research that considers using this algorithm in the context of wireless network selection. Second, we modify the algorithm so that it deals with scenarios where severe network uplink congestion is often experienced (e.g. at train stations, airports).