Estimating and predicting link quality in wireless IoT networks

Abstract

The use of poor-quality links in Internet of Things (IoT) networks leads to a bad quality of experience (QoE) with long delivery delays, low reliability, short lifetime of battery-operated nodes, to name but a few. In addition, network resources, such as bandwidth and node energy, are wasted by retransmissions. An accurate estimation of link quality will enable the network to better select the links used for data gathering. Hence, the number of retransmissions needed to achieve the required end-to-end reliability is decreased, leading to shorter end-to-end delivery times, a higher network throughput and an increased network lifetime. In this paper, different link quality estimators are reviewed with a more particular focus on those based on Received Signal Strength Indicator (RSSI) and Packet Delivery Ratio (PDR). We propose to go even further than link quality estimation with link quality prediction. The expected benefit of using link quality prediction is to anticipate link breakages and route changes before loosing packets. It should result in a better QoE provided by the network. For that purpose, we evaluate the performance of four machine learning techniques (i.e. Linear Support Vector Machine, Logistic Regression, Support Vector Machine and Random Forest) working on the traces collected from a real IoT network. They are compared in terms of per-class metrics as well as global metrics. In addition, issues dealing with the deployment of such machine learning techniques in IoT networks with limited resources and energy are presented.