Automatic identification of home IOT devices for traffic management in satellite networks

The widespread emergence of Internet-connected devices raises concerns about Internet of Things (IoT) traffic management in a consumer satellite network. For efficient IoT traffic management of such networks, detecting a device's semantic type is extremely important. This paper covers the challenges and solutions in identifying devices based on signatures of traffic flowing between locally connected devices and over-the-air sessions from devices behind Very Small Aperture Terminals (VSATs). Machine Learning (ML) techniques are used to predict device characteristics, and then IoT traffic can be carried via satellite channels according to the rule defined for each device type. For the first time, device-based IoT traffic classification is applied to a satellite network that greatly differs from terrestrial networks due to its resource-constrained return link. A supervised machine learning, Random Forest Classification algorithm is applied to multiple continuous datasets, and then resulting estimators are combined to produce a model using which the type and behaviour of an IoT device are predicted. Test results show an encouraging accuracy. The paper proposes a forward-looking federated learning model training that would produce better accuracy where a server in a central location trains a global learning model using machine learning parameters received from multiple edge VSATs.