UE-based Estimation of Uplink Data Rates
Managing heterogeneous networks presents a challenging task and is still a significant area of research. With the proliferation of Internet of Things (IoT) applications, gateway mediated architectures are becoming more common place, i.e. sensors, actuators and services deployed in the physical world utilise a gateway to access cloud services and enterprise applications. Gateways bridge the gap between the physical and digital worlds and are used for many applications in the vehicular domain, including but not limited to remote monitoring and maintenance, navigation, or value-added user services. Many of these services rely on reliable communications to effectively serve application needs, this not only constitutes a good downlink, but also a high quality uplink, for example to provide critical sensor data to safety systems. If the gateway has multiple communication interfaces, such as 2G/3G/4G/5G cellular, WiFi, or satellite, it must make an informed decision on which interface to use in order to best serve the application, and this decision should take into account the available resources such as the uplink capacity.
We use a passive estimation approach based on machine learning, e.g. Neural Network (NN) or Support-Vector Regression (SVR). Every mobile operator has different network characteristics, and for a user the maximum available uplink rate will depend also on many factors such as network load, type of subscription, core network architecture, etc. Therefore, offline machine learning techniques would require hundreds, even thousands of samples to achieve good performance. Gathering sufficient data in a reasonable amount of time would be very expensive and economically impractical for many applications. Hence we online learning approaches that can predict uplink data rates, while adapting to dynamic network conditions. Our algorithms are tested on both3G and 4G networks, and it is demonstrated that the approach can be deployed with minimal effort in a plug and play manner.