User Localisation in Massive MIMO Networks

Humankind has always experienced the need to explore its surroundings. To do so, already early on in the history of humankind, navigation methods were conceived. These methods relied on observing the sun, the moon and the stars. In the modern world, we have developed more sophisticated ways of navigating, still, these rely on the same principles of observing sources of electromagnetic waves. However, in this case, the celestial bodies have been replaced by artificial signal sources. The most widely used localisation system exists out of the GNSS systems. These rely on satellites orbiting the earth and broadcasting there location and time. In this way, we can globally localise ourselves within a couple of meters. However, soon autonomous robots will join humans in roaming around society. These robots require a much higher localisation accuracy to operate. In addition, both indoor and outdoor locations must be covered by a reliable fast localisation system. Furthermore, the system that will provide these localisation services preferably does not require us to deploy a large amount of proprietary localisation beacons. The upcoming 5G networks will use Massive MIMO (MaMIMO) base stations, which use many antennas to beam the communication signals towards the intended users. MaMIMO systems rely on measuring the Channel State Information (CSI) of the users to beam the power towards them using precoders. As the CSI can be used to spatially multiplex the users, it must contain information about the propagation environment and the location of the user inside this environment. This opportunity can be used to build localisation services on top of the communication network. As cellular networks have a large coverage, we can offer localisation services without the need for deploying extra infrastructure or altering the working of the already deployed communication system. In this PhD project, we study data-driven methods to localisation using a real-life Massive MIMO testbed. Data-driven localisation uses a database to compare the fingerprint of the user, in this case the CSI, with the fingerprints in the database. As a result, the accuracy of data-driven methods is bound to the quality of the database. We extend the KU Leuven MaMIMO testbed to construct large MaMIMO CSI datasets. By using robotical positioners to move the users, the measurement campaigns are automated while reaching an accuracy of the spatial label of less than 0.1mm. The first recorded dataset contains the channel of more than 250.000 user locations, for four different MaMIMO BS antenna topologies. The second dataset contains time series of captured channels while the environment is changing. The datasets have been published under an open-access license and have been used in several studies. The datasets are used to train Convolutional Neural Networks (CNNs) to localise users in the indoor environment. We study the different design aspects of data-driven localisation methods using the measured datasets.

Publication year:2022

Accessibility:Open