6G Sensing Systems: From Radio Channels to Machine Learning (wECHO2)

Sammanfattning

In the context of future 6th generation (6G) and Integrated Sensing and Communication(ISAC) applications, this thesis explores the possibility of using data fromthe S21 parameter for indoor detection of human presence in the 4 GHz to 16 GHzfrequency range. Physical measurements in Chalmers’ antenna lab were combinedwith simulations in a digital twin of the lab environment to analyze how a stationaryhuman affects the wireless channel. The Multiple Signal Classification (MUSIC)algorithm was applied to both physical and simulated data in order to analyze directionsof arrival (DoA) and reflections caused by human presence. The resultsshow that the MUSIC algorithm was able to detect changes caused by a humanin the wireless channel, although the accuracy depended on the environment, theplacement of the antennas as well as the assumed number of sources. Additionally,a Convolutional Neural Network (CNN) was trained on simulated and measureddata in several dataset combinations. As a result it could be observed that combiningsimulated and real data in training improved the models ability to adapt torealistic environments, while training on only simulated data resulted in reducedtransferability to the real measurements.