Deep Neural Network-Based Human Activity Classifier in 60 GHz WLAN Channels

In 6G systems, close integration of communication and sensing is envisaged. Human monitoring and in-air gesture recognition having a wide range of applications in smart homes, emergency systems or games, represents a typical sensing task where the monitoring of received communication signals has the potential to substitute currently used video cameras or dedicated sensors. Although there is still a long path until the real deployment of 6G will become available, it is already possible to test the main ideas of integrated sensing and communications within the current standards. The paper addresses Deep Learning based estimation of several typical people’s activity from WiFi-like signals in 60 GHz channel, that can be deployed in future smart access points. As shown in previous works, each activity generates a unique Doppler pattern that is extracted from the received signal. As the Convolutional Neural Network was used for classification, this paper also describes the method for obtaining a sufficient number of training samples for each activity. We also try to investigate classifier performance deterioration by varying distance of person from test-bed antennas and by testing on persons that have not participated during the classifier training.