Flexible and scalable software defined radio based testbed for large scale body movement

Human activity (HA) sensing is becoming one of the key component in future healthcare system. The prevailing detection techniques for IHA uses ambient sensors, cameras and wearable devices that primarily require strenuous deployment overheads and raise privacy concerns as well. This paper proposes a novel, non-invasive, easily-deployable, flexible and scalable test-bed for identifying large-scale body movements based on Software Defined Radios (SDRs). Two Universal Software Radio Peripheral (USRP) models, working as SDR based transceivers, are used to extract the Channel State Information (CSI) from continuous stream of multiple frequency subcarriers. The variances of amplitude information obtained from CSI data stream are used to infer daily life activities. Different machine learning algorithms namely K-Nearest Neighbour, Decision Tree, Discriminant Analysis and Naïve Bayes are used to evaluate the overall performance of the test-bed. The training, validation and testing processes are performed by considering the time-domain statistical features obtained from CSI data. The K-nearest neighbour outperformed all aforementioned classifiers, providing an accuracy of 89.73%. This preliminary non-invasive work will open a new direction for design of scalable framework for future healthcare systems.