A Comprehensive Ultra-Wideband (UWB) Dataset for Non-Cooperative Contextual Sensing

This repository presents the dataset accompanying the paper "A comprehensive ultra-wideband dataset for non-cooperative contextual sensing". The UWB dataset is intended for passive localization and Human Activity Recognition (HAR). The experiment was performed in an actual residential environment across several furnished rooms. The participant performed three main activities, namely, walking, sitting, and standing for an approximate experiment duration of 1.6 hours, including steady state (i.e., target not moving). Two UWB systems were used during the experiment. The first system (Qorvo's MDEK1001 kit) was used to obtain the ground truth position of the target while he/she carried a tag and moved within the monitoring area. Ten fixed anchor nodes were used for this purpose. The ground truth 2D XY coordinates reported by the tag were logged at an update rate of 10 Hz using Qorvo's DRTLS android app. The passive UWB system (Qorvo's EVK1000 kit) was solely used to capture Channel Impulse Response (CIR) data from the signals reflected off the target while he/she performed the different activities. It consisted of seven fixed modules installed in a multi-static configuration and operating as transceivers. The modules were programmed with a custom firmware so as to record CIR data on all modules. Each node acted as a transceiver, resulting in 21 bidirectional communication links. The average sampling rate was around 24 Hz between each transmit-receive pair (bidirectional) and therefore considering all 21 bidirectional links, the sampling rate of the passive UWB system amounted to approximately 504 Hz. Video ground truth (with milliseconds timestamps) was used to manually label the activities’ data (sitting, standing, walking and no activity). The dataset has been curated into 2 main dataset files: (i) The file “background_CIR” basically contains the CIR data collected in the environment using the multi-static UWB system when no target was present. (ii) The file “target_CIR” contains the CIR data recorded when the target was performing various activities such as walking, sitting and standing (including the no activity portions). Note that these two dataset files are available in both .mat format and .csv format. Users may choose the format that suits their needs. The dataset files basically consist of complex CIR data (50 samples per measurement), timestamps with milliseconds precision, details about the transmitting and receiving nodes, activity labels, ground truth XY coordinates of the target, as well as some other useful UWB parameters that are reported by the Qorvo's DW1000 chipset, for example, detected first path index and first path amplitudes of CIR measurements, first path and received power levels, CIR power value, preamble accumulation count value, maximum value of noise, standard deviation of noise, etc. This collection also contains a MATLAB script that allows users to load the dataset files (either in .csv format or .mat format) and plot the Channel Frequency Response (CFR) of the UWB data for a desired time segment (between a chosen pair of UWB nodes). Furthermore, this script allows the users to plot the first path power level (in dBm) and ground truth trajectory, and the accumulated and aligned CIR measurements.