Mobile contact tracing apps are -in principle -a perfect aid to condemn the human-to-human spread of an infectious disease such as COVID-19 due to the wide use of smartphones worldwide. Yet, the unknown accuracy of contact estimation by wireless technologies hinders the broad use. We address this challenge by conducting a measurement study with a custom testbed to show the benefits and limitations of Bluetooth Low Energy (BLE) in comparison to distance estimation by ultra-wideband (UWB). Our results confirm that BLE-based distance estimation is not sufficient in real scenarios where smartphones are shielded heavily by the users' bodies. Yet, multi-path signal propagation reduces the effect of body shielding. Finally, we demonstrate that UWB is more robust to the environment than BLE.

No abstract

A stepwise feature labeling method for UWB ranging is presented, which allows better separation of LOS and NLOS components in the training data. The packet-by-packet range error evaluation is used as input of the labeling function instead of the conventionally used double-sided two-way ranging result which relies on a cycle of three packets. To assess the packet-wise error, a two-step synchronization scheme is proposed. First, the clock model between anchor and tag is estimated by a least-squares approach. Second, the remaining bias is corrected by determining the time-shift between the channel impulse responses recorded by both nodes. The evaluation of measurement data shows a significant improvement in classification between LOS and NLOS, as well as slightly improved ranging accuracy when used to train a binary classifier.

UWB is one of the main technologies for localization in IoT applications. For range-based localization, it is crucial to secure UWB ranging by a suitable mechanism. Thereby, trustworthiness measures appear to be specifically attractive for constraints posed by IoT applications. In this work, a measure for data trustworthiness of the double-sided two-way-ranging estimate is proposed. The measure relies on features obtained from the channel impulse response and applies two machine learning techniques, namely a modified k nearest neighbour and a modified random forest, to infer an error correction term together with a trust value. To increase the number of trusted measurements, a more accurate stepwise labeling of the training data is used, and an optimum combination scheme of the resulting stepwise trust values is proposed. The results on experimental data show an improvement of 34% RMSE on the test set with 61% of the measurements considered trustworthy.

No abstract