Indoor Time-of-Flight Ranging Under NLOS Conditions In Uncharacterized Environments

Description

Wireless ranging is the estimation of the distance between two wireless nodes and is a core component of localization and tracking systems. Ranging measurements from multiple reference points to an object are combined to estimate an object’s current position, and to reliably do localization requires accurate ranging measurements. Time-of- flight(ToF) ranging with Ultrawideband(UWB) has received renewed attention due to growing prevalence of this technology, e.g. increased availability of chipsets supporting the IEEE 802.15.4a standard for UWB ranging, and the introduction of UWB in Apple iPhones and AirTags. However, UWB has long been considered a prominent wireless ranging technology, which has distinguished itself from other wireless ranging approaches, with its ability to achieve highly accurate and precise distance estimates. The expected accuracy of UWB ranging, however, is generally achieved in line-of-sight( LOS) ranging cases, with distance error increasing significantly in non-line-of-sight( NLOS) cases, i.e. when there is an obstacle in the ranging path that forces the signal to travel around or through. The impact of these ranging errors can be reduced if we can 1) detect if ranging was done under NLOS conditions (NLOS identification), and 2) correct the ranging distance (NLOS mitigation). Existing methods for NLOS identification and mitigation are built on classification and regression models, respectively, which are trained using channel impulse responses(CIRs) gathered from extensive measurement campaigns, i.e. characterization of each site where the ranging system will be deployed. This complicates system deployment to new sites, even if the environment is similar to previous sites, i.e. if we have characterized an office environment and we deploy the system to another office with different dimensions and layout, we would need to characterize the new site. It is also not ideal in dynamic environments, i.e. if a site layout changes, then it needs to be characterized again. We address this problem by developing methods for NLOS identification and mitigation that can allow adaptation to uncharacterized environments. Specifically, we will design and evaluate new classification and regression models that can be trained with measurements from one environment, but can then also be used in other environments with little to no additional measurement and retraining. Developing environment-independent models for NLOS ranging is a largely unexplored topic in related literature. Our project therefore has scientific merit, initiating early formative work on machine learning(ML) models designed specifically for NLOS identification and mitigation in uncharacterized environments, and the potential to improve current industrial UWB ranging systems.