Ultra-wideband technology is applied for indoor positioning systems and achieves a position accuracy in the order of decimeters. A trending approach in this context relies on combined angle-of-arrival and time-of-flight measurements, and enables localization using a single anchor, thereby reducing infrastructure overhead. Although analytical models already exist for Ultra-wideband-based distance estimation using time-of-flight, no model has been proposed for its angle-of-arrival counterpart. In this paper we cover this gap by investigating the use of 4 different machine learning regressors to generate such models. The models were trained with data from real-world experiments performed with commercial off-the-shelf Ultra-wideband modules. The models can be easily integrated in simulators, facilitating and even enabling the evaluation of scalable positioning systems using this technology. Among the tested regressors, the random forest regressor presented the best fit to the experimental data, with MAE of the “mean” parameter of 8°.

Hybrid time-of-flight and phase-difference-of-arrival ultra-wideband-based localization systems enable indoor position estimation of mo-bile nodes at a low infrastructure overhead. A known drawback in such systems is the decreasing localization accuracy over distance from node(s) with known location. In this paper we evaluated 3 ML algorithms: Random Forest Regression, LightGBM Regression, and Linear Regression to mitigate this issue. We evaluate the ben-efit of using those methods on experimental data obtained with commercial off-the-shelf transceivers. Our results show that the Random Forest Regression outperforms a typical linear regression correction by 0.49±0.01 (mean±std) meters in positioning accuracy at an computing overhead of nearly 0.34±0.003 [s].