Class Imbalance Learning for Identifying NLOS in UWB Positioning

Non Line of Sight propagation is an important reason effecting of the positioning accuracy of Ultra- wide Bandwidth system. It’s difficult to model and distinguish NLOS signal, as the characteristics of NLOS signal are closely related to the environment. Considering the characteristic that LOS and NLOS signals are very imbalance in UWB positioning system, a NLOS signal recognition method based on the class imbalance learning is proposed. In order to recognize NLOS signals which are rare but important, a Support Vector Data Description learning machine with negative is trained with the NLOS signal and the LOS signal with different misclassification cost. The simulation results show that the performance of this method is better than to Support Vector Machine, while the number of LOS signals and NLOS signals is extreme imbalance.