Unsupervised Learning Based Acoustic NLOS Identification for Smart phone Indoor Positioning

The NLOS phenomenon seriously impair the performance of acoustic indoor localization for smartphones in the real indoor environments. Through identifying and discarding the NLOS measurements, the positioning performance can be improved by incorporating only the LOS measurements. Even though the method based on SL and SSL can obtain a satisfactory accuracy, those methods are very hard to be apply because they need a large number of labeled signals. In this paper, an UL based acoustic NLOS identification approach is proposed. Based on the features extracted from the characteristics of acoustic channel, the performance of UL based methods, such as k-means, GMM and AGNES clustering algorithm, are evaluated and compared under accuracy criterion. The results show that the optimal size of sample set is 100, the accuracy of UL clustering algorithm based on GMM is higher than k-means and AGNES. The optimal feature combination is {tau(med), tau(rms), s, g(m), g(rms)}.