DeepRange

Abstract

Acoustic ranging is a technique for estimating the distance between two objects using acoustic signals, which plays a critical role in many applications, such as motion tracking, gesture/activity recognition, and indoor localization. Although many ranging algorithms have been developed, their performance still degrades significantly under strong noise, interference and hardware limitations. To improve the robustness of the ranging system, in this paper we develop a Deep learning based Ranging system, called DeepRange. We first develop an effective mechanism to generate synthetic training data that captures noise, speaker/mic distortion, and interference in the signals and remove the need of collecting a large volume of training data. We then design a deep range neural network (DRNet) to estimate distance. Our design is inspired by signal processing that ultra-long convolution kernel sizes help to combat the noise and interference. We further apply an ensemble method to enhance the performance. Moreover, we analyze and visualize the network neurons and filters, and identify a few important findings that can be useful for improving the design of signal processing algorithms. Finally, we implement and evaluate DeepRangeusing 11 smartphones with different brands and models, 4 environments (i.e., a lab, a conference room, a corridor, and a cubic area), and 10 users. Our results show that DRNet significantly outperforms existing ranging algorithms.