1: \begin{abstract}

2: We present a novel, robust sound source localization algorithm considering

3: 	back-propagation signals.  Sound propagation paths are estimated by

4: 	generating direct and reflection acoustic rays based on ray tracing in

5: 	a backward manner.  We then compute the back-propagation signals by

6: 	designing and using the impulse response of the backward sound

7: 	propagation based on the acoustic ray paths.  For identifying the 3D

8: 	source position, we suggest a localization method based on the

9: 	Monte Carlo localization algorithm.  Candidates for a source

10: 	position is determined by identifying the convergence regions of

11: 	acoustic ray paths.  This candidate 

12: %for a source position 

13: 	is validated

14: 	by measuring similarities between back-propagation signals, under the assumption that

15: 	the back-propagation signals of different acoustic ray paths should be

16: 	similar near the sound source position.  Thanks to considering

17: 	similarities of back-propagation signals, our approach can localize a

18: 	source position with an averaged error of 0.51~m in a room of 7~m by 7~m area with 3~m height

19: 	in tested environments.  We also observe 65~\% to 220~\% improvement

20: in accuracy over the state-of-the-art method.  This improvement 

21: %by the proposed

22: %technique 

23: is achieved in environments containing a moving source, an

24: obstacle, and noises.

25: 	%This improvement is mainly thanks to the analysis of the back-propagation signals in complex environments containing a moving source, an obstacle, and noises.

26: %	considering back-propagation signals.

27: %We observe that our approach is robust in complex environments containing a moving source, an obstacle, and noises; 65~\% to 220~\% improvement in accuracy over the state-or-the-art method.

28: \end{abstract}

29: