1 / 60

06-07-2010, University of Oldenburg, MEDI-AKU-SIGNAL Kolloquium

Signal Processing Algorithms for Wireless Acoustic Sensor Networks Alexander Bertrand Electrical Engineering Department (ESAT) Katholieke Universiteit Leuven. 06-07-2010, University of Oldenburg, MEDI-AKU-SIGNAL Kolloquium. Outline. Introduction Multi-channel Wiener filter (MWF)

ursa-cook
Download Presentation

06-07-2010, University of Oldenburg, MEDI-AKU-SIGNAL Kolloquium

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Signal Processing Algorithms for Wireless Acoustic Sensor NetworksAlexander BertrandElectrical Engineering Department (ESAT)Katholieke Universiteit Leuven 06-07-2010, University of Oldenburg, MEDI-AKU-SIGNAL Kolloquium

  2. Outline Introduction Multi-channel Wiener filter (MWF) Example: distributed MWF in binaural hearing aids DANSE in fully connected WASN Tree-DANSE Multi-speaker VAD Noise reduction Tracking of speech power

  3. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD

  4. Traditional sensor array DSP  known / fixed sensor positions  Sharp angle  centralized processing  #microphones is limited  Long distance(SNR drops 6dB for each doubling of distance) Sensor array DSP 4

  5. Distributed sensor arrays Wireless acoustic sensor network (WASN) • More spatial information • More sensors • Subset: high SNR recordings 5

  6. Challenges Distributed sensor arrays 4) Subset selection 3) Distributed processing 1) Unknown/changing positions, link failure  ADAPTIVE 2) Bandwidth efficiency 6

  7. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD

  8. Multi-channel Wiener Filtering (MWF) • Goal: estimate speech component in 1 of the N microphones • Output = sum of filtered microphone signals: W1 + W2 Clean speech W3 W4

  9. Multi-channel Wiener Filtering (MWF) • Goal: estimate speech component in 1 of the N microphones • Output = sum of filtered microphone signals: W1 + W2 Clean speech W3 W4

  10. Multi-channel Wiener Filtering (MWF) • Goal: estimate speech component in 1 of the N microphones • Output = sum of filtered microphone signals: • Needs: - N x N noise+speech correlation matrix Ryy - N x 1 clean speech correlation (column of Rdd) • Rddcan be estimated using Rdd= Ryy- Rnn using voice activity detection (VAD)mechanism W1 + W2 Clean speech W3 W4

  11. Multi-channel Wiener Filtering (MWF) • RECAP • Given: N microphone signals • Choose one (arbitrary) reference microphone • MWF computes optimal filters such that sum of outputs is as close as possible to speech component in target microphone

  12. Noise frame: destructive interference  Noise = electro music F1 + F2 F3 F4

  13. Speech frame: constructive interference  Noise = electro music F1 + F2 F3 F4

  14. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD • Subset selection • Conclusions

  15. Example: binaural hearing aids large bandwidth needed full matrix inversion = 2-node WASN Binaural link MWF left MWF right 15

  16. Example: binaural hearing aids + + Converges to optimum if single desired source (Doclo et al., 2007) Binaural link w11 g12 g21 w22 16

  17. Motivation for DANSE • > 2 nodes ?e.g. supporting external sensor nodes or multiple hearing aid users. 17

  18. Motivation for DANSE • > 2 nodes ?e.g. supporting external sensor nodes or multiple hearing aid users. 18

  19. Motivation for DANSE • > 2 nodes ?e.g. supporting external sensor nodes or multiple hearing aid users. 19

  20. Motivation for DANSE • > 2 nodes ?e.g. supporting external sensor nodes or multiple hearing aid users. 20

  21. Motivation for DANSE • > 2 nodes • Multiple desired sourcese.g. conversation monitoring. 21

  22. Motivation for DANSE • > 2 nodes • Multiple desired sourcese.g. conversation monitoring. 22

  23. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD

  24. DANSE • Previous requires more general framework:Distributed adaptive node-specific signal estimation (DANSE) • Allows for multiple nodes (fully connected topology) • Allows for multiple target sources: Estimating K sources requires communication of K-channel signals(DANSEK) 24

  25. DANSE • Considered here: • Fully connected WSN • Multi-channel sensor signal observations • Goal: each node estimates node-specific signal, but common latent signal subspace (dimension= # targets)

  26. 3 nodes, fully connected 26

  27. Binaural hearing aids (revisited) + + Binaural link w11 g12 g21 w22 27

  28. Binaural hearing aids (revisited) + + Converges to optimum if #desired sources ≤ 2 auxiliary channels(capture signal space) J=2, DANSE2 (K=2) Binaural link w11(2) g12(2) g21(2) w22(2) w11(1) g12(1) g21(1) w22(1) 28

  29. Binaural hearing aids (revisited) + + Converges to optimum if K= # desired sources J=2, DANSEK Binaural link 29

  30. Sequential updating Sequential round-robin update

  31. DANSE with simultaneous updating • Simultaneous updating: parallel computing • Sometimes convergence to optimal solution, but not always • Solution:relaxationyields convergence and optimality: 31

  32. DANSE with simultaneous updating Without relaxation (S-DANSE) 4 nodes, 3-6 sensors/node 32

  33. DANSE with simultaneous updating With relaxation (rS-DANSE) 4 nodes, 3-6 sensors/node 33

  34. DANSE audio demo (tracking omitted) Unfiltered Centralized MWF rS-DANSE 34

  35. Robust DANSE • Theory: DANSE == centralized MWF, but… 35

  36. Robust DANSE • Numerical errors due to: • Estimation errors in Rdd (especially at low SNR nodes)  ripple effect • Reference microphones are close to each other ill-conditioned basis for signal subspace • Solution: estimate speech component in communicated signals, preferably from high SNR nodes (= Robust DANSE or R-DANSE) • Convergence is proven under certain dependency conditions 36

  37. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD

  38. What if not fully connected?

  39. What if not fully connected? Nodes must pass on information from other nodes 1) Nodes act as relays(virtually fully connected): - huge increase in bandwidth if limited connections - routing problem 2) Nodes broadcast the sum of all filtered inputs: - no increase in bandwidth - no routing problem (?)

  40. What if not fully connected? 40

  41. What if not fully connected? FEEDBACK !!

  42. What if not fully connected? • Intuition • Theoretical analysis • Conclusion: feedback causes major problems • Direct feedback (one edge) vs. indirect feedback (loops)

  43. Direct feedback cancellation • Transmitter feedback cancellation

  44. Direct feedback cancellation • Receiver feedback cancellation

  45. What if not fully connected? • Intuition • Theoretical analysis • Conclusion: feedback causes major problems • Direct feedback (one edge) vs. indirect feedback (loops) • Prune to tree topology  T-DANSE (= still optimal output!!)

  46. Outline • Introduction • Multi-channel Wiener filter (MWF) • Example: distributed MWF in binaural hearing aids • DANSE in fully connected WASN • Tree-DANSE • Multi-speaker VAD

  47. Multi-speaker VAD speaker microphone - Goal: Track individual speech power of multiple simultaneous speakers or other non-stationary sources (VAD) - Exploit spatial diversity from WASN 47

  48. Multi-speaker VAD WASN’s ! • Ad-hoc microphone array • Assumptions: • Speakers in near-field • Speakers are independent • Limited noise/reverberance • Sources to track are well-grounded (= they attain zero-values) • Advantages: • Array geometry unknown • Speaker positions unknown • Energy-based low data rate  synchronization not crucial 48

  49. Data model

  50. Data model

More Related