0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
recover1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
Figure 2: Demonstration of the separation of two signals from one mixture. The first source is a thunder sound, and the second one is a fire sound. The green line are receivers;
red lines are sources and the blue lines are recoveries.
0 10 20 30 40 50 60 70 80 90 100
−1.5
−1
−0.5 0 0.5 1
1.5 Estimated A(1,1)
Estimated A(1,2)
Figure 3: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Figure 4: Demonstration of the separation of two signals from one mixture. The first source is a ocean wave sound, the second one is a rain sound, and the third source is a wind sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
Figure 5: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
Figure 6: Demonstration of the separation of three signals from two mixtures. The first source is a thunder sound, the second one is a water sound, and the third a source is fire sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
0 20 40 60 80 100 120 140 160 180 200
Figure 7: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
Figure 8: Demonstration of the separation of three signals from two mixtures. The first source is a thunder sound, the second one is a water sound, and the third source is a fire sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
0 20 40 60 80 100 120 140 160 180 200
Figure 9: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source4
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover4
Figure 10: Demonstration of the separation of four signals from three mixtures. The first source a is speech signal, the second one is a rain sound, the third source is a speech signal, and the final source is a thunder sound. The green line are receivers; red lines are sources and the blue lines are adjusted recoveries.
0 50 100 150 200 250
Figure 11: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
mixture
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
source2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
recover1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1
recover2
Figure 12: Demonstration of the separation of two signals from one mixture. The first source is a thunder sound, and the second one is a fire sound. The green line are receivers;
red lines are sources and the blue lines are recoveries.
0 10 20 30 40 50 60 70 80 90 100
−1
−0.5 0 0.5 1
1.5 Estimated A(1,1)
Estimated A(1,2)
Figure 13: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Figure 14: Demonstration of the separation of three signals from two mixtures. The first source is a ocean wave sound, the second one is a rain sound, and the third source is a wind sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
Figure 15: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
Figure 16: Demonstration of the separation of three signals from two mixtures. The first source is a thunder sound, the second one is a water sound, and the third source is a fire sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
0 20 40 60 80 100 120 140 160 180 200
Figure 17: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
Figure 18: Demonstration of the separation of three signals from two mixtures. The first source is a thunder sound, the second one is a water sound, and the third source is a fire sound. The green line are receivers; red lines are sources and the blue lines are recoveries.
0 20 40 60 80 100 120 140 160 180 200
Figure 19: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 mixture3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 source4
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover2
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover3
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
−1 0 1 recover4
Figure 20: Demonstration of the separation of four signals from three mixtures. The first source is a speech signal, the second one is a rain sound, the third source is a speech signal, and the final source is a thunder sound. The green line are receivers; red lines are sources and the blue lines are adjusted recoveries.
0 50 100 150 200 250
Figure 21: The estimations of mixing matrix A for each iteration of the null-space algo-rithm.
0 100 200 300 400 500 600 700 800 900 1000
100 200 300 400 500 600 700 800 900 1000
−10 0 10 source1
100 200 300 400 500 600 700 800 900 1000
−5
100 200 300 400 500 600 700 800 900 1000
−10 0 10 recover1
100 200 300 400 500 600 700 800 900 1000
−5
Figure 22: Demonstration of the separation of three signals from two mixtures. The green line are receivers; red lines are sources and the blue lines are recoveries.
0 100 200 300 400 500 600 700 800 900 1000
100 200 300 400 500 600 700 800 900 1000
−10
100 200 300 400 500 600 700 800 900 1000
−10 0 10 recover3
Figure 23: Demonstration of the separation of three signals from two mixtures. The green line are receivers; red lines are sources and the blue lines are recoveries.
References
[1] Bell, A. J. and Sejnowski, T. J. (1995). An in formation maximization approach to blind separation and blind deconvolution. Neural Computation, 7(6), 1129-1159.
[2] Bell, A. J. and Sejnowski, T. J. (1996). Learning the higher-order structure of a natural sound. Network: Computation in Neural System, 7, 261-266.
[3] Bell, A. J. and Sejnowski, T. J. (1997). The ’Independent components’ of natural scenes are edge filters. Vision Research, 37, 3327-3338.
[4] Box, G. E. P., Jenkins, G. M. and Reinsel, G. C. (1994). Time Series Analysis:
Forecasting and Control (3rd Edt.). Prentice Hall.
[5] Chen, R. -B. and Wu, Y. N. (2002). A Null-space Representation for Overcom-plete Independent Component Analysis, 2002 Proceedings of American Statistical Association, Statistical Computing Section [CD-ROM]. Alexandria, VA: American Statistical Association.
[6] Chen, R. -B. (2003). A Null-space Algorithm for Overcomplete Blind Source Sepa-ration. Ph.D. dissertation, Department of Statistics, University of California at Los Angeles.
[7] Comon, P. (1994). Independent component analysis - a new concept? Signal Pro-cessing, 36(3), 287-314.
[8] Godsill, S. J. (1997). Bayesian enhancement if speech and audio signals which can be modelled as ARMA processes. International Statistical Review, 65(1), 1-21.
[9] Lee, T. -W., Lewicki, M. S., Girolami, M. and Sejnowski, T. J. (1999). Blind source separation of more source then mixtures using overcomplete representation. IEEE Signal Processing Letters, 6(4), 87-90.
[10] Lee, T. -W., Girolami, M., Bell, A. J., Sejnowski, T. J. (2000). A unifying information-theoretic framework for independent component analysis. Computers and Mathematics with Applications, 31, 1-12.
[11] Lewicki, M. S. and Olshausen, B. A. (1999). A probabilistic framework for the adap-tation and comparison of image codes. J. Opt. Soc. AM. A: Optics, Image Science, and Vision, 16(7), 1587-1601.
[12] Lewicki, M. S. and Sejnowski, T. J. (2000). Learning overcomplete representations.
Neural Computation, 12, 337-365.
[13] Pearlmutter, B. and Parra, L. (1996). A context-sensitive generalization of ICA. In:
ICONIP’96, 151-157.
[14] Tanner, M. and Wong, W. (1987). The calculation of posrterior distributions bt data augmentation. Journal of the American Statistical Association, 82, 528-550.