文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > A Gauss-Newton method for blind source separation of convolutive mixtures
侵权投诉

A Gauss-Newton method for blind source separation of convolutive mixtures

In this paper we present several Gauss-Newton algorithms for blind source separation of convolutive mixtures. The algorithms can be interpreted as generalizations of two previous approaches due to Gerven-Compernolle [7] and Nguyen-Jutten [5]. Since they ar

Figure 3. Dependence Measure Φ versus Iterations.

A Gauss-Newton method for blind source separation of convolutive mixtures

We have considered two independent uniformly distributedwhite sources which are mixed through the convolutive channelA(z)= 1 -0.19+0.05z 1+0.00z 2+0.20z 3+0.69z 4

0.52+0.38z 1 0.32z 2 0.04z 3 0.33z 4 1

It is apparent that the mixing matrix satisfies the soft mixture

condition. We applied the GN feed-forward algorithm (9) with ={(1,3)}. Cumulants were estimated from 5000 samples of theobservations. After 2 iterations we arrived at the separatingmatrix

)= 1 0.190 0.043z 1 0.011z 2 0.189z 3 0.706z 4B(z

0.512 0.375z 1+0.323z 2+0.028z 3+0.347z 4 1

which is close to the optimum separation matrix

B(z)=

1 A 12(z) A .21(z)1

To measure the performance of the proposed method we useH

h2E=

∑

k

ij[k

, the energy matrix of the overall transfer

ij

matrix H. This matrix is normalized as follows in order that thediagonal elements are equal to one

E=

1 D 1HE+HED 1c

r

Dr=diag maxrows{HE} , Dc=diag columsmax{HE}

Perfect separation of the sources is achieved when E is the

identity matrix. In our computer experiment the initial and finalnormalized energy matrices of the overall transfer function were

S Initial Energy Matrix: 1

0.6277 E= 0.55471 S Final Energy Matrix: 1

0.0005 E= 0.00051

It is clearly seen that an exact separation is almost reached sincethe final energy matrix is close to the identity matrix. A residual

value appears in the non-diagonal elements because we are usingestimates of the cumulants instead of the true values.

Figure 3 illustrates the algorithm rate of convergence bypresenting the dependence measure Φ versus iterations. Thedotted line represents the case of using true cross-cumulantvalues (which corresponds to a quadratic convergence) whereasthe solid line represents the case of using cumulant estimates(which corresponds to a linear convergence). Finally, the dotted-pointed line represents the case of applying the feed-backwardalgorithm (11) using cumulant estimates. Observe that sourceseparation was achieved in all cases.

6. CONCLUSIONS

A new blind source separation algorithm for convolutive softmixtures has been presented. It is based on a Gauss-Newtonalgorithm that minimizes a dependence measure that involvescross-cumulants of the outputs. Exploiting the Jacobian matrixstructure, we found simplified versions of the algorithm withlesser computational complexity. We also showed that existingmethods [5][7] can be interpreted as particular cases of ourapproach. The proposed algorithms exhibit a fast convergence(typically three iterations) because they are of the Gauss-Newtontype. A convergence analysis was also carried out in thesimplified case of instantaneous mixtures of two sources and it isshown that the unique stable stationary point of the feed-forwardalgorithm corresponds to the separation solution.

7. REFERENCES

[1] Ackerman J.,“Sampled-Data Control Systems: Analysis and

Synthesis, Robust Systems Design”. Springer-Verlag, 1995.[2] Comon P., “Independent component analysis, a new

concept? ”. Signal Processing, 36, 287-314 (1994).

[3] Cruces S. and Castedo L. “Blind separation of convolutive

mixtures: A Gauss-Newton algorithm” Proc. of the IEEESP Workshop HOS’97, pp. 326-330,Canada, July 1997.[4] Golub G.H., Van Loan C.F., “Matrix computations”, Johns

Hopkins University Press, third edition, 1996.

[5] Nguyen-Thi H.L. and Jutten C., “Blind source separation

for convolutive mixtures” Signal Processing. 45, pp.209-229, 1995.

[6] Tong L., Liu R.W., Soon V.C., Huang Y.-F. “Indeterminacy

and identifiability of blind identification” IEEE Trans. onCircuits and Systems, 38, 5, pp. 499-509, 1991.

[7] Van-Gerven S. and Van-Compernolle D. “Signal separation

by symmetric adaptive decorrelation: Stability, convergenceand uniqueness” IEEE Trans. on Signal Processing, 43, 7,pp.1602-1612, 1995.

This work has been supported by the Spanish NationalResearch Plan, CICYT (Grant No. TIC96-0500-C10-08).

This work has been supported by the Grants: CICYT (TIC96-0500-C10-02) and Xunta de Galicia (XUGA 10502A96).

上一页第4页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email