A compensation method of blind additive stationary non-gaussian disturbance suppression

 

A. E. Manokhin

Ural Federal University, Mira st. 19, Ekaterinburg 620002, Russia

 

The paper is received on December 8, 2016

 

Abstract. In a work a method of blind additive stationary non-gaussian disturbance suppression based on blind signal extraction at the output of the neural network according to a criteria (maximizing the kurtosis), and then compensation from the mixing with the signal is presented. The main advantage of this method is the possibility of disturbance suppression without a priori knowing of the signals and disturbances properties apart from some assumptions (belonging to the class of signal and disturbance distributions, the absence of a signal mixed with the disturbance at a certain time, the difference of statistical characteristics of signals and disturbances, some correlation properties of signals and disturbances, etc). Blind disturbance extraction algorithms based on the maximization of absolute and normalized kurtosis, are used; their advantages and weaknesses are identified. Modeling confirmed the efficiency of the compensation method achieving a signal-to-noise power ratio at the output of the neural network from 9 to 17 dB depending on the input signal-to-noise power ratio and used blind algorithms.

Key words: blind disturbance suppression, blind signal extraction, kurtosis, Lagrange function, the rate of convergence.

References

1. A.Cichoki and S.Amari, Adaptive Blind Signal and Image Processing: Learning Algorithms and Applications, John Wiley & Sons, 2002.

2. A. Cichocki, R. Thawonmas, and S. Amari, “Sequential blind signal extraction in order specified by stochastic properties” Electron. Lett., vol. 33, ¹ 1, pp. 64–65, Jan. 1997.

3. Zhi-Lin Zhang, Zhang Yi, Extraction of a Source Signal Whose Kurtosis Value Lies in a Specific Range, Neurocomputing, vol. 69, ¹ 7–9, pp. 894–899, 2006.

4. Zhi-Lin Zhang, Zhang Yi, Robust Extraction of Specific Signals with Temporal Structure, Neurocomputing, vol. 69, ¹ 7-9, pp. 888–893, 2006.

5. S. Amari, “ICA of temporally correlated signals - learning algorithm,” in Proceedings of ICA’99: International workshop on blind signal separation and independent component analysis, Aussois, France, Jan. 1999. pp. 13­­–18.

6. K. Matsuoka, M. Ohya, and M. Kawamoto. A neural net for blind separation of nonstationary signals. Neural Networks, 8(3) : 411–419, 1995.

7. S. Amari. Natural gradient works efficiently in learning. Neural Computation, ¹ 10 : pp. 271–276, 1998.

8. Hyvarinen, A., Karhunen, J. and Oja, E. Independent component analysis. New York: J. Wiley, 2001.

9. P. Comon. Contrast functions for blind deconvolution. IEEE Signal Processing Lett., SPL-3, no. 7 : 209–211, July 1996.

10. O. Shalvi and E. Weinstein. Universal method for blind deconvolution. In S. Haykin, editor, Blind Deconvolution, pages 121–180. Prentice-Hall Inc., Englewood Cliffs, New Jersey, 1994.

11. S.C. Douglas and S.Y. Kung. Kuicnet algorithms for blind deconvolution. In Proc. IEEE Workshop on Neural Networks for Signal Processing, pp. 3–12, Cambridge, UK, August 1998.

12. Sharyi S.P. Kurs vychislitelnyh metodov [Course of computational methods]. — Novosibirsk : Institute of computational technologies of SB SRA, 2013. – 497 p. (In Russian).

13. Norenkov I.P. Ekstremalnye zadachi pri shemotehnicheskom proektirovanii v elektronike [Extremal problems in circuit design in electronics] — Minsk : BSS Publ, 1976. – 239 p. (In Russian).

 

For citation:
A.E.Manokhin. A compensation method of blind additive stationary non-gaussian disturbance suppression . Zhurnal Radioelektroniki - Journal of Radio Electronics, 2017, No. 2. Available at http://jre.cplire.ru/jre/feb17/4/text.pdf. (In Russian)