Research and application of noise suppression based on support vector machine

Support vector machine (SVM), built on statistical learning theory (SLT), was proposed by Vapnik in 1995. Being based on structural risk minimization (SRM) principle, SVM has a better generalization performance in comparison to those traditional methods on learning problems. Originally, SVM was used to construct classifiers for pattern recognition, and recently it has been extended to many fields, e.g. function regression and density estimation. In this paper, SVM is introduced into signal processing, namely, the generalization ability of SVM is utilized to suppress additive random noise. The principle of noise suppression is studied in both time domain and frequency domain. The signal resumed from noise with SVM is expressed in frequency domain. The effect on performance of noise suppression when choosing parameters /spl tau/ and /spl epsi/ is analyzed. Simulations are made to validate the theoretical analysis. At last a conclusion is drawn that, as a new method, SVM has a good performance on the noise suppression.