Reproducing Kernel Hilbert Space Models for Signal Processing

This chapter introduces a set of signal models properly defined in a reproducing kernel Hilbert space (RKHS). Also, it focuses on a class of support vector machine (SVM) for digital signal processing (DSP) algorithms that consists of stating the signal model of the time-series structure in the RKHS, and hence they are called RSM algorithms. The use of the theory of reproducing kernels can circumvent this problem by defining nonlinear algorithms by simply replacing dot products in the feature space by an appropriate Mercer kernel function. The most famous example of this kind of approaches is the support vector classification (SVC) algorithm. The chapter pays attention to the fundamental elements of the RSM approach and concentrate on particular signal structures which have been previously studied. Finally, it discusses the use of bootstrap resampling techniques (BRTs) in SVM for RSM models.