Modelling of the beat of a cardiac signal by Gaussians

The goal of this work is to break up the beat of a cardiac signal ECG into a linear combination of Gaussians with various averages and various standard deviations. For that we carry out a library of 132 Gaussians with 132 different averages and standard deviations. The research of the maximum of the scalar product of each Gaussian of the library by the beat signal ECG to be modelled, permits to find Gaussian the most relevant. Our results shows that the number of Gaussian does not exceed five and that the first Gaussian found in each case, is that which corresponds to wave QRS. At initialization step, the algorithm Orthogonal Forward Regression, achieves an error of 10^-4. The error is established between the real signal of MIT database ECG and the modeled signal. The projected gradient algorithm we achieved an optimization of the first order.