*2.4. Data Analysis*

Before being analyzed and compared, the spectra were numerically elaborated in order to eliminate background signal, limit the signal noise and normalize data. For this purpose, we used mathematical algorithms based on "wavelet" functions. Widely used in the signal analysis, "wavelets" provide an efficient basis for the hierarchical spectrum representation through the so called Discrete Wavelet transform (DWT) [19]. Biorthogonal wavelets based on the B-spline function were used in the framework of software package 'wavelet toolbox' of MATLAB program (by MathWorks Inc.). Low and high scale components of the signal DWT were not be included in the final reconstructed signal, taking away the background and part of the uncorrelated noise, respectively. Finally, the amplitude of the spectral data was normalized by using vector normalization, i.e., by scaling the spectral data set suitably to obtain the standard deviation with respect to the average value equal to 1. The spectra were analyzed in terms of convoluted lorentzian-shaped vibrational modes by performing a best-fit procedure based on the Levenberg–Marquardt nonlinear least-square methods.
