Method for superior denoising of UV/Vis/NIR transmittance spectra of thin films

UV/Vis/NIR transmittance spectra T(lambda) are often used for the characterization of thin films in both spectrophotometry and spectroscopic ellipsometry. T(lambda) are inherently noisy due to noise generated by the measuring equipment and the environment. Nevertheless, film characterizations are usually performed either without denoising T(lambda) or by smoothing it, which should limit the characterization accuracy. In this study is proposed a method, abbreviated as SMEDM, for denoising of UV/Vis/NIR T(lambda). The input to SMEDM consists of several intrinsic mode functions (IMFs) obtained from the decomposition of T(lambda) by the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN). In essence, SMEDM uses smoothed median envelopes of IMFs containing apparent noise features and computes the noise of T(lambda). Eight model spectra T(lambda) of a thin film on a thick substrate and two measured spectra T(lambda) of such samples are denoised by SMEDM and other methods most suitable for denoising of such spectra. It is demonstrated that, in all studied cases, the most accurate denoising of the model spectra is obtained by SMEDM utilizing CEEMDAN. Magnitude accuracy of the computed noise MACN > 70% is achieved even for model noises with a magnitude smaller than that of the two experimental spectra.