Iterative differential autoregressive spectrum estimation for Raman spectrum denoising

Although ultraviolet (UV) laser Raman spectroscopy offers the benefits of stronger signals, partial separation of fluorescence and Raman spectra, and increased eye safety, it suffers from excessive noise, poor resolution, low maturity level, and small intensities of remotely acquired signals and therefore needs to be used in combination with effective denoising techniques. Herein, a denoising approach denoted as iterative differential autoregressive spectrum estimation was developed relying on the assumption that more detailed Raman peaks can be obtained by dividing the Raman spectrum into multiple layers with different intensity levels and estimating the energy distribution of each layer. Specifically, each layer was computed from the difference between the upper layer spectrum and its autoregressive model estimation spectrum, and the energy distribution at progressively lower intensity levels was considered. Compared with traditional techniques, our method exhibited good noise suppression performance and an excellent Raman peak restoration ability while offering the advantages of decreased spectral resolution loss and stable robustness. Cutoff optimization strategies were proposed to improve convergence and noise suppression ability and thus decrease the calculation time to 0.18 s and meet the needs of remote Raman spectrometers for real-time denoising under the condition of long integration. The developed technique paves the way to Raman spectrum denoising based on power spectrum estimation, has a strong adaptive potential, and can be extended to other applications.