U-Net-based deep learning enabling denoised spectroscopy for terahertz quantum cascade laser.

Area of Science:

• Spectroscopy and Imaging
• Quantum Cascade Lasers
• Signal Processing

Background:

• Terahertz quantum cascade lasers (THz QCL) offer potential for high-resolution spectroscopy and imaging.
• Traditional Fourier techniques limit accuracy and signal-to-noise ratio (SNR) in THz QCL applications.
• Low SNR and accuracy hinder the practical use of THz QCL in sensing.

Purpose of the Study:

• To propose a novel hybrid denoising framework for enhancing multimode THz QCL spectra.
• To improve the accuracy and SNR of spectral data obtained from THz QCL.
• To address the limitations of traditional spectral analysis methods in THz sensing.

Main Methods:

• A hybrid denoising framework combining wavelet pre-processing and a peak-aware U-Net network was developed.
• Wavelet transform was used to remove high-frequency noise from the spectra.
• A U-Net network with a peak-aware mechanism reconstructed spectral details and enhanced peaks.

Main Results:

• The proposed method achieved an SNR enhancement of approximately 10 dB.
• A determination coefficient of 0.979 and a reconstruction error of 3.7 × 10^-5 were obtained.
• The framework significantly outperformed the original data and accurately recovered spectral information from methanol gas transmission spectra.

Conclusions:

• Deep learning-assisted signal processing shows strong capability in advancing THz spectroscopic sensing.
• The hybrid denoising framework offers high accuracy and robustness for THz QCL applications.
• This approach paves the way for more reliable and precise THz spectroscopic sensing and imaging.