High-sensitivity computational miniaturized terahertz spectrometer using a plasmonic filter array and a modified multilayer residual CNN

Abstract

Spectrometer miniaturization is desired for handheld and portable applications, yet nearly no miniaturized spectrometer is reported operating within terahertz (THz) waveband. Computational strategy, which can acquire incident spectral information through encoding and decoding it using optical devices and reconstruction algorithms, respectively, is widely employed in spectrometer miniaturization as artificial intelligence emerges. We demonstrate a computational miniaturized THz spectrometer, where a plasmonic filter array tailors the spectral response of a blocked-impurity-band detector. Besides, an adaptive deep-learning algorithm is proposed for spectral reconstructions with curbing the negative impact from the optical property of the filter array. Our spectrometer achieves modest spectral resolution (2.3 cm−1) compared with visible and infrared miniaturized spectrometers, outstanding sensitivity (e.g., signal-to-noise ratio, 6.4E6: 1) superior to common benchtop THz spectrometers. The combination of THz optical devices and reconstruction algorithms provides a route toward THz spectrometer miniaturization, and further extends the applicable sphere of the THz spectroscopy technique.

Keywords

• computational miniaturized terahertz spectrometer
• plasmonic filters
• adaptive deep-learning algorithm