EUVIMEDEUVIMED entdecken
Crossref

Machine Learning-Based Denoising for Single-Photon Detection With MKIDs

Viktor Makhrinov, Michael Arena, Aled Cuda, Mehdi Bagheri, Akhan Almagambetov, Aziret-Sultan Kiyazov, Ryan D. Martin, Bruce Grossan, George F. Smoot, Mehdi Shafiee

IEEE Transactions on Applied Superconductivity · 2026-11 · 36 (8) · 2402715-2402715

DOI: 10.1109/tasc.2026.3718447 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Applied Superconductivity, 2026-11. Verfasst von Viktor Makhrinov und 9 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: Ministry of Science and Higher Education of the Republic of Kazakhstan, Nazarbayev University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Microwave kinetic inductance detectors (MKIDs) enable single-photon detection with intrinsic energy discrimination by encoding photon energy in the amplitude of the transient phase response of superconducting resonators. In practical operation, however, noisy signals distort photon-induced phase pulses and limit achievable energy resolution. In this work, we investigate denoising of optically induced single-photon signals recorded from a 25-pixel MKID array under controlled illumination at 660, 780, and 1310 nm. We perform a comparative analysis of conventional signal-processing filters and machine learning approaches, including recurrent neural-network architectures trained using a Noise2Noise self-supervised strategy without clean reference targets. Denoising performance is evaluated using physically meaningful metrics related to detector spectroscopy, including peak depth, full width at half maximum (FWHM), and resolution-related parameters, in addition to reconstruction error. While traditional filters reduce baseline fluctuations, their effectiveness depends strongly on illumination wavelength, and they may distort pulse morphology. In contrast, a gated recurrent unit (GRU) autoencoder consistently preserves pulse shape and improves the resolution parameter R by a factor of 2, relative to the original signals, across all illumination wavelengths. These results highlight the importance of adaptive, time-series aware denoising techniques for realistic MKID photon measurements and demonstrate the potential of machine learning to enhance spectroscopy performance under varying experimental conditions.

Förderung

  • Ministry of Science and Higher Education of the Republic of Kazakhstan
  • Nazarbayev University

Zitieren

APA

Makhrinov, V., Arena, M., Cuda, A., Bagheri, M., Almagambetov, A., Kiyazov, A., Martin, R. D., Grossan, B., Smoot, G. F., & Shafiee, M. (2026). Machine Learning-Based Denoising for Single-Photon Detection With MKIDs. IEEE Transactions on Applied Superconductivity, 36(8), 2402715-2402715. https://doi.org/10.1109/tasc.2026.3718447

Vancouver

Makhrinov V, Arena M, Cuda A, Bagheri M, Almagambetov A, Kiyazov A, et al. Machine Learning-Based Denoising for Single-Photon Detection With MKIDs. IEEE Transactions on Applied Superconductivity. 2026;36(8):2402715-2402715. doi:10.1109/tasc.2026.3718447

BibTeX
@article{Makhrinov2026,
  author = {Makhrinov, Viktor and Arena, Michael and Cuda, Aled and Bagheri, Mehdi and Almagambetov, Akhan and Kiyazov, Aziret-Sultan and Martin, Ryan D. and Grossan, Bruce and Smoot, George F. and Shafiee, Mehdi},
  title = {Machine Learning-Based Denoising for Single-Photon Detection With MKIDs},
  journal = {IEEE Transactions on Applied Superconductivity},
  year = {2026},
  volume = {36},
  number = {8},
  pages = {2402715--2402715},
  doi = {10.1109/tasc.2026.3718447}
}
Kostenlos starten →