Accelerating the measurement of time-resolved emission line shapes with a denoising neural network

Single coherent photons on demand are essential inputs for many applications of quantum optics, including for linear optical quantum computing. One critical parameter when assessing the potential of a material to emit coherent single photons is the homogeneous linewidth, the measurement of which is often obscured by fast spectral fluctuations. Photon correlation Fourier spectroscopy (PCFS) provides a more accurate measurement of the homogeneous emission linewidth than traditional techniques by probing the system on timescales faster than spectral diffusion. However, PCFS is limited by the long integration times it requires, restricting its use to only the most stable materials and making it difficult to study large numbers of emitters within these select samples. Our group previously developed a machine-learning (ML) algorithm to denoise PCFS signals which showed promise on a single experimental dataset but was difficult to further validate. Here, we demonstrate the use of a similar ML model to accelerate PCFS experiments by an order of magnitude and verify its accuracy on a statistical number of samples. We use 10% of the photon stream from full PCFS experiments to replicate shorter measurements, resulting in data with prohibitively high levels of noise. By employing a denoising autoencoder neural network, we extract the underlying signal from the noisy truncated measurements and unambiguously verify the results of the reconstructions through direct comparison with the original experiments. We confirm the generalization of the ML treatment through application to >80 experiments from multiple samples including $\mathrm{CsPbB}{\mathrm{r}}_{3}$ and InP/ZnSe/ZnS nanocrystals. The large-scale verification of the ML reconstructions enables confident use of ML to accelerate future PCFS experiments up to 10\ifmmode\times\else\texttimes\fi{}, enhancing the practical capabilities of this technique and further promoting the application of ML to other spectroscopic methods.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.