Compressive ultrafast sensing enabled by programmable temporal fan-out

Ultrafast sensing lies at the heart of diverse applications including spectroscopy, nonlinear optics, and microscopy. Here, by developing a programmable temporal fan-out (TFO) gate, we propose and demonstrate a compressive temporal single-pixel imaging (CTSPI) scheme that can efficiently measure ultrafast pulses by surpassing the Nyquist rate. Benefiting from the spatio-temporal duality of light pulses, our TFO can convert the input ultrafast pulse into a programmable pulse train by using a digital micromirror device (DMD), resulting in an arbitrary control of the temporal structure. Compared to previous temporal ghost imaging (TGI) schemes, an advancement in temporal resolution by 3 orders of magnitude, down to 16.00$\pm$0.01 fs, enables the successful recovery of a 5 fJ terahertz pulse with over 97% fidelity at 30% compression ratio (CR). The robustness of CTSPI against temporal distortions on TFO is also illustrated. We apply the high sensitivity of our CTSPI to terahertz spectroscopy, aided by machine learning in order to accurately recognize samples at low signal-to-noise ratios (SNRs). This technique brings new possibilities for compressive spatio-temporal characterization, remote sensing, and high-data-rate optical communications.