The limitations of current imaging technologies in capturing extremely fast phenomena, such as ultrashort light propagation, molecular decay, and nonlinear interactions, are significant. Existing methods, like the pump-probe technique, require event repetition and cannot provide real-time observation. Real-time ultrafast imaging, defined as multi-dimensional observation without event repetition, is a long-standing challenge. While several trillion-fps imaging modalities have emerged, none have achieved speeds beyond 10 Tfps with sufficient sequence depth. Compressed Ultrafast Photography (CUP) showed promise, combining a streak camera with compressed sensing, but was limited by the streak camera's capabilities and sequence depth. This paper introduces CUSP, a novel technique designed to overcome these limitations by incorporating advanced concepts to achieve unprecedented imaging speeds and sequence depth.

The paper reviews existing ultrafast imaging techniques, including sequentially-timed all-optical imaging photography, frequency-dividing imaging, non-collinear optical parametric amplification, frequency-domain streak imaging, and compressed ultrafast photography (CUP). It highlights the limitations of each approach, such as trade-offs in imaging depth, sequential timing limitations, and speed constraints. The authors point out that CUP, while promising, is restricted by the streak camera's electron deflection capabilities and limited sequence depth (300 frames).

CUSP overcomes the limitations of previous techniques by employing spectral dispersion orthogonal to temporal shearing, enabling spectrotemporal compression. Pulse splitting enhances sequence depth. In active mode, a broadband femtosecond pulse is converted into a temporally chirped pulse train illuminating the dynamic scene. The scene is spatially encoded using a digital micromirror device (DMD) and spectrally dispersed, then temporally sheared by a streak camera. The resulting image is then reconstructed using compressed sensing algorithms. In passive mode, CUSP functions as a high-speed 4D spectral imager. The system uses two views: a time-unshared spectrum-undispersed image (u-View) from an external camera, and a time-sheared, spectrum-dispersed image (s-View) from the streak camera. Reconstruction involves solving an underdetermined inverse problem, leveraging the spatial encoding to recover the dynamic scene. The methodology involves characterizing both linear and non-linear optical phenomena using the active and passive modes of CUSP, respectively. Details on the illumination section, streak camera characterization, imaging system characterization, and reconstruction algorithms are provided in supplementary materials.

CUSP achieved 70 Tfps in active mode, imaging a systematically chirped pulse train and ultrafast light propagation in a nonlinear Kerr medium. The temporal resolution achieved was 240 fs, corresponding to a spectral resolution of 4.5 nm. Comparison with a state-of-the-art technique (T-CUP) showed that CUSP significantly outperforms it in both temporal resolution and sequence depth. In passive mode, CUSP enabled single-shot 4D spectral imaging at 0.5 Tfps, performing single-shot spectrally resolved fluorescence lifetime imaging microscopy (SR-FLIM). Experiments demonstrated a novel relationship between fluorophore concentration and lifetime at high concentrations, where lifetime decreases with increased concentration—a finding not previously observed. The authors also note that CUSP is superior to pump-probe and TCSPC methods because it does not require event repetition and can capture phenomena inaccessible to those techniques, such as high-energy radiation, self-luminescent phenomena, and chaotic dynamical states.

The results demonstrate that CUSP significantly surpasses existing ultrafast imaging techniques in both speed and sequence depth. The ability to capture real-time, multi-dimensional data without event repetition opens possibilities for studying a broad range of ultrafast phenomena previously inaccessible. The application of CUSP to study both linear and nonlinear optical processes showcases its versatility. The observation of the concentration-dependent fluorescence lifetime highlights the power of CUSP for uncovering novel insights into complex physical and biological processes. The comparison with conventional techniques like pump-probe and TCSPC emphasizes CUSP's advantages in terms of speed and the ability to image non-repeatable events.

CUSP represents a major advancement in ultrafast imaging, offering unprecedented speeds and sequence depth. Its ability to capture real-time, multi-dimensional data opens new avenues for research across various scientific fields. Future work could focus on further improving spatial resolution, exploring alternative encoding methods, and developing faster streak camera technology. The potential applications of CUSP are vast, ranging from fundamental physics research to advanced biomedical imaging.

The paper notes that compressed-sensing-enabled single-shot imaging often leads to compromised spatial resolution. While CUSP demonstrates a 2–3× degradation in spatial resolution, future advancements in lossless coding and multi-view projection could mitigate this. The reconstruction algorithm's proprietary nature prevents code sharing, which could limit broader adoption and verification. Also the temporal resolution in active mode is time-bandwidth limited.