We consider the problem of imaging a dynamic scene over an extreme range of timescales simultaneously — seconds to picoseconds. We propose a passive ultra-wideband single-photon imaging method that, without emitting any timing signals, achieves a frequency bandwidth spanning DC-to-31 GHz. Our approach uses single-photon avalanche diodes (SPADs) operating in free-running mode to capture asynchronous timestamps of individual photon detections. From these timestamps alone, we develop computational reconstruction algorithms that recover information across the full temporal bandwidth. We demonstrate capabilities including non-line-of-sight video, passive multi-source imaging, and direct visualization of the propagation of light itself.

Conventional imaging systems capture scenes at a single, fixed temporal resolution determined by the camera's frame rate. A standard video camera operating at 30 fps can resolve motion on the order of tens of milliseconds, while ultrafast imaging systems can achieve picosecond-scale temporal resolution but require active illumination sources such as pulsed lasers. The fundamental challenge is: can we build an imaging system that passively captures information spanning from slow scene dynamics to the speed of light propagation?

Single-photon avalanche diodes (SPADs) have emerged as a transformative sensing technology. Each SPAD pixel can detect individual photons with picosecond-level timing precision. While SPADs have been widely used in active time-of-flight (ToF) imaging — where a pulsed laser provides the timing reference — we observe that the raw photon timestamps themselves, even without an active illumination source, encode rich temporal information about the scene. This insight motivates our passive approach.

In this work, we present a passive ultra-wideband imaging framework that operates SPADs in free-running mode without any active illumination. Each pixel independently timestamps photon arrivals with picosecond resolution. We develop asynchronous timestamp-based reconstruction algorithms that analyze photon inter-arrival statistics, temporal correlations, and intensity fluctuations across multiple timescales. This enables us to simultaneously recover conventional video, high-speed transient phenomena, and ultra-fast light transport effects from a single passive capture.

Ultrafast imaging has a rich history spanning streak cameras, pump-probe techniques, and time-correlated single-photon counting (TCSPC). These methods achieve remarkable temporal resolution but invariably require active, synchronized illumination sources. Passive time-of-flight methods exploit ambient light correlations but are limited to specific scene configurations and narrow temporal bandwidths. Computational photography approaches such as flutter shutter and coded exposure recover temporal information beyond the native frame rate, yet remain constrained to millisecond timescales. Our work uniquely bridges these domains by providing a single passive sensor framework spanning from seconds to picoseconds.

Our imaging system operates a SPAD array in free-running mode, where each pixel independently detects photons and records their arrival times with sub-nanosecond precision. Unlike conventional TCSPC that requires synchronization with a pulsed laser, our approach treats the raw photon timestamps as the fundamental data modality. The resulting data stream is a set of asynchronous time-tagged photon events {(x_i, y_i, t_i)}, where each event records the pixel location and absolute arrival time. This representation naturally supports multi-timescale analysis without any predefined temporal binning.

We develop a suite of reconstruction algorithms operating at different temporal scales. At the slowest scale (seconds to milliseconds), we perform photon-counting integration to recover conventional intensity images and video. At intermediate scales (microseconds to nanoseconds), we analyze photon inter-arrival time statistics to detect and characterize periodic or quasi-periodic temporal modulations in the scene — enabling applications such as passive multi-source identification. At the fastest scale (nanoseconds to picoseconds), we employ temporal correlation analysis of photon pairs to reconstruct the scene's impulse response function, revealing light transport phenomena including multi-path propagation and non-line-of-sight light paths.

A critical component of our framework is the temporal correlation function, computed from photon pairs detected at the same or neighboring pixels. For a stationary light source with intensity fluctuations, the second-order correlation function g^(2)(tau) encodes the source's temporal coherence properties. By measuring g^(2)(tau) passively, we can reconstruct the temporal impulse response of the scene without any active illumination, effectively achieving time-of-flight imaging using only ambient light. The temporal resolution is limited only by the SPAD's timing jitter (approximately 32 ps), corresponding to the 31 GHz upper bandwidth limit.

We validate our framework through a series of experiments demonstrating imaging across the full temporal bandwidth. For conventional video reconstruction, we show that photon-counting integration recovers high-quality intensity images from passive SPAD data, comparable to conventional camera outputs. For passive multi-source imaging, we demonstrate the ability to distinguish and separately image multiple light sources operating at different modulation frequencies — a capability unavailable to any conventional camera. For the ultrafast regime, we demonstrate non-line-of-sight (NLOS) video captured entirely passively, as well as direct visualization of light propagation at picosecond timescales.

Our most striking result is passive non-line-of-sight imaging. By analyzing temporal correlations in photons scattered from a diffuse relay wall, we reconstruct video of objects hidden around a corner — without any laser or structured illumination. The reconstruction exploits the fact that photons traveling along different NLOS paths arrive at the sensor with path-length-dependent delays, which are encoded in the correlation function. We achieve temporal resolution sufficient to resolve centimeter-scale depth differences in the hidden scene, demonstrating that passive NLOS imaging is physically feasible.

We have presented passive ultra-wideband single-photon imaging, a framework that achieves imaging across an extreme temporal bandwidth — from DC to 31 GHz — without any active illumination. By operating SPADs in free-running mode and developing asynchronous timestamp-based reconstruction algorithms, we extract information across timescales spanning over ten orders of magnitude. Our experiments demonstrate conventional video, passive multi-source imaging, non-line-of-sight video, and visualization of light propagation — all from a single passive sensor. This work opens new possibilities at the intersection of computational imaging, quantum optics, and computer vision.