Laser-based light detection and ranging (lidar) has been widely used in various areas of science and technology, including autonomous driving, facial recognition, industrial process monitoring, and space debris tracking [1–4]. Among the various approaches, time-of-flight (ToF) ranging sets the benchmark for absolute distance metrology, offering high precision, long-range, and fast acquisition [5–9]. Moreover, single-photon detection has recently been incorporated into the ToF-based lidar system based on the time-correlated single-photon counting (TCSPC) technique [10–12]. Such photon-counting laser ranging is particularly pertinent to the low-light-level ranging and imaging context [13–15], for instance, satellite survey, altimetric measurement, and topographic mapping. Notably, further combination with advanced computational algorithms allows the demonstration of long-distance observation over the terrestrial atmosphere [16] or remote object identification beyond the line of sight [17]. However, these state-of-the-art single-photon ranging performances have so far been limited in the visible or near-infrared regions.

Nowadays, there is a significant impulse to extend the operation wavelength of single-photon lidar into the mid-infrared (MIR) region, pertaining to the unique features of the reduced photon scattering due to relatively long wavelengths [18] and the inherent chemical selectivity based on molecular rovibrational transitions [19]. Specifically, MIR lidar within the transparency windows of the Earth’s atmosphere favors remote sensing especially in adverse weather conditions such as mist, fog, and haze [20]. In virtue of free-space spectroscopy, MIR differential absorption lidar (DIAL) systems could facilitate range-resolved investigation of atmospheric gas concentrations [21], which holds great potential for pollution surveillance, leakage warning, and meteorological research [22–24]. In these envisioned scenarios, sensitive MIR detection is highly demanded to dramatically improve the working distance and/or reduce the transmitting laser power; however, this challenges existing infrared detectors based on narrow-bandgap semiconductors [25], bolometer sensors [26], or emerging low-dimensional materials [27]. Indeed, the attainable detection sensitivity is typically limited to about pW/Hz1/2, which is far away from the single-photon level. Recently, superconducting nanowire detectors have been engineered to access MIR photon-counting capability [28,29], albeit with a consequent trade-off for detection performances and additional system complexity of cryogenic operation. To date, it remains elusive to realize MIR single-photon ranging, which urgently calls for infrared photon-counting systems at high detection sensitivity and high time resolution.

In this context, frequency upconversion detection has been proposed to leverage high-performance silicon detectors by spectrally converting the infrared photons into the visible band, which offers the desirable features of high detection efficiency, low dark noise, fast time response, and room-temperature operation [30–32]. Consequently, the single-photon upconversion detector has served as an effective tool in a wide range of infrared applications, such as long-distance quantum communication [33], quantum-limited optical state characterization [34], high-sensitivity spectroscopic analysis [35–37], high-speed optical coherent tomography [38], and phototoxicity-free biological examination [39]. Furthermore, mode selective upconversion detection has recently been configured to facilitate noise-tolerant lidar [40,41], where only the backscattered signal photons in specific time-frequency modes are efficiently converted due to the intrinsic phase-matching requirement [42].

Generally, there are two main categories for the upconversion single-photon ranging. One relies on a continuous-wave pumping upconverter, which is suitable for collecting backscattered photons from noncooperative targets at an unknown distance. The recorded time correlation between the receiving signal and the trigger enables one to identify the absolute distance information [43,44]. But the ranging resolution in the TCSPC configuration is typically limited in the sub-centimeter scale, which is mainly determined by the timing jitter of the detector (about tens of picoseconds) [45,46]. An alternative scheme resorts to the pulsed pumping upconverter, where the detected photons are precisely time-stamped by the ultrashort optical pulse, thus significantly improving the axial resolution [47–50]. However, the available capture depth is usually confined within tens of centimeters due to a limited travel range of the involved mechanical scanner as the optical delay line [50,51]. Therefore, it necessitates the development of techniques to address the long-sought-after goal to realize a high-resolution MIR single-photon lidar over a wide operation range.

Here, we devise and implement what we believe, to the best of our knowledge, is a novel MIR upconversion ranging system based on asynchronous optical sampling, which features single-photon sensitivity, high timing resolution, and a wide scanning range. Our realization relies on a dual-comb configuration, where the probe and pump lasers are stabilized at slightly different repetition rates. The reflected MIR probe at 3.1 μm is asynchronously gated in a nonlinear crystal by the pair-wise ultrashort pump pulse at 1.03 μm through the sum frequency generation (SFG) process. Consequently, an all-optical temporal scanning is facilitated at a picosecond timing resolution and a kilohertz refresh rate over a long capture range of 46.45 ns, which results in a time-stretch cross-correlation trace at the SFG wavelength of 0.77 μm. The distance information can be monitored in real time from the waveform detected by a low-bandwidth silicon photodiode. Furthermore, MIR single-photon ranging is demonstrated by combining the TCSPC technique and dual-comb metrology, which permits an extremely low detected flux of 8×10−5 photons/pulse. Meanwhile, the achieved picosecond temporal resolution is improved about a hundredfold over the timing jitter of the used single-photon detector itself, which leads to a high positioning precision of 4 μm. As a proof-of-principle demonstration, the MIR photon-counting lidar is used to retrieve rich information about the thickness, reflectivity, and refractive index for a multilayered silicon wafer under a low-photon-flux illumination. Notably, our approach alleviates the limitation from the intrinsic timing jitter of the photon-counting system, which opens a path toward a sub-μm ranging precision at the single-photon level. Hence, the presented paradigm for the MIR laser range finder is promising to realize sensitive and precise infrared sensing at a long standoff distance.

We would like to note that the presented approach is inspired by the dual-comb architecture, where the asynchronous pulse trains are used to facilitate the fast and precise optical scanning in the time-of-flight measurement. In the conventional dual-comb scheme, the signal source and local oscillator are phase-locked to generate heterodyne cross-correlation through coherent mode-beating operation between two optical fields [6]. In contrast, our scheme requires no phase locking for the dual-color sources at disparate wavelengths. The cross-correlation is generated through a nonlinear parametric process and depends only on the optical field intensity. Moreover, the involved nonlinear optical gating also facilitates sensitive upconversion detection of MIR light. The precise gating of the temporal information and the ability to count infrared photons pave the way to realize the high-resolution MIR photon-counting lidar.

The core process of the proposed MIR laser ranging lies in an SFG-based nonlinear upconversion under asynchronous pulse pumping, as illustrated in Fig. 1(a). The involved probe and pump lasers are stabilized at a repetition rate around fr with a slight difference of Δfr. The MIR probe is split and transmitted to a local reference and a distant target. The reflected light is then spatially mixed with the pump within a second-order nonlinear crystal to perform the so-called asynchronous optical sampling [52–54]. The resultant cross-correlation trace at the SFG wavelength is finally recorded by a silicon photodiode, which allows for inferring the distance information through the time of flight.

Specifically, the underlying mechanism for the asynchronous upconversion sampling is further detailed in Fig. 1(b). Without loss of generality, the pulse interval for the MIR probe is here assumed to be shorter than that for the pump. As a result, there is a temporal shift for each pair of pulses, given by ΔTr=1fr−1fr−Δfr≈Δfrfr2.(1)

In the static reference frame of the probe, the pump pulse sweeping at a step of ΔTr facilitates an optical temporal scanning, which is an essential feature for dual-comb metrology [53,54]. The real time for an effective step of ΔTr is given by the pulse interval T=1/fr. Hence, the refreshing period for completing a full scan is derived as Tupdate=TΔTr×T≈1Δfr.(2)

It can be seen that Δfr corresponds to the updating rate. Similarly, the scaling factor between the laboratory time and effective time can be determined as M=TΔTr≈frΔfr.(3)

Through the nonlinear SFG interaction, a cross-correlation intensity profile is generated with a time stretching factor of M. Consequently, the time separation Δt between the target and reference peaks will be magnified to be M×Δt. It is the substantial time magnification that allows us to use a low-bandwidth detector to record the fast time response, which particularly benefits the high-precision MIR measurement where fast and sensitive detectors are currently challenging in this spectral region. Moreover, the stretched waveform also favors mitigating the limitation of the timing jitter for the single-photon detector in the photon-counting laser ranging scenario [55].

Figure 2 presents the experimental setup for the MIR single-photon lidar, which consists of laser source preparation, asynchronous upconversion sampling, and time-correlated counting. The involved light sources originate from an Er-doped fiber laser (EDFL, LangyanTech, ErPico RLocking) at 1.56 μm and an Yb-doped fiber laser (YDFL2, LangyanTech, YbPico RLocking) at 1.03 μm. The repetition rate of the two mode-locked fiber lasers can be tuned within the range of 21.53±0.01MHz based on an intracavity piezo-actuator mounted on motorized stage. With the help of a servo system, the repetition rates are stabilized to a reference about fr=21.53MHz with a difference of Δfr=1kHz. The output of EDFL is injected into a slave laser (YDFL1, LangyanTech, YbPico Elite) to realize an optical synchronization based on the cross-phase modulation effect, which offers a passive fashion to stabilize the relative repetition rate for the dual-color lasers [36]. The synchronized pulses are then focused into a periodically poled lithium niobate crystal (PPLN1, with a length of 40 mm and a poling period of 30.3 μm) to generate the MIR probe at 3.1 μm through difference frequency generation (DFG). The MIR probe is steered into a reference and a target, and the reflection is spatially mixed with the amplified pump. The combined beams focus into another PPLN crystal (PPLN2, with a length of 10 mm and a poling period of 20.9 μm) to perform the SFG-based upconversion, where the ultrashort pump serves as a rapid optical gate in the temporal domain. The resulting cross-correlation trace is detected by a silicon photodiode, which facilitates a high-speed ranging measurement. In the low-light situation, the upconversion light is recorded by a single-photon counting module (SPCM, Excelitas, SPCM-AQRH-54-FC), which allows the reconstruction of the photon-counting waveform based on a high-precision TCSPC device (Qutools, quTAG). The required timing trigger for the correlation measurement is provided by the SFG pulse that is generated by the two asynchronous lasers of EDFL and YDFL2. In our experiment, the peak conversion efficiency is estimated to be about 1.5×10−3 at a pump power of 130 mW. The conversion efficiency can be further improved by spectro-temporal engineering of the involved pulses and increasing the pump power via large-mode-area fiber amplifiers [56]. The background noise is measured to be about 5 kHz in the case of blocking the incident infrared signal at the entrance of the upconversion stage, which corresponds to a noise probability about 2.4×10−4 per pump pulse. The low-noise conversion process is essential to facilitate subsequent ultrasensitive MIR ranging.

First, the MIR ranging system is characterized to measure the relative displacement of the target. The target is mounted on a linear translation stage (Thorlabs, NRT150/M) with an on-axis accuracy of 2 μm. The cross-correlation trace is measured by a silicon avalanche detector (Thorlabs, APD410A/M) with a bandwidth of 5 MHz. Figure 3(a) gives the recorded trace at a fixed position, where two adjacent peaks are clearly identified for the reference and target. The scanning step ΔTr is calculated to be 2.16 ps, which suffices to sample the cross-correlation trace. The refreshing time Tupdate is measured to be 1 ms, which is dictated by the relative repetition rate. The optical sweeping technique allows for a large effective capture range of 46.45 ns, which is inaccessible for a mechanical scanner [51]. The full cover of the pulse period of the probe is essential to collect signals from any distance, thus eliminating the dead zone for the ranging. The cross-correlation peak for the target is zoomed in Fig. 3(b), which indicates a pulse duration of 0.211 μs in laboratory time. The effective duration is inferred to be 9.8 ps by using the conversion factor M=2.153×104, which agrees with the width of the convolution between the probe and pump pulses. The time-stretch operation greatly relaxes the bandwidth requirement for fast signal detection and processing.

For distance measurements, the SFG intensity profile for each peak is fitted with a Gaussian function, which allows for precise identification of the peak positions. The measured time separation ΔT between the target and reference peaks can be converted to the distance according to ΔL=cΔT/(2Mng), where c is the speed of light in vacuum and ng is the group refractive index of air. Here, the distance calculation is relatively simpler than that in conventional dual-comb metrology. Indeed, the distance measurement based on the temporal interferograms typically requires either Fourier transform for phase demodulation or Hilbert transform for carrier frequency elimination [2,55]. In contrast, the presented nonlinear optical sampling only depends on the optical field intensity. Such an incoherent operation not only enhances the measurement robustness with a resilience to phase variations, but also simplifies the laser preparation without the need to stabilize the carrier-envelope phase [53,54].

Figure 3(c) presents the measured distance as a function of the target displacement. The linear dependence indicates a high measurement accuracy, which is manifested in the corresponding residuals shown in Fig. 3(d). The ranging uncertainty is about 100 μm for a single measurement, which can be decreased to 5 μm for a longer averaging time, as seen from the Allan deviation in Fig. 3(e). In our proof-of-principle demonstration, the target is placed at an original distance L0=46.45cm. Actually, the nonambiguous range for the current experimental setting is about 6.97 m defined by c/(2frng). Similar to dual-comb ranging, the operation distance can be significantly extended by tuning the repetition rate for one laser source [5,54], or simply switching the repetition rates of the two combs in a separate measurement [2,6]. In the latter case, the nonambiguous range can reach 150 km, which is set by the laboratory dual-comb period of 1/Δfr=1ms. The underlying challenge for the long-distance ranging lies in collecting sufficient photons from a remote target, which merits the use of sensitive range finders, as demonstrated in this work. Furthermore, the combination of advanced computational techniques would enhance the signal-to-noise ratio (SNR) in photon-starved scenarios [16].

Now we turn to demonstrate the high-speed MIR ranging performance. An aluminum retroreflector as the target is mounted at a voice coil flexure scanner (Thorlabs, VCFL35). The scanner is driven by an arbitrary waveform generator (Rigol, DG4162). Figure 4(a) presents the reconstructed trajectory for the rapidly oscillating scanner, which is well fitted by a sinusoidal function. The 1 kHz sampling rate is sufficient to observe the dynamic position change. After a fast Fourier transform (FFT), the peak in the spectral domain clearly indicates the driving frequency of 15 Hz, as shown in Fig. 4(b). Furthermore, a more complicated trace in Fig. 4(c) can be generated by driving the scanner with a synthesized waveform. The corresponding FFT spectrum correctly reveals the frequencies and amplitudes for the three sinusoidal components set in the experiment. In contrast to mechanical scanners, the optical temporal sweeping is inertia-free, which thus favors rapid ToF distance measurements for a moving target. The refresh rate can further be boosted by increasing the repetition rate difference, while a higher repetition rate itself should be adapted to retain a suitable sampling step [2,55].

In the following, we proceed to investigate the high-sensitivity competence of the MIR photon-counting ranging. The implemented parametric upconversion detection leverages a silicon-based photon counter with high detection sensitivity and fast time response, which hence provides superior performances beyond that for currently available MIR detectors at room temperature [15]. To emulate the photon-starved scenario, the MIR probe is attenuated by a series of calibrated neutral density filters such that the detected photon number per pulse is much less than one [11,40]. Accordingly, the detector is replaced with a silicon-based SPCM. The detected events are connected to a stop channel of a TCSPC device, while the start channel is triggered by the generated pulses from the SFG between the YDFL2 and EDFL, as depicted in Fig. 2. The recorded time correlation between the receiving signal and the trigger enables one to identify the absolute distance information. The timing jitter in the root mean square of the TCSPC is specified to be below 6.4 ps. The bin width and the interrogation window are configured to be 100 ns and 1 ms, respectively. The MIR illumination energy is attenuated to be about 0.8 fJ/pulse. The detected photon number at the SPCM is estimated to be 8×10−5 per pulse after taking into account the duty cycle of 2.1×10−4 for the asynchronous sampling and the total detection efficiency of 3.2×10−5 for the upconversion detector. The total detection efficiency is limited by several factors in the experiment. The main loss is ascribed to the collection efficiency of 10% for the retro-reflected radiation from the target and the internal conversion efficiency of 0.15% within the nonlinear crystal. Additionally, the spectral filtering stage has a power transmission of 65% for the upconverted signal. The SFG signal is coupled into a single-mode fiber with an efficiency of 50%, before being recorded by a silicon photon counter with a detection efficiency of 65%. The total detection efficiency could be further improved with a synergic optimization on each stage from light collection, nonlinear conversion, and spectral filtering to photon recording [56]. The total acquisition time is set to be about 10 s for collecting sufficient photons to build a high-contrast histogram. The recorded coincidence histogram is presented in Fig. 5(a), which again shows two prominent peaks for the reference and target. A zoom-in illustration of the target peak is given in the inset. The histogram width is identical to the one obtained for analog optical detection. Notably, the effective time resolution of 9.9 ps is achieved, which is much smaller than the 830 ps timing jitter of the SPCM, as measured from the histogram shown in Fig. 5(b).

The ranging precision at the single-photon level is illustrated in the Allan deviation trace shown in Fig. 5(c). At an averaging time of 64 s, the precision reaches 4 μm, which is significantly smaller than previous demonstrations [45,46]. The presented time-stretch correlated counting technique overcomes the intrinsic timing uncertainty of single-photon detectors [57], which hence allows a precision to be obtained beyond the reach of conventional ToF single-photon ranging systems [12,16]. Furthermore, the SNR for the recorded histogram is investigated as a function of the detected power and the integration time, as shown in Figs. 5(d) and 5(e), respectively. The SNR is defined as the ratio between the peak amplitude and the standard deviation of the noise floor. The SNR is particularly important for single-photon ranging since the collected signal is extremely limited. A high SNR usually leads to a better precision. As expected, a stronger illumination power or a longer accumulation time enables more photons to be collected, which thus improves the SNR. Notably, tremendous effort has been devoted to developing advanced algorithms to realize photon-efficient ranging or imaging, to achieve a high SNR with a small number of photons [14,15].

In comparison to visible or near-infrared wavelengths, the MIR light allows penetration through semiconductor materials. For instance, the silicon and germanium substrates are opaque for the optical spectra below 1.1 and 2.0 μm, respectively. As a proof-of-principle demonstration, we finally use the MIR single-photon ranging system to perform a noninvasive investigation on stacked silicon wafers as a multiple-layer object. The backscattered photons from the interior interfaces provide rich information on the reflectivity and depth of the imbedded surfaces, as well as the refractive index of the medium [49,50]. A reference histogram in Fig. 6(a) is measured in the absence of the sample, where the peak position is defined as the origin. Because the refractive index of the silicon n2 is relatively higher than that for the air n1, the histogram peak in Fig. 6(b) is delayed after inserting a silicon wafer with a thickness of 3 mm. Specifically, two prominent peaks are observed, which correspond to the optical path configurations presented in the inset. From the measured relative distances, the geometric thickness and refractive index are inferred to be 2.94 mm and 3.49, respectively. Moreover, the power reflectivity at each surface is measured to be 0.304, which is consistent with the normal incidence Fresnel reflection given by (n2−n1)2/(n2+n1)2. In the presence of a 5 mm thick wafer, the recorded peaks have a larger delay, as illustrated in Fig. 6(c). As shown in Fig. 6(d), more peaks emerge with both inserted wafers, thus showing a depth-resolvable capability for multilayer structures. Pertinent to the transparency window for semiconductors and polymer materials [38,48], the MIR sensitive and precise depth ranging would open an effective way to perform nondestructive defect inspection or stand-off surface profiling in extreme scenarios for industrial quality control and process monitoring.

Although single-photon laser ranging is widely demonstrated in a variety of applications, the operation wavelength has long been restricted in the visible or near-infrared region due to the availability of sensitive and fast optical detectors [14,15]. Our work addresses the long-standing quest to extend the operation wavelength of photon-counting ranging into the MIR regime. The presented approach of the asynchronous upconversion sampling technique not only facilitates the optical temporal scanning at high speed and high precision, but also enables the sensitive upconversion detection for the gated infrared signals with a silicon-based visible photodiode operating at room temperature. The precise gating of the temporal information and the ability to count infrared photons constitute a key to the first realization of high-resolution MIR photon-counting lidar under low-light illumination.

In comparison to MIR upconversion lidar systems based on continuous-wave pumping [43–46], the involved pulsed pumping configuration favors the improvement of nonlinear conversion efficiency due to the high peak intensity and the reduction of background noise within a narrow time window. Meanwhile, the asynchronous pumping operation allows a long sweeping range while maintaining the ultrashort temporal resolution, which overcomes the limited scanning range of the mechanical delay line in the coincidence-pumping scheme [47,51,56]. Another notable feature for the presented approach is the time-stretch cross-correlation trace, which enables us to achieve a photon-counting temporal resolution much smaller than the timing jitter of the used single-photon detectors [44–46]. This unique feature significantly alleviates the bandwidth requirement for fast detection and processing typically required in traditional TCSPC architecture. We note that one remaining challenge for the presented photon-counting MIR ranging system is to shorten the acquisition time for observing dynamic scenes at the low-light level. The adoption of computational techniques may facilitate a photon-efficient lidar, where the target can be identified from a histogram with much fewer photons [14,16].

In conclusion, our work establishes an effective path to achieve high-performance MIR lidar with single-photon sensitivity, picosecond timing resolution, and a large nonambiguous range. The ranging resolution can be further enhanced to the nanometer level by using shorter optical pulses [2,6,9], without suffering from the timing-jitter restriction of the photon detection and counting devices. Moreover, enhanced performances with a higher refresh rate and a finer sweeping step can be obtained by properly increasing the repetition rates and their difference for dual-color lasers [55]. Notably, the involved asynchronous laser sources could be substantially simplified by resorting to recent advances of single-cavity dual combs [58]. In addition, it is feasible to extend the presented approach into longer-infrared or terahertz regions [35], where high-sensitivity and high-resolution distance measurements are highly demanded. We believe that the implemented MIR single-photon ranging system would promote a variety of low-light applications in remote sensing, environmental monitoring, meteorological observation, and defense surveillance.

Acknowledgment. This work was supported by the National Key Research and Development Program, National Natural Science Foundation of China; Shanghai Pilot Program for Basic Research; Natural Science Foundation of Chongqing; Shanghai Municipal Science and Technology Major Project; Fundamental Research Funds for the Central Universities.