Quantum Light Detection and Ranging

Light detection and ranging (LiDAR) systems with the ability to reach long distance at high speed and accuracy have emerged as a key technology in autonomous driving, robotics, and remote sensing Schwarz (2010). Today miniaturised LiDARs are integrated in many consumer electronics devices, e.g. smartphones. Moving beyond depth sensing, the LiDAR technique has been also used for non-line-of-sight imaging Velten et al. (2012); Gariepy et al. (2016); O’Toole et al. (2018); Liu et al. (2019a); Faccio et al. (2020), imaging through scattering media Lyons et al. (2019) and biophotonics applications Bruschini et al. (2019). A typical LiDAR system records the time-of-flight, $t$, of light back-reflected from a scene, which enables to estimate distance $d=ct/2$, where $c$ is the speed of light Sun et al. (2016). Thanks to their single-photon sensitivity, picosecond temporal resolution and low cost, single-photon avalanche diodes (SPADs) are widely used as detectors in LiDAR Shin et al. (2016); Tachella et al. (2019). In this respect, two well established techniques can be used to achieve the timing information at picosecond resolution: time-correlated single-photon counting (TCSPC) that operates by recording a time-stamp for each individual photon Zhang et al. (2018); Ronchini Ximenes et al. (2019); Seo et al. (2021) or time gating in which a gate window is finely shifted Ulku et al. (2018); Morimoto et al. (2020); Ren et al. (2018); Chan et al. (2019).
Despite recent advances, interference is a major challenge for robust and secure LiDAR applications through complex environments. In our work, the term ‘interference’ refers to the detection by the LiDAR sensor of any optical signals other than those emitted by the LiDAR source. These may originate from ambient light, other LiDAR systems operating concurrently and intentional spoofing signals. In addition to depth distortion such as degradation in accuracy and precision, interference could result in misleading information, causing the system to make incorrect decisions. Over the past several years, some approaches addressing LiDAR interference have been proposed. One technique isolates the signal based on temporal correlations between two or more photons Niclass et al. (2013), which effectively suppresses noise due to ambient light. Another technique based on laser phase modulation can reduce both ambient light and mutual interference Ronchini Ximenes et al. (2019); Seo et al. (2021). However, these approaches have limitations. For example, an external signal can still spoof the LiDAR detector if it copies the temporal correlation or phase modulation of the illumination source, which is easily achievable by placing a photodiode close to the target object. To date, there is no LiDAR system immune to all types of interference.
The use of non-classical optical states can also improve object detection in the presence of spurious light and noise. In a quantum illumination protocol, a single photon is sent out towards a target object while its entangled pair is retained and used as an ancilla Lloyd (2008). Coincidence detection between the returned photon and its twin increases the effective signal-to-noise ratio compared to classical illumination, an advantage persisting even in the presence of noise and losses. Recently, practical quantum illumination schemes have been experimentally demonstrated using spatially entangled photon pairs for target detection Lopaeva et al. (2013); Zhang et al. (2020) and imaging objects England et al. (2019); Defienne et al. (2019); Gregory et al. (2020) in the presence of background light and spurious images. These approaches rely on the ability to measure photon coincidences between many spatial positions, which is conventionally performed using single-photon sensitive cameras such as electron multiplied charge coupled device (EMCCD) cameras Moreau et al. (2012); Edgar et al. (2012); Defienne et al. (2018), intensified(i)CCD Chrapkiewicz et al. (2014, 2016) and SPAD cameras Eckmann et al. (2020); Ianzano et al. (2020); Ndagano et al. (2020); Defienne et al. (2021). However, while a handful of works have reported the use of photon pairs for target detection at distance Liu et al. (2019b); Frick et al. (2020); Ren et al. (2020), no imaging LiDAR experiments with absolute range distance and interference or background rejection have been reported.

In this work, we demonstrate a quantum LiDAR system immune to any type of classical interference by using a pulsed light source of spatially entangled photon pairs and a time-resolved SPAD camera. We use spatial anti-correlations between photon pairs as a unique identifier to distinguish them from any other light sources in the target scene. In particular, we show our LiDAR system successfully images objects and retrieves their depths in two different interference scenarios mimicking the presence of spoofing or additional classical LiDAR signals. In the first case, spurious light from a synchronised laser is used to demonstrate the robustness against intentional spoofing attacks. In the second case, the interference takes the form of asynchronous pulses imitating the presence of multiple background LiDAR systems running in parallel. The results show that our approach enables to image with high depth resolution while offering immunity to classical light interference.
Imaging system. The experimental setup is shown in Fig. 1a. Spatially entangled photon pairs are produced by type-I spontaneous parametric down conversion (SPDC) with a $\beta$-Barium Borate (BBO) nonlinear crystal pumped by a $355$ nm pulsed laser with a repetition frequency of $20$ MHz. The objects to be imaged are masks placed on a reflective mirror. One object $O_{1}$ (“skater”) is placed in the far field of the crystal thus the down-converted photon pairs are spatially anti-correlated at the object plane. Another object $O_{2}$ (“car”) is illuminated by a diffused $780$ nm laser pulsed at $20$ MHz as well to produce the interference. Both objects are imaged onto the SPAD camera SwissSPAD2 Ulku et al. (2018) (see Methods). Similar to a typical LiDAR scheme, the pump laser is synchronised with the camera, while the spoofing signal generated by a classical pulse laser can be synchronous or asynchronous.

As in conventional time-gated LiDAR, backscattered photons with specific time-of-flight are detected by scanning a gate window ($15$ ns wide) using $18$ ps time steps, which corresponds to a depth resolution of $2.7$ mm Ulku et al. (2018); Morimoto et al. (2020). At each gate position, a set of 8-bit frames ($\sim 10^{3}$ frames) is acquired to reconstruct two different types of images: (i) a classical intensity image, obtained by summing all frames, and (ii) a spatially-averaged photon correlation image computed by identifying photon coincidences in the frame set using a technique detailed in Defienne et al. (2018) (see Methods). The intensity image retrieves the shape of the objects in the scene at a given depth, while the spatially-averaged correlation image measures spatial correlations between detected photons to identify the presence of photon pairs. For example, if only reflected photon pairs are captured within the gate window (Fig. 1b), the intensity image shows the “skater” object and an intense peak is observed at the center of the spatially-averaged correlation image. The presence of such a correlation peak above the noise level confirms the presence of photon pairs among the detected photons. If only classical light is detected (Fig. 1c), the intensity images show the “car” object illuminated by the pulse laser and the spatially-averaged correlation image is flat.

Synchronous classical light interference. First, we consider the case of a spurious classical source of light that is synchronised with the SPAD camera i.e. photons reflected by both $O_{1}$ (“person”) and $O_{2}$ (“bike”) are synchronous with the camera (Fig. 2a). This scenario corresponds to a spoofing attack. To operate the LiDAR, the gate window is continuously shifted over a range of $27$ ns, which corresponds to 1500 gate positions. Figure 2b shows the intensity and spatially-averaged correlation images (zoom $9\times 9$ pixels in inset) measured at four specific gate positions $0.09$ ns, $7.2$ ns, $14.58$ ns and $23.4$ ns. At the early gate position ($0.09$ ns), there is only noise recorded by the camera such as dark count, crosstalk, afterpulsing and ambient light. As the gate window is shifted, $O_{2}$ appears in the intensity image ($7.2$ ns) and the absence of a peak in the spatially-averaged correlation image shows that it originates from classical light alone. When the reflected quantum light starts to be collected in the gate window together with the classical light ($14.58$ ns), $O_{1}$ and $O_{2}$ are superposed in the intensity image and a correlation peak becomes visible. For the late gate window ($23.4$ ns) the classical laser pulse vanishes while only quantum light is detected, as shown by the peak persisting in the spatially-averaged correlation image, and only $O_{1}$ is visible in the intensity image.
To acquire depth information and distinguish classical interference, the spatially-averaged intensity and correlation peak values represented in function of the gate position in Fig. 2c are analyzed. The two-step average intensity profile represents the double reflections from $O_{1}$ and $O_{2}$, while the correlation peak profile only reveals the trend of quantum light over the given time range. By locating the falling edges of the intensity profile, the arrival time information of all the objects can be obtained Morimoto et al. (2021); Chan et al. (2019). Whilst by just searching for the last falling edge of the correlation peak profile, the arrival time information of the quantum object is extracted. As shown in Fig. 2d the arrival time of the classical object of $16.110$ ns and its intensity image are obtained. The arrival time of the quantum object, $24.462$ ns, is located at the last fitted falling edge of the correlation peak profile and the corresponding intensity image is subtracted in Fig. 2e. Refer to the Supplementary Video for the scanned results over the entire detected range. The proposed dual-profile locating method enables locating and distinguishing objects illuminated by quantum light or classical light.
The anti-spoofing capability works as described, as long as the time delay between the two objects is larger than the temporal resolution of the SPAD camera. One may then enhance the removal of temporally overlapping interferences by increasing the number of frames (e.g. up to $\sim 10^{6}$ 8-bit frames) for each gate delay so as to retrieve a spatially-resolved correlation image instead of a spatially-averaged correlation image Defienne et al. (2018, 2019); Gregory et al. (2020); Defienne et al. (2021). An example is shown in Fig. 3: a spatially-resolved correlation image retrieves directly the shape of the object illuminated by photons pairs and remains insensitive to classical interference (classical background noise added in the experiment). In fact, such a spatially-resolved correlation image could potentially be measured at all gate positions of the LiDAR scanning. However, the acquisition time is much longer than that required to retrieve an spatially-averaged correlation image (a few hours instead of seconds for a single time gate delay) and it is therefore better to limit its use to gate positions for which the objects cannot be distinguished otherwise. In addition, note that because of the anti-symmetric spatial structure of photon pairs illuminating the object, each spatially-resolved correlation image contains both the object and its symmetric image, which means that the object must interact with only half of the photon pair beam to be imaged through correlations without ambiguity.
Asynchronous classical light interference. In real-world applications, another possible scenario is the interference coming from ambient light and other LiDAR systems. We therefore consider a classical source of light that is not synchronised with the SPAD camera but still running at the same repetition frequency ($20$ MHz) and illuminates the object $O_{2}$ “50 traffic sign” (Fig. 4a). Fig. 4b shows the intensity and correlation images at three example gate positions $2.16$ ns, $13.5$ ns and $24.66$ ns. The object $O_{2}$ is visible in the intensity images at all gate positions as background noise. When the gate is shifted to $13.5$ ns, the SPAD also captures photon pairs reflected by $O_{1}$ (“STOP traffic sign”) and both objects are superimposed in the intensity image. We now also observe a peak in the spatially-averaged correlation image which highlights the presence of photon pairs. By locating the falling edge of the spatially-averaged correlation peak shown in Fig. 4c, the time arrival information of the quantum object is located at $20.682$ ns, and its intensity image is also obtained by subtraction as shown in Fig. 4d. See the Supplementary Video for the entire measured results.
Discussion. We demonstrated a LiDAR system based on spatially entangled photon pairs showing robustness against interference from classical sources of light. In particular, we showed its successful use in the presence of (i) a spoofing attack (synchronous classical light interference) and (ii) of a background light and another LiDAR system operating in parallel (asynchronous classical light interference). Note also that because the quantum LiDAR harnesses anti-correlations between photon pairs to retrieve images, it is also immune to classically-correlated sources of light such as thermal and pseudo-thermal sources in which photons are position-correlated Valencia et al. (2005).
In our current implementation, time gate position is acquired in several seconds so that the full scanning takes several hours (5.6 hours for the synchronous case and 3.4 hours for the asynchronous case). This total acquisition time can however be significantly decreased by reducing (i) the acquisition time per gate position and (ii) the number of gate positions to detect the falling edge of quantum light (currently $1500$ steps scanned linearly). For example, in the case of the synchronous classical light shown in Fig. 2, the quantum illuminated object could be located by measuring only $300$ frames for $8$ different gate positions by using a correlation-driven scanning and falling edge fitting algorithm, which would reduce the total acquisition time to $7$ seconds (see details in the Supplementary Information). In addition, the speed of the SPAD camera in our experiment was limited to $370$ fps by the readout architecture, but it has been demonstrated that the similar cameras can be operated at frame rates up to $800000$ fps Gasparini et al. (2018), which would further diminish the total acquisition time and potentially reach real-time acquisition. Furthermore, in the current quantum LiDAR prototype the target object is a two dimensional ‘co-operative’ object attached to a mirror, which ensures enough photon pairs are reflected and collected by the camera. However, the proposed scheme can be extended to scattering materials with three dimensional profiles by using brighter photon pair sources and more sensitive SPAD cameras, which are currently under development. Looking forward, these results could enable the development of robust and secure LiDAR systems and more general time-resolved quantum imaging applications.

 
Acknowledgements. DF is supported by the Royal Academy of Engineering under the Chairs in Emerging Technologies scheme and acknowledges financial support from the UK Engineering and Physical Sciences Research Council (grants EP/T00097X/1 and EP/R030081/1). This project has received funding from the European union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 754354. H.D. acknowledges support from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 840958.

Authors contributions. D.F. and E.C. conceived the research. H.D., A.L. and J.Z. designed the experimental setup. J.Z. and A.L. performed the experiment. J.Z. and A.L. analysed the data. A.U. E.C. developed  SwissSPAD2 and H.D. developed the coincidence counting algorithm. E.C. supervised the project. All authors contributed to the manuscript.

Data availability. The experimental data and codes that support the findings presented here are available from the corresponding authors upon reasonable request.