1. INTRODUCTION

A single-photon avalanche diode (SPAD) allows high temporal resolution detection of very weak optical pulses, which in turn makes high range resolution LIDAR applications possible. The SPADs in the visible and near infrared (NIR), based on silicon (usable below 1100 nm), have low dark counts, high sensitivity, good temporal resolution, and can be operated near room temperature. On the other hand, fast SPADs and photon-counting devices for longer wavelengths are normally limited to wavelengths below approximately 2.5 µm [1] and require cryogenic cooling [2–4] due to the low photon energy. Some detectors still work at longer wavelengths, albeit with a significantly reduced detection efficiency, and still require cooling [5–7]. This is unfortunate since the mid-infrared (MIR) region is attractive for LIDAR for several reasons, such as gas concentration measurements in the molecular fingerprint region [8] and the higher transmission through scattering media, such as fog and smoke [9].

A solution to operate at these wavelengths is to convert the MIR radiation to the visible/NIR region through a nonlinear optical interaction. This allows the benefits from both the established silicon SPADs and MIR LIDAR features to be reaped. The most convenient nonlinear conversion for this application is sum frequency generation (SFG), since it allows for lower powers of MIR than second-harmonic generation (SHG). The SFG up-conversion can be done in either a single-pass configuration [10] or inside a laser cavity [11–13]. The latter allows for a higher intensity of the resonating laser, which improves the conversion efficiency significantly. It is of particular interest to exploit the quasi-phase-matching (QPM) scheme for the nonlinear conversion as it thereby is possible to obtain collinear interaction at the desired wavelength and bandwidth. A typical, 10 mm long, QPM crystal has an acceptance bandwidth [14] in the MIR of only a few nanometers (nm), which inherently filters the MIR background to only the laser spectrum and thus provides excellent optical signal to background ratio.

Single-photon-counting LIDARs around 1.5 µm have been realized previously, both through direct detection [15–17] and through the up-conversion technique [18,19]. Very recently superconducting nanowires were used in a proof-of-concept single-photon-counting LIDAR demonstration at 2.3 µm [20]. The work presented here was performed in order to extend the up-conversion LIDAR technology to longer wavelengths and to improve and determine the range resolution. The experimental setup consisted of a ${\rm Nd}:{{\rm YVO}_4}$ laser cavity with a 10 mm long periodically poled rubidium-doped ${{\rm KTiOPO}_4}$ (PPRKTP) inside. This allowed for intra-cavity SFG and a conventional Si-SPAD to be used. Furthermore, the conventional time-of-flight technique could be used to determine the distances to and between objects. The current system observed 30% broader temporal trace from targets separated by an optical distance of 1.9 mm than from a single target. It is therefore possible to determine whether the reflection was from a single target or from multiple targets. The measured temporal trace from a single target had a full width at half-maximum (FWHM) of 42 ps and was limited primarily by the response of the Si-SPAD.

2. EXPERIMENT SETUP

There are three essential parts for the time-of-flight up-conversion LIDAR: the MIR laser pulses, the conversion stage, and the detector unit with control electronics.

Sub-100 fs MIR pulses, with a spectral bandwidth of about 300 nm at 2.4 µm were provided by a Ti:sapphire laser pumped optical parametric amplifier at a repetition frequency of 1 kHz. In order not to saturate the detector, the pulse energy was attenuated down to the picojoule (pJ) regime.

These pulses were up-converted to 737 nm through SFG in an in-house fabricated PPRKTP crystal with a period of 23.5 µm and a grating length of 9 mm (physical length of 10 mm). The nonlinear crystal was positioned in the middle of a 1064 nm ${\rm Nd}:{{\rm YVO}_4}$ laser cavity as can be seen in Fig. 1. The cavity consisted of two plane mirrors (M1 and M2), one 200 mm radius of curvature mirror (M3), the uncoated PPRKTP crystal, and a ${\rm Nd}:{{\rm YVO}_4}$ crystal with one side high-reflection (HR) and the other side anti-reflection (AR) coated for 1064 nm. The total length of the cavity was 110 mm, which resulted in a beam radius of 200 µm inside the PPRKTP crystal. The incident pump power on the laser crystal was 2.5 W, emitted by an 808 nm laser diode. The resulting 1064 nm intra-cavity power was 5 W. This value is considerably lower than what could be obtained, as the PPRKTP crystal was uncoated and contributed to high loss in the cavity.

 

Fig. 1. Schematic sketch of the cavity used for the nonlinear up-conversion. The red, orange, blue, and yellow lines represent the generated visible, the MIR, the resonating 1064 nm light, and the 808 nm pump light, respectively.

Download Full Size | PDF

The center detection wavelength, Eq. (1), is set by the grating period, while the acceptance bandwidth [14], Eq. (2), of the nonlinear process is independently set by the grating length of the crystal, both of which can be freely chosen in the design. The equations governing the entre detection wavelength and bandwidth are the following:

By appropriate choice of period for the nonlinear crystal and laser wavelength for the cavity, the detection wavelength can be chosen as shown in Fig. 2. For example, by utilizing another laser medium, such as Yb:KYW, one can generate efficient tunable lasing from 997 to 1050 nm [21]. This would allow photons further into the MIR region to be detected with the same nonlinear crystal or with a crystal with a similar QPM period. Fine-tuning of the detection wavelength can be achieved by simply rotating the crystal or changing the temperature.

 

Fig. 2. Dependence of the required pump wavelength (wavelength in the cavity) on the MIR wavelength for PPRKTP crystals with two different periodicities (red curves). Corresponding dependence of the SFM wavelength (blue curves). The dashed and solid lines correspond to periodicity of 23.5 µm and 24.7 µm, respectively.

Download Full Size | PDF

This technique is applicable to any wavelength in the transmission window of the nonlinear crystal, which extends up to about 4 µm in RKTP [22]. Longer MIR wavelengths could be reached by using other QPM materials such as periodically-poled lithium niobate (PPLN) or orientation-patterned gallium phosphide (OP-GaP), which are transparent to 5 µm [23] and 12 µm [24], respectively.

To perform the time-correlated single-photon counting, a Si-SPAD (Micro Photon Devices, PDM-series) was used in combination with a time-tagging unit (PicoHarp 300, PicoQuant). The detector had a quantum detection efficiency of 23% at 737 nm, a dead time below 88 ns, and a dark count rate specified to be less than 200 Hz. The time-tagging unit allowed time bin widths down to 4 ps with a specified timing jitter of below 12 ps.

In order to prevent 808 nm pump photons from reaching the SPAD detector and acting as noise, a single grating monochromator was inserted after the up-conversion cavity. The photons exiting the monochromator were focussed by a lens on a 62.5 µm core-diameter graded-index fiber. At the output of the fiber the photon stream was recollimated, bandpass filtered to reduce ambient light leakage, and finally focussed on the detector. In the current setup this caused high optical coupling losses and hence a reduced photon detection efficiency.

To evaluate the system as a LIDAR, pulses from the MIR source were guided to the ranging setup using metallic mirrors and split with a microscope slide. The reflected beam hit two uncoated microscope slides separated by an adjustable distance $d$. These microscope slides acted as low reflecting targets for the LIDAR beam. The light reflected by the targets was collected and focused into the up-conversion cavity by a 250 mm lens. The up-converted light was then collimated by a 150 mm lens and sent through the monochromator before reaching the Si-SPAD as seen in Fig. 3. The signal was collected over an integration time of 100 s. In separate measurements, the signal from a single microscope slide was acquired. This acted as a reference for the signal from the two targets.

 

Fig. 3. Sketch of the experiment, where $d$ is the distance between the targets, i.e., two closely spaced microscope slides.

Download Full Size | PDF

The raw signal was fitted to two Gaussian distributions using the nonlinear least-squares method in order to determine the peak positions.

3. RESULTS AND DISCUSSION

The temporal FWHM of a single target was 42 ps, and the corresponding data and Gaussian fit can be seen in Fig. 4. This was primarily limited by the timing jitter of the Si-SPAD detector due to different absorption positions for different photons incident on the detector [25]. The contribution to the temporal FWHM from other sources, such as the optical pulse duration, timing jitter of the time tagging unit, and the bin width was small compared to the jitter of the SPAD.

 

Fig. 4. Measured signal from one specular target. The FWHM of the trace is 42 ps. The measured data is shown as circles, and the solid line is a Gaussian fit.

Download Full Size | PDF

The goal was to perform range measurements above 2 µm as well as determine how closely two separate targets can be positioned while still being distinguishable from each other. The traces from 5five different separations ranging from 8.4 to 0.4 mm of air $ + {1}\;{\rm mm}$ of glass can be seen in Fig. 5. The two targets can easily be separated for the larger separations. For the traces corresponding to 2.4 and 0.4 mm of air and 1 mm of glass between the targets, the FWHMs of the signals were 62 ps and 55 ps, respectively, and two Gaussian peaks could still be fit to the data. This should be compared to the 42 ps in the case of a single target. It is also worthnoting that this range resolution would be kept over long distances. The estimated air gap between the targets fits well with the measurement performed with a calliper as can be seen in Fig. 6. However, between 25% and 35% of all emitted pulses resulted in a detected signal. The high count rate will cause the LIDAR trace to be slightly distorted with reduced number of detections for the second target.

 

Fig. 5. Measured traces for physical separation of 1.4 mm (cyan), 3.4 mm (black), 5.4 mm (red), 7.4 mm (pink) and 9.4 mm (blue). The dashed lines are Gaussian fits, while the solid lines are measurements.

Download Full Size | PDF

 

Fig. 6. Estimated air gap between the two targets.

Download Full Size | PDF

The dark count rate was 480 Hz ($ \pm 5\% $) as the optical pump power at 808 nm was increased from 0 to 2.5 W, which corresponds to 0 to 5 W of intra-cavity circulating power. This indicates that the dark count rate was limited by ambient visible light leakage, since the dark count rate of the detector itself was below 200 Hz and up-conversion of background MIR photons would vary with pump power. The reason that no contribution from the MIR background was measured is likely the high optical losses in the coupling from the up-conversion cavity to the SPAD. The spectral bandwidth of the generated pulse was less than 3 nm, measured and limited by the monochromator, which is in good agreement with the theoretical value of 0.8 nm [Eq. (2)]. The combination of a low dark count rate and a narrow detection window allows for low noise measurements. The noise is further reduced by considering that the nonlinear interaction is only phase matched for vertical polarization, which reduces the random ambient background light by a factor of 2.

To put these values into comparison, it is reasonable to consider the work in Ref. [20], which presented superconducting nanowire single-photon detection (SNSPD) LIDAR at 2.3 µm. SNSPD and up-conversion detection are the best alternatives for single-photon LIDAR at these wavelengths. The authors in Ref. [20] reported a temporal FWHM response of 280 ps from a single target, while the FWHM reported here was only 42 ps, which shows the benefit of using well-established Si-SPAD in this work. The main difference is, however, practical: the SNSPD needs a liquid helium cryostat, whereas the up-conversion instead needs a near-IR pump laser and a more complex optical setup with an SFG stage and wavelength filtering. Which is more practical may depend on the application.

To prove the feasibility of this approach for a “real” LIDAR application, the reflected signal of a piece of paper was recorded. The position of the paper could easily be detected but, as expected, with a significantly lower signal due to the diffuse reflection. Nonetheless, MIR conversion efficiencies of as high as 20% have been reported previously [11]. Combining this with the detection efficiency of a SPAD, the overall detection efficiency could reach more than 5%. This means that by neglecting optical coupling losses, on average, 1 out of every 20 incident MIR photons could be detected. This would be enough for fast, real-world applications.

For practical 3D imaging applications, an integration time in the order of milliseconds per pixel is required. This could be achieved by increasing the repetition frequency of the laser to the megahertz (MHz) regime and allowing for a lower number of counts in the peak. To keep the necessary average laser power reasonable, the detection efficiency needs to be further improved. Possible improvements include the use of a narrower laser spectrum (only about 3% was phase matched), increase of the intra-cavity power by AR coating the nonlinear crystal, and optimization of the coupling efficiency from the conversion cavity into the SPAD.

A rough estimate using the laser radar equation [26] shows that at 1 ms per pixel integration time we could reach 10 m measurement distance with an attenuation length of 5.9 (attenuation coefficient of ${0.59}\;{{\rm m}^{ - 1}}$) using a 10 mW average power laser at 4 µm. The optical density has been measured to ${0.1}\;{{\rm m}^{ - 1}}$ at 4 µm wavelength in black smoke [9], corresponding to an attenuation coefficient of ${0.23}\;{{\rm m}^{ - 1}}$. Our estimate thus shows that the technology could be used to build a 3D lidar system bringing valuable capability to smoke diver applications.

4. CONCLUSION

In this work a MIR LIDAR above 2 µm based on up-conversion photon counting is presented for the first time to our knowledge. The temporal FWHM of the trace of the pulses reflected from a single target was 42 ps. Reflections from two targets, with an optical separation of 1.9 mm, exhibited a 30% broader temporal LIDAR trace. The system showed excellent temporal resolution and a low dark count rate, limited by the Si-SPAD and ambient light leakage. By altering the characteristics of the QPM process such as choice of crystal, period, length, and temperature, the bandwidth and wavelength of the detected signal can be freely chosen. Realistic real-world imaging requires a sampling time of in the order of milliseconds per pixel, which could be achieved by implementing the above listed system improvements.

The sensitivity of this system was reached through a combination of single-photon counting and up-conversion of the MIR signal. This alleviates the need for high NIR laser powers and enables other approaches to reach similar intensities, such as laser diodes and QPM waveguides. Furthermore, this would open up the possibility for detection of multiple MIR wavelengths in the same QPM structure by change of the pump laser.