1. Introduction

Frequency comb interferometry is a well established spectroscopic technique [1–4]. While most applications use combs to probe local samples, there has been some interest in remote sensing using frequency combs. Most proposed remote applications focus on high precision distance measurement from cooperative targets [5,6] or transmission measurements using a remote detector [7], but range resolved spectral characterisation has also been considered [8], although at a somewhat low spectral resolution.

Multispectral lidar has also been demonstrated for spectral reflectance measurement of hard surfaces using broadband pulsed lasers and diffraction gratings [9,10], where vegetation and man-made objects were ranged and discriminated from their low-resolution spectral features.

In this paper, we demonstrate chemical detection with a high-resolution hyperspectral lidar using two distinct frequency combs to probe the scene. The pulsed nature of frequency combs makes it possible to retrieve ranging information via time-of-flight measurements, while the sweeping delay between comb pulses yields an interferogram, which can be Fourier transformed to obtain spectral information. The overall result is the spectrum of the scene as a function of distance. Both spectral and range resolution are demonstrated by the simultaneous measurement of the HCN filtered backscattered signal from a water droplet cloud and a spectralon block.

2. The instrument

A schematic of the system is shown in Fig. 1 . The comb sources used are Menlo c-fiber and c-comb lasers. Each laser outputs 90 fs pulses with an average power of 20 mW at a 100 MHz repetition rate.

 

Fig. 1 Complete measurement setup. Both combs are sent to a pulse stretching module for chirped pulse amplification. Repetition rate is chosen using the pulse pickers, after which the pulses are amplified and combined. A local trace is taken for calibration purposes. The combined combs are then sent to the target and the return signal is collected by an APD in the image plane of a telescope.

Download Full Size | PDF

Before amplification, 400 m spools of SMF-28 fiber stretch both laser pulses. This is done to spread each pulse’s energy over a longer period of time, decreasing peak power and thus the effects of fiber nonlinearities after amplification.

Fiber couplers split the stretched pulses in two equal parts; half the signal power is sent to a passive optical referencing setup which tracks optical delay and phase difference between pulses for post correction purposes. The referencing system is described in detail in [3]. The other half of the pulse power is sent to the pulse pickers.

At a repetition rate of 100 MHz, the lidar unambiguous range is approximately 1.5 m, which is half the distance traveled by a pulse during the time between two pulses. To increase the unambiguous range, the repetition rate has to be lowered so that more time passes between successive laser pulses. This is the role of the pulse pickers. Each one is composed of an EOSPACE intensity modulator driven by an electrical pulse generator (HP 8130A and HP 8131A). The latter is triggered by the photodetected pulses from a comb and activates the modulator only during the pulses needed to get the desired repetition rate.

After pulse picking, the laser signals are amplified using two chains of optical amplifiers, each composed of an INO FAD-180 preamplifier and a Pritel high power amplifier (2 W and 5 W).

For a given amplifying chain, there is no loss of average output pulse train power when pulse picking up to a certain point. This is explained by the fact that the initial repetition rate is much higher than the charge carrier pumping response time of the amplifiers. Reducing the repetition rate increases the carrier buildup time between successive pulses, which results in a higher energy per pulse. With a low enough repetition rate, the amplifier reaches a steady state between pulses, where spontaneous emission rate matches pumping rate. At that point, the increase in pulse energy stops compensating for the decrease in repetition rate and the average pulse train power is no longer constant. That point was experimentally found to be at a repetition rate of a 2 MHz, as can be seen on Fig. 2 . This fact can be exploited to gain a better measurement signal-to-noise ratio in long distance measurements, where the return signal is weak and thermal noise is the dominant noise source. Indeed, increasing the pulse repetition interval by a factor of N concentrates the energy from N pulses into one pulse with N times the energy. This results in N times fewer noise samples needing to be considered for the same total signal energy, and thus in an increase in signal-to-noise ratio. This advantage is coupled with the drawback of higher peak power, which makes non-linearity management harder.

 

Fig. 2 Pulse energy at the output of one of the EDFAs as a function of pulse repetition interval. For all pumping currents, the pulse energy caps at approximately 0.5 µs, resulting in an optimal repetition rate of 2 MHz for long distance measurements.

Download Full Size | PDF

After amplification, the pulses are combined using a fiber coupler. Approximately 3 m of SMF-28 fiber lead to the couplers. One output arm of the coupler is sent to a Thorlabs PDB130C detector to generate a calibration interferogram. The second output arm sends the combined combs to the scene via a collimator and an adjustable beam expander, which generates a 24 mm wide beam and can be used to place the beam waist at the chosen depth in the scene. The length of fiber leading to the launcher is around 5 m, while the length on the calibration side is approximately 2 m.

Backscattered light from the scene is collected by a Meade LDX75 SN-8AT 8 inches telescope and focused on a Princeton Lightwave PLA641 avalanche photodiode (APD). An infrared camera is also used to facilitate alignment. The response time of the APD sets the range resolution of the system. Given the 2.7 ns response time of the PLA641, our system has a resolution of approximately 40 cm. The launcher and telescope system, shown in Fig. 3(a) ), has a coaxial launching scheme. It is designed for measurements at distances of up to 200 m. Thus, it is designed such that, when the waist of the 24 mm output beam is placed at 150 m, the total geometric attenuation is roughly constant for distances up to 200 m, where overlap factor increases stop compensating for smaller collection solid angle.

 

Fig. 3 a) Launcher, telescope and APD setup. b) Fog generating apparatus. Condensed vapor generated above the liquid nitrogen bath is pushed to the 1 m horizontal tube (labeled Interaction section) by the convection fan.

Download Full Size | PDF

The whole system, apart from the telescope module, fits on a 2.5 feet by 5 feet optical breadboard. The weight of the setup is dominated by the telescope and its mount, which weigh a total of 32 kg.

3. Experimental results

The scene used to generate the results presented here consists of a 1 m deep water droplet cloud approximately 1.5 m in front of a spectralon block. A 16 cm long room temperature gas cell filled with 100 Torr of hydrogen cyanide (HCN) is placed before the water cloud, providing high-resolution spectral content. The water cloud was generated by blowing room temperature air on a liquid nitrogen bath. The fog generating device is shown in Fig. 3(b)). The scene was situated at approximately 20 m from the launcher, due mostly to space constraints in the laboratory. When focused at 20 m, the spot size of the beam is 400 µm, which, coupled with the 80 cm focal length of the telescope and the 200 µm APD size, results in a unitary overlap factor. The measurement was done with the lights on, as the APD was unaffected by ambient light.

Data acquisition was done at 5 GS/s and 1 MS per trace on a LeCroy 725Zi oscilloscope. With a repetition rate difference before pulse picking of approximately 100 Hz between the pulsed lasers, this corresponds to an optical path difference (OPD) span of 200 ps and a spectral resolution of 0.04 nm. Pulse picking ratio was set at 1:3, resulting in 30 ns repetition period and an unambiguous range of 4.5 m, allowing for the full scene sample to be acquired without spatial overlap. The chosen pulse picking ratio results in a far lower repetition period than the optimal one found in Fig. 2. This is because of the aforementioned nonlinearity management problem, which degrades measurements for high peak pulse powers. The chosen repetition rate is a good compromise between signal-to-noise ratio optimization and nonlinearity mitigation. The high power amplifiers were set at 1 W. From the signal amplitude on the APD, the recovered power was found to be approximately 4 nW.

For this measurement, 30,000 traces were acquired, corrected and averaged to yield a single data matrix containing the scene interferograms as a function of distance. The acquired traces contain six seconds of data. However, because the repetition period of the probing combs is 30 ns, each interferogram generated by the beating between the combs spans a total OPD of 30 ns. Since the recorded OPD span 200 ps, the resulting measurement duty cycle, assuming every interferogram is recorded, is 0.67%. Because of that low measurement duty cycle, it takes 15 minutes to acquire 30,000 successive interferograms. With the current setup and acquisition scheme, which does not acquire every incoming interferogram, the measurement duration was about 45 minutes. By acquiring every interferogram and using a back and forth scanning technique, as demonstrated in [7,11,12], measurement time could be reduced to the aforementioned six seconds.

The measured range-resolved interferogram of the scene is shown in Fig. 4 . Two distinct interferograms can be seen; the first one is the result of the backscattered light from the water droplet cloud, while the second one is due to the spectralon block. Although it is not very apparent from the shown angle, the distributed nature of the cloud makes its trace slightly wider in the ranging dimension than the spectralon trace. However, its width does not correspond to the full 1 m length of the cloud, since most of the backscattered power comes from the front of the cloud.

 

Fig. 4 Interferogram as a function of distance from the launcher. About 3 m of the full unambiguous range of 4.5 m corresponding to a pulse picking factor of 3 can be seen. The graph features two distinct interferograms, corresponding to the distributed reflection from the aerosol cloud and the discrete but diffuse reflection from the spectralon block.

Download Full Size | PDF

A Fourier transform along the OPD dimension yields a spectrum as a function of distance, as shown in Fig. 5 . Once again, the range resolved nature of the measurement is apparent, with two distinct spectra from the water cloud and the spectralon block being clearly visible. HCN absorption lines can be seen on the highlighted slice at 20 m.

 

Fig. 5 Spectrum of the scene as a function of distance. Both the aerosol cloud and the spectralon block spectra can be seen. A slice at 20 m is highlighted in black to show the HCN absorption lines.

Download Full Size | PDF

Fig. 6 and Fig. 7 show the reference, scene and calibrated spectra of the water cloud and the spectralon block, respectively. Each trace corresponds to a slice taken at the peak of the two spectral responses shown on Fig. 5. Spectral signal-to-noise ratio remains better than unity in a 20 nm region around 1560 nm. This somewhat narrow optical bandwidth currently limits the usefulness of the method to rapidly varying spectral features in a relatively small spectral window, such as specific gas absorption lines. Amplification using high power EDFAs cuts out most of the spectral width of the combs, which have an initial wavelength span of approximately 80 nm. To be able to identify a wider variety of chemical species or slowly varying spectral features, such as the ones found in vegetation, a broader probing spectrum is needed. To this end, further work could include non-linearity management after amplification to coherently broaden the spectra, such as in [13]. This would allow atmospheric range-resolved spectroscopy of useful molecules, including water, methane and CO_2.

 

Fig. 6 Spectrum of the HCN filtered water droplet cloud. Absorption lines from the HCN cell can clearly be seen. Spectral features present in the combs are very well removed by the calibration spectrum.

Download Full Size | PDF

Fluctuations of relatively high amplitude can be seen on the reference spectrum, which makes calibration necessary. Those fluctuations are not present in the unamplified comb beatings. Amplification, followed by fiber propagation, results in nonlinearities that make the polarization state change within each pulse. Since the interference between a pair of comb pulses relies on their respective polarization to be aligned, the nonlinearity induced polarization fluctuations result in varying modulation efficiency throughout the pulses, which are seen as fluctuations in the measured spectrum. Nevertheless, it can be seen in Fig. 6 that those fluctuations are removed very well by the calibration process: the resulting spectrum is a relatively flat background with HCN absorption lines clearly visible and well resolved.

The calibrated spectrum from the spectralon block, shown on Fig. 7 , is not as smooth as the one from the water cloud. This is caused by speckle due to the irregularities of the spectralon surface, which results in wavelength dependent fluctuations in reflected intensity. Figure 8 highlights that fact by showing that spatial averaging through rotation of the spectralon sample during the measurement removes those fluctuations.

 

Fig. 7 Spectrum of the HCN filtered spectralon block. Some spectral features not due to the HCN remain after calibration. These features are caused by speckle from the spectralon block.

Download Full Size | PDF

 

Fig. 8 Calibrated spectralon spectrum for both stationary measurement and spatial averaging by making the spectralon block rotate while measuring. The spectral fluctuations are attenuated by spatial averaging, showing that they are caused by speckle due to surface irregularities.

Download Full Size | PDF

4. Conclusion

This paper shows that chemical detection from an aerosol-backscattered signal is possible with a dual comb hyperspectral lidar, getting both high-resolution spectral and distance information from the scene. Aerosol backscattering and HCN absorption lines are measured at up to 20 m. Further work will include ruggedizing the setup to carry out outdoors long distance measurements.

Although the presented system has enough resolution for absorption line identification, it lacks the optical bandwidth to be a full-fledged remote spectroscopic tool. Optical amplifiers, while necessary for long distance measurements, further narrow the spectral width of the frequency combs used as a source, from approximately 80 nm down to 20 nm. While challenging due to the coherence requirements of comb spectroscopy, non-linear broadening of the source lasers would greatly increase the usefulness of the technique.