3.1.

Setup

Depending on the operation mode of the laser, different laser drivers are used. For pulsed operation, a short-pulse voltage driver capable of delivering pulses with a pulse length between 50 ns and $10\mu \mathrm{s}$ and a current of up to 3 Amperes is used. A dedicated QCL low-noise current driver (Wavelength Electronics QCL2000) drives the laser chip in the case of cw operation. To measure laser intensity, photovoltaic mercury cadmium telluride (MCT) detectors (VIGO System) are used, thermoelectrically cooled to about 210 K, with a rise time of $<3\mathrm{ns}$. A gated integrator (Stanford Research SR250) with variable gate delay and a gate width of 100 ns is employed for time-resolved measurements, and spectra are taken by an FTIR spectrometer (Bruker Vertex 80) with a fast MCT detector that allows for boxcar-integrated, time-resolved spectral characterization of the laser source.

Fast spectroscopy measurements as shown in Sec. 4 are conducted with the laser in pulsed mode driven with repetition rates of 400 to 500 kHz, yielding 400 to 500 spectral sampling points within the oscillation period of 1 ms. To reconcile the very rapid tuning rate of the laser to the slow measurement acquisition of the FTIR spectrometer when measuring the absolute optical frequency produced by the laser source as a function of MOEMS grating position, the acquisition of the FTIR detector was gated with variable delay with respect to the trigger signal produced by the MOEMS positional output (Fig. 2). This allows for slower data acquisition methods and time-resolved spectral investigations using an FTIR.

Fig. 2

The time dependence of the angle of a typical MOEMS grating is shown in blue. The rectangle indicates the boxcar window gating the detector at variable time delays from the MOEMS trigger. When operating the laser in pulsed mode, the laser pulse is timed accordingly.

Figure 3(a) shows the tuning ranges of several chips in external cavity operation used for measurements shown in this paper, all with maximum peak power levels of several hundred mW. Figure 3(b) shows a transmission spectrum of a GaAs wafer with a thickness of $124\mu \mathrm{m}$ recorded with the MOEMS EC-QCL equipped with chip #2, exhibiting clearly visible interference fringes of the GaAs etalon. These fringes provide a quick and easy way to monitor the spectral tuning behavior, e.g., on an oscilloscope.

Fig. 3

(a) Power versus wavelength profiles of the chips used for the measurements reported here. (b) Transmission spectrum of a $124\text{-}\mu \mathrm{m}$ thick GaAs etalon recorded with the rapid-scan MOEMS EC-QCL equipped with QCL chip #2.

3.2.

Pulse-to-Pulse Fluctuations

For an evaluation of the laser source with regard to pulse-to-pulse intensity fluctuations, chip #2 is driven with 100 ns pulses at fixed delays to the MOEMS trigger (i.e., fixed wavelengths). The integrated pulse intensity is then recorded for time periods of 5 min. Allan deviation plots^24,25 of the data are shown in Fig. 4. As long as short-term noise dominates over long-term drift, the inverse dependence of Allan deviation on averaging interval length appears as a straight line on the log–log scale. It is apparent that up to averaging intervals of about 10 s, no drift effects affect laser noise performance. Note that for a laser source aiming at real-time applications, the envisioned averaging interval times are much shorter (typically below 100 ms). For such interval lengths, the normalized standard deviation of mean pulse intensity is on the order of ${10}^{-3}$. Allan deviation is measured in both static mode (i.e., with the scanner turned off) and dynamic mode. Using a variable gate delay of the boxcar integrator, multiple dynamic Allan plots are recorded at different wavelengths. Differences in overall noise level are tentatively attributed to the varying prevalence of mode hops at different grating angles.

Fig. 4

Allan plots: two-sample deviation (Allan) plots of pulse intensity are shown to differentiate between short-term pulse intensity noise and long-term drift—measured with both static MOEMS scanner (emitting at $1105{\mathrm{cm}}^{-1}$) and moving scanner. Starting at intervals of about 10 s, drift effects start to appear, corresponding to several thousand averaged spectra.

3.3.

Linewidth and Spectral Reproducibility

Spectral emission linewidth and wavelength reproducibility are among the most important characteristics of any laser source to be employed in spectroscopic applications. In pulsed operation, the typical spectral width²⁰ of its emission lies between 1 and $2{\mathrm{cm}}^{-1}$. Changing to a cw-capable laser [chip #3, see Fig. 3(a)] allowed us to significantly reduce the spectral width of its emission. The laser is driven with 800 mA of cw current and stabilized at a heatsink temperature of 20°C. Spectra are taken using the gated spectrometer and boxcar integrator, measured at different times of the scanner oscillation by tuning the time delay between MOEMS trigger signal and boxcar window (100 ns gate width). An exemplary spectrum of the cw MOEMS EC-QCL is displayed in Fig. 5 (blue), exhibiting a spectral width of $\mathrm{FWHM}=0.27{\mathrm{cm}}^{-1}$, limited by the resolution of our FTIR spectrometer. The spectrum taken with a static scanner (i.e., actuation of the scanner turned off, grating in its rest position) exhibits the same, instrument-limited linewidth. Obviously, the scanner movement does not impair the linewidth over the measurement time of about 30 s it takes to record one FTIR spectrum. From these measurements, it can be inferred that the spectral reproducibility of the MOEMS EC-QCL within that time frame is better than the spectral resolution of the FTIR ($0.27{\mathrm{cm}}^{-1}$).

Fig. 5

Lasing spectra in cw operation: Shown in blue is an example of an emission spectrum of the MOEMS EC-QCL in cw operation at a specific timestamp. For comparison, the spectrum of the static laser source is plotted in red. The linewidths ($\mathrm{FWHM}=0.27{\mathrm{cm}}^{-1}$) are the same, limited by the resolution of the FTIR spectrometer.

The time-dependent wavelength of the cw emission is investigated by varying the time delay between MOEMS trigger signal and boxcar window in small steps and plotting the center of the respective emission peaks as function of the delay, as shown in Fig. 6. Marked by black dots are the center wavelengths of the respective main emission peaks at a certain timestamp. The blue line represents the sinusoidal curve according to the grating equation (see Sec. 2) and the respective scan parameters. Note that amplitude and phase of the scanner oscillation were obtained by fitting Eq. (1) to the data. No indications for a deviation from the assumed sinusoidal motion of the scanner are found. A more detailed view reveals individual longitudinal mode-hops in steps of about $0.75{\mathrm{cm}}^{-1}$ during the spectral scan. This effect results from the coupled cavity that consists of the external resonator and the laser chip acting as a weak Fabry–Perot resonator (chip length of 2 mm, refractive index $n=3.33$).

Fig. 6

Mapping of the time-dependent emission wavelength: (a) center wavelengths of the main emission peaks are plotted against the respective delays between MOEMS trigger and boxcar window (black dots). The zero-crossing of the grating occurs at an arbitrary $321\mu \mathrm{s}$ after the trigger signal. The blue line shows the expected emission wavelength derived from Eq. (1). (b) Zooming into the plot resolves the longitudinal mode hops in steps of about $0.75{\mathrm{cm}}^{-1}$.

4.1.

Backscattering Spectroscopy

The first application to be discussed is contactless (“standoff”) identification of chemical substances on surfaces over a certain distance by diffuse reflectance spectroscopy.

In previous publications, we investigated this topic in a security-related context with the means of hyperspectral imaging using a slow-scanning EC-QCL for active illumination of the scene and a high-end MCT camera for collecting the backscattered radiation. With a tripod-mounted system with a large-aperture collection optics, small amounts of explosives and related precursors could be detected over distances of about 20 m with spectral scanning times on the order of 10 to 30 s.²⁰ Here, we present our first results obtained with a QCL-based handheld scanner designed to detect a wide range of substances on arbitrary surfaces over shorter distances of about 0.5 to 2 m. These developments are driven by the need of first responders for portable and simple-to-use detection hardware for fast identification of suspicious, potentially hazardous substances. Importantly, the technique works at eye-safe power levels. The MOEMS EC-QCL represents an ideal laser source for a handheld device, as one main challenge (in addition to the obvious requirement of a compactness and low weight) are slight, unavoidable movements of the hand holding the device and hence the laser beam. This issue can be solved by fast spectral scanning, as every single spectrum is collected in a time short enough to consider the beam stationary. Recently, we performed measurements in a comparable laboratory optical setup situation, where laser and detector were still fixed on an optical table, and the sample was moved manually during measurement to mimic the handheld-situation.¹⁹ Here, we present, for the first time, results using a handheld device.

The handheld demonstrator (Fig. 7) contains a MOEMS EC-QCL as light source, equipped with a QCL chip tunable from 1000 to $1300{\mathrm{cm}}^{-1}$. The laser is operated in pulsed mode at $\sim 500\mathrm{kHz}$, 100 ns. The sample is illuminated with a collimated beam of $\sim 5\text{-}\mathrm{mm}$ diameter. The beam is collinear with the optical axis of the receiver optics, a 4 in. $f/2$ mirror that collects and focuses the diffusely scattered light onto a fast, thermoelectrically cooled MCT detector (Vigo System). In this first demonstrator, the operating distance was fixed to 1 m, with the signal remaining above 85% within a total defocus range of $\sim 10\mathrm{cm}$. Beam position and working distance are indicated to the operator by two visible laser diodes crossing their beams in the focal plane of the receiver optics. A forward facing camera and rear-facing monitor are provided for operator feedback. The most recent constantly acquired spectrum is displayed in real time, along with the forward facing field of view.

Fig. 7

Portable demonstrator for contactless detection of solids. The MOEMS EC-QCL is located on top of the device, the single-element MCT detector at the bottom.

In Fig. 8, we show that diffuse reflectance spectra measured with this device are well comparable to those obtained by established reflectance spectroscopy using an IR microscope coupled to an FTIR.

Fig. 8

Diffuse reflectance spectra of bulk samples from (a) acetylsalicylic acid, (b) paracetamol and (c) riboflavin obtained with a MOEMS-ECL-based scanner (500 ms acquisition time, i.e., 500 averaged spectra, 1 m distance), and with an FTIR microscope (acquisition time of several minutes), respectively. The results are well comparable.

Samples as shown in Fig. 8 could be identified with our handheld scanner in quasi real-time by comparing the recorded spectra with a database, using a simple and very fast cross-correlation algorithm (not shown). While this approach is successful on bulk samples (i.e., strong coverage of the underlying substrate), detection of residual material on scattering substrates often poses the problem of background interference, i.e., significant contributions of the a priori unknown substrate to the recorded spectrum. In previous work on imaging detection, we showed that this can be accounted for by suitable data analysis, if a hyperspectral image (spatially resolved spectral information over a certain area) of the sample is available and contains uncontaminated pixels providing the respective background spectrum.²⁶ Here, such a dataset is obtained by scanning the beam manually over a sample of polyamide contaminated with the explosive pentaerythritol tetranitrate (PETN) for a few seconds. The sample and approximated path of the beam are shown in Fig. 9(a). Figure 9(b) shows color-coded time-resolved spectra taken during the recording process. Each column is treated as a pixel of a hyperspectral image by our target detection algorithms, described in detail e.g., in Ref. 26. Here, we want to point out that the only necessary input is the target spectrum while the unknown background is extracted from the dataset by the algorithm. The top row of Fig. 9(b) shows the time-resolved detection results. In Fig. 9(d), the spectra obtained in regions considered as “PETN contaminated” as well as in uncontaminated regions are shown. Despite strong contributions of the polyamide background present in all spectra, PETN is identified, and the spectrum of pure PETN is recovered from the data by spectral unmixing [Fig. 9(c)]. Similar results were obtained with other hazardous contaminations and substrates.

Fig. 9

(a) Sample: PETN contaminations on polyamide substrate. The arrow shows the approximated trace of the laser beam on the sample during manual scanning. (b) Color-coded evolution of the spectra obtained while the beam is scanned over the sample for $\sim 3.5\mathrm{s}$. In the top row, detection results obtained with the adaptive matched subspace detection algorithm²⁶ applied to the whole dataset are displayed. (c) Left: averaged spectra of regions classified as PETN contaminated (red), uncontaminated (green), and our PETN library spectrum used for detection. Right: the PETN library spectrum is recovered from the data after unmixing using our detection algorithms.

We note that the described approach does not deliver actual “real-time” results as in the case of cross correlation, analyzing single spectra. First, the spatial scan has to be completed, as our background-suppressing algorithm requires the entire dataset.

Current work on the optical part includes implementation of an autofocus for variable distance measurement and reducing the overall size. Furthermore, data acquisition and analysis, which took place on a separate computer, will be integrated in the device to provide a truly mobile scanner.

4.2.

Transmission Spectroscopy for Quality Control of Coatings

In our second example, we demonstrate how the MOEMS-EC QCL can be used to monitor coating quality. A potential, challenging, and timely example of in-line monitoring using our technology is the application of a coating such as glue or silicone to a thin foil, or the detection of contaminations (oil, silicone, etc.) on a foil before the coating process. Such applications with industrial context typically require high measurement speeds to achieve sufficient spatial resolution on fast moving samples.

We prepared a silicone coating with intentional defects (i.e., gaps) on a polyethylene foil by manual spray coating. The foil is glued to a chopper wheel, which allows moving the sample at predefined velocities. The foil is measured in transmission geometry with the laser beam being focused on the foil ($\mathrm{NA}\sim 0.05$) to ensure high spatial resolution. The chopper wheel has five segments containing [see Fig. 10(a)] A. PE foil, B. silicone on PE with a 6-mm-wide line defect, C. silicone on PE with 4-mm defect, D. silicone on PE with 3-mm defect, and E. silicone on PE with 2-mm defect. In Fig. 10(b), we show the measured transmission spectra versus time for a scan velocity of $0.5\mathrm{m}/\mathrm{s}$. Spectra recorded while the laser was blocked by the spokes of the chopper wheel were removed, as they contain no useful information (zero transmitted intensity). All spectra are referenced to a PE foil with no coating applied. The spectra from segment A are an indicator for the deviations of the PE foil thickness from the reference foil.

Fig. 10

(a) Sample: each of the five compartments of a chopper wheel contains a PE foil, patterned with a silicone coating with intentional line defects of width 6, 4, 3, and 2 mm, as indicated above. The sample was measured while moving at $0.5\mathrm{m}/\mathrm{s}$ at the beam position. (b) Color-coded evolution of the relative transmission spectra. The transmission is referenced to an uncoated PE foil. Spectra recorded while the laser was blocked by the spokes of the chopper wheel were removed. (c) Single spectral scan of the silicone coating using the MOEMS EC-QCL versus an FTIR reference measurement of the same sample. (d) Quality parameter constructed from the spectra in (b). A threshold can be chosen to define a faulty coating.

In Fig. 10(c), we compare a single transmission spectrum of the silicone coating measured with the MOEMS EC-QCL with an FTIR reference measurement for the same sample. The silicone coating exhibits several strong absorption bands in the considered spectral range of 1000 to $1300{\mathrm{cm}}^{-1}$. Using the strongest absorption bands, a simple quality parameter was constructed. In Fig. 10(d), we plot the quality parameter. All defect lines, including those with the smallest width of 2 mm could be identified. It becomes obvious that the coating thickness varies strongly throughout the sample due to our manual coating process. However, from the measured spectra, a limiting value distinguishing between a good and faulty or not existing coating can be defined.

We note that in Fig. 10(d), each point corresponds to a single spectral scan. The spatial resolution for a velocity of $0.5\mathrm{m}/\mathrm{s}$ is $\sim 0.5\mathrm{mm}$ along the moving direction using the full MOEMS scan period of 1 ms for the construction of one spectrum, as has been done here. During one such period the full spectral range is covered twice (once in either direction), therefore the above spatial resolution could be doubled. For the smallest defect of 2-mm width, this causes a deviation of the quality parameter in four consecutive IR spectra, which matches well with the experimental results. The evaluation of the full spectral dataset allows for an increased signal-to-noise ratio and less dependence on other changing parameters such as changes in the foil thickness due to the fast and fluttering movement of the sample.

4.3.

Transmission Spectroscopy of Gases

Our third example is real-time absorption spectroscopy in the gaseous phase.¹⁹ In its present form, the MOEMS EC-QCL is not intended for high-resolution spectroscopy of narrow gas absorption lines. While narrow linewidth can be provided in cw operation, length adjustment of the resonator during a wavelength scan is currently not implemented; therefore, the laser will undergo mode-hops during spectral scanning, as shown in Sec. 3.3. However, the absorption spectra of some gases consist of somewhat broader absorption bands, and these gases are well suited to be probed by a pulsed laser source with a spectral resolution on the order of $1{\mathrm{cm}}^{-1}$. The MOEMS EC-QCL allows fast tracking of changes in gas concentration as shown below. Note that also in a scenario where the main concern is not real-time monitoring but mere detection of some gaseous compounds, fast spectroscopy can still be very valuable. In turbulent media, for instance, spectra taken over longer periods than occurring changes in composition are very difficult, if not impossible, to interpret. These issues are avoided by fast spectral scanning, as every single recorded spectrum is meaningful.

Here, we chose isopropanol vapor as a model gas with two broad spectral features in the 7.5- to- $9\mu \mathrm{m}$ range. The beam of the MOEMS EC-QCL equipped with chip #1 is directed at an integrating sphere. Its intensity is recorded with a fast MCT detector placed at the exit port of the sphere. Air saturated with isopropanol vapor is blown through a small hole into the integrating sphere that also acts as a very simple means to increase the absorption length as the laser light undergoes multiple internal reflections before reaching the detector. The temporal evolution of the recorded absorbance spectra is shown in Fig. 11(a). In Fig. 11(b), the measured absorbance after saturation is depicted, showing very good agreement to literature data for isopropanol vapor (from Ref. 27).

Fig. 11

(a) Time-resolved color-coded representation of the measured absorbance spectrum, recorded while air saturated with isopropanol vapor is blown into an integration sphere (see text). Data have been referenced to the initial state (no isopropanol present). Changes in absorbance appear on short timescales of $\sim 10\mathrm{ms}$, saturation is approached after $\sim 0.5\mathrm{s}$. (b) Measured absorbance spectrum after 0.5 s (in red), showing a very good match to literature data in blue (from NIST,²⁷ scaled to our data to account for different path length and offset by 0.02 for better visual comparison).