2.1.

Experimental Setup

To characterize a filter with tunable CWL, a system that validates the spectral performance of static bandpass filters is first needed. The experimental setup consists of (1) an IR camera (Indium antimonide detector, FLIR SC8000 Series), (2) a radiation shield tube, (3) a contact hot plate (blackbody radiation source), and (4) the filter stack (Fig. 2). The IR camera is capable of imaging the MWIR waveband (2.5 to $5\mu \mathrm{m}$) with a resolution of $1344\times 784\text{pixels}$. The camera utilized a lens with a focal length of 50 mm. The distance between the lens and filter (target) was 122 cm. The IR images were recorded using FLIR ResearchIR software. The blackbody radiation source was an iron plate with emissivity $\epsilon \approx 1$ and a diameter of 17.8 cm. A tube between the camera and filter stack shielded against unwanted background radiation.

Fig. 2

Experimental setup. (a) A schematic of the experimental setup. The IR camera images the linear variable filter (LVF) with a fixed-wavelength bandpass filter behind it. They are both illuminated by the contact hot plate (blackbody). (b) A photograph of the experimental setup. The hot plate acts as a blackbody thermal radiation source. (c) The LVF is mounted in the IR camera’s field of view. (d) The bandpass filter is between the blackbody source and LVF.

A top-down schematic of the experimental setup is shown in Fig. 2(a), whereas the photograph of the entire setup is shown in Fig. 2(b). The LVF (Vortex Optical Coatings, Ltd.) and bandpass filter are shown in Figs. 2(c) and 2(d), respectively. The bandwidth of the LVF at 50% peak transmittance is specified by the manufacturer to be 2% of the peak wavelength with peak transmittance typically greater than 50% across the CWL band.¹⁶ The LVF also has out-of-band blocking within its spectral range of optical density greater than 2.5.¹⁶ The LVF has nominal dimensions of $15\mathrm{mm}\times 3.5\mathrm{mm}\times 0.1\mathrm{mm}$. The CWL of the passband ranges from 2.5 to $5\mu \mathrm{m}$, depending on the longitudinal distance along the filter.

The LVF was placed between the camera lens and as close to the bandpass filter as possible, minimizing any deviation due to converging/diverging beams. Similarly, the bandpass filter was placed between the LVF and the blackbody radiation source. The bandpass filters were manufactured by Andover Corporation and had CWL of 3.10, 3.37, 3.50, 3.60, 4.26, 4.50, and $4.70\mu \mathrm{m}$. These filters each had a transmission of 75% and an out-of-band blocking of optical density 3. The tolerance of the CWL of each filter was 40 nm and each filter was $1\mathrm{mm}\pm 0.2\mathrm{mm}$ thick. The bandwidths of the filters were between $120\mathrm{nm}\pm 30\mathrm{nm}$ and $140\mathrm{nm}\pm 30\mathrm{nm}$.

2.2.

Background on Linear Variable Filters

An LVF can be considered a FP-interferometer that consists of a tapered resonator between one flat and one tilted dielectric (or metallic) mirror.^17–20 This type of filter is also sometimes known as a Fizeau interferometer¹⁸ and for spectroscopic analysis an LVF can be considered a Fizeau interferometer illuminated with a broadband light source.^17,18 The thickness and particular material of the tapered layer enables LVFs with different wavelength ranges to be realized.^17,21 In the all-dielectric LVF case, for mirror reflectivities $<90\%$, the wavelength range (bandwidth) can be simplified to a conventional FP-model, which varies with the materials’ refractive indices according to the following reduced equation:¹⁹

If the incoming light to the LVF first passes through collimating optics, the following relationship results:¹⁹

The collimation of light can strongly affect the transmission and CWL of the LVF. A greater cone half angle (CHA) would decrease the transmittance of the LVF while blue-shifting the CWL.¹⁸ Because LVFs may be approximated to spatially varying FP filters, this behavior can be explained using the theoretical model of FP interferometers.¹⁸ For more details on the theory behind LVFs, the reader is referred to work by Ayerden et al.¹⁸

One purpose of this system is to track the changes in phase of the PCM during a reliability test. During a phase-change, the PCM filter may undergo a change from a $2.9\text{-}\mu \mathrm{m}$ CWL to a $3.3\text{-}\mu \mathrm{m}$ CWL for a change in CWL of $0.4\text{-}\mu \mathrm{m}$. In that case, the uncertainty of $0.085\mu \mathrm{m}$, although greater than other methods, is acceptable because is enables the clear detection of phase changes in the tunable optical filter.

2.3.

Experimental Results

Using the MWIR camera, the LVF was imaged with seven different bandpass filters behind it. Figure 3 shows the experimental results. The superimposed view of the IR intensity image and RGB image is shown in Fig. 3(a) for illustration purposes. As shown in Fig. 3, the CWL of the LVF varies along the length of the filter beginning at $2.5\mu \mathrm{m}$ and gradually increasing to $5\mu \mathrm{m}$ at the opposite end. However, due to variations in the manufacturing of our particular LVF, experimental data confirmed that it reached its $5\text{-}\mu \mathrm{m}$ peak in just 10 mm of filter length. The “scan line pixel number” in Fig. 3 refers to the numbering of the camera’s pixels along the horizontal axis of the LVF. The CWL here is defined to be the wavelength at which the transmittance is maximum. “Detector counts” from the IR camera are used as a proxy for transmittance; the location of peak detector counts is the location of the peak filter transmission. To find this location on the LVF, a single horizontal line of pixels is sampled in the center of the filter and parallel to the long axis. In Fig. 3(b), the locations of peak transmission can be identified as regions of higher relative intensity, where the brightness of the pseudo-color image corresponds to the number of counts recorded during a given detector integration time. For each bandpass filter, the particular integration time was determined experimentally to obtain the high-contrast image and depended on the temperature of the blackbody radiation source.

Fig. 3

Results of the bandpass filter characterization using the LVF. (a) An IR image of the LVF overlayed on top of a red–green–blue (RGB) image of the filter stack that includes the LVF and bandpass filter. (b) The IR images of the LVF when bandpass filters of different CWL are placed behind it. (c) The normalized detector counts versus the scan line pixel number for each bandpass filter, with CWL shown in the legend. (d) The CWL versus scan line pixel number of the bandpass filter. The correlation is linear such that the scan line pixel number can be used to predict the CWL of the optical filter.

In this experiment, we demonstrate that we can detect the CWL and accurately measure its location along the LVF. The plot of normalized detector counts versus scan line pixel number is shown in Fig. 3(c) for each bandpass filter. The experimental results confirm that the LVF accurately indicates the CWL of the filter that is placed behind. The correlation between CWL and the scan line pixel number is shown in Fig. 3(d). The pixel number of the scan line is a proxy for the linear distance along the filter.

For our experimental setup and detector/optics combination, the minimum horizontal spatial resolution for each pixel is $\sim 0.341\mathrm{mm}$. The instantaneous field of view (IFOV) of our imaging system represents the spot size on the LVF for an individual pixel and also corresponds to our estimated spectral measurement uncertainty. Using a 50-mm lens in combination with a $1344\times 784$, $14\text{-}\mu \mathrm{m}$-pixel-pitch sensor, we were able to sample the 15-mm LVF with $\sim 104\text{pixels}$ along the filter’s horizontal (longitudinal) axis. As stated previously, the LVF did not use the full 15-mm length to cover the 2.5- to $5\text{-}\mu \mathrm{m}$ wavelength span. We measured the $2.5\text{-}\mu \mathrm{m}$ CWL shift on the LVF in 10 mm rather than 15 mm. This slightly reduced the spectral resolution of the system. With our 10-mm (effective) LVF, we see a $0.25\text{-}\mu \mathrm{m}/\mathrm{mm}$ shift spatially along the filter rather than an expected $0.17\text{-}\mu \mathrm{m}/\mathrm{mm}$ shift.

2.4.

Error Analysis

When combined with the IFOV of our imaging system (0.341 mm), the spectral measurement uncertainty of our experiment can be computed as $0.085\mu \mathrm{m}$. A 0.341-mm spot size on the filter corresponds to a $0.085\text{-}\mu \mathrm{m}$ change in wavelength across a pixel. In Fig. 3(d), the CWL of the bandpass filter is the independent variable whereas the scan line pixel number is the dependent variable. If the CWL was unknown and the scan line pixel number was used to predict the CWL, the prediction error may be $\pm 0.085\mu \mathrm{m}$. The spectral measurement uncertainty is inversely proportional to the length of the LVF. However, the uncertainty is proportional to the change in wavelength of the LVF. It is computed as follows:

Table 1

Parameters of the system for experimental characterization.

2.5.

Determination of Center Wavelength

The location of the scan line, $y_{\text{scan}}$, is shown in Fig. 4. To find the location of the CWL, first, a region of interest (ROI) is drawn over the filter image. $y_{\text{scan}}$ was chosen to be in the center of the LVF. It was initially assumed that the horizontal position of the peak transmittance may depend on the chosen location of $y_{\text{scan}}$, due to fabrication tolerances. However, Fig. 5 shows otherwise, which shows the variation of the detector counts in the 2D pixel grid. Figure 5 shows that the location of peak transmission along $x$ does not change depending on the chosen scan line location, $y_{\text{scan}}$. Another method to determine the location of peak transmittance would be to scan every pixel in the 2D ROI for every CWL measurement. Compared to scanning a single line, this method is less computationally efficient, making it less suited for the real-time characterization application. The number of detector counts is evaluated within two bounds, where $x_{lb}$ and $x_{ub}$ are the lower and upper bounds, respectively. In Fig. 4(a), $x$ and $y$ refer to the numbering of the image’s horizontal and vertical pixels, respectively. The detector counts must be evaluated within bounds because the detector counts increase dramatically outside the range of pixel numbers that do not cover the LVF. This increase can be seen in Fig. 4(b), where beyond $x_{lb}$ and $x_{ub}$ the colors that correspond to the detector counts appear brighter. Figure 4(b) shows results for the $3.60\text{-}\mu \mathrm{m}$ filter with labeled spectral properties. The locations of the CWL, FWHM, and peak transmittance are shown. The CWL is a function of the scan line pixel number; the two variables vary according to the equation,

Fig. 4

The method for determining the CWL of the filter from raw pixel data. (a) The IR image used to obtain CWL. The scan line is located at $y=y_{\text{scan}}$. The upper and lower bounds of the horizontal pixel number are $x_{lb}$ and $x_{ub}$, respectively. (b) The spectral results of the $3.60\text{-}\mu \mathrm{m}$ bandpass filter. Spectral properties shown include the locations of the CWL, peak transmittance, intensity half-maximums, and bounds of the filter.

Fig. 5

A three-dimensional plot showing the variation of the detector counts with the horizontal and vertical pixel numbers that correspond to the area of the IR image that the LVF occupies. As long as the scan line passes through the entire length of the LVF, the location of peak transmittance does not depend on the exact location of the scan line.

2.6.

Comparison to Other Spectroscopic Techniques

FTIR spectrometers that utilize MEMS are cost effective, portable and lightweight.^23–26 However, commercial MEMS-FTIR spectrometers typically possess a much shorter spectral bandwidth ($\sim 1.1$ to $2.5\mu \mathrm{m}$),²³ capture light from a single point (typically through an optical fiber), and integration with pre-assembled wide-field imaging systems can be challenging. Diffraction grating-based spectrometers provide high performance and present the gold standard for spectroscopic analysis in a myriad of systems. However, as with MEMS FTIR, their utilization is application- and system-specific due to some disadvantages in relation to LVF-based spectrometers. These include: an increased optical train (system volume) due to the need for sufficient distance needed for the specific diffraction order to cover the sensor width and potential need for aberration correction mirrors; low optical throughput due to the utilization of only one diffraction order and narrow entrance slit, which means only light from a single point or averaged over one-or-two dimensions, and need for greater alignment precision compared to an LVF and subsequently lower stability.²⁷

Our approach presents spectral resolution disadvantages compared to the aforementioned approaches. However, it presents an attractive cost effective filter characterization solution for pre-assembled wide-field spectral imaging applications. Notable advantages include: (1), the light throughput of our system is high compared to other spectral measurement systems. (2), our system can straightforwardly measure spatial-spectral information; using the GUI we can study the two-dimensional spectral information of the filter. (3), our system can also easily be made more compact (the LVF is only 3.5-mm wide and 15-mm long). In our system, it is also easy (relative to other techniques) to add an LVF on top of a sensor. Most importantly, for real-world wide field remote sensing using spectral imaging, it’s straightforward to insert an LVF to check performance instead of using multiple systems or coupling into a fiber for diffraction grating. Table 2 provides a comparison of different types of spectrometers. The system in this work has better spectral range than any of the systems already mentioned; however, it has comparatively worse spectral resolution; we aim to improve the spectral resolution of our system in future implementations.

Table 2

Comparison of spectral measurement techniques.

3.1.

Overview

The PCM tunable filters are based on thin-film ${\mathrm{Ge}}_2{\mathrm{Sb}}_2{\mathrm{Te}}_5$ and fabricated in a laboratory at NASA Langley Research Center; the fabrication process is described in detail in other studies.^10,11,28 The evaluation of tunable PCM-based filters may be automated in two aspects: phase switching and performance characterization. A MATLAB^®-based application was custom-developed to control the hardware systems and analyze incoming data. Figure 6 details the main loop that runs to switch and characterize the PCM-based filter. The source code for the MATLAB^® application is available at https://dbombara.github.io/automated-pcm-characterization, as well as a description of the experiment and setup. More details on the developed MATLAB^® application are provided in Appendix A1.

Fig. 6

Operation flowchart: The general commands and processes executed by the MATLAB^® GUI. The program runs in a loop until the cycle limit is reached but can be terminated by the user at any time.

Once the user begins the experiment, the filter automatically switches between the amorphous and crystalline states while its CWL is automatically calculated and saved for each cycle. The sequence of functions that automates the system is shown in Fig. 6. The GUI allows the user to conveniently control and monitor the experiment’s progress. At the start of the experiment, MATLAB^® establishes connections to each instrument. The initial IR image is then loaded, with the ROIs displayed on the screen. At this point, the user may redraw the ROIs, based on the exact position of the LVF in the view of the camera. The selected ROIs are displayed on the screen and remain in view while the experiment runs. Before firing a laser pulse, first, the desired state is evaluated. Typically, the desired state alternates on each cycle between amorphous and crystalline. The desired state of the PCM determines the intensity and duration of the fired laser pulse. A closed relay switch instructs the IR camera to save its image. The laser waveform is then plotted for that cycle. Then, the file of the IR image is imported and displayed in the GUI. An image mask is applied that allows the CWL to be calculated and plotted within two bounds on the horizontal axis. All data from that cycle are then saved to a file. The experiment runs until the cycle limit is reached, which can be preset by the user or determined by the GUI based on the data from each cycle. More details on these processes may be found below.

3.2.

Phase Switching

The switching of the PCM between amorphous and crystalline states is triggered by the heat of an incoming laser pulse. A Quantel Evergreen HP pulsed laser is chosen in this study to deliver up to 340 mJ of energy at 532-nm CWL. The pulses are delivered at a rate of up to 20 Hz. The wavelength and energy level were chosen based on our previous work on GeSbTe phase-change metasurface spectral filters.¹¹ The pulsed laser output is controlled with a function generator and is viewed with an oscilloscope; this conveniently enables the user to monitor the waveform. The laser consists of two components: the power supply and the “head,” which emits the pulsed laser beam. In this study, the oscilloscope (Tektronix MDO4024C) and function generator (Tektronix MDO4AFG) functionalities are combined into a single instrument (MDO stands for mixed-domain oscilloscope). The particular oscilloscope and function generator were chosen due to their high-level programming commands that are available in MATLAB^®.

The laser, function generator, and oscilloscope are controlled within the MATLAB^® user interface. From MATLAB^®’s Instrument Control Toolbox™, Quick Control Oscilloscope commands are employed to issue high-level commands to the instruments. This aspect of the GUI builds upon the Oscilloscope App from MathWorks^® and incorporates similar functionality.²⁹ Serial commands can also be issued directly to the laser to adjust the settings of the instruments, such as the power level.³⁰

3.3.

Performance Characterization

The performance characterization is conducted in a way similar to the characterization in Sec. 2. To automate the image acquisition process, a USB-controllable relay (8-channel 5-Amp ProXR Lite, Relay Pros) triggers the camera to record the IR images. In this study, the switch is “dry;” the closure of the relay switch draws no electrical current. The FLIR ResearchIR software saves the image in a proprietary file format known as ATS. Then, the FLIR File Reader Software Development Kit (SDK) is used to import the ATS file into MATLAB^®. Functions from the SDK then extract the image’s metadata and raw pixel information so that the image can be displayed in the GUI. Using the imported image, the CWL of the tunable filter is calculated in the manner explained in Sec. 2, using Eq. (6).

4.1.

Reliability Experiment

The reliability of the PCM-based filters will be evaluated using the MATLAB^® GUI in future experiments. Figure 7 shows the schematic of the reliability test that will utilize the MATLAB^® GUI. The setup is similar to that in Fig. 2, but with added components to facilitate the laser-based tuning of the PCM-based filter. Because the laser will heat the PCM-based filter during the phase change, the IR camera must wait for that heat to dissipate before taking an image. This is because the IR camera cannot differentiate between surface heating of the filter and blackbody radiation. During the reliability experiment, the laser firing and the IR image acquisition will be separated for a duration that allows the heat of the PCM to sufficiently dissipate. However, the experiment will still be automated, since the filter may be potentially be switched for ${10}^3$ to ${10}^6$ cycles until it fails; the moment of failure can be identified with our approach. Previous PCM reliability studies have demonstrated long operational longevity, with one example demonstrating $1.4\times {10}^8$ switching cycles before failure.³¹ However, PCM-based optical filters face challenges over their resistive-based PCM counterparts. In optical PCM-based devices, the entire area of the tunable film must undergo a complete and reversible phase transition.¹² However, resistive-based PCM-based devices do not require complete crystallization of the tunable film to switch from “high” to “low.”¹²

Fig. 7

The reliability test setup: The reliability test will rely on the MATLAB^® GUI for automated laser control and automated measurement of the filter’s CWL. The CWL will indicate the current state of the filter.

To be used in NASA’s Space and Science missions, the filter must be reliable, with any drop-off in performance as a function of the tuning cycle known in advance. For example, one potential application of the tunable filter would be the Space Launch System (SLS) for the Artemis-1 Mission. The filter could enable accurate temperature measurement of the rocket’s core booster stage via radiometric techniques because of its nanosecond-scale filter switching speeds and narrow bandwidth. If the error margin of the temperature measurement is reduced, the maximum temperatures from the rocket may be smaller than originally designed for, therefore, the heat shield may not need to be as massive. The tunable filter could also be utilized for chemical and gas sensing in NASA’s Science missions. If the mission time is one year, for example, the tunable filter must work reliably for that duration.

4.2.

Cost-Effective and Compact Tuning System

Despite the advantages of PCM-based filters, there is a need for decreased size, weight, power, and cost; this would enlarge the range of possible applications for the tunable filter. Figure 8 shows the design of a cost-effective and compact tuning system for PCM-based filters that have detectable phase transitions in the NIR waveband, albeit with typically an increased absorption. The cost of components for this system was <$400. The schematic is shown in Fig. 8(a), the photographs of the initial build are shown in Figs. 8(b) and 8(c), and the comparison of input and output signals are shown in Figs. 8(d) and 8(e). The system consists of two main subsystems: the phase-switching subsystem and the imaging subsystem.

Fig. 8

The cost-effective and compact tuning system. (a) The schematic of the proposed system. (b) A prototype and (c) the wired connections on the breadboard. (d) The system takes an input pulse wave from the Arduino microcontroller. (e) The output pulse waveform is amplified nearly 30 times compared to the input pulse, with minimal distortion.

6.1.

A.1 Instruments Tab

It is crucial to ensure all instruments are properly connected before beginning the experiment. For that reason, the Instruments tab, shown in Fig. 9(c), includes information to connect and operate the oscilloscope and function generator. Options such as the resource (the instrument identifier), driver, channel, voltage range, and acquisition time are selected by the user. The Instrument tab also contains buttons for connecting to the relay and laser power supply via serial connections.

6.2.

A.2 Measurement Mode and Stop Condition Tabs

The Measurement Mode tab [Fig. 9(d)] and Stop Condition tab [Fig. 9(f)] both allow the user to select options for terminating the experiment when a particular condition, or combination of conditions, is met. These tabs are especially useful for hours-long reliability tests. In the Stop Condition tab, the user can choose to stop the experiment when either the transmittance, CWL, or both properties fall outside a predetermined limit. Similarly, the experiment may stop after a particular number of occurrences outside that limit, or when the number of occurrences reaches a set percentage of tuning cycles completed thus far. In contrast, the conditions in the Measurement Mode tab—total cycles and total time—are calculated in the GUI itself without relying on the connected hardware.

6.3.

A.3 Pulse Settings Tab

The required energy and pulse duration may change depending on the particular PCM that is under test. In the Pulse Settings tab [Fig. 9(a)], the user can set the desired laser voltages (V), pulse widths (ns), and delays ($\mu \mathrm{s}$) between pulses. The GUI then calculates the duty cycles (%), frequencies (MHz), and periods ($\mu \mathrm{s}$) of the pulses for amorphization and crystallization. The specific values of the laser voltage, pulse width, and delay are determined experimentally and vary depending on the particular materials utilized in the tunable filter.

6.4.

A.4 Tabs for Monitoring Raw Data

The top-right quadrant [Fig. 10(b)] of the GUI window is for displaying “raw’” data: in general, data that come directly from instruments, not values that are computed as a result of the instruments’ data. In this quadrant, there are two tabs: the “Laser (Osc/FGen)” tab and the IR Image tab. As the names imply, the former plots the input to and output from the laser, as determined by the function generator and oscilloscope. The input signal to the laser differs from the output signal; the laser requires two pulses of 5-V amplitude and $10\text{-}\mu \mathrm{s}$ duration, with $170\mu \mathrm{s}$ between pulses. However, the output pulse from the laser head is only nanoseconds-long. The IR Image tab displays the IR image and ROI that are selected before the experiment. As the CWL of the PCM-based filter is tuned, the IR image will change based on the indicator on the LVF.

6.5.

A.5 Tabs for Testing and Performance Characterization

The Testing tab [Fig. 9(g)] contains buttons for the user to test the functionality of the GUI before running the experiment. The relays can be turned on and off with the click of a button. Sample files can also be imported, after which the IR image can be plotted, and the spectral properties can be calculated. The Files tab [Fig. 9(f)] allows the user to specify file and folder names from which data will be imported and to which the data will be saved.

The bottom-right quadrant [Fig. 9(e)] of the GUI window contains plots of values computed from the IR image above. The cycle number is plotted on each abscissa whereas the CWL, FWHM, and transmittance intensity are each plotted in separate tabs on the ordinate axes.

The MATLAB^® GUI has much utility in the current experimental system, but in future work, the GUI will be applied to (1) an experiment to investigate the PCM-based filter’s reliability and (2) a compact, and custom-designed circuit to optically switch the PCM-based filter and characterize its phase changes in the NIR waveband.