In this paper we show the suitability of a miniaturized tunable diode laser spectroscopy (TDLS)-based carbon-monoxide (CO) sensor for fire detection applications. The sensor utilizes a vertical-cavity surface-emitting laser (VCSEL) and inherent calibration scheme with reference gas filled in the photodetector housing. The fire-detection experiments are carried out under realistic conditions as described in the European standard EN54. The CO generation of all class C fires (according to EN54) could be well resolved. The cross-sensitivity to other substances was found to be very low: the maximum CO false response from cigarette smoke, hairspray and general aerosols reaches a low value of a few μL/L and only if the substance is directly applied into the sensor gas inlet. Therefore this sensor overcomes the disadvantage of high false alarm rate given by smoke detectors and is also in small size which is suitable for household and industrial applications. Hence, the VCSEL-based TDLS sensor is shown to have sufficient performance for fire-detection. It has advantages such as capability for fail-safe operation and, low cross-sensitivities as compared to existing point fire detector technology which is presently limited by these factors.
© 2013 Optical Society of America
Fire detection is a very important application in industrial and domestic environments for personal protection and minimizing damage. Up to now, fire detection is predominately done using temperature sensors or smoke detectors which sense particles by either a scattering or an ionization effect. Although these detectors can be very sensitive and are the most frequently used fire detectors, they have several disadvantages. The false alarm rate can be very high (up to 50 % of all alarms, or in some areas up to 99 % according to ), due to unwanted cross-sensitivities to particles or other substances that are not necessarily generated by fires, e.g., water vapor, hairspray, insects, or general dust. Hence, lowering cross sensitivities of existing fire detector technology could result in significant cost savings or enable more sensitive detection with reduced detection thresholds. Therefore, in the literature alternate sensing principles for fire detection are routinely explored. Besides image processing based fire detection  a new approach to fire detection is the use of gas sensors . Most often prominent combustion gases like carbon monoxide (CO), nitrogen oxides (NO, NO2), or carbon dioxide (CO2) (or any combination thereof) are targeted. Among the systems reported in the literature are fire detection using hybrid suspended gate field effect transistors , metal oxide sensors [3, 5], and laser spectroscopic sensors based on photoacoustic spectroscopy [6, 7] or laser absorption spectroscopy [1, 8–12].
Smoldering fires usually generate high fluxes of CO while fires with visible/open flames generate less, but still significant, amounts of CO and also certain levels of nitrogen oxides. CO2 is toxic only in very high concentrations and also present in ambient air in non-trace concentrations (∼ 380μL/L in outside air). It is generated in large amounts by human beings and animals (∼ 40 mL/L in exhalation), so a CO2-only based gas sensor fire detector would suffer from an increased false alarm rate or reduced fire detection sensitivity. As for the nitrogen oxides, carbon monoxide (CO) is a toxic gas and affects human health whenever it is present in ambient air in high concentrations. However, the levels of nitrogen oxides generated by fires are often significantly lower than the concentration level of the generated carbon-monoxide. Please note, that only some organic substances like proteins, but not hydrocarbons or cellulose/wood, contain nitrogen. Hence, nitrogen containing gases are generally only emitted in limited cases or low amounts by fires (since nitrogen in air is relatively inert). So, a reliable and selective CO gas-sensor based fire detector is an interesting approach for fire detection with possibly reduced false alarm rates that has the further advantage of directly detecting the most dangerous gas component released by fires. Furthermore, CO poisoning is the most frequent cause of unintentional death at home in the USA with a death toll of more than 400 persons per year . However, the CO sensor should not exhibit cross-sensitivities to typical non-fire generated gases to give an improvement over conventional technology. This especially includes the substances typically causing false-alarms with smoke detectors like hairspray, water vapor, dust or insects. In addition to that, there should not be any new cross sensitivities to frequently used chemical substances like alcohols, solvents, or cleaning agents.
Solid state or electro-chemical sensors are known to show aging effects or posses unwanted cross-sensitivities to other gases because they are often sensitive to a set of similar molecules (e.g., reducing gases). Spectroscopic gas sensors on the other hand are known to have the lowest possible cross-sensitivity to other gases due to the spectroscopic measurement and due to the characteristic spectral fingerprints of gases. Furthermore, tunable diode laser absorption spectroscopy (TDLAS) enables fail-safe operation of the sensor as it allows for self-monitoring due to the dynamic characteristic of the optical output signal during each single scan. The signature in the recorded signals corresponding to gas absorption lines is only obtained if laser, detector, and driver/receiver electronics are working correctly. This is because the absorption line signal is a very narrow and unique feature that can not be simulated by malfunction of any component. The measurement itself is inherently insensitive to attenuation in the optical path by contamination since the absorption features of the gas are spectrally very narrowband compared to the wideband spectrum of typical contamination. The attenuation can rise by several orders of magnitude until reliable measurements become impossible. Furthermore, contamination, complete blockage of the optical path, failure of the laser, failure of the detector, or failure of the corresponding electronic circuits can be detected by comparing the absolute signal voltage from the detector circuit with the voltage applied to the laser driver. These mentioned advantages predestine TDLAS for sensors in safety applications or those where real-time and/or in-situ measurement is required. Note, that no other gas sensing method combines all of these advantages.
However, their wide use is hindered by the complexity and the resulting high price of existing TDLAS sensors. This is partly because of use of alignment sensitive optical components such as multi-pass cells, optical fibers, reference cells, and/or expensive mid-ir optics (including sources and detectors) [1, 6–12]. Although the mid-infrared allows for significantly more sensitive gas detection, working in the mid-ir has disadvantages. Besides expensive optics like windows and lenses (silica is not transparent in the mid-ir), lasers and detectors are also significantly more expensive and both require powerful thermo-electric cooling for equivalent performance of near-ir components. Hence, if a laser spectroscopic detection has sufficient performance in the near-ir, the near-ir it is strongly preferred over the mid-ir. Furthermore, the spectroscopic sensors for fire detection published in the literature are usually not evaluated for cross sensitivities and/or performance under realistic conditions. Hence, in this paper, a TDLAS CO sensor based on near-ir components (including a vertical-cavity surface-emitting laser) and with only the absolute minimum of optical components (no multi-pass cells, alignment sensitive lenses, beam splitters, optical fibers, etc.) is investigated for its suitability for fire detection. It will be determined if the performance of such a simplified sensor is sufficient under realistic test conditions. Also, cross sensitivities to substances typically causing issues with conventional smoke detector or solid state gas sensor technology will be evaluated because low cross-sensitivity would constitute a main advantage over existing technology.
This article represents chapter 5.4 from the dissertation .
2. Tunable diode laser spectroscopy with wavelength modulation spectroscopy
The TDLAS method employs light generated with a tunable semiconductor laser to measure the transmission of a gas sample . By tuning the laser wavelength the transmission spectrum of the substance can be measured which then is examined to find gas parameters like concentration, pressure, and temperature. Since the gas absorption lines are very narrow features with a halfwidth at half maximum (HWHM) of 1 GHz to 2 GHz at atmospheric conditions, lasers with spectrally single-mode emission (linewidth below 100 MHz) are needed (e.g., edge-emitting distributed-feedback (DFB) lasers or vertical-cavity surface-emitting lasers (VCSELs)). VC-SELs have advantages like low power consumption, circular beam emission, and wide current tunability. The output power is lower than for DFB lasers but sufficient for spectroscopic applications in the NIR because the laser noise dominates over shot noise for detected light powers of several 10 μW .
To sensitively detect the gas absorption features a technique called wavelength modulation spectroscopy (WMS)  is applied. A block schematic of the basic sensor functions is shown in Fig. 1. In addition to the ramp for scanning the wavelength a small sinusoidal modulation is applied and the second harmonic component in the received detector signal is determined with a lock-in amplifier. This second harmonic spectrum is similar to the second derivative of the transmission function  and allows for easy identification of the gas absorption features and the relevant gas parameters because a detection of a small change on a large signal is avoided.
3. Sensor Design
The sensor used for the experiments is described in . The optical cell, schematically shown in Fig. 2(a), implements a folded optical geometry (two-pass) and realizes an absorption path length of 10 cm. This is sufficient to achieve a sensor resolution in the lower μL/L range. Furthermore, a reference gas is filled in the photodetector housing, so that additional optical components, e.g., separate reference cell, second detector, and beam splitter are spared. These design choices were made so that with miniaturization of the electronics, as, e.g., in , the sensor has the potential to be of the size as conventional smoke detectors. A photograph of the utilized sensor prototype including electronics for data processing is shown in Fig. 2(b). The sensor’s total power consumption is around 1.4 W. The total weight of the prototype as shown in Fig. 2(b) is 2.7 kg, while the electronics and optical amount to about 700 g with 2 kg for the gray box.
Conventionally, a separate reference cell containing the target gas is used for stabilization of the laser emission wavelength to the center of the target absorption line [1, 8, 9]. Using the inline wavelength stabilization method developed in  this additional complexity is avoided. Adjacent absorption lines of methane (CH4), which is filled inside the photodetector housing, are used for wavelength identification. This allows for determination of both the absolute emission wavelength as well as the linear and quadratic tuning coefficient of the laser. Note, this technique takes advantage of wide current tunable lasers like VCSELs [20–22] so that both the CO and CH4 absorption lines at 2.365 μm can be included in a single scan as shown in Fig. 3(b).
The WMS modulation frequency is fm = 6 kHz with a spectral scan repetition rate of R = 10 Hz. The wavelength modulation amplitude is set to a value of approximately three times the half-width of the CO line at ambient conditions because this is the optimum setting when the spectral baseline is also fitted . Each recorded spectrum is curve-fitted with theoretical reference spectra by the microprocessor-based electronics and a concentration value is computed. The Allan plot for qualification of sensor stability is shown in Fig. 3(a). A sensing resolution of 2 μL/L in 1 s is obtained which corresponds to 7 · 10−6 in fractional absorbance. Up to averaging times of approximately five minutes the sensor is white noise limited and has a resolution of 0.4 μL/L at 60 s averaging time. This averaging time was chosen for the fire detection experiments to minimize the sensor noise at an acceptable time resolution for fire detection.
4. Experimental Setup for Fire Detection
The fire detection experiments were carried out under the regulations of the European standard EN54 for fire detection (part 7 for smoke detectors  and part 15 and 26 (draft) for gas sensors [25,26]). The room dimensions in which fires are created is 6 m × 10 m × 3.8 m whereas the sensor gas inlet is mounted at the ceiling on a 3 m radius circle around the test fire in the center of the room as shown in the experimental setup schematic in Fig. 4. The fire test is ended when the transmission of the air in the room falls below a certain threshold, i.e., a certain level of smoke is present. This is then followed by a purge of the room with fresh outside air. To verify the measured CO concentrations, a commercial CO reference analyzer (HORIBA model PG-250) was used in parallel during this experiment. The measurements of this reference instrument have been corrected for the different gas sampling delay time of about 1 min.
There are several fire types named TF1-TF6, optionally with suffix B or C denoting the size of the fire. These types simulate different fire incidents such as burning of different substances (e.g. wood, ethanol, n-heptane) under different conditions (e.g., smoldering fires or fires with open flames) as listed in Tab. 1. Clearly, class B fires generate much less gases and are, hence, more difficult to detect. According to EN54, for single point sampling detectors class C fires (large), with no additional air circulation in the room, are relevant, so in this experiment all class C fire tests were carried out without additional air circulation. Also according to EN54, the smaller class B fires had additional air circulation to ensure better distribution of the combustion gases inside the room during the fire. The fire test starts either with ignition of the fire (TF1, TF5, TF6), power-on of the hotplate (TF2), or the inflammation of the hanging cotton wicks (TF3).
5. Experimental Results and Evaluation of Cross-Sensitivity
5.1. Fires with high level of CO generation
Due to the high volume of the room, the CO concentration stays below 120 μL/L for the wood fire and the pyrolysis (TF2-C and TF3-C), as shown in Fig. 5(a).
The latter is a very incomplete burning with no flame and very high CO generation. The pyrolysis probably generates more CO than the wood fire with an open flame (TF1), but due to missing air circulation it does not distribute very well, which may explain the lower measured CO concentration. The steep falling of the CO concentration is caused by the purge of the room which ends the fire. High deviations between the sensor and the reference instrument are observed for class C fires. This is attributed to the missing air circulation of the room and different mounting positions of both sensor gas inlets. During the design of the experiment such an observed strong inhomogeneity of the gas distribution was not expected. In future experiments sampling at the same, or closely located, positions should be carried out.
These large deviations were not observed for the smaller class B fires shown in Fig. 5(b), which had additional air circulation in the room and thus much better homogeneity in gas distribution than the larger fires discussed previously. Even if a detection threshold as high as the Threshold Limit Value of 25 μL/L  is employed, all fires would have been detected in the required time, i.e., before the reference smoke detectors signal the end of the fire test.
5.2. Fires with low CO generation (TF5,TF6)
In case of fires with open flames like n-heptane or ethanol (EtOH), much less CO is generated as Fig. 6 shows. The ethanol fire is also typically the most difficult one to detect using smoke detectors because of virtually no particle generation.
In this case, the reference analyzer also only measured 3 μL/L CO compared to 6 μL/L of this sensor. This may be due to different air sampling points, the sensor baseline, or both. Reliability could further be improved by combining the gas sensor with a temperature sensor, to also detect the significant amount of heat (ΔT > 10 K) generated by open fires. Note, that the measurement cell can be made robust against this heating as, e.g., demonstrated by application of this sensor in an exhaust gas pipe of a gas furnace .
5.3. Evaluation of cross-sensitivity
Solid-state based carbon-monoxide sensors usually have cross-sensitivities to many reducing gases. This also applies to electro-chemical instruments used in chemical analysis. Particle detectors have cross-sensitivities to all kinds of particles like hairspray, water-vapor, or dust, whereas TDLS-based sensors generally have excellent selectivity. Nevertheless, the accuracy of TDLS-based sensors can be affected in three ways:
- A loss in absolute light transmission results in an increase of sensor noise. This happens if a broad-band absorber, e.g., dirt or some absorbing gas phase molecule with a broad absorption spectrum is present. At 2.3 μm this is the case for many hydrocarbons like butane, heptane, etc.
- Presence of molecules that have a structured transmission function in the wavelength range of interest. This can only be the case for molecules with a low number of atoms. Molecules with high number of atoms usually have a very smooth and broad-band absorption spectrum and thus do not contribute to the second harmonic spectrum, which is approximately the second derivative of the transmission. At 2.3 μm possible interferers are H2O, NH3, C2H6 and other hydrocarbons [28, 29].
- Sensitivity to interference on the transmission (“fringes”). This is depending on the chemical composition of the gas sample and the cause for drift and deterministic measurement errors.
The sensor has no cross-sensitivity to H2O. Simulations have shown, that even with 100 % absolute water vapor concentration (which is even impossible to reach at temperatures below 100 °C), the spectral interference due to absorption by water vapor is below the sensor resolution of 0.4 μL/L. However, sensor noise may increase when the absolute transmission due to condensed water vapor is low. Since the absorption strength of NH3 is an order of magnitude lower than CO and the spectral overlap to CO is also low, it was not considered to be relevant for inclusion in the sensor spectrum model. If NH3 might be present in air at higher μL/L concentrations this has to be reconsidered. In this case the wavelength stabilization method has to be modified to cope with the possible presence of NH3 absorption lines in the wide scan. Since NH3 is a flammable gas it is not expected to be generated by fires with open flames in high concentrations. It may be created by pyrolysis of protein containing organic substances, e.g., meat and is also contained in cigarette smoke due to the additive carbamide. Note, that cellulose which is the main constituent of wood and cotton does not contain nitrogen and thus even pyrolysis of this can not generate NH3. In Fig. 7 the measured sensor response to aerosols  and hairspray is shown. Tests were carried out by spraying in front of the gas inlet of the sensor (but not directed to) (legend ”room”) and by directly spraying into the gas inlet (legend ”direct”). Both hairspray and aerosols (which classical fire detectors most likely would generate a false alarm for) contain a high concentrations of hydrocarbons which are expected to absorb in the wavelength region around 2.3 μm. It can be seen, that during spectrum scanning a rather high amount of light is broad-band absorbed (see Fig. 8), summing to an absorbance of 3.6 % to 7 %. However, due to the differentiating nature of second harmonic detection this broad-band absorption is almost invisible in the measured second harmonic spectrum in Fig. 8. Only a small fine structure in the order of 10−4 of the broad-band absorber is present around 2.365 μm to 2.3655 μm, which, however, creates the slight false CO concentration signal observed in Fig. 7. This problem can be solved by identifying the substance that causes the fine structured spectrum and include it as an additional component in the spectral model. Note, that the cross sensitivity to hairspray and aerosols is not caused by the particles but most likely the solvent these are contained in.
In this paper, a simplified laser spectroscopy based carbon-monoxide sensor is tested for fire detection under conditions described by the European standard EN54. The reduced complexity provides potential for the sensor to be of the size of conventional smoke detectors. Additionally, a separate reference cell or alignment sensitive multi-pass cell are avoided. It turns out that, using an averaging time of 60 s, all class C fires can be reliably detected. The sensor is thus well suited as a point-sampling fire detector. Although TDLS-based sensors typically have the lowest possible cross-sensitivities to other gases among all sensing principles, a cross-sensitivity analysis with critical substances (hydrocarbons) was performed. Furthermore, it has been shown experimentally that direct application of hairspray or aerosols into the sensor gas inlet gives false CO concentrations of only a few μL/L. This is a very low value, and might be further reducible by identifying the relevant substance and the inclusion of its spectrum in the sensor spectral model used for curve-fitting.
The smaller class B fires or the ethanol fire have also been detected successfully, but may require a higher sensing sensitivity for very reliable detection. This can be achieved by a) lowering the sensor baseline by improving the spectral model or the laser source, b) increasing of the sensor optical path length to enlarge the gas absorbance, or c) combining the sensor with other sensing principles like temperature measurement or a smoke (particle) detection.
The authors gratefully acknowledge the financial support by the European Union (Project ’NE-MIS’, No. 031845) and the company VERTILAS for providing the 2.3 μm VCSEL diode.
References and links
1. S.-J. Chen, C. Hovde, K. A. Peterson, and A. Marshal, “Fire detection using smoke and gas sensors,” Fire Saf. J. 42, 507–515 (2007). [CrossRef]
2. T.-H. Chen, P.-H. Wu, and Y.-C. Chiou, “An early fire-detection method based on image processing,” in Proceedings of the 2004 International Conference on Image Processing(2004), pp. 1707–1710.
3. D. Gutmacher, U. Hoefer, and J. Wöllenstein, “Gas sensor technologies for fire detection,” Sens. Actuators B 175, 40–45 (2012). [CrossRef]
4. R. Pohle, E. Simon, R. Schneider, M. Fleischer, R. Sollacher, H. Gao, K. Müller, P. Jauch, M. Loepfe, H.-P. Frerichs, and C. Wilbertz, “Fire detection with low power FET gas sensors,” Sens. Actuators B 120, 669–672 (2007). [CrossRef]
5. U. Hoefer and D. Gutmacher, “Fire gas detection,” Procedia Eng. 47, 1446–1459 (2012). [CrossRef]
6. P. Stefanski, R. Lewicki, J. Tarka, Y. Ma, M. Jahjah, and F. K. Tittel, “Sensitive detection of carbon monoxide using a compact high power CW DFB-QCL based QEPAS sensor,” in Conference on Lasers and Electro-Optics 2013, paper JW2A.68.
7. A. A. Kosterev and F. K. Tittel, “Advanced quartz-enhanced photoacoustic trace gas sensor for early fire detection,” SAE Int. J. Aerosp. 1, 331–336 (2009).
8. R. Engelbrecht, “A compact NIR fiber-optic diode laser spectrometer for CO and CO2: analysis of observed 2f wavelength modulation spectroscopy line shapes,” Spectrochim. Acta, Part A 60, 3291–3298 (2004). [CrossRef]
9. D. S. Bomse, D. C. Hovde, S.-J. Chen, and J. A. Silver, “Early fire sensing using near-IR diode laser spectroscopy,” in “Diode Lasers and Applications in Atmospheric Sensing,”, Proc. SPIE 4817, 73–81 (2002). [CrossRef]
10. E. A. Fallows, T. G. Cleary, and J. H. Miller, “Development of a multiple gas analyzer using cavity ringdown spectroscopy for use in advanced fire detection,” Appl. Opt. 48, 695–703 (2009). [CrossRef] [PubMed]
11. J. L. Bradshaw, J. D. Bruno, K. M. Lascola, R. P. Leavitt, J. T. Pham, F. J. Towner, D. M. Sonnenfroh, and K. R. Parameswaran, “Small low-power consumption CO-sensor for post-fire cleanup aboard spacecraft,” Proc. SPIE 8032, 80320D (2011). [CrossRef]
12. B. Scherer, H. Hamid, J. Rosskopf, and S. Forouhar, “Compact spectroscopic sensor for air quality monitoring in spacecrafts,” Proc. SPIE 7945, 79450S (2011). [CrossRef]
13. Center for Disease Control and Prevention, “Carbon Monoxide Poisoning: Fact Sheet,” (CDC, 2013) http://www.cdc.gov/co/faqs.htm.
14. A. Hangauer, “Detection schemes, algorithms and device modeling for tunable diode laser absorption spectroscopy,” Ph.D. thesis, Technische Universität München (2013). In Selected Topics of Semiconductor Physics and Technology, Vol 164, ISBN 978-3-941650-64-0. http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:91-diss-20130412-1113114-0-7.
15. J. Reid and D. Labrie, “Second-harmonic detection with tunable diode lasers - comparison of experiment and theory,” Appl. Phys. B: Lasers Opt. 26, 203–210 (1981). [CrossRef]
16. A. Hangauer, J. Chen, R. Strzoda, M. Ortsiefer, and M.-C. Amann, “Wavelength modulation spectroscopy with a widely tunable InP-based 2.3 μm vertical-cavity surface-emitting laser,” Opt. Lett. 33, 1566–1568 (2008). [CrossRef] [PubMed]
17. J. Chen, A. Hangauer, R. Strzoda, and M.-C. Amann, “VCSEL-based calibration-free carbon monoxide sensor at 2.3 μm with in-line reference cell,” Appl. Phys. B: Lasers Opt. 102, 381–389 (2010). [CrossRef]
18. S. So, E. Jeng, C. Smith, D. Krueger, and G. Wysocki, “Next generation infrared sensor instrumentation: remote sensing and sensor networks using the openPHOTONS repository,” Proc. SPIE 7808, 780818 (2010). [CrossRef]
19. A. Hangauer, J. Chen, R. Strzoda, and M.-C. Amann, “Laser wavelength stabilization using gases with complex spectral fingerprint,” presented at International Conference on Field Laser Applications in Industry and Research, Garmisch-Partenkirchen, Germany, 2009.
20. M. Ortsiefer, G. Böhm, M. Grau, K. Windhorn, E. Rönneberg, J. Rosskopf, R. Shau, O. Dier, and M.-C. Amann, “Electrically pumped room temperature CW VCSELs with 2.3 μm emission wavelength,” Electron. Lett. 42, 640–641 (2006). [CrossRef]
21. G. Böhm, M. Grau, O. Dier, K. Windhorn, E. Rönneberg, J. Rosskopf, R. Shau, R. Meyer, M. Ortsiefer, and M.-C. Amann, “Growth of InAs-containing quantum wells for InP-based VCSELs emitting at 2.3 μm,” J. Cryst. Growth 301, 941–944 (2007). [CrossRef]
22. A. Bachmann, T. Lim, K. Kashani-Shirazi, O. Dier, C. Lauer, and M.-C. Amann, “Continuous-wave operation of electrically pumped GaSb-based vertical cavity surface emitting laser at 2.3 μm,” Electron. Lett. 44, 202–203 (2008). [CrossRef]
23. J. Chen, A. Hangauer, R. Strzoda, and M. C. Amann, “Tunable diode laser spectroscopy with optimum wavelength scanning,” Appl. Phys. B: Lasers Opt. 100, 331–339 (2010). [CrossRef]
24. European Standard, EN54-7: Fire detection and fire alarm systems - Part 7: Smoke detectors - point detectors using scattered light, transmitted light or ionization (Beuth, 2001).
25. European Standard, EN54-15: Fire detection and fire alarm systems - Part 15: Point detectors using a combination of detected fire phenomena (Beuth, 2006).
26. European Standard, EN54-26: Fire detection and fire alarm systems - Part 26: Point fire detectors using carbon monoxide sensors (Beuth, 2008).
27. American Conference of Governmental Industrial Hygienists, 2005 Threshold Limit Values and Biological Exposure Indices (American Conference of Governmental Industrial Hygienists, 2005).
28. L. Rothman, D. Jacquemart, A. Barbe, D. C. Benner, M. Birk, L. Brown, M. Carleer, C. Chackerian Jr., K. Chance, L. Coudert, V. Dana, V. Devi, J. M. Flaud, R. Gamache, A. Goldman, J.-M. Hartmann, K. Jucks, A. Maki, J.-Y. Mandin, S. Massie, J. Orphal, A. Perrin, C. Rinsland, M. Smith, J. Tennyson, R. Tolchenov, R. Toth, J. Vander Auwera, P. Varanasi, and G. Wagner, “The HITRAN 2004 molecular spectroscopic database,” J. Quant. Spectrosc. Radiat. Transf. 96, 139–204 (2005). [CrossRef]
29. S. W. Sharpe, T. J. Johnson, R. L. Sams, P. M. Chu, G. C. Rhoderick, and P. A. Johnson, “Gas-phase databases for quantitative infrared spectroscopy,” Appl. Spectrosc. 58, 1452–1461 (2004). [CrossRef] [PubMed]
30. An aerosol mixture for testing of smoke detectors was used. The exact contents are unknown.