Characterisation and potential for reducing optical resonances in FTIR spectrometers of the Network for the Detection of Atmospheric Composition Change (NDACC)

Although optical components in Fourier transform i nfrared (FTIR) spectrometers are preferably wedged, in practice, infrared spectra typically suffer from the effects of optical resonances (“channeling”) affecting the retrieval of weakly absorbing gases. This study investigates the level of channeling of each FTIR spectrometer within the N twork for the Detection of Atmospheric Composition Change (NDACC) . Dedicated spectra were recorded by more than twen ty NDACC 35 FTIR spectrometers using a laboratory mid-infrared source and two detectors. In the InSb detector doma in (1900 – 5000 cm-1), we find that the amplitude of the most pronounced c hanneling frequency amounts to 0.1 to 2.0 ‰ of the sp ctral background level, with a mean of (0.68 ± 0.48) ‰ and a median of 0.60 ‰. In the HgCdTe detector domain (700 – 130 0 cm-1), we find https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c © Author(s) 2020. CC BY 4.0 License.


Introduction
Ground-based FTIR (Fourier transform infrared) spectroscopy is a widely used technique for measuring column abundances of a variety of trace gases in the atmosphere. Within the Network for the Detection of Atmospheric Composition Change 50 (NDACC), this technique is used at about twenty sites covering a wide range of geographical latitudes. The NDACC data are used to study short and long-term variability of the atmosphere as well as for satellite data validation (De Mazière et al., 2018).
For both applications, high data quality and station-to-station consistency are of utmost importance. Ground-based FTIR spectroscopy provides data of high quality (e.g. Schneider and Hase, 2008). However, several key instrumental characteristics need to be addressed. These parameters such as detector non-linearity (Abrams et al., 1994), instrumental line shape (ILS; 55 Hase et al., 1999), intensity fluctuations (Keppel-Aleks et al., 2007), precise solar tracking (Gisi et al., 2011), and sampling error (Messerschmidt et al., 2010;Dohe et al., 2013) have been studied in some detail and need to be taken into account.
In this paper, channeling -the presence of instrument-induced periodic oscillations of spectral transmission resulting from internal optical resonances -will be investigated and discussed. In the past, each site or each new spectrometer was tested for channeling individually. This paper describes a network-wide exercise for characterizing channeling in FTIR spectrometers. 60 Channeling is caused by interference of reflections of the incoming light at parallel transmitting surfaces of optical elements.
In practice, the resulting channeling amplitudes are less than 1% in signal. Thus, the retrieved data for species with strong absorption signatures, as for example ozone and many others, are less critically affected. However, the retrieved trace gas amounts of weak absorbers can be substantially disturbed. In such cases, channeling becomes an important component of the total error budget. 65 Recently, time series of column abundances of formaldehyde (HCHO) were retrieved from NDACC FTIR sites (Vigouroux et al., 2018(Vigouroux et al., , 2020. The studies of Vigouroux also includes an error characterisation of the HCHO product. Within the network, two retrieval codes are in use: SFIT4 and PROFFIT. While the retrieval codes were inter-compared and show consistent results (Hase et al., 2004), the assumed error budgets differ slightly. The stations using PROFFIT include an error contribution due to channeling while the stations using SFIT4 do not. The result is a larger total error for HCHO data retrieved with PROFFIT as 70 https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c Author(s) 2020. CC BY 4.0 License. compared to SFIT4 (Vigouroux et al., 2018). In the PROFFIT error calculation, a set of typical channeling frequencies and amplitudes is taken into account. More specifically, channeling amplitudes of 0.5 ‰ for four frequencies are assumed: 0.005, 0.2, 1.0, and 3.0 cm -1 . The resulting error contribution doubles the total error of HCHO columns amounts.
In order to make this assumption more robust and to quantify more carefully the differences from spectrometer to spectrometer, an exercise was performed to measure channeling frequencies and amplitudes of NDACC FTIR spectrometers. Since 75 atmospheric spectra are densely populated with absorption signatures interfering with the signal generated by channeling; the test was designed using spectra collected in a laboratory setting. Section 2 briefly describes the origin of channeling, Sect. 3 the setup of this exercise, and Sect. 4 shows the results followed by a discussion. Finally, to reduce the channeling amplitude, the investigation of a modified beam splitter design is presented in Sect. 5, and lastly, Sect. 6 gives the conclusions.

Spectral transmission of a Fabry-Perot cavity
In an FTIR spectrometer, the transmitted light passes through several optical components such as optical windows, optical filters and a beam splitter, typically comprised of a beam-splitting layer system deposited on a transparent substrate and a compensator. At the transmitting surfaces of these components, the optical beam is partially reflected. In the case of parallel surfaces, each pair of surfaces defines a cavity (Fig. 1a) in which multiple reflections occur. Due to interference of the reflected 85 light, a standing wave is created (Fig. 1b). This effect is called the Fabry-Perot or etalon effect or channeling. The optical length of the cavity defines the free spectral range ν(FSR) as with n refractive index and d thickness of the optical component (Hecht, 2017). θ is the angle between incoming light beam 90 and the normal of the optical surface (Fig. 1a). Equation (1) is used to assign the optical element responsible for a certain channeling frequency. Table 1 gives a few examples of ν(FSR) for optical materials commonly used in FTIR spectrometers. The Fabry-Perot etalons generated by these undesired parasitic effects naturally have rather low etendue, so the resulting spectral transmission is well described by assuming an harmonic oscillation.
In order to reduce or avoid channeling, optical components need to be wedged or installed with a large tilt. A large tilt is not 95 feasible in many cases. Thus, components are normally wedged, which requires a special design and limits compatibility with non-wedged devices. Furthermore, some components such as detector elements are not available as wedged versions (the partially transparent detector element can also act as optical cavity). Therefore, in practice it is challenging to design an FTIR spectrometer that is completely free of channeling.

Experimental setup
In atmospheric spectra, channeling can be difficult to see due to the presence of complex atmospheric signatures. Therefore, laboratory spectra are used for this exercise, recorded either with a mid-infrared globar or with a black body of at least 1000 °C temperature. Since these types of sources do not include a window, no additional channeling is added to the spectra. A 115 temperature of 1000 °C is required to record spectra with a sufficient signal-to-noise ratio in a reasonable amount of time.
Within NDACC, two detectors and the NDACC filter set are used. The NDACC filters have a wedge of 10 arc min and therefore, if properly oriented, do not cause channeling. Therefore, not all filters but both detectors were included in this exercise. More specifically, NDACC filter #3 (2400 to 3000 cm -1 spectral range) for the InSb detector and NDACC filter #6 (700 to 1300 cm -1 spectral range) for the HgCdTe detector were used. Some sites use filter #7 (700 to 1000 cm -1 spectral range) 120 and #8 (1000 to 1400 cm -1 spectral range) instead of filter #6. In this case, filter #7 was used for this exercise. Filter #3 was selected since this filter range is used for the retrieval of HCHO column abundances.
Multiple reflections within optical components such as optical windows or beam splitters typically show channeling frequencies of a few tenths of a wavenumber up to a few wavenumbers. In general, higher frequency channeling with wavenumbers below 0.1 cm -1 might occur when different optical components form the surfaces of the resulting cavity, e.g. in 125 the Bruker 120HR spectrometer the rim of the entrance field stop is part of a resonator of about 1 m length. However, this is seldom the case in an FTIR spectrometer and secondly, due to the high frequency, easily detectable even in atmospheric spectra.
In order to focus on channeling due to multiple reflections inside optical components and to achieve a very good signal-tonoise ratio, a spectral resolution of 0.05 cm -1 was chosen. This resolution allowed us to add thousand interferograms within a 130 few hours, thereby achieving signal-to-noise ratio that allowed channeling amplitudes to be detected and quantified on a per mille scale.

Analysis of channeling test spectra
To quantify channeling frequencies and their amplitudes, an FFT (Fast Fourier Transform) analysis of the spectra was conducted. First of all, a spectral interval was chosen with a nearly constant intensity: 950 to 1000 cm -1 for HgCdTe and 2550 135 to 2600 cm -1 for InSb spectra. This step was carried out using OPUS™, a software package from Bruker Optics to control FTIR spectrometers (Fig. 2a). Then, the background was normalized and a straight line was subtracted using Origin™ software ( Fig. 2b). Finally, an inverse FFT was conducted also with Origin™ software (Fig. 2c  In this section, the results are presented for more than twenty spectrometers. Table 2 provides the list of spectrometers included in this study. Please note that a few spectrometers do not include an HgCdTe detector: Garmisch, Karlsruhe, and Sodankylä. https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c Author(s) 2020. CC BY 4.0 License.  Tables 3 and 4 list the detected channeling frequencies and their amplitudes in spectra recorded with InSb and HgCdTe detectors, respectively. 160 Figure 3 shows the detected channeling frequencies and their amplitudes in InSb spectra analysed at about 2600 cm -1 . Most spectrometers show the expected channeling frequencies: about 0.9 cm -1 and 0.11 or 0.23 cm -1 . These frequencies are consistent with (i) the gap between beam splitter and compensator plate (0.9 cm -1 ), and (ii) the beam splitter substrate (0.23 cm -1 ; Table 1).

InSb detector domain
A frequency of 0.11 cm -1 corresponds to a resonator due to both substrates, the beam splitter and the compensator plate. 165 A few spectrometers show an additional channeling fringe with a frequency of about 3 cm -1 . This is due to the detector window that is often made of sapphire or calcium fluoride (CaF2). Also in Izaña, this channeling frequency was detected in 2018. In December 2018, the detector was exchanged because of decreasing sensitivity. The new detector (Izaña-2019) shows much less channeling. Detectors purchased in the 1990s sometimes had a detector window with insufficient wedge. Figure 4 shows the amplitude of the strongest channeling frequency of each spectrometer. In most cases, channeling caused 170 by the gap of the beam splitter is the most pronounced one. The amplitudes range from 0.1 to 2.0 ‰ with a mean of (0.68 +/-0.48) ‰ and a median of 0.60 ‰. These mean and median are consistent with the PROFFIT error estimate of 0.5 ‰ as used in Vigouroux et al. (2018). However, the channeling amplitude differs strongly from spectrometer to spectrometer and a few spectrometers show an amplitude of up to 2 ‰.    https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c Author(s) 2020. CC BY 4.0 License. Table 4 lists major channeling frequencies and their amplitudes in spectra recorded with an HgCdTe detector at about 1000 cm -1 . As for the InSb detector, most spectrometers show two dominant channeling frequencies: about 0.9 cm -1 and 0.1 or 0.2 cm -1 caused by the beam splitter (Table 1). Two spectrometers show an additional channeling frequency of 2.17 and 3.85 cm -1 , indicating that the wedge of the detector window is not sufficient in these cases. Figure 5 shows the amplitude of the strongest channeling frequency of each spectrometer. In most cases, channeling caused 195 by the gap of the beam splitter is the most pronounced one. The amplitudes range from 0.3 to 21 ‰ with a mean of (2.45 +/-4.50) ‰ and a median of 1.2 ‰. The amplitude is even larger as compared to the InSb domain. At several sites, a reduction of channeling amplitudes would be desirable in order to improve trace gas retrievals of species with weak signatures, in particular from HgCdTe spectra, e.g. of ClONO2, HNO3 or SF6.

HgCdTe detector domain 190
As for the InSb domain, channeling amplitudes differ strongly from spectrometer to spectrometer. Figure 6 shows spectra with 200 different levels of channeling of the same frequency (about 0.9 cm -1 ) demonstrating the need of increasing the wedge of the gap and for narrowing the tolerances of wedges in the manufacturing of the beam splitters. https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c Author(s) 2020. CC BY 4.0 License.

Investigation of a modified beam splitter design for reducing channeling 205
This test exercise has found that the channeling amplitude differs strongly from spectrometer to spectrometer. A few spectrometers (at Altzomoni, Izaña, Karlsruhe and Kiruna) use customer-specific beam splitters with an increased wedge of 1.75° for the air gap and 10 arc min for the CaF2 substrate and 8 arc min for the KBr substrate. Their channeling amplitudes are the lowest among all the spectrometers studied in this paper. Unfortunately, this type of beam splitter is not a standard device and is not compatible with standard beam splitters, as it requires a realignment of the interferometer. Namely due to its 210 incompatibility with far-infrared pellicle beam splitters, the manufacturer Bruker adheres to the standard design with lower substrate wedge. To avoid the need for strongly wedged substrates, a different approach is proposed here. We focus on the wedge of the gap between the beam splitter and the compensator plate. Since the largest channeling amplitude (at 0.9 cm -1 frequency) is caused by the air gap, an increased wedge of this gap has the potential to reduce channeling significantly. The typical air gap wedge 220 for the Bruker beam splitter is 0.5°. Different spacers with wedges of 0.5°, 1.27° and 2.2° have been manufactured by Bruker and tested. Figure 7 (upper panels) shows the resulting channeling test spectra recorded with an HgCdTe detector. Similar to most of the NDACC spectrometers, the spectrum of the 0.5° wedged beam splitter shows a pronounced channeling with an amplitude of 5.7 ‰. In contrast, the 1.27° and 2.2° wedged beam splitters are (nearly) free of channeling with an amplitude of 0.46 and of 0.87 ‰, respectively, that is close to the noise level of these spectra . Analysed in the 850 to 900 cm -1 spectral 225 range, the amplitude is 8.9, 3.3 and 0.6 ‰ for a wedge of 0.5°, 1.27° and 2.2°, respectively. For InSb spectra, the 0.9 cm -1 channeling generates amplitudes of 0.9, 0.45 and 0.19 ‰ for beam splitters with wedges of 0.5°, 1.27° and 2.2°, respectively.
To ensure compatibility between different beam splitters, the wedge should be limited to 0.8°. This design will be implemented in future Bruker HR spectrometers. Figure 7 (lower panels) presents test spectra with an air gap wedge of 0.5° and 0.8°. In the 850 to 900 cm -1 spectral range, even the slightly increased wedge reduces the channeling by nearly 50 % (from 10 ‰ to 6 ‰). 230 In the 950 to 1000 cm -1 range, however, the effect is smaller.
Moreover, this exercise demonstrates that a wedge of about 2° on the air gap eliminates channeling even without a larger wedge of the beam splitter substrate. However, such a spectrometer completely free of channeling would result in incompatibility with beam splitters having a smaller air gap wedge and therefore, the need to realign the spectrometer after a beam splitter exchange. 235

Conclusions
Firstly, this paper documents the channeling amplitudes for nearly all of the FTIR spectrometers used in NDACC. Such a systematic performance analysis is needed for improving the trace gas retrievals and for calculating complete error budgets.
Within NDACC, laboratory test spectra of about twenty spectrometers were recorded and analysed. The derived channeling amplitudes range from 0.1 to 2.0 ‰ and from 0.3 to 21 ‰ in the InSb and HgCdTe domains, respectively. These values are 240 not negligible when constructing the error budget of minor trace gases. A reduction of the channeling amplitudes is highly desirable for the analysis of gases like ClONO2, HNO3, HCHO, and SF6.
Secondly, this study shows the potential to reduce channeling in several spectrometers and to improve the homogeneity within the network. The channeling frequencies allow us to determine the responsible optical component. A few instruments show channeling with a frequency of a few wavenumbers due to insufficiently wedged detector windows. Switching the detector 245 window or, more easily, the entire detector including dewar and detector window, will help reduce channeling in these cases.
Finally, we found that most spectrometers show two dominant channeling frequencies with about 0.1 or 0.2 cm -1 and 0.9 cm -1 corresponding to beam splitter substrate and beam splitter air gap. In most cases, the channeling caused by the gap of the beam splitter is the leading one. The option of reducing this channeling contribution was investigated by adjusting the wedge angles on a test beam splitter. Increasing the wedge of this gap significantly reduces the channeling at 0.9 cm -1 and therefore, such a 250 https://doi.org/10.5194/amt-2020-316 Preprint. Discussion started: 4 September 2020 c Author(s) 2020. CC BY 4.0 License. beam splitter design offers the promise of further reducing channeling. Switching to this modified beam splitter design would contribute to further homogenization of the spectrometers operated within NDACC.
Data availability. Channeling test spectra used in this study are available on request from the corresponding author (thomas.blumenstock@kit.edu).