End-to-end ground calibration and in-flight performance of the FIREBall-2 instrument

The FIREBall-2 gondola has a height of $\sim 5$ meters hand a weight of $\sim 2.3$ metric tons (including the $500$-kg ballast)[8]. This payload is designed to reach an altitude of 35 km attached to a stratospheric balloon filled with $1.1$ million cubic meters of helium. Because the atmosphere absorbs most of the UV light coming from the cosmos, the FIREBall spectrograph has been optimized to operate in the narrow atmospheric transmission window around $2000$ Å where there is a dip in the atmospheric UV absorption above $35$ km[9]. The nominal UV band-pass for this window ($1990$ - $2130$ Å) transmits nearly half of the incident UV light and the residual continuum atmospheric background is comparable to or lower than the extragalactic UV background. Within this atmospheric window, there is a contaminating emission band between $2044$ - $2062$ Å from NO-$\delta$, which is emitted at altitudes well above that of the balloon, but the zodiacal contribution is negligible [10]. The typical atmospheric transmission is $50\%$ when the balloon is at float altitude and the pointing has an elevation of $50^{\circ}$. In the subsequent modeling, we have assumed a total sky background (extragalactic $+$ atmospheric) of $500$ continuum units¹¹1Continuum Units : $n_{photon}\cdot s^{-1}\cdot cm^{-2}\cdot sr^{-1}\cdot\mathring{A}^{-1}$.

The optical design of FIREBall-2 relies on a low-inertia $1.2$-meter sidereostat mounted on a controlled gimbal system that stabilizes the reflected beam in the gondola’s frame and directs it to a fixed $f/2.5$ paraboloid that in turn focuses it at the entrance of the $f/2.5$ instrument (see Figure 1). The object selection is achieved with a series of pre-installed precision slit mask systems that also feed the guider camera (see Figure 2). The detector is a Teledyne-e2v electron multiplying CCD (EMCCD) anti-reflection-coated and delta-doped by the Jet Propulsion Laboratory (JPL) [11] and Caltech.

The one-meter telescope shows very good performance at a cost compatible with suborbital programs achieved by using an innovative thin mirror/hierarchical-kinematic support and telescope pointing control.[12] The telescope is mounted on a low-mass gondola made of aluminum and carbon-fiber. This gondola provides four-axis guidance (a coarse azimuth control for the gondola to face the field, fine elevation and cross-elevation to target the exact center of the field despite gondola pendulum, and field rotation) over a range of $40$ - $70$ degrees in elevation. During the 2018 flight, in degraded conditions (see Section 3.1), the pointing stability performance on each axis was $\sim 0.5$ arcsec.

Inside the tank, The narrow field of view of the parabola alone is extended to $\sim 800$ arcmin² by a Schwartzchild-type two-mirror field corrector $f/2.5-f/2.5$ (Figure 1). The corrector is designed to introduce a field curvature that cancels the native curvature of the Schmidt-based spectrograph to produce a flat focal surface at the detector. The slits masks of the spectograph have been fabricated at the same curvature and are mounted on a rotation stage. The spectrograph itself consists of identical $f/2.5$ Schmidt collimator and camera mirrors, the aspheric correction being provided by the $2400$ g/mm reflection grating obtained by double replication of a deformable matrix using the Lemaitre method [13]. Flat folding mirrors are used to reduce the volume of the vacuum tank in which all the optics and detector are encased. The Invar spectrograph structure and the Zerodur (TM) low-expansion glass make the instrument robust against temperature changes. A layout of the overall payload and the instrument are respectively presented in Figure 1.

In addition to the science objectives described in the introduction, the flight of the FIREBall-2 instrument also validated many processes and state-of-the-art hardware components (UV EMCCD [14], CNES pointing system [15], cooling system, large UV aspherical grating [13]).

The FIREBall-2 instrument complexity and the intricacy of its sub-components require a hierarchic adjustment and calibration procedure. An especially demanding specification is the absolute guidance to 1 arcsec ($\sim 1/6$ slit width) during the $\sim 2$ hours of observation per field. This goal is made still more challenging by the slight mask-to-mask shape differences (see section 2.1) that require a mask-dependent calibration.

The optical fine adjustment procedure has been split into two major blocks, each containing several alignment procedures.

FIREBAll-2 was launched on a $1.1$-million-cubic-meters balloon from the Fort Sumner Columbia Scientific Balloon Facility on September $22^{\textrm{nd}}$ at 10:20 AM local MDT. The flight sequence is detailed in Hamden et al. (2020). [8]

After $\sim 3$ hours of ascent and $\sim 3$ hours at float altitude, the balloon began to lose altitude around sunset, most likely due to a hole in the balloon. Because of this anomaly, the balloon dropped to its minimal mission requirement altitude after 45 minutes of science acquisition instead of the $>6$h required. Balloon altitude is important for science as between 30 and 40 km altitude, each km of altitude loss represents a reduction of transmission of $\sim 5\%$.

In addition, weather constraints resulted in a launch the day before a full moon (moon $94\%$ percent full during the flight). The deshape of the balloon due to the balloon’s deflation also impacted science data by concentrating a specular reflection of the moonlight over the balloon directly at the entrance of the payload. This boosted our background levels by $>1000$ times what was expected. The resulting scattered light produced $\sim 1$ event/pix/50s frame background, nearly an order of magnitude too high to efficiently perform photon-counting.[20]

Aside from these considerations, almost all subsystems performed nominally. Some flight anomalies are detailed in Hamden et al. (2020) [8] as well as the four fields observed. In order to focus on the instrument performance, we will only detail the reduction and analysis of the first field (DEEP 2: RA=$253.06$, DEC=$34.97$, $\theta=20^{\circ}$) when the instrument was at the nominal altitude.

In order to analyze flight data in real time, we built a python FIREBall dedicated DS9 extension similar to the DS9 [21] Quick Look plugin (https://people.lam.fr/picouet.vincent/pyds9plugin). Indeed, DS9 provides easy communication with external analysis tasks and is extensible via XPA. This extension was designed for real time processing in order to support in-flight decisions as well as assembly, integration and tests (AIT). During the flight it was mostly used for cosmic ray (CR) removal, background subtraction, help with object recognition, and stacking.

This extension proved its worth during the AIT and the flight, so the same DS9-oriented method was used to design the entire reduction pipeline to process data after the flight. This extension is divided into three main parts, each of which fills a specific need:

AIT: Supports assembly, integration, and test. This include throughfocus and through-slit visualization and analysis, XY calibration analysis, radial and encircled energy profiles, etc.

Flight reduction pipeline: Contains every part of the flight data processing: Cosmic ray removal, over-scan correction, background subtraction, photon-counting processes, and image/object stacking.

Detector: Supports EMCCD characterization prior to and after the flight: gain computation from fluctuation analysis or histogram fitting, computation of the different noise sources ( clock-induced charges (CIC), read-out noise, etc.) and other tests: auto-correlation, column-line correlation, smearing analysis, etc.

The ground-based calibration for absolute pointing and XY positioning described in Section 3 assumes a stability better than 1 arcsec of the guider-to-mask relative positioning until the end of the flight, in an environment $40^{\circ}$C colder than during the calibration.

The continua of the three UV-bright GALEX stars in the science field were detected, with a SNR $\sim 18$ for the brightest one ($M_{FUV}$=17.8). This proves that the targets were positioned close to their nominal position, one of the toughest challenges of FIREBall-2. Flight data do not allow us to access the residual decentering between the source and the exact slit center. Photometric measurements are performed in Section 4.5.

In addition to its impact on the throughput and flight duration shown in the previous section, the loss of altitude also impacted the balloon’s (and consequentially the gondola’s) motion because of the varying wind direction with respect to altitude. This intensified the different gondola excitation modes, requiring better damping and causing a decrease in the pointing performance. Despite that, the pointing performance stayed well within the technical specification [8].

The in-flight raw guider error signal is found to be $0.7^{\prime\prime}$ RMS per axis, an upper limit to the guidance error, as it includes the guider signal noise and the field rotation guiding error of $\sim 0.3^{\prime\prime}$ RMS. The latter number derives from a $1.5$ arcmin field rotation 1D guidance error, and a typical distance guiding stars-targets of 12 arcminutes[15].

An in-flight spatial resolution of $7^{\prime\prime}$ has been derived from a gaussian fit along the spatial direction of the GALEX brightest star continuum in the first field. We are currently working at LAM on improving the focus of the spectrograph in order to obtain an in-flight $5^{\prime\prime}$ FWHM angular resolution all over the field for the upcoming flight in 2021. This is of course a very important gain in resolution and it would also decrease all the sources of noise in counts/PSF/hour as the PSF gets sharper (this is taken into account in Table 3 and detailed in Section 5.2).

FIREBall-2 incorporates a state-of-the-art $2$Kx$1$K, $13$-$\mu m$-pixels electron-multiplying CCD 201 from Teledyne-e2v processed at JPL to increase its UV sensitivity with a delta-doping procedure, which yields a $100\%$ internal quantum efficiency device limited only by the reflection losses, and a multi-layer antireflection coating to further enhance external QE[11]. The stochastic nature of the amplification process in EMCCDs generates an Excess Noise Factor (ENF=$\sqrt{2}$ at high gain) that can be removed under certain conditions ($Flux<<1$e^-/pix/exp and $G_{amp}>10\times RN$) by applying a thresholding process known as photon-counting [20]. The different characteristics of the detector are summarized in Table 2. These state-of-the-art devices require special parameter tuning in order to take advantage of their performance, as well as specific data reduction such as cosmic ray removal or photon-counting processing.

The present model of FIREBall-2 detector has been calibrated on the ground and shows a $55\%$ peak efficiency in the nominal UV instrument bandpass and $>50\%$ in the entire bandpass. This value is compared to flight data in Section 4.5. The detector performance is detailed further in Kyne et al. (2020)[14].

Relevant to the present analysis, the detector shows sub-optimal charge transfer efficiency (CTE) due to the unusually low temperature selected to minimize the dark current. We think this can be improved by further optimization of the controller/parameter tradeoff for the next flight.

At flight temperature, this average CTE induces a charge smearing that yields a performance similar to an electron multiplying (EM) gain three times lower. This increases the noise contribution and causes event loss when attempting to perform photon-counting thresholding. This effect combined with the high background prevented us from using photon-counting. Smearing inversion algorithms can be applied, but there is an intrinsic limitation as these inversions always generate noise amplification [14].

The total efficiency was computed based on the flux measured on Bright Star 1 using a fitting-based photometric algorithm. The measured continuum was found to be $30\%$ lower than expected from the ground-based calibration and atmospheric model, which is well within statistical fluctuations derived from applying the same photometric algorithm at random locations. The main contributors to this uncertainty are: target centering, photometric measurement, atmospheric transmission model, focus, instrument throughput and detector calibration. In addition, the photometric measurement on BS1 shows a decreasing flux during the observation, which result probably from defocus or target decentering rather than atmospheric model bias.
No emission line has been detected at a statistically significant level, neither by searching at the predicted individual emission line locations, nor by stacking the target galaxies or a subset of the brightest ones. In order to measure the sensitivity limit for the total Ly$\alpha$ emission line of a target, we have performed profile fitting of a 2D gaussian matched to the PSF at random locations in the image. We avoided areas where target signal can be expected. The derived $5\sigma$ sensitivity limit for an unresolved $z\sim 0.7$ source for the available 30-min observation is $\sim 1.5\times 10^{43}$erg/sec.This limit is found consistent with a rather simple but quite complete analytical calculation of the total sensitivity of the instrument. As can be seen in Figure 7 of Augustin et al. (2019) [23] FIREBall-2 is presently lacking a factor of 3 in sensitivity to constrain the models at $z\sim 0.7$.

For the diffuse CGM Ly$\alpha$ emission, the instrument-limiting flux $\sim 7^{\prime\prime}$ away from a single object center, using one resolution element on each side of the central one, translates to a surface brightness detection limit of $~{}1.7\times 10^{-16}$ergs/cm²/s/arcsec² (721,000 LU).

Limiting flux performance is, of course, lower than what was expected without the spurious background but can be used as a starting point for evaluating the instrument’s ultimate sensitivity.

Table 3 describes the different sources of noise as observed in 2018 and expected in 2021. It shows that the major offender to the sensitivity is the high sky level, mostly due to the moon-scattered light, as discussed in Section 4.1. Because of the very short turnover wind period, launch opportunities are scarce and can occur very close to the full moon, as it happened in 2018. A launch far from full moon phase will always be preferable, but stray light mitigation has to be sufficient so that even in the worst scenario, sky noise would not impact FIREBall-2’s performance [24].

Our simple SNR model shows that to achieve this goal, it is necessary to cut stray light to the UV detector by a factor larger than 200. In this case, the sky background will be less than $3$ counts/PSF/hour. Refinements of the stray light/off-axis reflections analysis are on-going but it is already clear that three baffling stages need to be set up: a gondola top cap combined with a lower baffle will prevent direct view of the balloon from the spectrograph entrance (direct view of the moon was already baffled). It will be aided by an improved system of baffles internal to the tank. A more stringent moon-angular-distance limit and extensive testing of the baffling will ensure in-specs background in the worst expected conditions.

We experienced some additional background noise which we distinguish from the detector dark noise (Table 3). Indeed, during the flight we observed a signal, when the shutter is closed, an order of magnitude higher than the ground-based dark at the same temperature. Even though it is not totally understood, we think it is not an increased detector dark but rather a consequence of either Cerenkov emission coming mostly from the back of the FF2 mirror, a guider camera leak or (less likely) a shutter failure that would have let $\sim 1$cm² of light through. We have decided to reduce any possible Cerenkov contribution by masking all the mirrors’ backs and are currently addressing the other possible causes. We have also added more comprehensive baffling around the immediate detector environment. It is important to stress that the level of additional background (shutter closed) might come from other unknown causes as it is the very first flight of these-state-of-the-art devices (large number of very low energy cosmic rays or longer than estimated smeared cosmic ray tails).

In addition, a new detector system will be used for the next flight. It will equipped with a new version of the detector controller (Nüvü v3), developed specifically by Nüvü Cameras for these EMCCD detectors.[14] The expected lower read noise provided by this controller combined with a slightly higher temperature should make the photon-counting algorithm more efficient. Data acquired with this new setup will allow further optimization of the temperature/smearing/gain parameters in order to increase the SNR.

The improvements in the spectrograph alignment and guidance algorithm also promise an image quality increase from $7^{\prime\prime}$ to $\sim 5-6^{\prime\prime}$, which improves our chances of disentangling CGM signal from the galaxy signal and beginning to spatially-resolve CGM emission.

In order to analyze the advantage of using photon-counting, we compare here two scenarios for the next flight: with [Counting] or without [Analog] thresholding photon-counting process. We assume a 200-factor sky reduction and that the additional background level (as defined in Section 5.1) is mitigated by a factor of 10. To compute the SNR, we use one resolution element on each side of the central one for 144 exposures, amounting to 2 hours observing time⁶⁶6The effect of cosmic ray mentioned in Section 3.3.1 is taken into account. It is important to remind that when stacking over a long period of time, the SNR does not increase as fast as $\sqrt{t}$. We accounted for that with the same $1.4$ factor observed between $50$sec and $30$min..

The noise contributions for these two cases are reported in Table 3 so that they can be compared with 2018 values.

In both scenarios, FIREBall-2 appears to be photon starved: the biggest offender (highest noise contribution) is the shot noise. In the photon-counting scenario, stacking 10 targets with similar galaxy properties/morphologies would decrease the shot noise to a level comparable to the other noise sources. This would make the major noise contributors (CIC, sky, background, shot noise) very similar, prohibiting any further gain.

As expected in photon-counting mode, the read noise contribution becomes much smaller and part of the CIC noise disappears as the serial CIC is only amplified by an average gain of 200 instead of 1400 e^-/e^-. The Noise was computed assuming $30\%$ count loss and only $85\%$ read noise reduction because of smearing when applying the photon-counting thresholding process. These values are representative of the expected improvement of the charge transfer efficiency (CTE) with respect to the 2018 values.

The fact that, in both scenarios, the detector dark noise is not dominant might allow for a higher detector temperature, which would improve its CTE and facilitate photon-counting.

For these two 2021 cases, the respective limiting $5\sigma$ total Ly$\alpha$ emissions are $\sim 4.3\times 10^{41}$ and $5.3\times 10^{41}$erg/sec for a $\sim 5^{\prime\prime}$ unresolved $z\sim 0.7$ target. For the diffuse CGM Ly$\alpha$ emission limit and using the same definition as in Section 4 (with $5^{\prime\prime}$), the surface brightness detection limits are respectively $~{}1.4\times 10^{-17}$ and $1.8\times 10^{-17}$ ergs/cm²/s/arcsec² (61,000 LU and 75,000 LU). As we can see, these limiting fluxes for the two scenarios are very close despite very different total noise. This is due to the fact that the high photon-counting threshold used to mitigate the read noise and the serial CIC, coupled with smearing, generates event loss and therefore lowers the total efficiency by $\sim 30\%$.

In order to compare FIREBall-2’s performance with that of the first generation of the instrument, we gathered in Table 4 the characteristics and performance of FIREBall-1 and 2 for their respective flights. In Table 4, the sensitivity limit is computed for 2 hours acquisition on an extended $30^{\prime\prime}$ object such as a bright Quasi Stellar Object QSO.

For total flux (unresolved source), according to GALEX observations and simulations (Figure 7 of Augustin et al. (2019) [23]), the $\sim 41.7$ $log(L_{Ly\alpha})$ limit we find for both scenarios in Section 5.2 would be constraining by an order of magnitude for the Ly$\alpha$ integrated flux at $z\sim 0.7$. Stacking 10 objects per flight with similar galaxy properties/morphologies, using a very accurate continuum subtraction, should decrease this limit to $\sim 41.1$.

Concerning diffuse CGM emission, observations at higher redshift with MUSE [4] revealed some Ly$\alpha$ haloes at $z\sim 3$ reaching $4^{\prime\prime}$ ($\sim 30$ kpc) from the galaxy at their limiting surface brightness of $\sim 10^{-19}$ erg s^-1cm² arcsec^-2.⁷⁷7The discussion in Sullivan et al. (2019)[3] suggests that average $Ly\alpha$ surface brightness in the vicinity of QSOs might decrease with time (study from $z\sim 3$ to $z\sim 2$). With the assumed angular resolution of $5^{\prime\prime}$ for the 2021 flight, the first FIREBall-2 resolution element after the center is already at $\sim 35$ kpc at $z\sim 0.7$. The same physical distance at $z\sim 3$ ( $4.4^{\prime\prime}$) is just beyond Wisotzki et al. (2015)[4] published radius.

For a spectrally unresolved object, the favorable $\sim\times 30$ cosmological emission line surface brightness dimming at $z\sim 3$ with respect to $0.7$ makes the FIREBall-2 single-object limit translate to $\sim 14.7\times 10^{-19}$ erg s^-1cm² arcsec^-2 at $z\sim 3$, a value five times brighter than MUSE’s [4]. Stacking 25 or more galaxies [23] is thus expected to provide a detection surface brightness limit at $z\sim 0.7$ comparable to that of MUSE at $z\sim 3$ for individual galaxies. FIREBall detection of extended emission would thus only be possible in presence of large spectral offsets due to strong winds. Because of this, quasars, which have a brighter and more extended diffuse Ly$\alpha$ emission, are an even more interesting candidate for the 2021 flight [2].