An Integral Field Unit for the Binospec Spectrograph

Binospec (Fabricant et al., 2019) has been in operation at the f/5 focus of the 6.5m MMT telescope since early 2018. Binospec was designed for rapid survey spectroscopy and imaging at wavelengths between 360 and 1000nm with twin 8^′ (wide) by 15^′ (slit length) fields of view. Binospec was designed to use exchangeable aperture masks but its long slit length (totaling 1800^′′ or $\sim$300mm also lends itself to use with an optical fiber integral field unit (IFU). For operation with Binospec the main challenge is designing and assembling a compact IFU that fits within the confines of the aperture mask frames $\sim$260 by 290mm and $\sim$25mm thick to fit within a single aperture mask slot in the existing mask exchange mechanism. The other principal design goal is to use a hexagonal lenslet array to provide continuous spatial coverage with minimal gaps. Allington-Smith et al. (2002) describe an early example of such an IFU for the Gemini GMOS spectrographs. Prieto et al. (2000) describe a similar IFU for VIMOS on ESO’s VLT and Schmoll et al. (2004) an IFU for IMACS at Magellan. There are many examples of IFUs with dedicated spectrographs, see e.g. the introduction in Roth et al. (2023). Designing an IFU to coexist with multiple operating modes of an imaging spectrograph is a specialized problem due to space constraints.

The 12^′′ x 16^′′ IFU region may be completely filled by an extended source so we provide an array of 80 sky fibers offset by $\sim$50mm or $\sim$300^′′ from the center of the IFU field. The fibers in the IFU and sky fiber array are oriented parallel to the aperture mask frame for compactness and view the sky through small fold mirrors. Prisms at the output end of the optical fibers direct light into the spectrograph as a pseudo slit at the correct angle and focal position to reproduce direct illumination from the MMT’s f/5 optics and the concave focal surface (Fabricant et al., 2004). The sky and IFU fibers are type Polymicro FBP150170195 with 150$\mu$m cores, corresponding to 0.90^′′ at the input of Binospec.

The basic design of the Binospec IFU assembly is shown in Figure 1. Science target light enters at the location labeled “Science fiber pickoff mirror”, and passes through the magnifier optics and lenslet array. These are bonded to the fibers on the back side, shown in orange in Figure 1 and in the photo in Figure 2. Fibers are then routed in bundles of 20 to one of the 90-degree prisms arranged linearly along the length of the mask frame to form the pseudo slit (32 science plus 4 sky on each side). Off-target light for sky background subtraction enters at its own pickoff mirror, but passes through without additional focusing optics. Sky fibers are randomly divided into bundles of 8, which are evenly dispersed in between the science bundles on the output side. Figures 3 and 4 show the final assembled unit, back and front, respectively.

After all fibers were routed from the lenslet array to the fiber block prisms, but before installing covers, the light path through the fibers was tested. A light source was scanned along the back spectrograph-facing slits of the assembly, and a video camera was set up to record the light emerging on the science lenslet side. This allowed us to verify and record the mapping between lenslet position in the focal plane and the fiber output position on the CCD side. This mapping is essential for proper reconstruction of a data cube from the reduced data of the individual fibers. We identified 4 dead fibers which passed no light, and found 3 fibers which were swapped from their expected ordering on the CCD side. (See Figure 7).

After final assembly and shipment to the telescope, the fit of the IFU into Binospec was carefully tested, first by manually sliding the unit into an open slot in the Binospec mask elevator to verify clearances. After slight modifications, we verified that Binospec’s mask arm could grab the IFU unit and precisely slide it into place. Once that mechanical operation was verified, we were ready to proceed with science commissioning.

We created a dedicated version of the Binospec mask design software, BinoMask, to help potential users visualize the IFU placement over their target of interest, and verify suitable guide stars, WFS stars, and placement of the sky fiber bundle.¹¹1https://scheduler.mmto.arizona.edu/BinoMask/index.php?ifu=true

At the telescope, queue observers align the IFU using a technique similar to that used for long slit or mask observations. First, the telescope operator slews to the IFU target. Then, the observer executes a command that offsets the telescope to place the target within the IFU field of view. Expected guide star positions are calculated for the offset position, and then final alignment is achieved by applying pointing offsets to place the detected guide stars at the pre-calculated coordinates. Thanks to Binospec’s stiff and well-calibrated guider mechanism and optics, this procedure achieves alignment better than 0.1^′′, far better than required for the IFU.

The dimensions of the guide star cutout regions in the IFU mask frame are stored in the Binospec database (Kansky et al., 2019). These dimensions are slightly different than those used in Longslit or Imaging mode, and so we must store and use the proper boundaries when selecting available guide stars for any given instrument pointing.

The bobserve instrument control software has special templates for IFU observations, which require somewhat longer exposure times for calibration frames. Also, because of the offset of the science IFU block with respect to the center of the mask assembly, there is a shift of the central wavelength compared to long-slit and MOS observations, which is taken into account while selecting a suitable arc line for the active flexure control system. This shift varies with the selected grating and nominal central wavelength. For PIs and observers this is completely transparent but managed automatically by the OnDeck staging and the automatic queue of bobserve allowing to take series of images and calibrations if necessary.

Currently, the guider software allows for dithering offsets up-to $\pm$7.5 arcsec in both directions and enables us to observe small mosaics up-to 2$\times$2 FoVs (30$\times$24 arcsec on the sky) without re-acquiring guide stars. The dithering pattern and amount is set by the observer in OnDeck or the automatic queue. In the future, we plan on extending the mosaicing mode to handle larger offsets along the long side of the IFU FoV, which is aligned with the direction of guide star boxes in the IFU mask assembly. In principle, we should be able cover the sky area of up-to 240$\times$24 arcsec without re-acquiring guide stars to map extended targets (e.g. edge-on galaxies or long filaments in Hii regions or supernova remnants.

The commissioning and science verification of Binospec-IFU was performed in December 2023. The commissioning tasks included:

• •

  Confirm the focus position and the calibration of central wavelength versus grating rotation. We obtained arc and internal flat field spectra and found that there is no significant focus offset between the focus for IFU fibers and MOS aperture masks. The central wavelength calibration offsets agree with the Binospec optical model predictions within two pixels.

• •

  Confirm the field orientation and offsets of the IFU relative to the guiders and measure the IFU angular scale. We observed a four-star asterism at low Galactic latitude and confirmed the spatial scale to be within 1% of 0.6^′′/lenslet. The location of the science IFU FoV with respect to the guider probes was established to better than 0.1^′′.

• •

  Determine the IFU fiber illumination correction and line-spread function. We obtained day-time sky flats using sunlight entering through the partially open rear MMT dome shutters. We compared the sky and internal flats and found that the illumination corrections for all three Binospec gratings and central wavelengths are consistent. From the same sky flats we determined the spectral line spread function as a function of wavelength and spatial position by fitting a high-resolution Solar spectrum to spectra in each fiber. We use a Gauss-Hermite parametrization with 3-rd and 4-th order Hermite polynomials $h3$ and $h4$.

• •

  Test the data reduction and image reconstruction algorithms. We observed NGC 2392 (catalog ), a planetary nebula with a bright central star and compared reconstructed images in several optical emission lines (Oiii, H$\alpha$ and others) to smoothed narrow-band images obtained with the Hubble Space Telescope. A cutout from the raw image of NGC-2392 (catalog ) is shown in Figure 8. The image shows the individual fiber traces for two 20-fiber science blocks, as well as a single 8-fiber sky block. Nebular emission lines are clearly visible in the science fibers, distinct from the sky emission lines visible in the sky block.

Science verification targets were added to the Binospec queue, interspersed with long-slit, MOS, and imaging observations. The main goals of science verification were:

• •

  Measure the Binospec-IFU throughput in a wide range of grating/central wavelength combinations. We observed several spectrophotometric standard stars in photometric conditions, and compared the observed count rates to absolute spectrophotometry in the literature.

• •

  Assess the instrument performance for bright emission-line targets by observing two Seyfert galaxies with radio jets.

Data reduction recipes for the Binospec IFU are integrated into the official Binospec data reduction pipeline (Kansky et al., 2019) maintained by the Telescope Data Center at the Center for Astrophysics (https://bitbucket.org/chil_sai/binospec/wiki/Home). Some of the IFU-specific algorithms in the pipeline are based on those from the IFU data reduction package developed for the Multi-Pupil Fiber Spectrograph (MPFS) at the Russian 6m telescope described in Chilingarian et al. (2007b). This package was enhanced to reduce VIMOS-IFU (Chilingarian et al., 2009) and Potsdam Multi-Aperture Spectrograph (Chilingarian & Bergond, 2010) data. For the Binospec IFU we modified the spectral tracing and extraction routines. The pipeline computes and propagates variance frames from photon statistics to provide realistic error estimates. We determine relative illumination corrections and matched spectral line spread functions for the sky IFU sky fibers (also used in the GMOS-IFU (Allington-Smith et al., 2002), MPFS (Afanasiev et al., 2001), and SCORPIO2-IFU (Afanasiev et al., 2018))

Overall, the data reduction steps in the IFU pipeline are similar to those in the multiple slitlet reduction excepting that IFU flat fielding is performed after spectral extraction, not before as with slitlets. The reduction steps are:

• •

  CCD overscan removal: we apply a non-linearity correction and correct for the different gains of the four amplifiers in each Binospec CCD.

• •

  Geometrical distortion mapping: in each Binospec channel we cross-correlate the spatial profile in seven segments at 32 wavelength positions with a reference profile extracted at the central wavelength. We use a 3$\times$3 order 2D-polynomial to fit the distortion map.

• •

  Fiber profile tracing and scattered light subtraction: this step is carried out in two iterations. First, the fiber traces are identified using a peak detection algorithm, and then their shapes and positions are fit by 3rd to 6th order Gauss-Hermite functions plus an additive offset. We start by fitting the first fiber in each sky or science prism block and subtracting the best-fitting function from the flat field profile. Then we fit the subsequent fiber and also subtract it from the profile, then refit the current fiber in a wider window. This approach allows us to better handle cross-talk between neighboring fiber traces, which overlap at 5–7% of the peak intensity. From the result of the first iteration of tracing we determine the gaps between fiber blocks and build the scattered light model, which we then subtract from all 2D frames (science and calibrations). Then we proceed to the second iteration which is identical to the first one but the Gauss-Hermite functions now are fitted without the additive term. Here we also assume a smooth behavior of Hermite coefficients as a function of both wavelength and position on the pseudo-slit and fix them in the fitting procedure leaving free only width and central position of each trace peak.

• •

  Spectral extraction: we use scattered light subtracted frames and the normalized Gauss-Hermite profiles of spectral traces obtained at the previous step to extract the signal at each position along the wavelength using optimal extraction, i.e. applying the weights that correspond to the profile shape. We do not extract individual fibers within an extraction window as is commonly done. Instead, we extract the signal from all the fibers within a fiber block of 8 (sky) or 20 (science) fibers using a constrained linear inversion to force the positivity of an extracted signal from each fiber. This step is implemented by reducing a constrained linear fit to a positively defined quadratic programming problem. We use NaN (not a number) to mask the flux values in case a given trace has more than 70% of pixels affected by cosmic ray hits or CCD imperfections. Our approach automatically handles the crosstalk between neighboring fibers assuming that we derived accurate trace profiles at the previous data reduction step. Through the rest of the pipeline reduction, the sky and science blocks of extracted spectra are treated very similarly to MOS slitlets but without applying a distortion correction because the extracted stacked spectra are rectified by construction.

• •

  Flat field correction: we use an internal incandescent lamp + LED flat field source for field illumination and vignetting correction. Pixel-to-pixel sensitivity variations are negligible after the extraction that averages five pixels at each wavelength position. We normalize the flat field shape to that derived in the central part of the SN1 Binospec channel to cross-calibrate the two Binospec channels. We apply the sky flat illumination correction.

• •

  Wavelength calibration: we treat each fiber block as a MOS slitlet in the original Binospec pipeline. We use the Binospec optical model to calculate an initial wavelength solution accurate to two CCD pixels. We then identify Ar lines in calibration exposures or airglow lines in the observed spectrum (where possible) or a combination of the two. We fit a preliminary wavelength solution using a low-order two-dimensional polynomial (Order 3 to 4 in the wavelength direction and order 2 to 3 in the spatial direction). We slightly adjust the zero-point of the wavelength solution in each fiber within a block by cross correlating a linearized calibration spectrum with a reference spectrum from the center of the pseudo-slit. This procedure accounts for the small fiber offsets of individual fibers. We typically achieve a wavelength solution better than 1/20th of a pixel RMS.

• •

  Secondary illumination correction: we measure fluxes in one or more airglow lines (e.g. [Oi] $\lambda=5577$ Å) by fitting the lines with Gaussian-Hermite profiles and computing a secondary fiber-to-fiber illumination correction assuming that the airglow line flux is constant across the IFU field of view. When several lines are used, we can correct wavelength dependent corrections. However, we find that these corrections are $<$1.5%. We compute the barycentric correction and apply it to the wavelength solution.

• •

  Sky background subtraction: we use a variant of the sky subtraction technique proposed by Kelson (2003) modified for MOS and fiber-lens IFU data reduction to use a global sky background model as described in Chilingarian et al. (2015). The sky model is computed with extracted spectra prior to the wavelength linearization. We use a $b$-spline parametrization of an oversampled sky spectrum assembled from all sky background fibers and evaluate the model at each wavelength and pseudo-slit position. The low-order Legendre polynomial terms are added to the $b$-spline parametrization along the pseudo-slit to handle smooth variations of the sky background originating from line-spread-function variations and residual flat-fielding imperfections. This approach allows one to eliminate an extra re-sampling step and, therefore, yields much better emission line sky subtraction. We apply a partial parametric deconvolution to the sky model to account for the $\sim$10% FWHM LSF difference between sky background and science fibers. This is done by convolving the signal with the kernel in the form: $[-a,1+a+b,b];0\leq a,b\ll 1$ where $a$ and $b$ are of the order of 0.05–0.1 (the values were determined empirically for each spectral setup). We reach sky subtraction close to the Poisson statistical limit.

• •

  Flux calibration and telluric correction: If a spectroscopic standard star is observed, the pipeline performs absolute spectrophotometric calibration and calculates the instrument throughput. By comparing our pipeline products against flux calibrated spectra from large surveys such as Sloan Digital Sky Survey, we estimate that our absolute flux calibration quality is $\sim$3% in photometric conditions when the standard star is within 30 deg of the target. Any a0v non-variable star with low Galactic foreground extinction can be used for less precise flux calibration. We can run a postprocessing code to correct for telluric absorptions. We fit a grid of atmospheric transmission models at the MMT elevation with a star or galaxy spectrum. The approach is described in Borisov et al. (2023) and corrects better than 2% in spectral regions with $<50$% telluric absorption (the entire spectral range at Binospec resolution).

The final data products are presented in the fits format either in the form of linearized flux-calibrated stacked spectra with a table for individual fibers or as a regularly gridded oversampled 3-dimensional data cube. The fits World Coordinate System (WCS) keywords are provided for the calibration of both wavelength and sky coordinates. We encourage scientific users to use stacked spectra for scientific analysis while the data cubes are useful for data quality assessment in quick look tools, e.g. SAOImage ds9 (Joye & Mandel, 2003).

The commissioning of the Binospec IFU significantly expands the utility of this workhorse spectrograph. Although the total throughput of the integrated IFU is $\sim$50% lower than a wide Binospec slit ($5^{\prime\prime}$), the throughput is comparable to a 1^′′ slit in good seeing and considerably higher in poor seeing. The throughput of the Binospec IFU compares favorably to many operating IFUs as discussed in Section 7. The effective resolution with the IFU is 25% higher than with a 1^′′ slit as described in Section 8. The Binospec IFU is particularly useful for measuring the kinematics and stellar population of galaxies compatible with its 12^′′ by 16^′′ format. A fully functioning pipeline is available to produce complete processed datacubes.