PRAXIS: Low-Background NIR Spectrograph

• PRAXIS is a fibre-fed near-infrared spectrograph that uses fibre Bragg grating (FBG) suppression to remove atmospheric OH lines before dispersion, lowering scattered light.
• It employs photonic lanterns and a modular, cryogenic design with fixed-format optics to boost throughput and minimize both detector dark current and thermal emission.
• By integrating optimized IFU sampling and low background technology, PRAXIS is predicted to achieve significant SNR gains over its predecessor, GNOSIS.

PRAXIS is a second-generation, fibre-fed near-infrared spectrograph designed specifically to exploit fibre Bragg grating (FBG) suppression of the atmospheric OH airglow lines while remaining sky-noise limited after suppression. Conceived as the successor to GNOSIS, it couples pre-dispersion OH filtering with a very low background spectrograph architecture, emphasizing a cryogenic fibre slit, low detector dark current, low thermal self-emission, fixed-format optics, and high throughput in the J and H bands (Horton et al., 2013, Content et al., 2014).

1. Scientific motivation and observational regime

Ground-based near-infrared spectroscopy is dominated by bright, narrow OH airglow emission lines from the upper atmosphere. In conventional spectrographs these lines not only dominate where they fall spectrally, but are also scattered by the spectrograph optics into the interline regions, raising the apparent background across the bandpass. PRAXIS is built around the premise that suppressing the OH lines before dispersion removes both the direct line flux and the spectrograph-scattered wings that would otherwise contaminate the interline continuum (Horton et al., 2013).

This pre-dispersion logic is the key physical distinction between FBG suppression and post-dispersion masking or subtraction. If the OH lines never enter the spectrograph, the spectrograph cannot scatter them into adjacent wavelengths. The 2014 design paper further notes that an astrophysical line close to an OH line can become visible provided the suppression is applied at higher resolution than the spectrograph and the astrophysical line is separated from the OH line by more than a little over half the suppression notch width (Content et al., 2014).

PRAXIS was proposed because existing near-infrared spectrographs were designed for the much brighter unsuppressed sky. Once the OH lines are removed, detector dark current and instrument thermal self-emission can dominate unless the entire instrument is redesigned for very low background and high throughput. GNOSIS had already demonstrated on-sky OH suppression, but it did not deliver a net sensitivity gain: in practice it “broke even,” with throughput losses roughly compensated by the background reduction (Content et al., 2014).

2. OH-suppression technology and the GNOSIS legacy

The suppression system uses aperiodic FBGs written into single-mode fibres. In this application, the gratings are engineered as multi-notch filters that reflect the OH line wavelengths while transmitting the interline light with high efficiency. The demonstrated performance cited for the FBG technology is more than 100 notches, resolution around $R \approx 10{,}000$, notch depths exceeding $30\,\mathrm{dB}$, Butterworth-like $n=5$ notch profiles, and interline throughput above 90% (Horton et al., 2013).

Because seeing-limited telescope beams do not couple efficiently into single-mode fibres, PRAXIS uses photonic lanterns to convert a multimode astronomical input into an array of single-mode channels, each containing the FBGs, and then reconvert back to multimode output. This lets the OH-suppression filters operate on normal telescope-fed fibres. In the GNOSIS-style implementation, each science fibre is split into 19 single-mode fibres; the later PRAXIS design also describes a planned 55-core multicore-fibre implementation for larger field of view per fibre (Content et al., 2014).

GNOSIS provided the immediate empirical justification for PRAXIS. Its background measurements showed that detector dark current dominated the observed background below about $1.67\,\mu\mathrm{m}$, while instrument thermal background dominated above about $1.67\,\mu\mathrm{m}$. The IRIS2 PACE HAWAII-1 HgCdTe detector had a measured dark current of $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$, and the warm external fibre slit and relay optics introduced substantial thermal background. PRAXIS was therefore framed as the instrument that closes the loop opened by GNOSIS: OH suppression alone is not enough unless the downstream spectrograph is exceptionally clean (Horton et al., 2013).

3. Optical architecture and modular system design

The full PRAXIS concept is modular. It consists of a spectrograph, a short fibre bundle to an OH suppression unit, the OH suppression unit itself, a long fibre bundle from the telescope focal station, an integral field unit (IFU), and a fore-optics package. This arrangement allows the heavy cryogenic spectrograph to remain at a gravity-invariant location while only the lighter fore-optics and IFU are mounted at the focus, and it was intended to let PRAXIS operate as a visitor instrument on any large telescope (Horton et al., 2013).

The spectrograph proper is bench-mounted, fixed-format, fully cryogenic, fibre-fed, and based on volume phase holographic (VPH) gratings. The 2013 concept was designed around $50\,\mu\mathrm{m}$ core fibres with output focal ratios as fast as $f/3$, corresponding to a maximum field of view per fibre of $0.87''$ on a 3.9 m telescope and $0.43''$ on an 8 m telescope. The baseline operating wavelength range was $30\,\mathrm{dB}$0–$30\,\mathrm{dB}$1, with an explicit J-band upgrade path; the later detailed design optimized the main science mode for $30\,\mathrm{dB}$2–$30\,\mathrm{dB}$3 and added an alternate $30\,\mathrm{dB}$4–$30\,\mathrm{dB}$5 J-band mode by changing the grating and refocusing the spectrograph (Horton et al., 2013, Content et al., 2014).

The concept study described a compact transmissive design chosen to minimize size and cost: because the target resolution was modest and the wavelength range limited, the required beam size was under 50 mm, the collimator and camera focal lengths could each be below 150 mm, and the entire spectrograph could be only about 300 mm in size. In the later realization, the spectrograph had a 19-fibre slit, a four-element collimator, swappable H- and J-band gratings, and a three-element camera, with all optical surfaces spherical. The detailed design adopted a spectral resolution of $30\,\mathrm{dB}$6, sampled at 2 pixels per FWHM resolution element, and used a $30\,\mathrm{dB}$7 HAWAII-2RG detector with $30\,\mathrm{dB}$8 pixels (Content et al., 2014).

At the telescope focal plane, the fore-optics reimage the telescope focal plane onto an IFU and include field and pupil stops to suppress scattered light. The later instrument description specifies 19 hexagonal microlenses glued to corresponding fibre ends, with seven central OH-suppressed science lenslets and twelve outer reference/acquisition fibres on a 250 $30\,\mathrm{dB}$9 pitch. The earlier concept had emphasized an exchangeable microlens-based IFU, based on a $n=5$0-pitch double-microlens telecentric design, precisely because photonic-lantern acceptance depends on lantern format (Horton et al., 2013, Content et al., 2014).

4. Background control: dark current, thermal emission, and detector footprint

A central design principle of PRAXIS is that after OH suppression the instrument must remain sky-noise limited. The technical drivers are dark current, thermal background, and throughput. On the detector side, the concept paper contrasted older InSb and first-generation PACE HgCdTe arrays with newer molecular beam epitaxy HgCdTe devices, noting measured dark currents of $n=5$1 for $n=5$2 cutoff devices and stating that, with sufficient cooling, $n=5$3 should be possible for $n=5$4 cutoff arrays. The proposed detector was therefore a single MBE HgCdTe array with at least $n=5$5 pixels, intended to achieve about an order-of-magnitude reduction in dark current relative to the detector used with GNOSIS (Horton et al., 2013).

The spectral format was also optimized to reduce dark current per resolution element. The concept study stated three requirements: the resolving power should be no higher than the science demands; the pixel scale should critically sample the fibre image, with a target image diameter of about 2.5 pixels; and the number of pixels in the spatial direction should be minimized. The specific concept value was 2.8 pixels assuming $n=5$6 pixels. The later $n=5$7 design retained the same underlying principle of minimizing the detector footprint per spectrum (Horton et al., 2013).

Thermal-background control is the most pervasive engineering theme of PRAXIS. The decisive architectural change relative to GNOSIS was moving the fibre slit inside the cryogenic spectrograph dewar. This reduces or eliminates thermal emission from the slit itself, removes the need for a dewar window in that part of the optical path, and reduces the number of warm optical surfaces. The instrument was therefore designed as fully cryogenic, bench mounted, and intended to operate at a gravity-invariant location with no cryogenic mechanisms; only the fibre vacuum feedthrough and detector electrical connections penetrate the cryostat (Horton et al., 2013).

The later design quantified the thermal model by treating each component as a greybody with temperature $n=5$8 and emissivity $n=5$9, with photon background

$1.67\,\mu\mathrm{m}$0

where $1.67\,\mu\mathrm{m}$1 is in $1.67\,\mu\mathrm{m}$2, $1.67\,\mu\mathrm{m}$3 is in $1.67\,\mu\mathrm{m}$4, and $1.67\,\mu\mathrm{m}$5 is in K. The design goal was that the thermal self-emission of the instrument should not exceed that of the telescope itself. The fore-optics cold stop and tube operate at about $1.67\,\mu\mathrm{m}$6, the spectrograph optics enclosure at $1.67\,\mu\mathrm{m}$7, the spectrograph optics are cooled to at least 200 K, and the GNOSIS grating unit was to be used at $1.67\,\mu\mathrm{m}$8, with a colder multicore unit planned. The detector-side baffling also includes a cold filter that cuts wavelengths longer than $1.67\,\mu\mathrm{m}$9, and the spectrograph window material was changed to ZnSe because the original Schott glass would have emitted thermally at ambient temperature (Content et al., 2014).

5. Throughput optimization and photonic-lantern design constraints

Throughput was treated as a co-equal design driver because the suppressed interline signal is intrinsically faint. The surface-relief grism in IRIS2 had an H-band peak diffraction efficiency of about 40%, while VPH gratings were expected to provide more than twice that. PRAXIS therefore adopted VPH gratings, optimized coatings, low-incident-angle operation, and low-absorption glasses over a restricted wavelength interval rather than broad all-purpose coverage (Horton et al., 2013, Content et al., 2014).

The étendue per fibre is constrained by the photonic lantern. The concept paper gives the approximation

$1.67\,\mu\mathrm{m}$0

where $1.67\,\mu\mathrm{m}$1 is the number of single-mode fibres in the lantern, $1.67\,\mu\mathrm{m}$2 is the wavelength, and $1.67\,\mu\mathrm{m}$3 is the telescope diameter. Using GNOSIS as an example, with $1.67\,\mu\mathrm{m}$4, $1.67\,\mu\mathrm{m}$5 over $1.67\,\mu\mathrm{m}$6–$1.67\,\mu\mathrm{m}$7, and $1.67\,\mu\mathrm{m}$8, the maximum field of view per fibre is about $1.67\,\mu\mathrm{m}$9, which the authors note is smaller than ideal for typical seeing. They further point out that 61-core lanterns had already been demonstrated, giving $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$0 larger $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$1 per fibre (Horton et al., 2013).

Laboratory measurements of photonic lantern behaviour were then folded directly into the PRAXIS performance model. Those measurements showed that lantern throughput as a function of input angle is neither Gaussian nor flat-topped like a conventional highly multimoded fibre. The empirical lantern model proved particularly important in choosing the on-sky field of view per fibre and in modelling thermal background. In the PRAXIS case study, sky counts rise with field of view per microlens but then level off because lantern throughput rolls off at larger incident angles; compact-source throughput has a true maximum and then declines as the useful étendue saturates. The adopted design choices were $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$2 per microlens for 19-core lanterns and $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$3 for the planned 55-core multicore-fibre lanterns (Horton et al., 2014).

The same end-to-end study assumed $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$4, a sky surface brightness of $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$5, and a single 1800 s exposure. Under those assumptions, predicted sky SNR per resolution element reached roughly $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$6 for the 19-core PRAXIS configuration and $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$7 for the 55-core configuration, whereas a GNOSIS-like system would not exceed about $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$8, and at the GNOSIS-used field of view of $0.015\ \mathrm{e^-\,s^{-1}\,pixel^{-1}}$9 would achieve only about $50\,\mu\mathrm{m}$0 in a single exposure. The lantern model also informed the thermal-emissivity budget and supported the decision to cool the OH suppression unit (Horton et al., 2014).

6. Predicted performance, astrophotonic role, and development trajectory

The later PRAXIS paper presents the instrument as the transition from proof-of-concept OH suppression to practical sensitivity gain. Its main quantitative claim is that the gain in interline sky-background SNR relative to GNOSIS may be as high as 9 when PRAXIS uses the existing GNOSIS OH suppression unit, and as high as 17 when it uses the planned multicore-fibre unit. These gains arise not from stronger OH suppression alone, but from the system-level combination of higher throughput, lower detector noise, better-optimized IFU sampling, and much lower thermal background (Content et al., 2014).

The observing mode is intentionally fixed-format rather than broadly configurable. The concept paper presents this not as a general-purpose limitation but as a deliberate optimization: because the spectrograph only needs to support OH-suppressed J/H spectroscopy at moderate resolution, it avoids the losses and complexity associated with multiple camera modes, moving mechanisms, broad cross-dispersion formats, or very large optics. This specialization also made full cryogenic implementation feasible within a compact package (Horton et al., 2013).

PRAXIS was also positioned as an astrophotonics testbed. Its modularity was designed to allow different suppression units and front ends, including future multicore-fibre and integrated-optics lanterns, and the authors explicitly state that PRAXIS could serve as a platform for other astrophotonic technologies. In the 2012 status reported by the concept paper, funding was available for the spectrograph and fibre components, prototype work on vacuum feedthroughs, the cryogenic slit, and fibre connectors was planned to begin that year, and the goal was completion and commissioning in 2013. By 2014, the instrument was described as using the GNOSIS OH suppression unit at present, with a multicore-fibre Bragg grating upgrade intended soon afterward (Horton et al., 2013, Content et al., 2014).

In this sense, PRAXIS occupies a specific place in near-infrared instrumentation. It is not merely a spectrograph with an external suppression module, but a purpose-built low-background system whose architecture is defined by the suppressed-sky regime. The instrument’s significance lies in that systems-level recognition: once the OH lines are removed, dark current, thermal emission, photonic-lantern étendue, and detector footprint become first-order design variables rather than secondary details.