The case for OH suppression at near-infrared wavelengths

1.5 MB·English

by  S. C. Ellis

Mon.Not.R.Astron.Soc.000,000–000 (0000) Printed3February2008 (MNLATEXstylefilev2.2)
The case for OH suppression at near-infrared wavelengths
S. C. Ellis⋆1 and J. Bland-Hawthorn2
1Anglo-Australian Observatory, P.O. Box 296, Epping, NSW 1710, Australia
2Institute of Astronomy, School of Physics, Universityof Sydney, NSW 2006, Australia
8
0
0
Accepted......Received.....
2

n
a
ABSTRACT
J
We calculate the advances in near-infrared astronomy made possible through the use
5
of fibre Bragg gratings to selectively remove hydroxyl emission lines from the night
2
sky spectrum. Fibre Bragg gratings should remove OH lines at high resolution (R =
10,000),with high suppression(30dB) whilst maintaining high throughput (≈90 per
]
h cent)betweenthe lines.Devices currentlyunder constructionshouldremove150lines
p in each of the J and H bands, effectively making the night sky surface brightness
- ≈ 4 magnitudes fainter. This background reduction is greater than the improvement
o
adapative optics makes over natural seeing; photonic OH suppression is at least as
r
t important as adaptive optics for the future of cosmology.
s
a We present a model of the NIR sky spectrum, and show that the interline con-
[ tinuum is very faint (≈ 80 photons s−1 m−2 arcsec−2 µm−1 on the ecliptic plane).

We show that OH suppression by high dispersion, i.e. ‘resolving out’ the skylines,
1
cannotobtain the requiredlevelof sensitivity to reachthe interline continuum due to
v
scattering of light. The OH lines must be suppressed prior to dispersion.
0
We have simulated observations employing fibre Bragg gratings of first light ob-
7
8 jects, highredshift galaxiesand cool,low-massstars.The simulations are ofcomplete
3 end-to-endsystemsfromobjectto detector.The resultsdemonstrate thatfibre Bragg
. grating OH suppression will significantly advance our knowledge in many areas of
1
astrophysics, and in particular will enable rest-frame ultra-violet observations of the
0
Universe at the time of first light and reionisation.
8
0
Key words: infrared:general–instrumentation:miscellaneous–atmospheric effects–
:
v cosmology:earlyUniverse
i
X
r
a
1 INTRODUCTION
1.8
Thefutureofcosmology is dependentin part on theability
to achieve deep observations at near-infrared (NIR) wave- H
1.6
lengths. As we strive to observe theearly Universe, our ob-
servations must be performed at increasingly longer wave-
lengths, as thediagnostic optical spectroscopic features be- 1.4
comemoreredshifted.Atveryhighredshifts,observationsin
theopticalbecomefutileduetotheredshiftedLymanlimit. J
Figure 1 shows the observed wavelengths of several useful 1.2
spectroscopic lines as a function of redshift, illustrating the
requirement for NIRobservations. 1 Y
TheneedforNIRobservationsiswelldocumented.Over
the last few years there have been several major reviews of Z
thecurrent status of ourknowledge of theUniverse, and in 0.8
0.1 1 10
particular of gaps in our understanding, serving as a start-
ingpointforstrategicplansforfutureobservations(e.g.the
AustralianDecadalPlanforAstronomy2006-20151,theUK Figure1.Theobservedwavelengthofseveralusefulspectrallines
as afunction of redshift.The shaded areas show the wavelength
coverageofstandardzJHfilters.
⋆ [email protected]
1 http://www.atnf.csiro.au/nca/DecadalPlan web.pdf
2 Ellis and Bland-Hawthorn
of the NIR background spectrum suggest that in principle
Table 1.Typicalsky-brightnessesatagoodobservingsite.
itshouldbepossibletoobtainverydeepobservationsinthe
Surfacebrightness/magarcsec−2 NIR.Firstly the OH spectrum comprises nearly all ( 99.9
Band Natural NoOHemission ≈
per cent from our model, see 2) of the J and H band
§
U 21.5 – background. The interline continuum, due mainly to zodi-
B 22.4 – acal scattered light, is expected to be very faint, e.g. the
V 21.7 – models of Kelsall et al. (1998) give a flux density of only
R 20.8 21.7 77photonss−1 m−2 arcsec−2 µm−1 ontheeclipticplane.
I 19.9 21.7 S≈econdly, the OH lines are intrinsically very narrow (Turn-
J 16.0 22.1
bull & Lowe 1983), meaning that large sections of the NIR
H 14.0 22.2
sky spectrum will be very dark if the OH lines can be effi-
ciently removed, thusenabling deep observations.
PPARCroadmap2,theUSNationalResearchCouncil’s‘As-
tronomy and Astrophysics in the New Millennium’ NRC
There are several methods which can, in principle, ex-
2001 and ‘Connecting Quarks to the Cosmos’ Committee
ploit the dark interline continuum for deep NIR observa-
On The Physics Of TheUniverse et al. 2003).
tions, including narrow-band filters which allow observa-
Thesereviewsprovidesimilarlistsofthe‘bigquestions’
tions of a very short bandpass between lines (e.g. DAzLE;
for thefutureof astronomy,showing aconsensusof opinion
Horton et al. 2004), or space-based telescopes (e.g. JWST;
at an international level. Some of the key issues they iden-
Gardner et al. 2006). Here we discuss ground-based OH
tifyare: Whatisthenatureofdarkenergyanddarkmatter?
suppression, by which we mean the selective removal (by
How and when do the first stars form? How and when did
meansoflowtransmission,reflectionetc.)ofverynarrowre-
the Universe reionise? How and when do galaxies assemble,
gionsofthespectrumcorrespondingtotheOHlines,whilst
and how do they evolve? What is the relationship between
maintaininghighthroughputintheinterlineregions. There
super-massive black-holes and galaxies? How do stars and
have been several methods proposed to achieve OH sup-
planetary systems form? How common are planetary sys-
pression, including spectroscopic masking at high disper-
tems capable of supporting life?
sion (e.g.Maihara &Iwamuro2000), rugatefilters(Offer&
The success of answering thequestions aboveis reliant
Bland-Hawthorn 1998) and multi-notch holographic filters
onNIRobservations.Thefirsthalfofthelistdependsupon
(Blais-Ouellette et al. 2004). We discuss these methods in
theability to observe thehigh redshift Universe, whilst the
section 5.1; eachofferspotential improvementsovernatural
problemsconcerningplanetarysystemswillalsobenefitfrom
conditions, but in practice there are problems in achieving
dust-penetrating NIR observations, especially since plane-
efficient OHsuppressionbyanyofthesemeans.Byefficient
tarysized bodiesemit most of theirradiation atNIRwave-
wemeanachievinghighsuppressionofthelineswhilstmain-
lengths.
taininghighinterlinethroughput,foralargenumberoflines
Observations of high redshift galaxies carried out in
(>100).InthispaperwemakethecaseforusingfibreBragg
the NIR offer significant advantages over optical observa-
g∼ratings(FBGs,seeBland-Hawthornetal.2004)toachieve
tions: extinction due to local dust is smaller in the NIR; k-
OHsuppression( 5.2); FBGsdonotsufferfrom anyofthe
corrections are smaller, and therefore less critical; adaptive §
drawbacksthathavepreventedsuccessfulapplication ofthe
optics is easier to achieve in the NIR. However, there are
methods mentioned above. There is now a real case for be-
also significant disadvantages in NIR astronomy, the chief
lievingthathighlyefficientOHsuppressionisclosetobeing
concern beingthe high background level. Table 1 lists typi-
demonstrated.
calsurface-brightnessesforagoodobservingsite,asusedin
theESO ELT exposure time calculator3.
Atwavelengths longerthan theH band,thermalemis-
sionfromthetelescopeandsurroundingsdominatetheback- Before discussing methods of OH suppression, we first
ground emission. Therefore the detector and camera must review the background sources contributing to the NIR in
bedesignedtooperateatcryogenictemperatures–asignif- section 2. We then address the scattering properties of dis-
icant technical challenge. At shorter wavelengths the ther- persive optics ( 3), the proper understanding of which is
§
mal background is less problematic but instead thereis the essential to achieve efficient OH suppression. The study of
supplementary frustration of very bright hydroxylemission scatteringshowsthattheOHlinesmustbesuppressedprior
(Meinel 1950; Dufay 1951), resulting from the vibrational to dispersion; the combination of the extremely bright OH
decay of an excited OH-radical, formed from the combina- lines and the strong scattering wings, that are an unavoid-
tion of a hydrogen atom with an ozone molecule. Further- able consequence of any diffraction system, make attempts
moretheintensityoftheOHemissionishighlyvariable(see to remove or ‘resolve out’ the OH skylines after dispersion
section 2.1.2 below), making accurate sky subtraction very futile. We then make the case for OH suppression, i.e. se-
difficult (e.g. Davies 2007). lectively removing the OH lines prior to dispersion ( 4),
§
The intensity and variability of the NIR background including a review of previously proposed methods ( 5.1),
§
have traditionally restricted the depth of NIR observations and their limitations. We then show that fibre Bragg grat-
compared to optical observations. However, two attributes ingsoffertheprospectofcleanlyandefficientlyremovingOH
linesphotonically ( 5.2). Wecalculate therequiredlevelof
§
suppression and spectral resolution to achieve scientifically
2 http://www.so.stfc.ac.uk/roadmap/rmQuestionsHome.aspx useful observations. We summarise our case in section 5.4,
3 http://www.eso.org/observing/etc/bin/ut3/conica/script/eltsimu andconclude with some remarkson thefuturein section 6.
The Case for OH Suppression 3
vibrationallyexcitedOHmolecule(andanoxygenmolecule)
2.6 witharadiativelifetimeofafewmilliseconds(Content1996,
and references therein). The emission then results from the
vibrational decay of themolecule, i.e.,
2.4 H+O3 OH∗+O2. (1)
→
The intrinsic width of the lines is determined from the
Doppler shift due to the thermal motions of the molecules,
frompressurebroadening,andfromnaturalbroadening.The
2.2
formereffectdominatestheinnerwidthofthelineandhasa
Gaussian form, whereas the latter has a Lorentzian profile,
giving thelines broad wings, albeit very faint ones.
TheDopplerwidthcanbeestimatedfromthefollowing
2
argument.Foragasinthermalequilibriumthegasparticles
will have a distribution of speeds, but the most probable
speed will be when the kinetic energy, mv2/2 is equal to
the thermal energy kT. Thus the light emitted from the
1.8
gas particle will be Doppler shifted by an amount ∆λ/λ =
vr /c. Thus,
220 240 260 280 300 ±| |
T/ K
2λ 2kT
∆λDB= c r m , (2)
Figure 2. A lower limit on the wavelength at which the OH
emission dominates the NIR background as a function of envi- wheremisthemassoftheparticles,andT isthegastemper-
ronmentaltemperature.Thiswascalculatedasthewavelengthat ature.ThusforOHmoleculesatatemperatureofT =200K
which the blackbody spectrum from a typical telescope exceeds (Rousselot et al. 2000), ∆λDB 5 10−6µm.
≈ ×
thatofthezodiacalscattered lightatβ=60degrees. Pressure broadening istheresult of interactions or col-
lisionswithneighbouringparticlesinthegas.Theeffectcan
be estimated by considering the average time between col-
2 THE NIR BACKGROUND
lisions or encounters, ∆t0, which is approximately equal to
Wenowconsiderthevariouselementsthatcontributetothe themeanfreepathdividedbytheaveragespeedofthepar-
NIR background, focussing on the wavelength range 0.9– ticles, i.e.
2.0µm, as the region in which OH suppression will be most
l 1
effective.FollowingTokunagainchapter7ofCox(2000),we ∆t0 = , (3)
≈ v nσ 2kT/m
divide the NIR background into the following components:
theOHairglow;thethermalemissionfromthegroundbased where σ is thepcollisional cross section and n is thenumber
telescope; the atmospheric thermal emission; the zodiacal density.ItisassumedthattheshapeofthelineisLorentzian,
scatteredlight;zodiacalemission;galacticdustemission;the which is the result of an emitting atom undergoing simple
cosmic microwave background radiation. Each of these will harmonic motion. Substituting equation 3 into equation 7
bediscussedinturn.OurfinalmodeloftheNIRbackground yields,
spectrum is shown in Figure 9.
λ2nσcross 2kT
∆λPB= cπ r m . (4)
2.1 Hydroxyl emission lines
2.1.1 Emission mechanism tWhee dadiffoeprten∆ceλPiBn ≈wav3e×nu1m0b−e7rµmof,<an0.u0p0p2ercmli−m1itqbuaosteedd obny
The OH emission (Meinel 1950; Dufay 1951) is by far the Turnbull& Lowe (1983).
largestcomponentoftheNIRbackgroundatλ<2.0µm.At NaturalbroadeningisduetotheHeisenberguncertainty
longer wavelengths the background is dominat∼ed by ther- in the particles energy,
mal emission, from the telescope and dome etc. The exact ¯h
∆E = , (5)
wavelength where the thermal emission becomes dominant ∆t
depends on the temperature of the environment, as illus-
where∆tisthedurationofthechangeinenergy,suchthat,
tratedinFigure2,whichshowsthewavelengthatwhichthe
blackbody spectrum from the telescope exceeds the black- λ2
∆λNB . (6)
body spectrum of the zodiacal scattered light, i.e. a lower ≈ 2πc∆t
limit to the wavelength at which OH emission dominates The full calculation shows that the FWHM is in fact twice
the background. Thus for very cold sites, such as Dome C thisvalue,
in Antarctica with an average year round temperature of
λ2
≈effe2c2t0iKve a(steweaBvuerletnongthetsaλl.<2020.45µ)mO.H suppression would be ∆λNB= πc∆t. (7)
TheOHemissionarise∼smainlyfromthecombinationof Thusforavibrationally excited statelastingafew millisec-
hydrogenandozonefromalayerabout9kmthickatanalti- onds,thenaturalbroadeningis 5 10−13µmatλ=1µm.
≈ ×
tudeofaround87km.Thisreactionleavesarotationallyand FromtheaboveestimatestheDopplerbroadeningisby
4 Ellis and Bland-Hawthorn
200
150
100
50
0
0 2 4 6
Figure3.ThetheoreticalloggedVoigtlineprofileduetoDoppler Figure 4. The distribution of doublet separations for OH lines
and pressure broadening, of a very bright OH line line with an withwavelength 0.9–1.8µm.
integrated flux of 740 γ s−1 m−2 arcsec−2 assuming T = 200K
andγ=1.5×10−13m.TheLorentzian wingsaresignificant out
toatleast1˚Afromthelinecentre. is determined by the separation of these doublets. Using
the models of Rousselot et al. (2000), we have determined
the doublet separation for all lines in the wavelength range
far the most dominant contribution to the intrinsic width 0.9 6 1.8, shown in Figure 4. The distribution is strongly
of the line. However, due to the much broader wings of peaked at separations of 0.1–0.2 ˚A, which gives a practical
the Lorentzian pressure broadening, this mechanism is not limittothelinewidths.Notethough,thatwemodeltheOH
negligible, as it will contribute much more to the region spectrum by treating each line individually.
between the line peaks. The natural broadening can be
safely neglected. Therefore in modelling the OH emission
line spectrum we include both the Doppler broadening and 2.1.2 Variability
theLorentzian profile, i.e. we use a Voigt profile,
TheOHemissionisnotonlyextremelybright,butitisalso
+ −x′2 highly variable in the spatial and temporal domain. This
V(σ,γ,x)=Z ∞ σe√2σ22ππ((x−xγ′)2+γ2)dx′, (8) hbaascksgigrnouifincdanfrtocmonosbesqeurevnacteiosnfso.rDtheseprietmeiotvsasltorofnthgevnarigiahbtilsiktyy
−∞
a few trends with time and space are known. The intensity
where σ = 2∆√λ2DlnB2 and ∆λDB is the full-width at half max- of thelines increases with zenith distanceduetothelonger
imum of the Doppler broadened line given by equation 2,
path length through the emitting layer of the atmosphere.
and γ is the half-width at half maximum of the pressure
The intensity changes as,
broadened line given by ∆λPB in equation 4.
2
In our models we assume a temperature at 87km of I(θ) rsinθ 2 −12
s2h0o0wKnaonvderγa=ran1.g5e×of10−1013−m3µ.mThineFreigsuulrtein3gfoVrotihgetbprriogfihlteesist I(0) =(cid:20)1−(cid:16)r+h(cid:17) (cid:21) , (9)
±
doubletsintheH band(whicharenotresolved).Theplotis whereθisthezenithangle,I(θ)istheintensityatθ,ris
logged to emphasise theLorentzian wings which are signifi- theradiusoftheEarthandhisthealtitudeoftheemitting
cant out to 5 10−4µm eithersideof theline centre(c.f. layer abovetheground (Content 1996).
≈ ×
zodiacal scattered light described below). However in prac- Thetemporalvariationisduetotwomaincauses:diur-
tice the interline continuum is dominated by instrumental nalvariationintemperature,anddensitywavesintheupper
scatter. We includethis effect in ourmodel, butin practice atmosphere.
line widths are dominated by the instrumental profile (see The diurnal variation has been modelled by Shimazaki
3) and doublet separations. & Laird (1970), taking into account reactions between 14
§
Notethatalthoughtheintrinsicwidthofanyindividual molecular species and diffusion. We show their model in
lineisextremelynarrow,asshowninFigure3,theOHspec- Figure5,andsummarisetheirdescription ofthevariations.
trum is actually composed of close doublets. The doublets The features result from two emission layers, a broad layer
areofequalintensity(Osterbrocketal.1996) andarisedue at 45km altitude and a narrow layer at 80km. The height
toΛ-typedoubling(seeHerzberg1939)fromtheinteraction of the lower layer rises in the afternoon, and partially dis-
between the rotation of the O and H nuclei and the orbital appears around sunset, causing a dip in the OH emission.
angularmomentumoftheelectron.Thepracticallinewidth Abouttwentyminutesaftersunset,bothlayersriseandthe
The Case for OH Suppression 5
1
400
0.8
300
0.6
200
0.4
100
0.2
0
0
0 2 4 6 8 10 0
Time / hrs Theoretical relative line strength
Figure5.ThenighttimevariationofthemodelofShimazaki& Figure 6. The correlation between the theoretical relative line
Laird(1970)bythecontinuousline,andouranalyticapproxima- strengths of Rousselot et al. (2000) and the measured line
tionbythedottedline. strengths of Maihara et al. (1993b). The closed symbols are for
the J band and have a slope of 1.4×10−5. The open symbols
are for the H band, with a slope of 3.7×10−6. The dashed line
intensity of the emission is greatly enhanced due to an in- betweentheJandtheHbandcorrelationsisthegeometricmean
creaseintheozonecontentatsunset;thiscausesthesudden of7.2×10−6,whichwasusedtocalibratetheRousselotlist.
rise in the models. After sunset the lower layer disappears
completelybecauseoftheremovalofhydrogen,andtheup-
mained unresolved. Since the OH lines are close doublets
perlayerdecreasesslowlyduetothedecreaseofO3.Thereis
we combine all theoretical lines which have a separation of
also a rapid dimming and brightening around sunrise. This
∆λ < λ/17000, before comparing the theoretical and ob-
modelisconsistentwithobservations,whichshowadecrease
served line strengths. Figure 6 shows the correlation be-
by a factor 2 in the OH emission throughout the night
∼ tween the theoretical line strengths and the measured line
(Ramsay et al. 1992; Content 1996). The model is well ap-
proximatedbythefunction1/(1+t)0.57,wheretisthetime strengths.ThereisasignificantdifferencebetweentheJand
H band data, and the H band data shows a stronger corre-
in hours, as shown in Figure 5. We assume that a typical
lation. This is indicative of the transient nature of the OH
nights observing lasts for 8hrs, so for observations longer
lines.WehaveusedthegeometricmeanoftheJandHband
than than this we reset the OHbrightness every 8hrs.
correlations to calibrate theRousselot data.
Aside from this long term decline in OH line strength
throughout the night, there is also short term variability
caused by density waves (Ramsay et al. 1992; Frey et al.
2.2 Other lines
2000). Thesearetypicalperiodicvariationsoflength 5–15
minutes, with an amplitude of about a tenth of the flux. There are a number of O2 and N2 lines in the J band that
We assume that all OH lines vary in unison for our are not included in the models of Rousselot et al. (2000).
simulations. In fact the real situation is more complicated; These lines are self-absorbing and dim rapidly after sunset
lines within thesame transition will be correlated, but sep- (Content 1996). We have included line strengths for these
aratetransitions will not.However,toareasonable approx- taken directly from Maihara et al. (1993b). For simplicity,
imation, lines from a particular transition all lie within the we have assumed they are broadened according to exactly
same region of the spectrum, with very little contribution thesame prescription as theOH lines.
fromothertransitions(Davies2007).Hencetheassumption
ofunisonwillbetrueforsectionsofspectrum 0.1µmlong.
∼ 2.3 Thermal emission from the telescope
Therewillbesignificantthermalemissionfromthetelescope
2.1.3 Line list
andtheinstrumentemployed.Theinstrumentdesigncanbe
Rousselot et al. (2000) have modelled the emission spec- optimisedtocoolasmanypartsaspossibletherebyminimis-
trumofthenight-skyOHlines,givingaccuratewavelengths ingitscontribution tothebackground,butsome parts,e.g.
for 4732 lines, and approximate relative intensities. In or- opticalfibresmaybeexternaltothemaininstrumentdewar
derto usethis atlas we havenormalised thetheoretical rel- and therefore unable to be cooled. Even so, the main con-
ative intensities to the fluxes measured by Maihara et al. tribution will come from the much larger telescope, and we
(1993b) (after first removing the O2 lines). The lines were modeltheemissionasablack-bodyspectrumatT =273K,
measured with a resolving power of R 17,000 and re- with an emissivity of ǫ=0.02 (chapter7, Cox 2000), i.e.,
≈
6 Ellis and Bland-Hawthorn
2hc2
B =ǫ , (10)
λ λ5(ekhλcT −1) 1
which can be expressed as a surface brightness, Kelsall (1998)
B Butterworth
Nφ = ǫλµm hcλ approximation
1.41 1016ǫ 0.8
× , (11)
≈ λ4µm(e1λ4µ3m87T.7 1)
−
in unitsof photons s−1 m−2 µm−1 arcsec−2.
0.6
2.4 Zodiacal scattered light
Zodical light is emission from sunlight scattered by inter-
planetary dust. As such it has a solar spectrum, multiplied 0.4
bythescatteringefficiencyasafunctionofwavelength.Indi-
vidualmeasurementsintheoptical(e.g.Levasseur-Regourd
& Dumont 1980) and the NIR (e.g. Noda et al. 1992; Mat-
sumoto et al. 1996) seem to show a solar spectrum when -1 0 1
considered individually, but have different normalisations
when compared, with the NIR zodiacal spectrum being
twice as bright dueto theincreased scattering efficiency off Figure 7. The variation of zodiacal light with ecliptic latitude
largerdustparticlesatlongerwavelengths(Matsumotoetal. from Kelsall et al. 1998 shown by the continuous line and our
1996). modelasgivenbyequation 12shownbythedotted line.
Thebrightnessofthezodiacallightobviouslydecreases
as a function of ecliptic latitude. Kelsall et al. (1998) have
modelled thezodiacal light indetailinordertoremovethis because Rayleigh and Mie scattering are both less efficient
componentwhenanalysingdatafromtheCOBEDiffuseIn- atlongerwavelengths;Rayleighscatteringhasaλ−4 depen-
frared Background Experiment (DIRBE). We approximate dence and Mie scattering has a λ−1.3 dependence.However
theKelsallmodelwithaLorentzianfunction,whichenables in theregime of OH-suppressedobservations theseassump-
afastercalculationwhilstmaintainingmostoftheaccuracy. tions need to be re-evaluated.
In Figure 7 we show their model’s dependence on ecliptic Krisciunas & Schaefer (1991) present a model of the
latitude, β, for 1.25µm emission and ourfit to it given by, brightnessofscattered moonlight thatis accurateto23per
cent. However, this model is only appropriate for observa-
0.75
I = +0.25, (12) tions in the V band. Therefore we have adopted the proce-
(1+(0.27β43)2) dure described in the CFHT Redeye Manual4 to adapt the
model to NIR wavelengths. The principal changes are al-
where β is in radians.
lowing for the change in colour assuming a solar spectrum,
For the majority of astronomical applications of FBGs
andallowingforthewavelengthdependenceofthescattering
we anticipate a preference for high galactic latitudes, and
function.WeshowtheresultsofthemodelinFigure8.Com-
thereforeawiderangeofeclipticlatitudesdependingonthe
paringtoestimatesofthezodiacalscatteredlightshowsthat
galactic longitude, which could change the zodiacal bright-
forobservationatafull,ornearlyfull,moon,thenscattered
ness by a factor of 2.
∼ lunar light would be significant out to 50 degrees from
For the purposes of the model we normalise a solar ≈
spectrum (using equation 11 with ǫ = 1.08 10−13 and themoon.OHsuppressedobservationsshouldthereforenot
× be performed around the full moon, though ‘grey nights’
T =5800K) to theKelsall et al. (1998) model valuefor the
should be acceptable. We note that the effect of scattered
ecliptic plane. Several measurements and models of the zo-
moonlight couldhavehadaneffectonpreviousattemptsto
diacal scattered lightexist intheliterature.Thesearecom-
measuretheinterlinecontinuum,ifthesewereperformedin
piledinTable2.Notethatwhilethemodelspredictthezo-
bright time and the contribution due to moonlight was not
diacalscatteredlightshouldbefainteratlongerwavelengths
taken into account.
(sinceithastheformofaT =5800K black-body)themea-
sured data disagree with this. Note too, that the ground
based measurements are rather higher than the models. In
2.6 Other components
section3wesuggestthatforthegroundbasedmeasurements
scattering of OH light may have a significant contribution
There are a number of other components that make im-
to these values.
portantcontributionstotheIRbackgroundatlongerwave-
lengths(>2µm),butarenegligibleintheregimeconsidered
here. These are: the thermal contribution from the atmo-
2.5 Moonlight
sphere, zodiacal emission from interplanetary dust at the
Moonlight is known to be an insignificant source of back-
groundwithconventionalNIRobservations.Thisisbecause
theOHskylinesdominateoverscatteredmoonlightandalso 4 Seehttp://www.cfht.hawaii.edu/Instruments/Detectors/IR/Redeye/Manual/chapt
The Case for OH Suppression 7
Table 2. Measurements and model predictions of the zodiacal scattered light compiled from the literature. aIncludes
instrumental background. bModel based on COBE DIRBE measurements. cFrom a STSI report by M. Stiavelli, see
http://www.stsci.edu/hst/nicmos/documents/handbooks/current NEW/c04 imaging8.html#309385
Source Measurmentor Flux/γ s−1 m−2 arcsec−2 EclipticLatitude/
model J H degrees
Maiharaetal.(1993b) Measurement – 590 –
Cubyetal.(2000) Measurement 1200 2300 –
Thompsonetal.(2007)(NICMOS) Measurement 72 150 -45(HUDF)
Thompsonetal.(2007)(NICMOS) Measurement 74 220a 57(HDF-N)
Thompsonetal.(2007)(IRTSNIRS) Measurement — 110 –
Kelsalletal.(1998) Modelb 43 – 46(Lockman Hole)
Kelsalletal.(1998) Model 99 – 0
NICMOShandbookc Model 130 77 0
NICMOShandbook Model 50 31 45
Stockmanetal.(1998)JWST Model – 3 1AUorbit
2.7 Atmospheric transmission
1000
-1m The dominant absorption features of NIR light are those
Μ
--22marcsec 10 dopzuhoeontteoominwettarhticeerwa,ticmnadorobswpohsn.erWdei,oexwhihdaviec,ehnudisteerfidonutehsetohmxeiodJde,,eHlmtareantndhsaKmneibssaainnondd
-1s profileatMaunaKeaassuminganairmassof1.0andaWa-
ns 0.1 tervapourcolumn of 1.6mm in our default sky background
o
hot model (Lord 1992, obtained from Gemini Observatories5).
P
SeeFigure 9.
0.001
0 50 100 150
Distance(cid:144)degrees
(a) J band 3 SCATTERING OF DISPERSIVE OPTICS
BecausetheOHlinesaresobrightcomparedtotheinterline
1000 continuum (see Figure 9) they cannot be cleanly removed
-1m through high dispersion spectroscopy. This is because even
Μ
-2arcsec 10 aintsemrlainlleacmoonutinntuuomf s.cTatotesreeedthOiHs, cliognhstidwerillthsweaemxapmtphleetorfuae
-2m ruled diffraction grating. The theoretical normalised inten-
-1s sity distribution of a grating with N lines is given by,
ns 0.1
hoto sinβ 2 sinNγ 2
P I =I0 , (13)
(cid:18) β (cid:19) (cid:18)Nsinγ(cid:19)
0.001
0 50 100 150 where,
Distance(cid:144)degrees
kb
(b) H band β= sinθ, (14)
2
Figure8.Thebrightnessofscatteredmoonlightasafunctionof kh
γ = sinθ, (15)
angular distance away from the moon. The different curves are 2
fordifferentlunarphases from0to180instepsof20degrees. It and k is the wavenumber, b is the slit width, h is the slit
isassumedthatthemoonisatthezenith.
separation, and θ is the angle through which the light is
diffracted.
Woodsetal.(1994) showthatequation13mayberep-
ecliptic pole, Galactic emission from from interstellar dust, resented as a Lorentzian plusa constant background,
thecosmic background radiation. ω2
It is possible that there is also a faint continuum or Yfit = (λ λ0)2+ω2 +B, (16)
pseudo-continuum composed of line emission from NO2 or −
other atmospheric models (Content 1996). Such emission whereYfit is theratio of intensity at wavelength λ tointen-
couldperhapsexplainsomeofthediscrepancybetweenmea- sityatλ0,B isaconstantbackgroundtermduetoRayleigh
surements of the zodiacal scattered light and the ground scattering and
based measurements of the interline continuum. However λ0
ω= , (17)
therearenoreliable measurementsof thiscontinuum,since
Neπ√2
previous estimates do not take into account the instrumen-
tal scattering profile ( 3), therefore we do not include any
§
such continuum in our model. 5 http://www.gemini.edu/sciops/ObsProcess/obsConstraints/ocTransSpectra.html
8 Ellis and Bland-Hawthorn
Figure 9. The NIR background model. The top panel shows the atmospheric transmission. The middle panel show the contributions
from the OH spectrum (blue), the zodiacal scattered light (red), the telescope thermal emission (black). The bottom panel shows the
finalresultingspectrum.
The Case for OH Suppression 9
and it is assumed that λ λ0 << λ0 and N > 100. Equa-
−
tion 16 is normalised by a factor 1/(πω+2hB∼), where h is
1
thegroove spacing.
Themostimportantfactorindeterminingthelinecon-
tribution is Ne the number of effective lines. In theory the
numberof effectivelines should bethenumberof lines illu-
0.1
minated. However, Woods et al. (1994) find that Ne is al-
waysmeasuredtobemuchlowerthanthistheoreticallimit.
In practice Ne can be taken to be the smallest number of
consecutive grating grooves with no imperfections. Woods
0.01
et al. (1994) report values that are typically 0.25–0.5 the
theoreticalupperlimit,butoccasionallyaslowas0.16.Fur-
ther from the line centre the constant background AB also
becomes important.
Woods et al. (1994) fit profiles to 10 different types of 0.001
gratings including holographically and mechanically ruled
gratings and find the scattered light is indeed well fit by
a Lorentzian plus constant. Our own measurements with
IRIS2(a grism spectrometer) and AAOmega (a VPH spec- 0.0001
trometer) arc spectrum confirm this; we find that thespec- -40 -20 0 20 40
tral line profile can befit bya function,
(a) AAOmega,a=0.97andf =5.5.
e−(λ−2σλ20)2a (1 a)fσ 2log(2)
I =I0 √2πσ + π (λ −λ0)2+p2f2σ2log(2) ,(18)
{ − }
  1
which is simply a Gaussian plus a Lorentzian, with a con-
tribution of factor a from the Gaussian and 1 a from the
−
Lorentzian, and the Lorentzian has FWHM f times larger
thanthatoftheGaussian.Wefindtypicalvaluesoff =4–6
and a = 0.89–0.97. In Figure 10 we show a fit to the IRS2
and AAOmega arc spectra.
0.1
Wenowshowtheeffectsofthesescatteringwingsonthe
OHspectrum,andindoingsoweaddressthelarge discrep-
ancybetweentheground-basedobservationsoftheinterline
continuumlisted inTable2andthemodelvalues.Weshow
theeffectofscatteringontheOHlinesaroundλ=1.665µm,
whereMaiharaetal.(1993b)measuredtheinterlinecontin-
uum. 0.01
We have modelled the scattering from the University
ofHawaii2.2mCoud´espectrograph,withthe600lines/mm
grating, and λblaze = 1.32µm used for the H band observa-
tionsbyMaiharaetal.(1993b).WehaveassumedB=108,
-40 -20 0 20 40
the value for the Bausch & Lomb grating used by Woods
et al. (1994).
(b) IRIS2,a=0.89andf =4.0.
We use equation 16 to model the scattering profile as
it can be normalised exactly. Note that this will provide a Figure 10. The line profiles of two spectrographs, AAOmega,
slightly conservative estimate of thescattering far from the aVPHspectrograph andIRIS2 agrismspectrograph. Themea-
linecentres,asthewingsinequation16dropofffasterthan sureddataareshownbytheopencircles,andafitoftheformof
in thefull equation. equation18isshownbytheline.
In Figure 11 we show the scattered light from all H
band lines reported in Maihara et al. (1993b) in the win-
dow 1.660–1.675 (c.f. their figure 2) whence the continuum between 35 and 280 photons s−1 m−2 arcsec−2 µm−1 at
emission was measured. It is unclear what the theoretical 1.665µm. Thus it is possible that scattered light could ac-
value of Ne should be, i.e. the number of illuminated lines, count for a significant part of the difference between our
but it will be much less than 120,000 (e.g. Woods et al. predicted value of the continuum at λ=1.665µm and that
1994).TherealvalueofNe,i.e.thesmallest numberofcon- measured by Maihara et al. (1993b), if the number of im-
secutive lines without imperfections will be smaller again. perfections in thegrating is high.
Therefore we show plots for four different values of Ne: for The scattering behaviour of dispersive optics thus pre-
N = 120,000 = 20cm 600 lines mm−1, i.e. the absolute ventsobservationsreachingtheleveloftheinterlinecontin-
×
maximum number of lines, and for values of Ne = 60,000, uumregardlessofinstrumentbackgroundorspectralresolu-
30,000 and15,000. Thescattered wingsarehigherforfewer tion.ForexampleinFigure12weshowtherelativestrengths
lines. These estimates give a scattered light intensity of of the OH spectrum with a resolution of R = 10000 (using
10 Ellis and Bland-Hawthorn
must remove a section of finite wavelength, ∆λ, from the
incidentspectrumforeachlinetobesuppressed.Aftersup-
pression, these regions will be unsuitable for any scientific
analysis. Therefore OH suppression comes with thepenalty
ofreducingtheuseful bandpassof thespectrum,which will
vary with the number of lines suppressed and the size of
the‘notches’,∆λ.InpracticetheregionsaroundbrightOH
lines are not usually suitable for scientific analysis in any
case, since the variation of the OH emission hampers the
successofsky-subtractionandresidualskylinesareusually
observed.
IndesigninganOHsuppressionsystemitisthusneces-
sary to choose howmany lines to removeand at what reso-
lutioninordertomaximisetheefficiencyofthesystem.We
nowaddressthisissuefornotchesfixedat∆λ=1˚A,3˚Aand
10˚A (λ/∆λ 10000, 3000 and 1000), for varying numbers
Figure 11.Thescattered lightdueto allmeasured OHlines in ≈
of lines. The lines are suppressed in order of their relative
the H band. The different curves arefor values of N =120,000,
brightness, with the brightest lines removed first. We mea-
60,000, 30,000 and 15,000. The scattered wings are higher for
fewerlines. sure the efficiency of the system by the fraction of useable
spectrum remaining and the brightness of the faintest line
suppressed(asanestimateoftheexpectedsensitivityofthe
observations).NotethatinpracticemanyOHlinesarevery
close doublets, and therefore these can be removed with a
single 1˚A notch.
Figure13showstheresultsforupto200notchesacross
the J and H bands. Even with 200 lines of width 1˚A, sup-
pressing down to a level of 4 10−3 photons s−1 m−2
arcsec−2 µm−1 (far fainter than×would practically be nec-
essary), only 8 per cent of the J band has been removed,
and only 7 per cent for 200 lines in the H band. Thus,
by virtue of the intrinsic slenderness of the OH-emission
lines,suppressionathighresolutionisahighlydesirableob-
jective, maintaining a large fraction of scientifically useful
wavelength range, whilst vastly increasing signal to noise.
Figure 12. The relative strength of the OH spectrum at R =
5 METHODS OF OH SUPPRESSION
10000(thinline)comparedtothezodiacalscatteredlightatβ=
60 degrees (thick line). The scattered OH line light reaches the Thereareseveralmethodswhichhavebeenproposed,andin
interlinecontinuum for11percentofthespectrum.
some cases employed, to achieve OH suppression. However,
to date no instruments exist which can suppress OH light
a scattering profile of the form equation 18 with a = 0.90 athighefficiencywhilstmaintaininggoodinterlinethrough-
and f = 4) and the interline continuum at an ecliptic lati- put.Inthenextsectionwewillreviewproposedmethodsof
tude β = 60 degrees over the range 0.9 6 λ 6 1.8µm. The OH suppression and address their limitations. This will be
interline continuum is reached for only 11 per cent of the followedbyadiscussionofFBGs,inwhichwewillarguethat
spectrumbetween0.96λ61.8µm,and3percentbetween thereisnowarealcaseforbelievingthathighlyefficientOH
1.46λ61.8µm. suppression will finally be acheived.
5.1 Previous methods of OH suppression
4 THE CASE FOR OH SUPPRESSION
5.1.1 High dispersion masking
By OH suppression we mean the selective removal of OH
linesfrom theincidentlightwhilst allowing lighteitherside Maiharaet al.(1993a) presentamethodofOHsuppression
of the deleted line to pass through with high efficiency (as using high dispersion and a ruled mirror. The basic idea
opposed to OH avoidance by narrow band filters e.g. Hor- is to disperse the light at high resolution, whereupon it is
ton et al. 2004, which selects light between OH lines, and reflected from a mirror-mask, which has a series of black
is therefore restricted to narrow bandpasses). We discuss lines coincident with the position of the OH lines, which
how this may be achieved using fibre Bragg gratings below arethereforenotreflected.Thereflectedlight,with theOH
( 5.2). First however, we address the desired performance lines removed, then proceeds back through the camera and
§
of a general OH suppression system. gratingsystemintheoppositedirectiontoitsinitialjourney,
The OH spectrum contains hundreds of bright lines andhenceisrecombinedintoa‘white-light’beam.Thisfinal
(Rousselot et al. 2000; see 2). An OH suppression system beam may then be dispersed at low resolution as required.
§