Integral field spectroscopy
This is a brief description of the principles of integral field
spectroscopy, mainly using fibre-lenslet techniques. It is meant to
serve as a link to the GEMINI Multiobject
Spectrographs and Durham
Astronomical Instrumentation Group: Spectroscopy Programme pages
which contain a list of current instrumentation projects .
It is not meant to be comprehensive and apologies are offered for not
giving credit or references to the the many other workers in this field.
Click here for James
Turner's IFU data reduction software page
For a summary, please see the paper
given at the 1998 SPIE large telescope conference in Kona Sweden in
March 1998 (gzipped postscript 0.5Mb). An older paper given at the 1996 SPIE large
telescope conference in Sweden in May/June 1996 (Postscript 0.3Mb) is
A general paper on sampling and
background subtraction (0.4Mb, gzip postscript). in fibre-lenslet
IFUs has been accepted by PASP (Allington-Smith and Content).
Beyond longslit spectroscopy
Traditional spectroscopy is based on dispersing the image of a slit
(single or multiple) so that a spectrum is produced for whatever
fraction of the light from the target of interest falls within the
aperture defined by the slit. If the slit is extended in length beyond
the confines of the target, then it is also possible to record the
spectrum of adjacent sky to subtract from that of the object -
particularly important if the object is fainter than the sky, which is
very frequently the case. While this is satisfactory for many
applications, it makes poor use of the incident light when the object is
extended, either intrinsically or due to poor seeing. In these cases,
what is really required is the ability to record a spectrum from each
part of an extended object.
This cannot be done with a longslit except in one dimension defined by
the length of the slit. However the longslit can be stepped in position
across the target by moving the telescope and recording separate
exposures for each position. But this is time-consuming since the
effective exposure time is multiplied by roughly the ratio of the object
size to the width of the slit.
Other techniques are available such as Fabry-Perot scanning. This allows
a large object to be surveyed in a single exposure but only at a single
wavelength (which depends on position within the field) so that the
required data volume with axes labelled by x-position, y-position and
wavelength must be built up via a series of exposures. As with stepped
longslit spectroscopy, this is an inefficient use of telescope time.
Techniques which record spectra from each part of an object
simultaneously are termed Integral Field Spectroscopy
The terms two-dimensional spectroscopy or three-dimensional
imaging are also used, although, strictly, they include
non-simultaneous techniques as well.
There are three main techniques.
As will be seen below, it is possible to combine some of these
techniques to get the best combination of features.
- Lenslet arrays. The input image is formed at the input
surface of a microlens array (MLA). These form images of the telescope
pupil which are then dispersed by the spectrograph. The pupil images are
smaller than the aperture of each microlens so that light from each
segment of the input image is concentrated into a dot. Because the dots
are small, it is possible to tilt the MLA about the optical axis of the
system so that the spectra do not fall on top of each other. This
technique allows the input image to be sampled contiguously and is not
subject to FRD so that the throughput and spectral resolution can be
optimised, but the length of spectrum that can be produced without
overlapping is very small.
- Fibre bundles. The input image is formed at the
entrance to a bundle of fibres which transfer the light to the slit of
the spectrograph. The flexibility of the fibres allows a round field of
view to be reformatted into one (or more) slits so that the spectra are
obtained without wavelengths shifts between them. The disadvantages of
this techniques are: (a) the sampling of the sky is not contiguous since
there are gaps between the fibre cores and (b) the fibres do not work
efficiently at the slow focal ratios at which most telescopes work.
The latter problem occurs due to Focal Ratio Degradation (FRD) which
causes the focal ratio of the beam emerging from the fibre to be faster
than that at the input. This problem is especially bad for slow beams
(e.g. f/16 for GEMINI) but can be negligible for fast beams (f/5 or
faster) provided that fibres of the best quality are used. FRD
translates into a violation of the Etendue (solid angle - aperture
product) invariant in optical systems with the result either the
throughput or spectral resolution is degraded. Fibre
throughput is not generally a problem since the fibres are usually quite
short, unless working far in the ultraviolet.
slicers . The input image is formed at a segmented in thin
horizontal sections which are then sent in slightly different
directions. A second segmented mirror is arranged to reformat the slices
so that, instead of being above each other they are now laid out end to
end to form the slit of the spectrograph (actually a virtual slit). The
advantage of this technique is that FRD is avoided and the slicing
arrangement gives contiguous coverage of the field. Because this system
uses only mirrors, it is especially suitable for the infrared since it
is inherently achromatic and can be cooled to cryogenic
temperatures. Potential disadvantages are: (a) that the sampling
along the slices is the same as that provided naturally by the telescope
so there is reduced scope to optimise for use with a spectrograph that
must also work in a normal slit-spectroscopy mode and (b) the optical
system might be bulky and difficult to fabricate.
In some cases, these techniques require the telescope focus to be
magnified so that either the physical size of the devices can be made
more manageable and/or to adjust the focal ratio of the input telescope
beam. This requires the use of fore-optics , generally a pair of
Integral field spectroscopy in Durham
Integral field techniques which are being studied by the Durham
Astronomical Instrumentation Group (AIG) include.
Durham is currently embarked on three projects to provide integral
field units (IFUs). The most ambitious is that for the GEMINI Multiobject
The above discussion concerns only the technical implementation. At
this stage we need to consider the general science requirements
. Naturally enough, these vary from project to project but, taking
some sort of sum over these, it appears that the basic requirements are:
More specifically, GMOS's requirements are for 0.2 arcsec sampling, a
field of view with an area of at least 50 arcsec-squared and separate
object and background fields separated by about 2 arcmin.
- Field of view: around 10 arcsec in linear dimension (roughly
square or circular)
- Spatial sampling: such as to adequately sample the expected image
size at the input. For the case of an AO-corrected 4-m telescope or
the GEMINI telescopes, the expected image size is typically 0.4 arcsec
FWHM so the sampling needs to be around 0.2 arcsec/element. Rarer
conditions of enhanced seeing (e.g. 0.25 arcsec FWHM at 10th percentile for
GEMINI) would benefit from better sampling around 0.1 arcsec/element.
- Background subtraction: Unless the field is very large there will
be times when the object of interest completely fill the field of view.
In these cases, a separate field dedicated to sampling the sky
background, the background field , is required. This should be
placed on the sky as far away as possible from the
- Field of view. Since the technique works by
reformatting a 2-D area into a 1-D line, the area of the field of view
is limited by the product of the number of detector pixels in the
spatial direction and the number of output slits divided by
the allowable spacing (in pixels) between adjacent spectra.
In principle, multiple slits, each of which occupies the full spatial
extent of the detector, can be provided by offsetting them in the
dispersion direction so that the resulting spectra are arranged in
multiple tiers (Figure 4).
The spectra will not overlap if band-pass filters are used.
The spacing of the spectra is dictated by technical issues such as
sampling, aberrations and the allowable cross-talk between adjacent
spectra. Cross-talk can be tolerated if neighbouring elements of the
slit are also adjacent on the sky
(Figure 5). Therefore intensity variations
between adjacent parts of the sky will be reflected in the way in which
the light from adjacent slit elements overlap spatially in the
spectrograph. In fact, the optimum sampling strategy is that each FWHM
of the input image at the input to the MLA should be sampled by two slit
elements. This ideal situation is generally severely limited by other
considerations but the general message is that the spectra should be
packed together as closely as possible so that the distribution of light
on the detector is similar to that which would have been obtained from a
longslit sampling the sky directly.
The detection process is summarised in Figure 6.
In some cases, adjacent spectra must be separated by a clean gap. This is
true for elements which sample a background field which may be well
separated on the sky from those elements sampling the object. Also it
will be necessary to have clear areas of the detector from which to
estimate the background signal caused by scattered light within the
- Background subtraction. As far as possible, the object and
background fields must share the same optical path. One way to do this
is to duplicate the entire optical train, linking them only at the
output slit. Within the context of a microlens/fibre system, there would
be two separate fore-optics and MLAs whose fibres would be mixed
together at the spectrograph slit. The background and object fibres
would be grouped together in blocks to avoid cross-talk between
unrelated regions of the sky, but be alternated in position within the
slit so that any errors related to the position within the spectrograph
field cancel out. In general, these blocks will be matched to continuous
regions of the object or background fields (Figure
Since the total number of slit elements is fixed by the detector size,
number of output slits and spectrum spacing, only a small proportion of
the total number of elements can be dedicated to the background field
without restricting the size of the object field. Roughly speaking, we
may allow a 90:10 split between the object and background fields,
leading to a negligible reduction in the linear size of the object
There are subtle problems with sky-subtraction using fibres arising
from possible variations in the details of the point-spread function
between fibres due to high-order FRD-type effects. These may affect the
detailed line profiles so that the subtraction of a bright sky-line in
one spectrum from that in another may give a significant non-zero
residual even when variations in the throughput (as a function of
wavelength) of each element is calibrated out. One way to eliminate
this effect is to beam-switch between the object and background
fields, which we shall now label A and B
(Figure 7). The observation
is broken down in to pairs of exposures 1 and 2.
By forming linear combinations of the signal obtained from each element
in exposures 1 and 2, the true object signal can be obtained. Since the
same physical elements have been used to sample both object and sky, the
background signal can be completely eliminated.
Since the background field is smaller than the object field, the
field over which true beam-switching can be done is only as large as
the background field. However, sky-subtraction will still be possible
over the rest of the field, but the precision may not be as great. In
some cases, where the object is relatively small (e.g. a distant galaxy
a few arcsec in extent), it will also be acceptable to derive the
background signal from elements within the object field itself,
generally at the edge.
- In exposure 1, field A samples the object and B samples the
- The telescope is then offset by an amount equal to the spatial
separation between fields A and B. The result is that in exposure 2,
field B samples the object while field A samples the background
(actually on the opposite side of the object from the first background
11 Jan 1996 - last revision 30 June 1998