PACS is an instrument of the Herschel Space Observatory. The Herschel mission ended in 2013, but all the data collected are available from the Herschel Science Archive (HSA). You can download raw to fully reduced PACS data. In some cases it will be necessary to partially re-reduce the data, in other cases you can take the pipeline end-products from the HSA and use them for science.
For full details of PACS spectroscopic data and how to work with them, check out the various documentation provided on the Herschel/PACS web-site (here). Start with the PACS Data Reduction Guide (available also directly via the Herschel software environment: HIPE) if you are going to reduce data, but even if you are only planning to work with the archived and already-reduced data you will find the first chapter of that guide useful. The PACS Observer's Manual should be consulted for all the technical details of the instrument.
The PACS spectrograph had a blue and red side which operated simultaneously: ask for a wavelength range in the red and you get a "free" range in the blue. PACS offered different spectral sampling modes: cover the entire SED with a Nyquist spectral sampling, or sample smaller (either pre-defined or user-defined) sections of the spectral range at a higher spectral sampling.
The PACS observing mode most used was chop-nod, which is typical for the IR regime and used to allow one to remove the “telescope” background spectrum (the Herschel telescope mirror was not cooled). For sources located in regions of the sky where there was no blank "sky" to nod to, an "unchopped" mode was offered: here the "nodding" was in spectral space rather than on the sky. The difference between the modes is important only if you want to re-reduce the data yourself; the end products that you can get from the HSA are the same.
There were three pointing modes offered: a single pointing, spatial dithering, and a mapping mode. The single pointing was intended for observations of point sources, spatial dithering was for observations where the best possible spectral sampling was required, and the mapping mode was for observations of larger fields-of-view, where offset single pointings could be patched together.
PACS design and the effect on its spatial sampling
A single pointing PACS observation produced a cube of 5x5 spaxels. The IFU is an image slicer, employing reflective optics to re-arrange the sliced 2d field-of-view to fit into the 1x25 pixel entrance slit of the grating. Amorphic re-imaging optics was used to match the spatial and spectral resolution of the system to the pixels of the detector arrays. The pixels of these detector arrays (one for the blue and one for the red) were a grid of 16x25 GeGa crystals: 16 pixels to cover the entire requested wavelength range of the 25 spaxels. The result is data with the Nyquist or better spectral sampling at each wavelength (and hence the dispersion gets larger with wavelength) but with the same spatial sampling at all wavelengths: the spaxels of the IFU are almost contiguous squares of 9.4" on a side. The spaxels are not actually physically square in shape: the spatial coverage is rather formed by light cones that feed the light onto the detector pixel arrays.
So, a single pointing will give you a field-of-view of about 47x47", formed by the 5x5 spaxels. However, this 5x5 is not a perfectly square grid: in particular one row of spaxels is offset slightly from the rest. Because of this, any image-slice you take from your cube (i.e. any 2d slice at a single wavelength) has a non-linear grid in sky-coordinate (RA, Dec) space. Traditional viewers have a hard time dealing with this.
Each spaxel does has its own (RA, Dec) coordinate, but you cannot define a traditional WCS for the whole cube, i.e. you cannot define a reference RA and Dec and spatial offsets for each spaxel (the FITS keywords crval, crpix, and cdelt). To get around this, the spectral viewer in HIPE considers only the "spaxel-WCS", i.e. the coordinates of the spaxels are defined by an integer that is their spaxel position in the cube (effectively we are setting: cdelt1|2=1, crval1|2=1, crpix1|2=1). Treat the cube like that and you can display spectral slices in a traditional viewer. However, the image you will see is not an accurate representation of the distribution of light of your source, on the sky, at that wavelength: the skew of the cube's sky footprint is not properly accounted for.
An alternative approach would be to set the keywords crval, crpix, and cdelt by creating an "average" WCS using the actual spatial coordinates of the spaxels. In this way you effectively assign an almost-correct coordinate to each spaxel, and by giving the cube a regular spatial grid it can be loaded easily into any cube viewer (e.g. ds9). Another alternative is to resample the spatial plane from 5x5 (slightly-skewed) spaxel grid to a >5x5 grid with a proper sky WCS, and with spaxels smaller than 9.4". This differs from the first approach because you are actually resampling the image plane, at each wavelength, rather than just fiddling with the WCS. However, if doing this then consider that while the spaxels are smaller than the beam of PACS (which varies from about 9" to 13" over the spectral range), they are not small enough to Nyquist sample the beam. This means that you are missing information about the spatial distribution of the flux in your field-of-view, and so from a single pointing you do not really have the information to do any spatial regridding of the flux plane. Any attempt to reproduce this morphology will not be quantitatively accurate, although it may not be too different, and certainly if done with care it should be quite close. There are tasks provided in HIPE that can do this, and the PACS documentation will explain the pros and cons as well as their use.
It is exactly to get around this issue of spatial undersampling that PACS offered a dithering observing mode. Slight spatial offsets between individual pointings on the source allow for a full Nyquist sampling of the beam (at each wavelength). The individual pointings are then reduced through the pipeline and the cubes spatially resampled and mosaiced together to create a cube with 3" spaxels. Image slices taken from these cubes then reproduce accurately (to the spectral resolution of PACS) the morphology of the field-of-view.
Data reduction pipelines and the end products
A number of pipelines are offered to reduce PACS data, all via the HIPE Herschel interactive software (which is an entire data reduction and data analysis software package, based on java/python). These differ mainly according to the observing mode used when collecting the data, but there are also pipelines with different ways of removing the telescope background and flux calibrating the data, and pipelines aimed at point sources. The end result of any pipeline ("Level2/2.5" products in Herschel-speak) is either: a single point source spectrum (for single pointing observations on a point source); a 5x5 cube with native spatial sampling (for the single pointing mode); a cube with smaller spaxels, 3" (for the dithering mode); or a larger mosaic of offset single pointings (for the mapping mode).
You can ask for all the levels of data from the HSA, but unless you want to re-reduce the data yourself, or if you just first want to see what the data look like, you should first look at the Level 2 and (if available) 2.5 products. By far the easiest way to view these products is in HIPE, but they can, with some effort, be viewed with any software.
The FITS files
The pipeline end-product cubes can be gotten as FITS files from the HSA. However, these are more complicated than you may be used to. Unfortunately the more complex issues with PACS FITS files, and a translation of all the PACS- and Herschel-specific headers, have not yet (as of April 2014) been nicely documented, but once they are they should be linked from the PACS web-site. Here we offer a brief summary of things to consider when working with these FITS files.
1) For cubes arising from a single pointing, the flux information is held in an Image extension, and in separate extensions you will find the RA, Dec, and wavelength data, these being organised as a table or a 3d dataset. There is no WCS with a linear wavelength or spatial grid (for reasons explained above) attached to any of the extensions: the spatial coordinates are the integer numbers of the spaxel coordinates (as explained above) and the wavelength dimensions are not defined. For cubes which have been mosaiced toegther (mapping or dithered observations) there is a WCS with sky dimensions assigned, but no wavelength dimensions here either.
2) The spectral sampling is not even: the optimal spectral sampling for the long wavelength range (50 — 220 microns) of the PACS spectra changes linearly with the wavelength, since the resolution also changes with the wavelength. This is why the wavelength part of the WCS of PACS cubes is also not defined: a reference wavelength and a dispersion value cannot reproduce the actual wavelength grid. Instead, the wavelength information is held as a table in an extension of the FITS files.
For cubes of short wavelength range you could consider resampling the wavelength grid to a linear dispersion, and hence create a wavelength entry in the WCS. The difference between the old and the new grid would be very small. For longer wavelength stretches, you can also do this but note that this will result in non-optimal spectral sampling for considerable parts of the spectrum.