Last updated on by Attila Kovács

 

WARNING!

This document is now outdated. It was written for the original, proprietary, SCUBA-2 module for CRUSH, which provided only limited support for SCUBA-2. That module is now deprecated, and has been replaced with a new public licensed (GPLv3) module, which is now packaged with CRUSH as of version 2.30. The new module supports SCUBA-2 with any and all subarrays, and you are free to use it as you see fit, without any restrictions or binding agreements. You can find more information on the re-newed SCUBA-2 support in the Quickstart guide.

 

CRUSH for SCUBA-2

Table of Contents

   

Introduction and Conditions for Use

CRUSH can reduce data from SCUBA-2 just as simply as for the other instruments. However, the SCUBA-2 modules of CRUSH are not part of the public release of the software. Nor it is expected to become public unless there is such a request from the SCUBA-2 consortium (in exchange for some appropriate compensation for maintenance and support).

If you would like to use CRUSH with your SCUBA-2 data, you can obtain a private copy of the required modules directly from Attila Kovács (attila[AT]submm.caltech.edu), provided you agree to

You can also have a look at the detailed user license. In the sections below you will find some information that may help you decide if CRUSH is worth a try...

 

Requirements and Performance

CRUSH requires modest computing resources only. It can run on any recent desktop or laptop computer (it reduces SCUBA-2 data on my seven-year-old laptop just fine!).

For every hour of SCUBA-2 data you wish to reduce in one go, CRUSH will require between 1 to 2 GB of free RAM (the exact number depends somewhat on the details of the observing pattern you used to obtain the data). The reduction is reasonably fast: an hour of SCUBA-2 data is reduced in 9–16 minutes (depending on waveband, scanning speeds, input file format, etc.) on my newer laptop with a dual-core i5-520M CPU, including over a 100 processing steps in the pipeline. Thus, with a decent quad-core desktop PC or Mac, you can expect to crunch through an hour of SCUBA-2 data in about 3–8 minutes (more or less)...

 

CRUSH vs. SMURF

Point Sources

Below is a comparison of the 450μm stamps of a point-like source (IRC+10216) reduced by SMURF (left panel; courtesy of Jeyhan Kartaltepe) and by CRUSH (right panel), inside a 90" × 90" box centered on the source (scan #36 taken on 2010-03-01). The CRUSH image has been produced on the same grid as the SMURF one. Otherwise the CRUSH image is representative of its default reduction pipeline.

SMURF reduction CRUSH reduction
Figure 1. IRC+10216 with SMURF (left) and CRUSH (right).

The default reduction of CRUSH produces an image that is up to 30% deeper than the SMURF reduction of the same data set (with S/N around 34 vs. 44 at this pixelization). However, the images become comparable (with S/N of 78 vs. 82) after beam smoothing, hinting that the excess noise in the SMURF image are present only at the high spatial frequencies.

Recovery of Large Scale Emission

One of the strengths of CRUSH, compared to other submillimeter data reduction packages, is its ability to recover large scale emission. Below are the 450μm images (SMURF is left, CRUSH is right panel) of OMC-1, obtained with SCUBA-2 in a 19 second integration in average weather (τ225GHz~0.075; scan #39 taken on 2010-02-19). Contours are plotted around -10, 0, 10, 30, 100, and 300 Jy/beam levels.

SMURF cal image SMURF science image SMURF tweaked image CRUSH image
Figure 2. Reductions of OMC-1. Images are from (from left to right): (1) the SMURF standard calibration pipeline from the public archive, (2) SMURF science reduction (courtesy of Ed Chapin), (3) Tweaked SMURF reduction (Tim's image), and (4) the standard CRUSH pipeline. All images are smoothed the same amount, and display matching contours. (Approximately 4' × 5' shown).

All images apart from the SMURF calibration reduction produce a lot of the expected large-scale emission, even for such a shallow exposure. Longer data sets would likely reveal even more of the faint large-scale structure that lies beneath.

Comparing the CRUSH image to the SMURF reductions, the CRUSH image has flatter background than either the standard science pipeline or the tweaked pipeline of SMURF. The wavy background is visible as the bright stripe on the bottom of the SMURF science image, or as significant negatives (contours in the black areas), especially to the East of OMC-1, in both SMURF images.

 

Instrument Characterization

Pointing

There is no published information about the details of the SCUBA-2 array, such as the orientation and position of its subarrays in the focal plane, or the spacing of pixels. Thus, these parameters have been reverse engineered using available data on bright sources. The subarray position angles used by CRUSH are expected to be accurate to a few tenth of a degree for each subarray. The array positions are probably good to 0.1" rms, and the pixel separations are accurate to around 10 mas rms.

The absolute positioning of the array has been determined by matching the position of point sources to the the images from the standard SCUBA-2 data reduction pipeline. The resulting images are thus expected to have accurate astrometry to within 0.1" rms when compared to the official SCUBA-2 pipeline. Thus the systematic astrometric error of CRUSH is negligible when compared to the typical 2–3" pointing accuracy that can be achieved during the observations.

Calibration

The calibration of CRUSH is based on the available public data on Uranus (ca. 30 scans), which is perhaps the best primary calibrator source in the submillimeter bands. Observations at different elevations and in different weather conditions (τ) allow the separation of the instrumental calibration factor (counts to incident power) from the determination of in-band opacity scaling relations.

CRUSH generates images in Jy/beam units. The images are crudely calibrated. The blind aperture calibration of CRUSH is estimated to be accurate to within 7% rms for 850μm, and 23% rms at 450μm. These uncertainties are consitent with an error resulting from the imperfect knowledge of the atmospheric opacities. Specifically, the calibration error at both wavelengths is consistent with a measurement error of approximately 0.01 in the 186GHz radiometer value.

Peak flux measurements are expected to exhibit slightly larger calibration uncertainties, due to variations of beam shape with elevation and focus quality. While the 850μm beam is generally very close to Gaussian, with an effective main beam efficiency (i.e. peak flux/integrated flux) approaching 100%, the 450μm beam has significant sidelobes, which can contain 20%–40% of the integrated point source flux.

The calibration can be improved either by specifying more accurate tau values (e.g. from daily polynomial fits) and/or by using order-of-unity corrections based on observed calibrator sources.

Point Spread Function (i.e. Beam Shapes)

Below you can see 45" × 45" postage stamps of IRC+10216 (same scan as earlier), at 450μm and at 850μm, resulting from the default CRUSH pipeline, but smoothed very slightly for better visual appearance. Contours are at 5dB, 10dB, 15dB, 20dB, and 25dB (from inside out). These shapes probably trace the error-beam pattern of SCUBA-2 in its two bands, although the underlying structure of the not-quite point-source could play a part also.

PSF images
Figure 3. PSF at 450μm (left) and at 850μm (right).

Sensitivity

The typical instrumental point source sensitivity of SCUBA-2 is determined to be ~90 mJy s1/2 at 850μm, and ~350 mJy s1/2 at 450μm. These are average point source sensitivities over the working pixels of the two subarrays, which can be reached by CRUSH in the deep reduction mode.

The sensitivity degrades rapidly with emission scale, due to the strong 1/f noise of the SCUBA-2 detectors, which dominates below ~1 Hz. The larger the observed structure, the more noise one has to put up with (see the README for details). Thus, reaching the maximal sensitivities of SCUBA-2 is possible only for point-like or very compact sources.

Noise Properties

The detectors of SCUBA-2 exhibit a strong 1/f noise profile, with a knee frequency around 1 Hz. Above that, the typical detector spectrum is flat, with occasional resonances (e.g. 50 and 60 Hz pickups etc.). Such spectral features are effectively dealt with by CRUSH. However, the 1/f is ultimately limiting the recovery of scales, which are larger than what the detectors can scan over the 1/f stability timescales of ~1 s.

450μm noise spectra 850μm noise spectra
Figure 4. Representative noise spectra of a handful of pixels from the 450μm subarray (left) and the 850μm subarray (right).

By the end of the reduction, CRUSH is expected to remove all signals, leaving essentially independent white noise residuals. One way to check on this is to look at the pixel-to-pixel covariance matrix of the unmodeled residual timestreams. Below, you can see a section of the full 1280x1280 covariance matrix (with the trace removed). The diagonal stripes are consistent with higher order sky-noise, very similar to that observed by the SHARC-2 camera at 350μm (see Kovács PhD thesis, Caltech 2006). These typically exhibit as spatial correlations on ~35" FWHM scales that cannot be removed without filtering away the astronomical structures on the same scales. However, because these correlations are faint (at a fraction of the detector noise levels), they do not add a significant noise excess. If necessary, an appropriate spatial filtering of the map can be performed post-reduction to remove any noise excess caused by these correlations.

covariance matrix section
Figure 5. A section of the normalized pixel-to-pixel covariance matrix of the residuals, with the greyed-out horizontal/vertical bands corresponding to bad pixels. The diagonal traces are consistent with sky-noise on ~35" scales.

Apart from the structure attributed to sky noise, the covariance matrix appears consistent with independent (non-covariant) white noise. A histogram (log-y scale) of the covariance matrix confirms this (see below). The overwhelming fraction of the matrix elements (the 99% interval is indicated by the red and green markers) follow what appears to be a Gaussian (parabolic on the log-y plot) distribution, which sits on top of a wider distribution that is consistent with the spatial sky-noise discussed above.

distribution of covariance values
Figure 6. The distribution of elements in the covariance matrix (log-y scale).

This serves to convince that CRUSH enables to reach the fundamental noise limits of SCUBA-2, effectively proving that no reduction aproach could possibly go much further or deeper (certainly not by more than a few percent) in the same data.

 
Copyright © 2010 Attila Kovács (attila[AT]submm.caltech.edu)