Документ взят из кэша поисковой машины. Адрес оригинального документа : http://www.naic.edu/alfa/ealfa/meeting1/minutes/hipass.html
Дата изменения: Mon May 8 23:01:35 2006
Дата индексирования: Sun Dec 23 01:56:20 2007
Кодировка:

HIPASS Results and Lessons

Next: HIDEEP/HIJASS Up: Minutes of the 1st Previous: Pulsar Surveys

Minutes of the talk on :
HIPASS Results and Lessons

Lister Staveley-Smith (ATNF)

A more detailed version of this talk by the speaker is available in HTML, PDF or PPT format.

The HI Parkes All Sky Survey (HIPASS, see for example, the Barnes et al. 2001 paper) used the Multibeam receiver on the Parkes 64-m telescope. The outline of this talk:

HIPASS results
Bright Galaxies and the HIMF
The ALFA advantage: speed, depth, resolution
Things to get right: on-line & reduction software, observing strategies

The HIPASS Bright Galaxy Catalog

HIPASS has detected about 7000 galaxies south of declination +25 deg. The BGC, which is about to be published, is defined to contain the 1000 HI-brightest galaxies with Dec. < 0^o and a minimum peak flux density S > 116 mJy.

Previously uncatalogued objects included in the BGC were presented in Ryan-Weber et al. 2002; they number 87 out of the 1000 galaxies. Of the BGC galaxies, 138 have redshifts for the first time. The newly identified galaxies mostly look like smudges.

Zwaan et al. 2003 (see astro-ph/) use a bivariate maximum likelihood technique to measure the space density of galaxies using the BGC. In this method, the detectability is not just based on the peak flux but also the velocity width. This robust method is called the 2 dimensional stepwise maximum likelihood method, and is insensitive to the effects of large scale structure. They derive a HIMF with a faint-end slope alpha of -1.30.

To use this method, we look at the variation of M_H with log W. If you integrate over log W, then you can solidify the mass function below 10⁸ solar masses. Zwaan et al. find a slope of -1.3, fairly consistent with previous determinations. It is quite a bit lower at the low mass end than Rosenberg & Schneider found. We really need ALFA to address that faint end.

The ALFA Advantage

Speed: ALFA will be about 4 times faster than Parkes for mapping. This estimate accounts for the smaller beam size, the number of pixels, and the available solid angle.
Sensitivity:
- HIPASS (with an integration time of 250 sec per pixel) was complete only for M^* galaxies at cz < 4000 km/s
- ALFA (with the same intergration time) would be complete to 15,000 km/s
- A deep ALFA (few hours per pixel) could go to cz ~ 0.1 to 0.15. That would be really interesting to look at possible evolution of the HIMF.
Resolution: For a filled aperture, the linear resolution is always the same at the maximum survey depth $D_{\rm max} \propto (A/T)^{1/2}$ .
Large solid angle: We have looked at the cosmic variance for HIPASS BGC dwarfs and asked: does it vary? While HIPASS covered a large solid angle, the number of 10 Mpc cells nearby is still very small, making it hard to detect many objects at the lowest masses.

So ALFA has the potential to make great contributions.

Things to get right:

Correlator: The ideal spectrometer would have:
- 4 to 8 bit sampling
- 200 MHz bandwidth
- 8192 channels
- 2 polarizations
- 7 beams and
- interference
The present plan is only for 1.5 bits, 100 MHz, 4096 channels, 2 polarizations, 7 beams. That's rather disappointing.
On-line software: Should hide its complexity but allow much flexibility.
Reduction software: Should run in real-time.
- At HIPASS there was no distinction between on-line software and reduction software.

An illustrative example:

Go give an example of what we are talking about for ALFA, suppose ALFA remapped a 64 square degree HIPASS field. Let's adopt the HIPASS sensitivity (13 mJy/beam) and require Nyquist sampling of the sky with all 7 beams. Assume 100 MHz bandwidth, 7 beams and the planned 4096 channels. This gives some interesting numbers.

The integration time would be only 1 sec per beam for a total time of about 2.6 hours. To achieve 1 sec beam, the drive rate would have to be 20 deg/min, which is not possible with Arecibo. The correlator dump time would 60 ms, again not possible. The minimum data volume would be 36 GB and the minimum data rate would be 4 MB/s, 200 times the HIPASS data rate.

The Parkes LiveData software would not be able to handle this. The JCMT/DRAO ACSIS 32-processor Linux cluster might. So we really have to worry about this sort of data reduction bottleneck!

Interference mitigation

A promising method for dealing with RFI is the technique of post-correlation RFI cancellation with reference horns as discussed in the Briggs et al (2001) paper.

Things become interesting when the PSF varies with position, beam orientation, frequency, beam number, etc. Then you need deconvolution, and that is not a trivial matter.

Questions/Discussion:

Riccardo	What is H_o used to derive the HIMF?
Lister	I want to say 75 but I'm not sure. Often people use 100 and then scale, but here I think 75 was used.
Phil	In order to do the cross-correlation for interference mitigation, you need lots of subcorrelators, true?
Lister	Yes.
Riccardo	Where do you get 60 ms?
Lister	For Nyqust sampling along the scan directions for each of 7 beams, you need to scan this fast to repeat.
DJ	What is the limitation on the LiveData data rate?
Lister	GLISH is quite slow. It can only get up to about 20 KB/sec. A Single LINUX processor can probably only handle rates less than 1 MB/sec.
Steve	Is there a real problem if you don't Nyquist sample all beams?
Lister	Then you would have a problem for (a) uniformity, and (b) RFI. You don't want to have to examine everything by eye. Rather, you want to be able to do robust processing.
Eli	On your factor of 4 for ALFA's improved speed of single pointings, I can see only a factor of 2. Where does extra factor come from?
Lister	You get a factor of 2 in time, but you also divide by reference spectrum, so you introduce another root 2, but then with multi beams, you gain back.
Eli	Is interference mitigation incorporated into your correlator design?
Lister	You need cross-correlation between reference horns and each beam on the array. That multiplies the correlator requirement by 3.
Phil	The problem is that there are not enough of them. If you want to do complete RFI mitigation you would need 3 times more WAPPS. The functionality is in the boards, but you need more of them.
Eli	If you have to stick to the 100 MHz current design, how will we do it?
Phil	I would not use the word "if".