|
Australia Telescope Compact Array |
 |
|
|
||
|
The Compact Array Broadband Backend (CABB) upgrade has provided a new broadband backend system for the ATCA. The maximum bandwidth of the ATCA has been increased from the 128-MHz (in each of two IF bands) to 2-GHz, a factor of 16 improvement. This, coupled with higher level data sampling, has improved the continuum sensitivity of the ATCA by at least a factor of four as well as providing a greatly enhanced spectral line performance, particularly at the higher observing frequencies. The wider bandwidth will allow projects studying spectral lines within a 2-GHz band to be carried out simultaneously.
The MOPS spectrometer in use at Mopra is largely a by-product of the development being undertaken for the CABB upgrade project.
CABB was installed in March/April 2009. Since then, a single mode has been available for general observing, with two 2.048 GHz bandwidths and 2048 channels across each band, corresponding to a 1 MHz spectral resolution. (In addition, a mode with a single 64 MHz channel has been available for VLBI and NASA tracking.)
A wiki and a discussion forum have been set up for observers to discuss matters relating to CABB observing and data reduction.
Zoom mode implementation has been delayed. The correlator section is advanced however the development of channel selection, finite impulse response (FIR) filter and FFT has been more problematic. Data transport issues have also contributed to the delay. A fully commissioned mode is unlikely to be available before the end of April 2010.
The next zoom to be provided will need discussion after the lessons of the 1 MHz mode have been considered. Technically, a 64 MHz mode may represent a challenge compared with 16 MHz and 4 MHz modes.
The Call for Proposals for 2010APR provides the best information available at mid-November. As of early March 2010, while good progress toward zoom modes has been made (see figure above), a zoom mode is unlikely to be available before the end of April 2010. The 2010APR semester is being scheduled based on this assumption.
Observers are reminded that, although correlator cycle times as short as 1 second can be used with the ATCA is special circumstances, the use of the default cycle time of 10 seconds is recommended. In the centimetre bands, the averaging option can still be used (with a default of 3 cycles) to minimise the size of output data-files. One exception is mosaic-ing observations, for which a shorter cycle time may be contemplated; however, please read the Current Issues page for the most recent information for recommended cycle times in mosaic observing.
A spectrum over a full 2GHz is ALWAYS observed and recorded (strictly, 2.048GHz, since all BWs and frequency separations in this document are precisely 2^n x 1-MHz). All 4 polarization products are computed. Auto-correlations are routinely computed as well as the interferometer output. Note that the 2-GHz band edges are defined by an analogue filter (as with the bands of the current ATCA correlator), so a region of roughly 32MHz at the edges of the band should not normally be used.
For this basic 2-GHz spectrum, the maximum resolution is with channels of width 1 MHz; each channel has a 'square' response and thus there are 2048 independent channels across the full spectrum. The basic 2-GHz spectrum will in the future alternatively provide larger channel widths with fewer channels, the other options being 512x4MHz, 128x16MHz and 32x64MHz - see next section. Because the channel shapes are square (with no ringing) Hanning smoothing which is often used to remove ringing, will not be needed. Internal to the correlator, an output of 4096 channels is routinely computed (i.e., with additional channels interleaved with a half-channel shift) so as to allow flexibility in positioning of zoom bands (see below).
When CABB is fully implemented, zoom bands will be available in addition to the 2-GHz spectrum. Within the 2-GHz band, we will ultimately have available, simultaneously, 16 zoom bands. If you wish to use a zoom mode, you will need to consider the alternative outputs from the basic 2-GHz band, in order to accommodate your preferred zoom mode. There will, in time, be four configurations:
In each zoom band, the output will always be 2048 independent non-overlapping channels. Interleaved channels are not computed, but the channels will still be clean 'square' channels, with negligible spillage (no ringing from narrow lines and thus Hanning smoothing will not be necessary to reduce ringing). As described above, the zoom bands will have a choice of 4 widths: 1MHz, 4MHz, 16MHz, 64MHz. Each of the 16 zoom bands MUST be the width of the channel width selected for the basic 2-GHz spectrum.
The separation of selected zoom bands within the 2-GHz band can be any integral multiple of half the zoom band width. Subject to this single constraint, we will be able to distribute them over the 2-GHz band any way that we wish. In particular, we can have an overlapping bunch at each of several different parts of the 2-GHz spectrum. (In the overlap region of overlapping zooms, the output from the spectrometer will be seamless and not need further blending by the user.)
The method for selecting zoom modes has required modifications to the atcasched software that have been implemented and will be tested before 2009JULS commences. Some preliminary advice on how to select zoom modes follows. More examples will be added to this description later.
If you calculate your desired spectral resolution (=channel separation in view of the square spectral response), and it is 1 MHz or more, then you do not even need the zooms. This may be the case for much continuum work, and even for extragalactic 3-mm spectroscopy.
If 1-MHz resolution is not adequate, then you investigate which zoom width (with 2048 channels) will just achieve the desired resolution. You then decide how much contiguous frequency coverage is needed, and thus how many overlapping zoom bands are needed. If you need less than 16, then there are spare zoom bands to do additional science.
Example 1. Consider OH observations at frequencies of 1665.4018, 1667.359, 1720.530 MHz, 1612.231 and HI at 1420.4058 MHz, Suppose we require for each transition a frequency resolution of 0.5 kHz (to study narrowband maser emission and narrow HI absorption) and velocity coverage of about 180 km/s corresponding to 1 MHz. This combination of coverage and resolution is provided by each zoom of the CFB 1M-0.5k configuration.
All lines lie within 2 GHz. Good.
There will for 2009 OCT be limit of 4 zooms per IF, and since we will need more than 4 zooms, some will be at freq 1 and the others at freq 2 (which can be set to be the same value as freq 1).
Settings for the centre frequency at each IF are constrained to integer MHz and, although not fundamental, this is likely to remain so. Note that at 20cm, our range of center frequency is limited from 1701 to 1805 MHz due to CABB lo requirements. Since the band is 2 GHz wide, we may as well use the recommended standard setting of 1750 MHz.
Suppose we set the freq 1 to 1720 MHz to center our favourite line. This might not be a good idea since there are birdies located one-quarter (channel 513). one-half (ch 1025, centre) and three-quarters (ch 1537) across the 2-GHz band. So we're better off sticking with 1750 MHz.
In the new sched, after specifying in 'freq1 setup' the desired freq 1 of 1750 MHz, you will notice that, if this is selected as a line frequency, its channel number is now labelled 2049, not 1025. 2049 is the centre channel of the 4096 channel centres spaced at 0.5-MHz intervals which are available for fine positioning of the zooms (and allowing concatenation of adjacent zooms). So the selected line must now not correspond to ch 2049 (if we wish to avoid the birdie).
We use the velocity calculator to see which frequency we prefer to centre our first line and, as an example, suppose this to be 1665.4 MHz. Entering this for line 1 will round it to 1665.5 MHz and show that the appropriate channel for the zoom centre is 2218. A width of one zoom will be adequate, with only a small shift of centre velocity (by 18 km/s) from the desired one. Entering 1667.4 and 1720.5 will show appropriate zoom channels for these two transitions of 2214 and 2108. Note that in this band the frequency decreases with increasing channel number. The scheduler takes these spectral inversions into account. If we desire a 4th zoom centre at 1612.23 MHz it will be rounded to 1612.0 or 1612.5 and a single zoom of 1-MHz will not be well-centred, so we may find the velocity coverage is not adequate and require 2 adjacent zooms. But we only have 1 of our 4 zooms left at freq 1, so we now choose freq 2.
At frequency 2, also choose centre at 1750 MHz (say).
Enter 1612.23 as the line frequency and rounding will show 1612 is at channel 2325. 1612.5 is at ch 2324. If we choose a width of 2 zooms, the channels will be 2324 and 2325 and the output will be centred at 1612.25 and cover 1.5 MHz giving vel width 279 km/s in a single seamless spectrum. The scheduler will always try to get the center of the concatenated spectrum as close as possible to the sky frequency of the line. You may see the channel number jump by 1 when switching from odd to even zoom width and vice versa. You can specify the sky frequency using the built in velocity calculator or by typing it into the line frequency box for the appropriate zoom. The number displayed will be rounded to the nearest possible spectrum centre, but internally kept at full precision.
We can now choose another line fequency, e.g 1420.4058 and can decide whether the rounded frequency gives velocity coverage sufficiently well centred, or prefer a width of 2 zooms, where the centre frequency of the combination is 1420.25MHz.
Example 2 (for zooms that will not be implemented until at least 2009oct). Consider the 7-mm case to mimic observations reported in ATNF newsletter #62, 2007June, page 16, which were made by Maxim Voronkov with the current (pre-CABB) correlator and used a channel separation of 64kHz (i.e., 256 channels across 16MHz and thus strictly 64/1.024 kHz).
With CABB, we can use zoom bands of 64 MHz which will give us channel separation of ~32 kHz. This is a factor of 2 better than we need, but is the nearest one available.
If we want to cover as large a contiguous band as possible for a line search, we set the 16 zoom bands to overlap, and achieve a coverage of 8.5x64 MHz = 512 (+32) MHz.
We thus need 4 observations to cover the 2-GHz band.
For comparison, with the old correlator, a 2.5-GHz band was covered with 96 observations (making use of both IFs) and had spectral resolution worse than CABB by a factor of 2.
In this simple case, we estimate that CABB is faster by a factor of at least 19 (and when 2 IFs are implemented, will be faster by a factor of 38).
Enhancements are being made to MIRIAD to enable CABB data to be reduced in a similar manner to the existing ATCA data reduction.
First fringes with the interim CABB system were obtained on 23 May 2008 between CA02 and CA03. On 12 August 2008, a three-baseline system was first demonstrated with the addition of CA05, with fringes obtained on a single polarisation with a 128 MHz bandwidth, enabling further testing to be carried out. In early November, full 2GHz autocorrelation spectra were being obtained from CA01 to CA05. In early December, the first CABB image was obtained (5 antennas, single IF, 128 MHz bandwidth, full polarisation). Comparison with data from the existing correlator indicated the CABB system was working well. In late December 2008 and early January 2009, the CABB bandwidth was gradually being increased toward the full 2 GHz, with additional imaging observations being made to test the flow of data from the correlator and through miriad. In February 2009, the first observations utilising the full 2048 channels were made with the interim CABB system (five antennas, single IF). On 23 February, CA06 was taken off-line and preparations started for the full CABB installation. A six-week shutdown for the full CABB installation ran from March 2nd to April 14, 2009, and be followed by a week of scientific commissioning, with scheduled observing starting on April 22. The CABB shutdown diary provides updates on progress with the CABB installation. CABB fringes were progressively obtained to more antennas as the shutdown proceeded, with first CABB fringes to all six antennas obtained on April 1st.
Emails sent to ATCA PIs for this the first CABB semester, 2009APR, on March 31st and on April 27th with more details about observing with the new system.
The 2009APR semester was the first to offer the full 2-GHz CABB bandwidth in both frequency bands and on all six antennas. Due to the staged implementation of CABB modes, the 2009APR semester for the ATCA ran from April 22 to July 15. For 2009APRS, a single mode was available: this has 2.048 GHz bandwidths with 2048 channels across each band, corresponding to a 1 MHz spectral resolution. Proposals not completely scheduled in the 2009APR semester had to be resubmitted for consideration in the 2009JUL semester.
For 2009JULS, it was expected that the first "zoom" modes will be available, offering significantly higher spectral resolution. The implementation of zoom modes has been delayed, however, and proposals requiring zoom modes have been postponed.
The CABB project can be broadly divided into four parts: a wideband conversion system, wideband sampling, wideband optical data transmission and a complementary wideband filterbank/correlator. An overview of the sampling and data transmission is given in a February 2007 ATNF News article. The filterbank/correlator is outlined in an October 2006 ATNF News article. A description of the CABB upgrade is given in a paper by Ferris & Wilson (though note some details have changed in the final implementation of CABB).
This page will be updated regularly.
Modified: Phil Edwards (5-Mar-2010)
| © Copyright CSIRO Australia, 1994-2010. Legal Notice and Disclaimer, Privacy Statement. |