Observing with LBW for ALFALFA: Background

Version date: January 2012


  • Discussion of the ALFALFA followup LBW observing modes



    In this document, we discuss how we set up the receiver (the "frontend"), the spectrometer (the "backend"), conduct observations and finally, produce a flattened (instrument/sky subtracted) spectrum for observations with the ALFALFA followup observations made with the L-band Wide Receiver.

    The L-Band Wide Receiver (LBW)

    Because the objective of the followup observations is to obtain spectra with lower noise than the main ALFALFA survey at specifically targeted positions, we will use the more sensitive single-pixel L-band Wide (LBW). Although the LBW receiver operates in the frequency range from 1120-1730 MHz, we will insert a filter (called "1320-hipass") to limit it to 1280-1470 MHz to cut down the impact of RFI on our observations.


    The Spectral Bandpass

    Remind yourself what the ALFALFA total power bandpass looks like. The ALFALFA bandpass is discussed in Question 4 of the ALFALFA Introduction to ALFALFA drift scan data (and Scavenger Hunt #2 at the UAT workshops).

    The image to the right (a display during the setup of the routine BPD) shows the bandpass (the response over the full spectral bandwidth) for one beam over a 10-min ALFALFA drift. For ALFALFA, the bandpass covers 100 MHz centered at 1385 MHz. Since we already know the frequency of the candidate signal in the followup observations, we can use a narrower bandpass of say 25 MHz covering only the frequency of the candidate so that we avoid known sources of RFI.. In practice, we will use a set of "standard" frequency ranges so that we can get better statistics on low level RFI (and other annoyances).

    Notice the frequency structure within the bandpass and the sharp followoff at the high and low frequency edges; those are due to the bandpass filters and we'll have the same issues when we go to the narrower bandwidths.

    Click on image to see larger version.


    Basic Total-Power Position-Switched (ON-OFF) Observations

    Like most observations, measurements of the HI line emission from galaxies require removal of the contribution to the noise from the instrumentation, the sky and other sources other than the target. ALFALFA and its followup observations use different variations of the basic "position switching" technique to remove these other (more-or-less predictable, at least over short times) contributions to the power over the bandpass.

    The drift technique adopted by ALFALFA is designed to allow for adequate bandpass subtraction using the drift scans themselves with no manipulation of the data taking stream or electronics during a drift, e.g. with "minimum intrusion". We make use of the fact that most of the time, nearly all of the bandpass contains no signal. In ALFALFA, the bandpass subtraction is performed in the LOVEDATA routine called BPD. A linear fit is calculated along the time dimension for each frequency channel. The rms noise for each channel is calculated, and a record is kept of what fraction of the time sequence deviates less that 2 times the rms. Outliers are then excluded, and a final bandpass value for each frequency channel is taken as equal to the zeroth order coefficient of the linear fit. The adopted "OFF" bandpass then is the normalized bandpass times the system temperature Tsys. Removing the noise contribution in this way works very effectively for sources that are not very large; it does not work for very extended features such as galactic hydrogen. But, for nearly all ALFALFA galaxies, this is quick and easy and quite acceptable.

    The purpose of the LBW observations is to increase the integration time ON-source, so, rather than letting the sky drift across the beam, we actually track the target. Therefore, in order to perform the bandpass/sky subtraction, we need an equivalent OFF-source observation, i.e. one in which we expect the instrument+sky noise at each channel to be identical to that contained in the ON-source spectrum. Thus, we need to acquire an appropriate OFF-source bandpass that can be subtracted for each ON-source observation; we end up then with matched ON-OFF "pairs". This method is very common in radio astronomy and is called "total power" or "ON-OFF" observing.

    Let's define a few terms:
    • Trx(f): The receiver noise temperature at each frequency f
    • Tsource(f): The source noise temperature at each frequency f (during the ON observation only)
    • Tconf(f): The contribution to the noise from "confusion" (caused by lots of weak unresolved sources) within the main beam during the OFF observation; this will normally be negligible and negligibly different from the general contribution during the ON-source observation
    • Tsky(f): The sky noise temperature at each frequency f, which includes the CMB contribution and the galactic background
    • Tspill(f): The contribution to the noise temperature from other sources such as ground pickup ("spillover", atmospheric noise, etc.)
    • Grx(f): The receiver gain (from the all elements, including the backend, mixers, etc)
    • Gmainbeam(f): The telescope gain, i.e. the response to cosmic radiation in the main telescope beam (from the aperture plane until the receiver)

    Click on image to see larger version. (Thanks, BK!)

    The signal then from the ON-source observation is

    ON(f) = Grx(f) * (Trx(f) + Tsource(f)*Gmainbeam(f) + Tsky(f)*Gmainbeam(f) + Tspill(f))

    Somewhere in the middle of all that noise, there will be our HI line source. It will not, of course, be present in the OFF-source observation. Similarly, the signal from the OFF-source observation is

    OFF(f) = Grx(f) * (Trx(f) + Tconf(f)*Gmainbeam(f) + Tsky(f)*Gmainbeam(f) + Tspill(f))

    The normal practice of tracking ON-OFF position switching is to undertake the OFF-source observation immediately after the ON-source one and covering the exact same track on the sky (e.g. in azimuth and zenith angle). In those circumstances, we expect that Trx(f), Tsky(f), Tspill(f), Grx(f) and Gcos(f) will not change significantly between the two observations. Furthermore, under most conditions, Tconf(f) will be negligible and negligibly different from the general noise level during the ON-source observation (except in the case of a strong continuum source; that's a complication for another day...). Hence, the quantity we want, i.e. the derived total power difference spectrum, is:

    Thus, our standard procedure will be to conduct paired ON-OFF observations for each HI target. The first observations, the ON-source scan, will track the position of the target for few minutes (nominally 3). The telescope will then move back to the starting Az,ZA of the ON-scan and expose the OFF-source scan for the same length of time. The track of each scan in the pair will cover the same Az,ZA (the same orientation with in telescope coordinates). We will then construct the total power difference spectrum from each pair as above.

    Following each pair of ON- and OFF-source observations, a set of calibration scans will be taken with the built-in noise diode (the "cal") alternately turned ON and OFF for 10 seconds each; these are the CalON and CalOFF scans. It's a bit of a waste of time to fire the diode for so long, but there is no option to make it shorter, as this is a standard part of the "standard ON-OFF" CIMA mode.

    Bottom line: Each target observation, including both the ON-OFF target pair and the calibration scans, will take about 8 minutes plus slew time.


    The Spectrometer (aka the "Backend")

    For the main ALFALFA survey observations (program A2010), we use the WAPPs as the backend. For the LBW observations, we used what is called the "Interim Correlator". BTW, it's been "interim" for more a decade....

    The correlator has lots of flexibility, actually more than we need, so as always, we will try to "keep it simple". The basic parameters to control, set by the mysterious terms "sbc", "lags", and "subcorrelator" (="sbc") (not to mention "interleaved", "Stokes" and "double Nyquist"), are the total bandwidth per spectrum, the number of spectra (spectra of different bandwidths could be take simultaneously if we wanted to, which we don't), and the number of channels per spectrum. There is also a choice between 3-level and 9-level sampling; this describes the number of levels of encoding of the signal autocorrelation function. The more the better for dynamic range (e.g. in the presence of RFI) and precision. As the AO page says, "9-level operation achieves 96.8% of the signal-to-noise of analog correlation, whereas 3-level achieves 81%." Under some circumstances, e.g. where spectral resolution is important, it is better to sacrifice the 9-level, and doesn't make much difference. These subtleties are left to those who take advanced radio astronomy....

    Like ALFA, LBW samples orthogonal polarizations which are essentially means that we record two independent spectra. So we will set up the correlator to treat the two polarizations independently.

    Another trick learned by young radio astronomers (c.f. H&G 1984, AJ 89, 758) on the Arecibo staff in practically-prehistoric days (to many of you, at least) is to double-up the spectra analyzed by the backend. Even though they are not wholly independent, recording this extra set of spectra (both pols) can recover some of the noise introduced within the backend (amplifiers, mixers, etc).

    We fully understand that some of the very low signal-to-noise ratio sources are spurious. Some of the high S/N ones may also be bogus, especially because of RFI. Hence, we are going to set up the spectrometer to use some of its extra (unnecessary for our needs for spectral resolution and frequency coverage) capability to monitor RFI within the band and/or to look for signals at velocities different from the source candidate.

    For the Jan 2012, we have decided on two observing modes one which we will use for the large number of very low velocity sources (cz < 2000 km/s) and, a second, for sources at higher redshift (cz > 2000 km/s). See the discussion here.

    The Radiometer Equation

    Take a look at the detailed discussion of radiometers under the web site Essential Radio Astronomy by Jim Condon and Scott Ransom at NRAO.

    For us, what we want to know is how the noise in the HI spectra depends on the integration time, and the properties of the receiver/spectrometer setup. To derive the rms noise per spectral element, ΔSrms, in Jy, the relevant parameters are:
    • the system temperature (total noise contribution; see above), Tsys, in K
    • the system gain, G, in K/Jy
    • the bandwidth per spectral pixel (i.e. spectrometer channel, or the final spectral resolution if we smooth), β, in Hz
    • the integration time, τ, in seconds
    • C, an efficiency factor that depends on the correlator sampling level, of order 1, but about 0.8 for 3-level and more like 1.0 for 9-level.

    Note right away, that since the system temperature of LBW is lower than that of ALFA and its gain higher, observations of the same integration time and spectral resolution made with LBW will be more sensitive than those made with ALFALFA. And, with the LBW observations, we will integrate 3 minutes ON-source. So we really are digging deeper into the noise (which is the whole point!).

    Because we integrate for 3 minutes both ON- and OFF-source and also combine both polarizations, the above equation doesn't exactly work. In ON-OFF position switching, the line is observed only half of the time, but the noise is there all of the time. So, the equation, we actually use is per polarization, or when both polarizations are averaged to obtain the final spectrum. Here, τ refers to the total time (ON+OFF). You can read more about this at the Arecibo observer's guide.




    This page maintained by the members of the Cornell ExtraGalactic Group and their friends, the other members of the ALFALFA observing team
    Last modified: Sun Jan 15 15:42:34 EST 2012 by martha