Attenuation and strong sources

General description

Recent test observations of the powerful source Cygnus A have revealed a remarkable difference in the apparent flux density of that source between the 'B/D' and 'A/C' IFs. In short, the EVLA antennas are not properly adjusting the gain of the electronics -- a necessary action for the VLA correlator to provide correct estimates of the correlation coefficient -- with the result that for both IFs, large errors in the computed visibilities can result. The problem will be corrected with improved procedures for adjusting system gain, and when updated hardware is installed.

The current situation is a bit complex, and is described in some detail in Section 'Explanation of the Problem' below. Those who want to know the full story can read on. Those who are more interested in 'how bad it is' should skip to Sections 'How bad is the problem?' and 'What observations are affected'.

Explanation of the Problem

To provide correct estimates of the source visibility, we must have both a correct measure of the correlation coefficient and a correct measurement of the system temperature. A minimum condition for both is that the full electronics path be operating linearly over the range in powers provided by target sources. For the VLA's correlator, the linear range is very narrow, so there are gain adjusting circuitry both in the antenna (to prevent saturation in the waveguide and backend electronics) and before the correlator. The soon-to-arrive WIDAR correlator has a much wider linear operating range, so the need to make fine and continuous adjustments to the system gain is much relaxed.

In the meantime, however, we need to replicate on the EVLA what the VLA antennas are doing for gain adjustment. The back-end electronics with the final AGC remain in the signal path, while at the front end, the T304s have -- for the A/C IFs only -- a variable attenuator which can be adjusted on the basis of a total power detector's value to decrease the power level provided to the 8-bit sampler, so as to not exceed the gain adjustment range in the backend electronics. The attenuator setting is adjusted on the basis of the power measured in a ~1-second duration over 1GHz bandwidth, taken about 2 seconds after the subreflector has completed any necessary motion. The intention is to permit correct level changes to be made following a band change, as the different receivers produce rather different power levels. Note that there is no adjustment of the attenuators for change of power due to change in source or sky brightness. For nearly all target sources, this is not a problem, as the variation provided by such sources or sky is only a few Kelvin on top of the ~30K system temperature -- and this can easily be handled by the back-end AGC.

For the B/D IFs, the necessary capability for automated attenuator settings is not yet available, so a 'table lookup' is used -- generally sufficient when changing bands.

A significant problem with the current system can arise when we attempt to observe an object or region which contributes much more power than the electronics noise power. Such an object is Cygnus A, which at L-band adds about 150K in equivalent noise -- roughly multiplying the total power by a factor of ~6. This is not a problem for the 8-bit samplers -- as the increase corresponds to about 2 bits -- but can be a problem for the VLA's backend electronics, especially if these are running at a higher power level -- close to where some component saturates. The expected effects differ for the two IFs:

  1. On the A/C side, because the attenuator settings are not aware of antenna location, an incorrect (or no) adjustment of the attenuators will result if the antennas are not on-source within ~2 seconds of the time when the subreflector has ceased moving. This will likely happen if the preceding observation is of a calibrator. If, however, the preceding observation is of the same source (perhaps at a different band, perhaps not), the correct power will be seen, and the appropriate correction made.
  2. On the B/D side, the wrong levels are guaranteed, as the system currently knows only what the table values say.

How bad is the problem?

To answer this question, Cygnus A was observed at 1415 (A/C) and 1455 (B/D) MHz, with calibration on a cold-sky source. Each source (both the calibrator and Cyg A) was observed twice in succession, to enable estimation of the effects of incorrect power level setting. This was done both in spectral line and in continuum. Six cycles of 'cal-cal--source-source--cal-cal' were done to determine how repeatable the effects are. The calibrator employed was 2007+404 -- located only about 1 degree away from Cygnus
  1. Results from A/C IFs

    As expected, the observed correlation coefficients on Cyg A for baselines including one or two EVLA antennas rose dramatically between the paired observations of Cyg A. (No VLA-VLA baseline showed any problems). The magnitude of the jump varied considerably from antenna to antenna-- from no change at all to factors of nearly 10! The latter values indicate electronics components -- presumably the VLA's backend -- which have been pushed strongly into compression. This interpretation is supported by the observation that every jump in correlation coefficient following the implementation of the correct attenuator setting was accompanied by a decrease in measured system temperature. This effect is not real, but due to compressed electronics depressing the incremental noise power from the switched cal, resulting in an overestimate of the Tsys. (Curiously, this helped offset the error in the estimated visibility!)

    Only about one half of all EVLA antenna/IFs showed a jump in correlation coefficient. And an antenna/IF which changed in one observation did not necessarily change in any, or all, of the remaining five! This somewhat randomized behavior was not expected, but can be plausibly explained by noting that the calibrator is so close to Cyg A that the 2-second gap preceding the total power determination might have been sufficient to enable the antenna to get close enough to the source for the attenuator adjustment to be made in some cases.

  2. Results from B/D IFs

    For the B/D side, the situation is very straightforward: With a single exception (see below), the correlation coefficients for EVLA-included baselines are a factor of up to ten below those of the A/C side (as measured by the 2nd half of the pair observations, when the attenuator levels are correct). Only EVLA antenna 17 deviates from this rule. For a reason as yet unknown, the correlation coefficients on the B/D IFs for this antenna are a factor of 2 to 3 higher than those in A/C, even after the A/C amplitudes were adjusted to (presumably) the correct values.

  3. Effect on Flux Scale, Closure and Imaging

    The underestimates of the correlation coefficients are a result of electronics saturation. At a minimum, these will cause significant underestimation of the source total flux by a factor very difficult to predict. It is unclear if the closure relations are affected. If not, then self-calibration can correct for the error, providing the total flux is known in advance. This will also greatly improve the image quality. If closure is affected, the data cannot be corrected by antenna-based methods, and the images will be severely degraded. In all cases, the sensitivity is strongly affected, as we are in effect reducing the depth of the sampling to something close to zero bits.

What observations are affected, and what can be done about this?

Only Cyg A was utilized in this test, but we estimate that the problem will potentially affect any observation which increases the backend power (as passed by the D-rack filters) by more than a factor of two or three. For continuum sources at L-band, this means objects of more than ~500 Jy -- Cas A, Cyg A, Tau A, Sag A. Objects near the galactic center could also be affected, due to the high brightness temperature of that region. (This effect is due to total power in the beam, not to high correlation). In spectral line, the effect is diluted by the ratio line-width over the 1 GHz over which the total power is measured, so even masers of 10s of thousands of Jy are unlikely to be affected.

Without changes in the software or hardware, the effect can be avoided (in the A/C IFs only) by adding a 'dummy scan' of short duration prior to any source for which a large change is expected in Tsys: Following a calibrator when going to one of the objects named above, and after one of those named objects when returning to a calibrator, or cold sky.

It would be better if the software could be modified to have the system wait for the antenna to get on source before the power level is taken. This should remove the problem for A/C IFs now, and for the B/Ds when the new electronics are installed -- some time in 2009.




Modified on Tuesday, 29-Jan-2013 13:57:48 MST by Gustaaf van Moorsel