ATCA Internal RFI (prior to March 2009)

The dominant source of internal interference, prior to March 2009 was from the 128 MHz sampler clocks used to digitise the incoming radio emissions. This clock was phase locked to the observatory clock, so as far as the correlator sees it, it is a very strong, very pure and very precise signal. Historically, the even harmonics were much stronger than the odd harmonics. The 1408 MHz birdie is the 11th harmonic.

Neil Killeen wrote the following report on the matter.

Dealing with Birdies in ATCA data

Neil Killeen, 28 March, 1995

ATCA data suffers from self-interference at integer multiples of 128 MHz; these signals (which are generated in the samplers) are very narrow. However, the frequency spectrum is convolved by the Fourier Transform of the lag weighting function, so in certain correlator configurations these ``birdies'' ring through the spectrum. Because the signal is quite phase stable, it will cause phase centre artifacts if not eradicated. A particularly annoying birdie occurs at 1408 MHz, as it is in the external interference free part of the 20cm (L band) spectrum that we all like to use.

Ideally, we would like to eradicate these birdies by hardware changes. However, this is a non-trivial exercise and cannot be done easily. In the mean time, it is possible to deal with the birdies reasonably successfully during the data reduction in the most common observing configurations.

The character of the ringing that occurs depends upon what spectral setup the correlator is configured to. That is, the lag weighting function changes with the correlator setup.

In standard 128 MHz 32-channel mode, the lag weighting function is a triangle, and its Fourier Transform is a sinc squared function. This sinc squared function has a straightforward character though. If the nbirdie occurs in channel N (and it will be an integer channel because we can only set the band centre to the nearest MHz) then the sinc squared function convolving this signal will have zeros at channels N+/- 2, N+/-4, N+/-6 and so on. All other channels are corrupted by the non-zero sinc squared signal. Thus, if we flag out or drop channels N, N+/-1, N+/-3, N+/-5 and so on, the birdie will be fully eliminated. This causes virtually no loss in sensitivity because adjacent channels are not independent in this correlator configuration anyway. The miriad task "atlod" has been changed to offer a new option to eliminate the corrupted channels in this way. Alternatively, you can flag out the bad channels yourself. Additionally, the ATCA scheduling program "SCHED" now warns you if a birdie is expected to be in your band.

At other bandwidths, the situation is not so straightforward. As we decrease from 128 MHz to 4 MHz bandwidths in steps of 2, the shape of the lag weighting function changes gradually from a triangle to a rectangle (top hat).

The Fourier Transform of the top hat (appropriate to a 4 MHz bandwidth) is a sinc function, rather than sinc squared, and it has zeros every channel (other than channel N of course). Thus, with the 4 MHz bandwidth, the birdie will not ring at all, but will simply show up in one channel that needs to be flagged out as bad. At the other bandwidths between 4 and 128, the situation is somewhere in between, where the zeros of the Fourier Transform of the lag weighting function will not occur at integer channels. In these cases, the birdie will ring through the spectrum in such a away that it cannot be flagged out.

In summary, at 128 MHz and 4 MHz bandwidths, any birdie can be fully eliminated by eradicating the appropriate channels with no significant loss of sensitivity. At 8 MHz bandwidth, the lag weighting function is sufficiently close to a top hat that a birdie will generally not ring substantially. At other intermediate bandwidths, it is not possible to fully eliminate birdies from the spectrum by post processing.

Original: Derek McKay (3-Apr-1995)
Modified: Michael Dahlem (5-Mar-2007)