What is the dynamic range of GMRT?


Dynamic range of a telescope is its "ability" to detect weak sources in the presence of strong sources. Commonly used definition of dynamic range is the ratio of the peak flux to the rms noise in the image. This is true if the rms noise remains same at all regions in the image. If due to systematics, correlated noise and calibration errors deconvolution is not proper and ripples appear surrounding the strong sources in the image, then this definition of dynamic range may not be fully meaningful. Even in this kind of images, rms noise quite far from the strong source will be very small and will look "impressive". The ratio of peak flux to the noise in this region could be large - but we will not be able to detect faint sources close to the strong source. This "source dependent noise" where the noise is large close to the strong source tells us that all is not well.

We have attempted to understand the dynamic range of GMRT at L-band (1280 MHz) using the very strong source 3c48. The source was observed at a continuous stretch for about 7.5 hours at 1280 MHz using 16 MHz bandwidth. Moderate flagging, based on single channel, was applied to the entire data. Initially, the quality of the image did not improve as much as expected. Later, all far away arm antennas are flagged (all ARM-4,5,6). Bandpass was determined for every 2 min and was applied while averaging about 90 channels. The channel collapsed uv data was self calibrated twice with phase only and once with amplitude and phase.

Please click here for the final image. The peak flux of 3c48 at this frequency is 17.22 Jy, but the grey scale of the image covers from 0 to 0.1 Jy (linear) for better contrast. The expected thermal noise is about 55 microJy. Away from 3C48, the noise in my final image is 155 microJy and the peaks are +650 and -690 microJy respectively. This means, the peak to noise ratio is 0.11 million. Very close to 3C48, the noise is 400 microJy and the noise peaks are +2.4 and -2.2 milliJy respectively (ripples). The peak to noise ratio now is about 43,000.

What is the dynamic range of GMRT at L-band? is it 0.11 million or about a third of this? We are able to detect very faint sources far away from 3c48, but we will not be able to detect a source of similar strength (very) close to 3c48. Why did the image quality improve after ignoring far away antennas? By flagging far away antennas the UV coverage became less than 40 kilo lambda, for larger baselines, 3c48 is partially resolved. This may not be the major
reason, because even if 3c48 has structures for larger baselines, the self-calibration will take care of it.

We may be affected by genuine errors with baselines involving these arm antennas. Since good dynamic range can not be achieved when the calibration is not accurate, we need to understand why we are not able to calibrate the data in such a strong source to highest accuracy. There could be baseline errors, bandpass instability, amplitude and phase fluctuations at short time scales (shorter than integration time), mild RFI, etc. could limit the accuracy of final calibration. We are looking into the data for further understanding these problems.

Effect of bandpass

The image quality becomes poorer, if average band pass was applied to the data, instead of estimating the band pass for short intervals of a few minutes. This suggests that all is not well with the band pass. Please visit Nimisha's home page for her work on bandpass instability.

Normalised vs Unnormalised

We have analysed unnormalised data to check whether unnormalised data improves the dynamic range. There was no noticeable difference between the two. We also need to check whether ALCs OFF give better results.