When testing the new fix to the Tsys Variations
, it was noted that in position switched data a dimple exists in the center of strong continuum sources, as shown by the sharp drop in antenna temperature just above 8.58 GHz in the figure below. Note that position switched data use two scans, one on source and one off source, which are calibrated using the formula ([on/off-1]*Tsys - where Tsys is a scalar).
This dimple is fixed in channel space, i.e. it is located at the center of the bandpass regardless of LO setting or VEGAS mode (figure below - raw tp scan, bin number vs. counts).
Upon inspection of the individual scans, it becomes obvious that the negative dimple in position switched data is actually caused by a bump in the individual scans which does not appropriately calibrate out. To understand why it does not calibrate out, Ron Maddalena looked at the height of the dimple vs total power around the dimple, and found a surprisingly nice linear relation, as shown in the plot below.
This non-zero y-intercept in the fit of the line is likely the cause of the dimple in the position switched data. The dimple does not appear to be resolution dependent. Below shows the dimple in two total power scans in modes 7 and 9 (L1/LBW1 100 MHz bandwidth, 32768 and 131072 channels, Fs=2e8). The baseline has been fit and removed around the center of the band, leaving a decent approximation of the dimple. The data have also been smoothed with a 100 bin boxcar. The height and width seems reasonably comparable between the two modes.
When using a different bandwidth (and thus different sampling frequency), the scale of the dimple changes drastically. Below shows the same two dimples with the dimple from mode 6 overplotted (L1/LBW1 187.5 MHz bandwidth, 131072 channels, Fs=3.75e8). The mode 6 dimple is much larger.
One of our initial thoughts was that this could be a windowing function acting on the center Fs/4 spike. As an intial check of this, I looked at the spike height vs input power, though this trend goes in the opposite direction as the dimple height: The spike intensity decreases with total power in the spectral window (each color is a different scan, the individual dots represent individual integrations within the scan - note the logarithmic x axis):
This was a troubling relation for me. If I understand the Fs/4 spike correctly, it is the clock running the four ADCs leaking into the data. This to me would suggest that the spike height should be constant. I would also think that the spike height and the center bump would follow similar trends, but not the possibly linear and logarithmic trends shown above. What could be the cause of this bump? Could it be a drift in the clock driving the ADCs (though would that create the DC offset seen in the dimple height?)?