CCD Readout Noise Study

Starting in mid-February 2014 we began to archive individual CCD readout noise statistics as well as backplane averages in the the telemetry DB. For each CCD the recorded noise value is the average from amp A and amp B. It is obtained as the mean of the row by bow averages in a subsection of the overscan regions. In order to avoid charge blooming effects caused by bright objects predominantly near the shift registers, the first 200 rows of each amp are excluded. It's probably easiest to just look at the code segment from in the IB product:

def rauschen(p):
    max_row = 3900
    min_row = 200
    max_col = 50
    min_col = 12
    for i in range(len(p)):
        if i == 0:
        if i>62:
            max_row = 1850
        na = p[i].data[min_row:max_row,min_col:max_col].std(axis=1).mean()
        nb = p[i].data[min_row:max_row,-max_col:-min_col].std(axis=1).mean()
        bkp = str(p[i].header['FPA'].split('_')[1]).lower()
        n.append({'CCD' : str(p[i].header['EXTNAME']).lower(),'BKP' : bkp, 'N_A' : na, 'N_B' : nb})
    return n

The noise information is available in the ccd_readoutnoise table of the telemetry DB. We have backfilled this table for most of the images that were still available in the 3-day store at CTIO.

Telemetry Plots

For now, i.e. before we complete a more detailed study here are just a few characteristic plots from the telemetry database.
If anybody is interested, the complete noise file (all CCDs and the backplane averages for DES season 1 (2013-8-31 - 2014-2-23) is attached to this page.

The following plots show the entire 1st season (plus a few extra nights) for different exposure types. The left plot includes backplanes 1, 3, 4 and 5 and for the set of plots on the right backplane 6 (focus and alignment) was added. The November noise incident (sync cable) is easy to find.

Zeros (bkp1-5) Zeros (bkp1-6)
Flats (bkp1-5) Flats (bkp1-6)
Darks (bkp1-5) Darks (bkp1-6)
Objects (bkp1-5)

The readout noise for backplanes 1, 3 and 5 is fairly stable around 1.5 ADU. In most plots the difference between these backplanes is hard to see as the dots/datapoints overlap. Of the 3, backplane 1 has the largest number of outliers. Backplane 6 is stable as well but the readout noise is slightly higher, around 2.5 ADU. The readout noise on backplane 4 is not only higher than for the other backplanes but it also changes with time. Actually, as we will see below it changes in the course of a night. This poor(er) noise performance seems to have been present for the entire season. No step indicating the onset of the different noise behavior can be seen in the data.
In general, the larger noise variations and large value outliers are expected for the object exposures since nearby bright objects affect the overscan regions and the simple algorithm we are using cannot distinguish a gradient from larger noise.

The following set of plots shows the noise performance for a few more or less randomly selected nights. All exposures for the particular SISPI instances are included in the plots. In most cases we start with zeros and flats, followed by object frames and the morning calibration exposures. For the night of Dec. 13 we only have zeros and dome flats and for Jan. 14 the noise data exists only for object exposures. The behavior of the bkp4 noise for these two nights appears to differ. We will look at a few more nights from the Dec-Jan time frame to figure out what - if anything - has changed (see plot below).

2013-8-31 (1-5) 2013-8-31 (1-6)
2013-10-28 (1-5) 2013-10-28 (1-6)
2013-12-13 (1-5) 2013-12-13 (1-6)
2014-1-14 (1-5) 2014-1-14 (1-6)
2014-1-29 (1-5) 2014-1-29 (1-6)
2014-2-22 (1-5) 2014-2-22 (1-6)

The plot below shows the noise behavior just for December and January to check if something might have changed for backplane 4 (see previous plots for DECam_20131213 and DECam_20140114). That does not appear to be the case. Some increase in noise can be seen for Dec 28. This is probably related to work done by Marco et al on the CCD voltages. See eLog entry 150076 for details:

In the afternoon telops did some calibration work while varying the voltage on the CCD chips. 

Readout Noise (Dec-Jan)

All of the backplane 4 noise lines appear to start out more or less flat, i.e. constant noise, and have a fairly sharp feature where the jitter increases. The following plots show that this increase in variance coincides with the start of object exposures.

BKP4, 2013-8-31 BKP4, 2013-10-28
BKP4, 2014-1-29 BKP4, 2014-2-22

So what is different for object exposures? We can check if our noise algorithm is to blame by looking at individual backplane 4 CCDs. If the increase in jitter is caused by bright objects close to the shift registers we should see single CCD outliers. This doesn't appear to be the case. The noise on CCDs N4 - N7 is coherent.

BKP4 CCDs (N4-N7), 2013-08-31 BKP4 CCDs (N4-N7), 2013-10-28

What else could cause this bkp4 noise behavior in object exposures? Two things that are different for object exposures are: the guider is running (for exptime > 20 seconds) and the telescope could be slewing during readout. But as is demonstrated by the next two plots neither possibility seems to be the source.

BKP4, 2013-8-31, Exposure Time BKP4, 2013-8-31, Slew Angle