Reference Pixel Subtraction

Reference Pixel Subtraction

In an attempt to improve linearity of the ramp and noise over the full set of scans (rather than having to ignore the first one) reference pixel subtraction was explored. Initially the simple step of averaging all reference pixels in a scan from both the leading and trailing stripe of the array gave no real improvement. The second attempt was to subtract row by row the average of the 8 reference pixels in that row. This improved the linearity of the ramps, as shown below, but worsened the noise. Scan average is plotted vs. scan number in the data cube.

First using a set of noise data, 5 ramps of 8 scans for Fowler-4 noise, that was taken during grounding scheme optimization at 2V applied bias, the initial linearity was poor (data 413 from CD016) and the noise was 1.336 DN (47 electrons) full frame, all 5 scans, or 1.168 DN full frame only the last 4 scans.

By subtracting the reference pixels row by row, the observed linearity in the following plot was improved, especially for scans 1 and 8 of each ramp, but the noise increased to 1.56 DN (55 electrons) full frame all 5 ramps.

Now trying a different set of data taken at zero applied bias to eliminate any photon generated charge noise, the initial data is a 7K run in a set of noise-temperature measurements (data 1794 from CD016). The initial ramps had larger (0.4 DN) offsets of the first and eighth scans in each ramp as shown below, and the noise was 1.336 DN.

After subtracting row-by-row reference pixels, the offsets of the first and last scans improved a lot but the noise increased to 1.544 DN (54 electrons).

So, although the scan averages fall on a better line, the fact that we are not averaging thousands of reference pixels before our subtraction makes the subtracted data noisy and contributes to the overall noise significantly.