I'm performing several dark current measurements to characterize my ZWO ASI1600MM Pro. Taking all images in RAW16/.FIT format.
The results are very inconclusive from my first run comparing various temps and exposure times. I'm keeping GAIN and BRIGHTNESS at 0, as I didn't want to add any bias.
Although, now I'm convinced the data looks so strange because the 0 offset is causing the low-signal images to be "clipped" at 16DN (but divide by 16 because it's a FIT image, so this = 1 DN), and there is supposedly data that is below this floor.
I believe I need to modify BRIGHTNESS or PEDESTAL to avoid this clipping. From my research, these seem the exact same to me. Does anyone have experience with the affects of either?
An offset is required. Otherwise you WILL clip to black. The bias offset is the BRIGHTNESS setting, and that MUST be greater than zero. For the ASI1600, I recommend setting it to 50 and forgetting about it.
Why is the offset required? Because read noise causes random deviations from the "baseline mean" of the camera, which IS the offset. In other words, if your offset is zero:
...| | |
1| | | | | | , | , | |
You might have "noise" like the bars above the zero line in this little graph. Those bars represent the random oscillations causing the signal to DEVIATE from that "mean" of zero. Note, however, that noise is random, and will deviate above and below the mean. Here, there is nowhere for negative deviations from the mean to go...they clip to zero. If we instead have a non-zero bias offset:
...| | |
11| | | | | | , | , | |
9|`| | || | | | ` | | |
...| | | | |
You lave room for all the deviations from the mean (the offset) both positive and negative to occur without clipping. A bias offset is essential. Not only is does clipping increase the noise in your result, it prevents proper calibration from ever being possible. Because in addition to the random oscillations due to read noise, the bias offset itself, as well as any dark offset, is actually non-uniform itself across the sensor, and due to the differences in pixel response, the DFPN itself may cause pixels to clip.
A PEDESTAL is similar to a BIAS OFFSET, however they are not the same thing. A Pedestal is added during pre-processing to ensure that ONCE the bias offset is subtracted (either by subtracting a master bias, or by subtracting a master dark), you ensure that you still won't clip any of the data. Calibration of each pixel must be performed as a single operation, with real numbers that allow negatives to temporarily exist. So usually, during calibration, the bias or dark will be subtracted, leaving negative numbers, then the flat will be divided, potentially lifting some of those negatives back up (potentially), then finally the output pedestal is added to shift all remaining negative values positive. A pedestal should be large enough to ensure that any pixels that would have clipped once the bias offset is removed will remain positive after the pedestal is added. The easiest way to ensure that, is to make your pedestal the same size as your bias offset. So, if the bias offset is 50 ADU, then the pedestal should be the equivalent of 50 ADU. For a 12-bit camera like the ASI1600, you usually need to convert the bias offset in ADU, to a 16-bit number...so 800 DN. An output pedestal of 800 16-bit DN will guarantee that you will never clip anything during calibration.
Edited by Jon Rista, 27 January 2021 - 09:48 PM.