Stop Using Zero Second Bias Frames — Here's Why 0.2s is Better

TL;DR up front: Depending on your camera, taking short exposures can cause your camera to enter a different mode. Often, these modes produce bias values different from those in your long-exposure images. So we recommend you stick to exposure times of at least ~0.2s for your Bias and Flat frames.

Bias frames are essential for calibrating your astronomical images. They subtract the fixed electronic readout noise from your Lights and Flats taken with the same sensor. If you look up just about any guide for taking Bias frames, they'll tell you to set your camera to the shortest exposure it can handle. But this can cause problems. Here's an evidence-backed explanation of why you should set your Bias exposure time to ~0.2 seconds instead:

The Evidence

Here are two autostretched bias frames taken on a QHY Minicam8. The first image was taken with the minimum exposure time allowed by the camera. The second image was taken with a 0.25-second exposure. Everything else was the same, and the sensor cover was installed to prevent any light from getting in:

They're similar, but pay attention to the right side. Do you see the gradient as it gets darker towards the right side of the 0-second (first) image? This gradient disappears if you set your exposure time to 0.25 seconds or longer -- suggesting that it's not dark current (as that would be expected to continue to change for longer and longer exposure times). 

Here are some more discrepancies between the two images, again keeping the 0-second exposure on top and the 0.25-second exposure on the bottom:

Notice how the noise pattern in the 0-second bias is not uniform. Up close, it looks almost like a fabric's weave pattern. 

Lastly, here are the unedited histograms of each image, zoomed in on the same spot:

This shows that the noise pattern on the 0-second exposure has a much wider distribution.

Why Does This Happen

Good question. It doesn't happen on every camera, and some cameras exhibit different behavior (banding, etc) than the patterns shown here. I wish I had a good explanation for you — if you know more, then please feel free to educate us in the comments. But here's what I do know:

I was discussing this online, and a NINA developer said that in the QHY driver, short exposures cause the camera to switch into what the driver refers to as "Stream" mode. This is distinct from the readout mode; the Stream mode is different. They also noted that the ZWO ASI1600 exhibits similar behavior at exposure times <0.2 seconds, but that it's not caused by a similar "stream" mode. 

What To Do

Many (most?) cameras do not behave this way. But it's unclear what this behavior is dependent on: Camera sensor? Driver version? Gain? Offset? USB Limit? Temperature? 

Assuming that you're taking bias frames with the lens cap (or sensor cover) installed, there's no harm in switching to 0.2 second bias and flat frames. At worst, you spend a couple more seconds taking your bias and flat frames. At best, you save yourself from improperly corrected images and nonuniform noise patterns that can throw off noise reduction techniques (which often assume a uniform distribution).

If you test your camera for this issue, do leave a comment to tell us what you find — or if you don't find anything at all.