9 replies to this topic

[Balcony_Astronomer]

I recently purchased the camera above from Germany. I did not try it yet, but I am noticing what I think is abnormal noise in my dark frames. It would be great if somebody could take a look and either confirm my thoughts or convince me that my camera is actually working as expected.

 

For this test I am collecting 180s dark frames at -10C (or as much as it can cool in my apartment) in APT and Nina. I tried different gain and offset setting, HCG and LCG. "low-noise" mode is on. I tried with the native and Ascom driver.

You can find them at this link, I hope file names are self explanatory.

https://www.dropbox....t=w8lyrouz&dl=0

 

I would expect my dark frames to be narrow and to be centered at very low counts. Instead I observe a prominent tail on the high-counts side. The extent of the tail changes with temperature and setting used. I attach a screenshot where I compare a dark frame of 180s taken with my old ASI2600MM and one taken with the new camera. I think the difference in the histogram tail is evident. Both dark frames are taken with gain 100, despite they likely indicate different conversions. For the ASI2600MM, gain 100 is a very popular value. For the ToupTek I could not find a recommended gain, but I noticed this is the gain where the tail is weaker.

 

Next, I count hot pixels. For this I use the simple script provided here: https://pixinsight.c...t-pixels.16830/

 

This calculates the number of pixels above a threshold:

// RGB/K:
nhp += $T>hpt;
$T>hpt;

// Symbols:
hpt=0.8;nhp=global(+,0);

They are 5 for the Touptek and 9 for the ASI. Impressive, also considering there are 2.4X more pixels in the Touptek.

 

If I instead calculate the pixels that are above the median of neighboring pixels with:

// RGB/K:
nhp += $T>hpt*medfilt($T);
$T>hpt;

// Symbols:
hpt=3;nhp=global(+,0);

I obtain a staggering 7725 warm pixels for the Touptek versus 32 in the ZWO. This is consistent with the visual appearance of the dark frames, which is plenty of non-saturated white pixels.

 

What do you think? Is the camera defective? Am I using the wrong settings?

I reported my doubts to the seller on the same day I received the camera. Eight days have passed and so far they could not reply on this issue.

Thank you in advance for your help on this matter.

 

 

Attached Thumbnails

• 

[sharkmelley]

 

If I instead calculate the pixels that are above the median of neighboring pixels with:

// RGB/K:
nhp += $T>hpt*medfilt($T);
$T>hpt;

// Symbols:
hpt=3;nhp=global(+,0);

I obtain a staggering 7725 warm pixels for the Touptek versus 32 in the ZWO. This is consistent with the visual appearance of the dark frames, which is plenty of non-saturated white pixels.

 

What do you think?

I think that metric is complete nonsense because it will change depending on the level of the offset (bias).  Therefore it is probably giving a false impression.

 

A more sensible metric is what percentage of pixels have values more than 3 standard deviations from the mean.  Using that metric, the dark frames from your camera have a similar percentage to dark frames from my camera which uses the IMX410 sensor.  I'm not seeing anything in your dark frames to cause me any concern.

[Balcony_Astronomer]

Thanks for the reply. It is really appreciated. Please let me ask you more questions so that I can feel better about this camera.

 

I agree that the metric above depends on the offset level. However, I used a reasonable offset. I can imagine that I will use the camera as similar offset values in the future. Is it not of concern that several pixels acquired a signal three times higher than the median? Or do you think the script is not doing what I think it is doing, that is to calculate the pixels that deviate a lot from the median?

 

Regarding your suggestion to count the pixels above 3 standard deviations: I think it is normal that this procedure gives the same result (about 1% I suppose) irrespective of the camera used. If we want to quantify noise, we are more interested in the value of the standard deviation, not on how gaussian the distribution is. Could you please post a screenshot of the histogram of your dark frames? In the example above, I see a huge difference in the distribution between my new camera and my old one.

[sharkmelley]

I agree that the metric above depends on the offset level. However, I used a reasonable offset. I can imagine that I will use the camera as similar offset values in the future. Is it not of concern that several pixels acquired a signal three times higher than the median? Or do you think the script is not doing what I think it is doing, that is to calculate the pixels that deviate a lot from the median?

I think you want is to identify the outlier pixels.  These are the pixels outside a certain number of standard deviations.  You want to know if there are too many outlier pixels - especially bright pixels in the tail of the histogram.  The PixInsight formula you used doesn't even calculate a standard deviation.

 

Regarding your suggestion to count the pixels above 3 standard deviations: I think it is normal that this procedure gives the same result (about 1% I suppose) irrespective of the camera used. If we want to quantify noise, we are more interested in the value of the standard deviation, not on how gaussian the distribution is. Could you please post a screenshot of the histogram of your dark frames? In the example above, I see a huge difference in the distribution between my new camera and my old one.

With a bit of further analysis we can calculate the total noise in your dark frame (in electrons) and see if it makes sense given the read noise and dark current.

 

I don't know how you produced the weird-shaped histogram in your original post but the histogram of one of the darks you provided looks like this:

 

 

It looks perfectly OK to me.  The file name of the dark is in the image.

[sharkmelley]

I've done some preliminary analysis which suggests the darks are noisier than expected.  The dark current of the ZWO ASI6200 is 0.001e/pixel/sec, so I would expect yours to be similar. So 0.18 electrons of dark current should accumulate in 180sec which means the noise in a dark frame should be very similar to the bias frame noise i.e. the read noise.  But I'm measuring noise much higher than the ToupTek specified read noise of 2.8e at gain 100 HCG (0.78e/ADU) when taking a pair of dark frames.

 

The best way for me to take this further is to upload a pair of bias frames (gain 100 HCG) and a pair of darks at -10C (all taken in the same session) to see if the extra noise is already in the bias.

 

Edit:  A pair of flat frames would also be good, in order to check if the 0.78e/ADU gain is correct.

Edited by sharkmelley, 03 March 2025 - 10:49 AM.

[sharkmelley]

Two of your files resolve the issue of dark current.  The files are are Native_HCG_10C_Gain100 and Native_HCG_27C_Gain100 taken at -10C and 27C respectively.  Their respective means are 504.3 and 533.9, which is a difference of 29.6ADU.  Using a gain of 0.78e/ADU this is 23.1e, which has accumulated over the 180sec exposure.  This gives a mean dark current of approx 23.1/180 = 0.128e/pixel/sec at 27C.  ZWO quotes 0.125e/pixel/sec at 30C for the ASI6200, which is very similar.

 

The extra measured noise in the darks (even when subtracting a pair of similar darks) is totally a result of the dark current non-uniformity (a.k.a. thermal fixed pattern noise), which makes those brighter pixels a lot noisier than the average pixel.

 

So my revised opinion is that your sensor is behaving quite normally.  

[Balcony_Astronomer]

Thanks for taking the time to look at the files. I find your analysis very compelling, and I am now convinced that the dark current makes sense and the main peak is where it should be.

 

If you are interested, I uploaded more dark frames in the sub-folder "Second_Test". I cannot take flats right now, as all my gear is in a different location.

 

What I am still unsure about is the background noise on the right-hand side of the main peak.

In the attached files I plotted the histograms of the new dark frames (log scale, so that the huge main peak does not mask the little signal). What confuses me is the tail that develops on the right of the peak, which results in the appearance of several sparse warm pixels in the Touptek.

 

Attached Thumbnails

• 
• 

[sharkmelley]

I don't know the precise interpretation of the "Intensity (kau)" axis in your plots but I certainly agree that the darks from your SkyEye62AM have a far greater proportion of outlier pixel values than the equivalent darks from your ASI2600MM.  This is unexpected because both sensors use the same sensor technology with identical pixel size.

[Balcony_Astronomer]

"kau" should have been k ADU, that is 1000 ADU. The axis arrives up to 65535, but there is basically nothing above 10000.

I am glad you agree with me on this point. The seller agreed too that there is something strange with the camera and I am sending it back to them. I will keep you posted.

[Balcony_Astronomer]

An update on the story. My camera was deemed faulty by the German seller and sent back to China for repair. After about a month, I got tired of waiting and request a new camera instead. After all, I informed them about the noise issue on day 1. I got a new ToupTek SkyEye62AM Gen. II and guess what... the second camera had a similarly bad histogram. That is, also the second sensor had thousands and thousands of pixels with a large dark current.

At this point I returned the camera for good to Germany, and instead purchased a Player One Zeus M directly from the PO website. It arrived in Switzerland within a few days and it is doing great.

 

I attach two plots, both showing the histograms of a 600s, -10°C dark frames. The plots have a log vertical axis and only differ by the zoom in. Here I compare my good old ASI2600MM, the second SkyEye62AM, and the PO ZeusM.

In order to keep the comparison as fair as possible, the gain had to be adjusted on each camera to be in the same read mode and having a similar e/ADU conversion.

 

Touptek: gain 300, HGC, low noise

ASI2600MM: gain 100

ZeusM: gain 125, low noise

 

To quantitatively compare the histograms, one has to calculate them on the same number of pixels. On the Touptek and PO cameras (full frame) I therefore analyzed just a sensor surface equal to the ASI2600MM (APS-C). Finally, I horizontally shifted two histograms to have the three maxima aligned. You can clearly see how the histogram of the Touptek extends far to the right, and includes several pixels that collect high intensity. They show up as white dots on light frames. Another interesting fact is the width of the main peak. The Touptek camera gives a standard deviation of 227, the PO of 34.

 

I guess one could tackle the warm pixels with dark frames, dithering and cosmetic corrections. I am afraid that a sensor with so many warm pixels will require one to constantly maintain dark frames library. As my setup will be remote, this will be challenging. Also, I am unsure on why the width of the main histogram peak is so broad. Maybe read noise is also out of specifications.

Attached Thumbnails

• 
•