IMX camera capabilites #88
Replies: 12 comments 2 replies
-
|
Hello, For the IMX290 I get: 
A gain of 40 at a exposure of 1 or 10 is set to analogueGain 22,3 .. Whatever is requested, the RPICAM stops at 22,26. But at higher exposures (e.g.10000000) gain 40 is set to analoque 22,3 and digital 1,8 ...
The hotpixel vs. digi gain gain 40 at 1000000 is OK, gain 7 at 600 sec is OK, gain 1 at 10000000 OK, 1000000at gain 25 OK. Above the hot pixel counts jump up .. 
On the min pixel value vs the gain, the normal behavior is observed, like seen with 219 or 477. 
I will rerun the exposures to see, if the gain 7 has a reason ... Best regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hello all, I did this night the DARK run with IMX290 and IMX477. Both with an additional set of black cover to be on the safe side. As well I did 8 shots for each setting to get a distribution. The IMX477 stays stable on analogue gain vs. gain set but the 600 secs exposures show errors no stacking will be able to cover: On the 290, the 100 sec exposures show constant higher stddev: Interesting is the digital gain vs. the gain: 100 us exposures are using a higher digital gain than 10 or 1000 us. That's strange. For the hot pixels, the 1 sec exposures show strange behavior when digital gain kicks in, but it's 3-4 pixels only! So, the aren't really hot, but tend to overreading. On the min pixel value vs. gain, a gain of 17 is reducing the error, but (I ASSuME), that likewise the hot pixel, error goes down as we press all pixel against the lower bound: The IMX477 temp changed by +1 deg, what not increasing over time. Seems to be mostly a measurement error ... Now you have all at your hands! My conclusion:
IMX290: (My seeker/plate solver/tracker):
Best regards, |
Beta Was this translation helpful? Give feedback.
-
|
Dear all, based on the comments of @illgr I reran the tests.
The digital gain varies with the exposure time: IMX290 is lowest at gain 1 and 17 with both 1 sec., but gain 17 will add hot pixels (3); for the IMX477 10 sec at gain 1 or 3 are OK (0 hot pixel), or 1 sec at gain to 40 (with digital gain > 1 🙄). 
Reg. the min pixel value, the gain for the IMX290 should be lower than 10, the IMX477 the gain should be lower than 7. I plotted avg normal vs. the digital gain and the effect is quiet obvious: The stddev against the digital gain shows the effect of the gain kicking in (again: these are darks!!): My conclusion: Stay away 600 sec exposure time and gain larger 10! Even if you have a supersonic transwarp harmonic drive IMX290: (My seeker/plate solver/tracker): IMX477 sensor temp has been 17-19 deg celsius uncooled. |
Beta Was this translation helpful? Give feedback.
-
|
I added more tests to support some of the questions: On the IMX477 I see exposures of 600 sec with gain 22 with exploding hot pixel count, but with 100 sec the gain 22.3 has the lowest digital gain if compared to 22,.0, 22.1, 22.2, 22.4, 22.5. |
Beta Was this translation helpful? Give feedback.
-
|
With the scope reduced, I ran another batch of dark shots during night with sensors covered.. 
The number of hot pixels is increasing with exposure for me significantly as soon as digitial gain kicks in slightly and exposure time gets above 0.1 sec.
The stddev normal is flattening out against the gain selected, so higher gain should be better for SNR.
IMX290: On my main IMX477, a gain of 1 seems to be best overall ... yes, hard to see, if you see something, but taking the best picture. |
Beta Was this translation helpful? Give feedback.
-
|
@tschoenfelder Thank you for this detailed analysis. You did a huge amount of work! Thanks. |
Beta Was this translation helpful? Give feedback.
-
|
@scriptorron Dear Ronald, It's not as nice for the IMX290: The middle DigitalGain==1 steps don't exist: 
For the IMX477 the stddev get's larger, the digital gain getting back to 1.0 makes gain above 23 seemingly a sweet spot. But seeing the curve, I wonder if gain1 loses dynamic range, but wins by reduced noise. 
From avg normal, gain 12 something looks nice ..
Best regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hi Torsten, each analog gain value would need a hardware implementation in the camera circuit. I can imagine, that the camera IC has only a few analog gain values implemented. It seems, the kernel driver makes some magic to allow intermediate gain values by a combination of analog and digital gain. I still do not like the digital gain: it takes the 12 bit ADC value, multiplies it by something and rounds or cuts the result to 12 bit. This rounding/cutting adds noise! There is a good reason to do post-processing of astro-images with floating-point arithmetic. In your diagram "IMX477: Digital Gain vs Gain" the digital gain becomes "noisy" for gain values above 11. This is strange, because the gain "magic" in the kernel driver should be deterministic and reproducible. Do you have an idea where this digital gain "noise" comes from? It would be good to know, which gains are done analog only. Can you make a list of IMX477 gain settings which lead to digital gain = 1? It does not need to be complete, only a few recommended values would be enough. Do I understand your last post right, that gains 1, 2, 4, 8, 16, 22.26 are done with digital gain = 1? Regards, |
Beta Was this translation helpful? Give feedback.
-
|
Hello Ronald,
|
Beta Was this translation helpful? Give feedback.
-
|
Torsten, have you seen this: https://forums.raspberrypi.com/viewtopic.php?t=346873 I found this interesting:
I did a google search to find out, if "DGAIN" is already used in the FITS header for other purposes. It does not seem so. In the next driver version I will add "DGAIN" to the FITS. Regarding the "noisy" digital gain: https://datasheets.raspberrypi.com/camera/picamera2-manual.pdf states:
Can it be that the "noise" in your diagrams comes from a wrong metadata of the first image in your sequence? |
Beta Was this translation helpful? Give feedback.
-
|
Hello @scriptorron , thanks for finding this.
This has been my understanding prior starting the test. Regarding the noisy: |
Beta Was this translation helpful? Give feedback.
-
|
DGAIN is implemented in main branch. |
Beta Was this translation helpful? Give feedback.
-
|
Btw: I am using the standard tuning files inside the PISP directory: and IMX477: |
Beta Was this translation helpful? Give feedback.
-
|
Hello @scriptorron , Both sensors/base drivers have gain settings, which lead to the wished state of digital gain == 1: The IMX477 has several sweet spots: The larger error bars on the larger gains are due to the fact, that in this area the drivers are mapping different gain settings to some analogue gains and different digital gains: By this, the errors bars at higher gains should be taken with a grain of salt. But the average normal vs the gain requested via EKOS shows a behavior, which seems to be relevant to me: 
Whilst for the IMX290 as a guide cam, the max expo time is limited (and higher gains/noise) max be acceptable, the IMX477, when used as main camera, should be used with more thoughts, I think ...
Gain 21.3 seems to be much better than 21.6. As this is a difference by gain 0.1 differences, staying with gain 16 seems to be better than 23.2 .
16.0 is _011-.its or _012.fits:
For the IMX290, a gain below 12,88 seems to be a goal, but without any distinct sweet spots ... Best regards, |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Dear all,
I ran some checks on IMX219 and IMX477 to get a better feeling on good parameters to use. On the EXIF-Tool report they both state black level 4096 (which would be 1 bit only based on the 12 bit ADC).
I took darks with both sensors with changing gain and expo time via rpicam-still (rpicam-still --raw --immediate --mode {camera['width']}:{camera['height']} --width {camera['width']} --height {camera['height']} -n " +
f"--denoise=off --awb=custom --shutter={shutter} --camera {camera['camera']} --gain {gain} --awbgains 1,1 "+
f"-o pic_dark_{camera['name']}{gain}{shutter}raw{run}.jpg ; due to having the look running in dark/cool night, the IMX219 is called with too high gain and exposure time).
On all pixel values below 62258,25 I calculated average and stddev.
On the IMX219 these numbers look weird. The average gain increases, but not directly correlated with the exposure time. Again, mind the scale ...
On the IMX477, it looks OK, the 600 sec exposure increases may be due to some light leak (even so in the middle of the night, but perhaps ...)
Taking these darks the IMX219 shows a flattening starting above 10 whilst the still stays in the linear range. For the IMX477, the stddev goes up by 4,5 when the expo goes up by 6. This seems to be an improved SNR. BUT: Fr the IMX219, increasing the expo time from 1 to 10 sec. (times 10, the stddev increases by 1,02 only.
But as well the hot pixels are interesting:
Def Hotpixel: all pixels with a value > 62258,25 (2^16 - some percent).
Recall: In the night, 219 covered with black dust cap of the seeker telescope, the ones you know ...
Coming from 0 at short exposures getting to many pixel being just white. What you are not able to see, it's mostly the same pixels, not just random.
It isn't really better with the IMX477:
With 600 sec. the same pixels overshoot regardless Gain 20 or 7.
My theorem is:
I received the IMX290, building a case with peltier cooling. But I am not sure, what 1) the improvements will be and 2) if dew will just kill everything ..
BTW: Does anyone know how to get the sensor temp when using rpicam-still?
Best regards,
Torsten
Beta Was this translation helpful? Give feedback.
All reactions