I am new to the SBIG world, and am lucky enough to have a 16803. I am familiarizing myself with the camera, and as part of that effort, have been running CCDParameters in Pixinsight in order to match the results up against the camera specifications. There are a couple of places where the results from Pixinsight differ markedly from the camera specifications, and I was hoping someone could help me reconcile the differences or explain where I may have gone wrong in my approach. The primary differences are in Full Well Capacity (where Pixinsight generates s figure of between 77k and 87k e vs camera specifications of 100k e) and in dark current at 0C, where Pixinsight generates a figure of 0.038e/second vs camera specifications of 0.8e/p/second. Both the Pixinsight report and a link to the test frames are below: https://www.dropbox.com/sh/whqu4o7i8bvu8fw/AAAiDACOy5jxXhOyetwR2_Ija?dl=0
The full well depth quoted is the nominal number from the sensor datasheet. The datasheet actually quotes minimum 85,000 and typical 100,000. Also please note that we optimize the EGAIN for each camera. In an ideal world we would set the converter so that it saturates slightly under 65,535 to maximize dynamic range... but if we do this then customers complain that they don't get their full 16 bits. So we set it to saturate the converter a little early. Linearity is degraded at that point anyway. Dark current varies greatly from sensor to sensor, so it is specified conservatively. What is your camera's serial number?
Thanks for the quick response! I am travelling today, so will post the serial number when I get back this evening. I am going to display my ignorance here, and ask for your patience in educating me, but my understanding of gain is that it is the rate of conversion of e (not really but, for the sake of simplicity) to ADUs. Based on the 1.234 gain measured by Pixinsight in the frames above, the 80,865 FWC hits 65,530 ADUs, which is pretty close to the 65,535 max of 16 bits. If the FWC was 100,000, then the gain would be roughly 1.52? Is that correct? If I were to set the gain to be 1:1 (unity gain), then I would be using less than the FWC of the sensor? If all that is correct, is EGAIN something different? And what relationship to the question about FWC vs the datasheet?
Yes EGAIN is photoelectrons per ADU. The software is estimating the gain based on the photon shot noise in your flat fields. They are then calculating EGAIN * (65535 - bias level) to back out the approximate full well depth.
You're on the right track, a little subtly here. There are always trade-offs in camera design. The output of the CCD sensor goes through an amplifier that multiplies the signal by some amount (EGAIN) before it is digitized by the Analog to Digital Converter. CCD signal -> Amplifier multiplies x a gain factor -> ADC converts value from 0 to 65535 full scale. If the Full Well Capacity was 65535, then the ideal gain would be 1.0x to correspond to the ADC range. If FWC is higher than 65535, then the camera designer has a choice of lowering the gain of the amplifier to allow the FWC to be represented OR the tradeoff of detecting closer to individual photos (1 photon per ADU). So, if FWC = 100,000, gain might be dialed back to 65536/100,000 = 0.65, or converted to e-/ADU as PI expresses it, 100000/65536 = 1.52 as you say. The other consideration is there is usually a bit of noise and an offset that gets factored in there. CCDs are primarily analog devices, and each device has its own unique characteristics, within a range. Pixinsight's estimate of about 80800 is in the right neighbourhood, however you can't consider it a rigourous test, done under reproducible condistions at the factory. Plus, we don't know what algorithm or mathematics they are using to come up with their numbers, but as an estimate is close to what is expected. The PI dark current number is really good (maybe a bit too good?) - again, without knowing how they calculate, it's hard to comment. If you really want to dig into this, the datasheet for the ON Semiconductor KAF-16803 sensor is here: https://www.onsemi.com/pub/Collateral/KAF-16803-D.PDF As @Doug mentions, Table 5, Specifications, has the Ne- SAT values, ranging from a minimum of 85,000 to 100,000 nominal, and as indicated by Note 11, it is tested during production. Table 5, Integration Dark Signal value is nominally 3 e-/pixel/second (with no minimum), and PI is saying 0.038. So I'd be pretty happy with that.
Very happy with the dark current, assuming that is accurate. Still trying to understand why FWC is less than the minimum for the sensor, and whether that matters.
Our test data shows the EGAIN is 1.31. That would indicate full well of approximately 85,000 e-. I can't comment on the accuracy of third party software. I do trust our production test software to produce accurate results.
Thanks!. No worries - I have no doubt that your production test suite is more accurate than anything I am doing at home. What did the production test show for readout noise? That said, I am still trying to understand the information. When I researched this camera before purchasing it, the FWC was listed as 100k e- on the product overview, and 100k e- typical on the specifications page. Your production test shows 85k e-?
Bear in mind that these parameters are dominated by the characteristics of the individual sensors. These devices push the limits of silicon detection technology and do vary a bit from chip to chip. If you would like to review the sensor datasheet, it is available here: https://www.onsemi.com/pub/Collateral/KAF-16803-D.PDF The table on page 7 lists the performance parameters including typical and maximum/minimum values where applicable. We measured read noise for your camera at 11.1 e-. We don't directly measure well depth, as it is not a critical parameter. In demanding applications you avoid approaching saturation to ensure sensor linearity. You can calculate well depth from EGAIN and saturation minus bias level in ADUs. If you really want to optimize well depth, we could log into your camera remotely and adjust the gain setting in the camera. It could be set so the sensor saturates a bit short of 65,535 ADU, in which case you'll see the absolute maximum sensor well depth for every pixel. But there's really not much point in doing this.
I'm not focused on chasing specs for the sake of specs. Really just trying to understand all of this so that I am a better owner/user of the equipment. I am not entirely sure I understand. I believe that the ADUs are really the mathematical results of applying a factor to the photoelectron counts that come out of reading the photosites? In my camera, the gain was set by SBIG to 1.31. So for each electron coming in, the ADU count rises by 1/1.31 = 0.7634. That would imply that the saturation point of 65,535 is roughly 85,850 e-? Which is above the measured FWC? I am sure I am missing something basic, but that looks to me the camera is already using its FWC and has not been set to avoid approaching saturation? Wouldn't you need to decrease the gain so that 65,525 ADUs was hit well beneath the FWC to do that? And isn;t it currently set to hit 65,535 above the FWC? Again - apologies for what must be basic questions, just trying to wrap my head around all of this.
Due to customer preference, we set the gain so that the A/D converter saturates slightly before the CCD sensor does. People complain if they don’t see 65,535 coming out of their camera. The gain is adjusted individually for every camera because the sensor full well varies from chip to chip.
So is that optimal Understood. But if you measured FWC at 85k on my chip, and set the gain to 1.31, doesn't that imply that the AD convertor only saturates at 85.85k e-, which is above the level at which my chip saturates? If I am understanding your correctly, wouldn't you have set the gain to cause the ADU count to hit 65,535 before the FWC hits 85k? Separately, if you are doing that just so customer see 65,535 levels, is that the optimal setting for astroimaging? If not, what would be the optimal setting?
Yes, we set the converter gain to cause the ADU count to hit 65,535 before the wells saturate. Not at 85k or 100k electrons or whatever - we don't measure that - but at whatever voltage output level that the particular sensor saturates. ON Semi guarantees the corresponding well capacity will be at least 85k e- as part of their extremely thorough sensor production testing. Your sensor saturates somewhere slightly above 85.85 ke-. I don't know the exact number because we don't measure it. In these cameras the limiting factor for dynamic range is the sensor, not the A/D converter. You want to oversample the noise at the bottom end, with at least 2-3 bits worth of ADUs just recording the noise background of the sensor; otherwise if you don't do that the A/D converter quantization noise adds to the overall system noise floor. At the top end I would ideally set it up so the CCD sensor hits saturation (full well) slightly before 65,535, so that we're guaranteed to get the full dynamic range of the sensor. So by adjusting the converter to always hit 65,535 we're very slightly compromising the dynamic range and full well capacity of the camera. Unfortunately if we do adjust it to fall short of 65,535, then customers complain that they're "not getting the full 16 bits". It's impossible to explain to people that it absolutely doesn't matter. (And if I may point out... I'm having trouble explaining this to you LOL!)
Information disparity. You understand it so well it is hard to know how little I understand it. That seems pretty clear. As a matter of curiosity, roughly how much of the FWC on average is sacrificed to ensure that the full 65,535 ADUs are used? And isn't this something the user can adjust themselves by adjusting camera gain? Or is that not a user adjustable parameter with the 16803? Also, I have failed to follow up on your earlier statement that linearity degrades at some point. Would you mind explaining that? Really appreciate the education here.
We do have a software tool that can be used to adjust the gain and offset of the camera; however, we only make it available when absolutely necessary to correct a problem. The tricky part is that changing the gain also changes the offset - the two things interact. The camera has to evenly illuminated to a certain level while you use it. It's very easy to muck up your camera's settings and make a working camera not work. Historically speaking we've had cases of people passing around adjustment tools despite being asked not to, and that can result in a lot of broken cameras. For that reason we only make the tool available when the alternative is to send the camera in for service. As for linearity, all cameras obviously go nonlinear when they saturate - they no longer respond to light. Also most sensors have anti-blooming technology, and that includes the KAF-16803. Blooming is what happens when a bright star oversaturates the sensor; it results in a big saturated tail going up and down the sensor. This happens because the big electric fields generated by all those excess photoelectrons overwhelm the electrodes used to read the charge out of the sensor. ABG chips have an overflow drain that removes the excess electrons before the bloom happens, which limits the damage to the image. Unfortunately this introduces a little bit of nonlinearity as the charge levels approaches saturation. Historically speaking ABG chips would go slightly nonlinear at about 60-70% of saturation - the sensitivity would trail off slightly - which meant that people doing photometry avoided them. The KAF-16803 is a more modern sensor and does not exhibit significant nonlinearity, though I would avoid doing photometry above the 90% level just to be safe.
