设为首页| 收藏本站|

开源天文论坛  astrocn.org

 !forgotpw!
 !register!

QQ login

One step, quick start

!show!: 144|!reply!: 4

[冷冻CCD] 搬运-CMOS设置和校准

[!share_url_copy!]
  • TA的每日心情
    慵懒
    7 dd
  • 签到天数: 3 天

    [LV.2]偶尔看看I

    1

    viewthread_userinfo_threads

    3

    viewthread_userinfo_posts

    85

    viewthread_userinfo_credits

    初级会员

    Rank: 2

    viewthread_userinfo_credits
    85
    !poston!: 2018-2-12 16:30:03
    | !thread_show_all! |!read_mode!
    本帖最后由 SaharaC 于 2018-2-12 16:36 编辑

    搬运两篇关于CMOS设置和校准的文章,欢迎各位大神积极点评~
    http://astrojolo.com/gears/cmos-setting-challenge/
    CMOS sensor based cameras are becoming more and more popular among both new astrophotogtraphers but also amog current CCD users. Many people in this latter group wonder if the work with these new electronic devices will look the same like before. Well, in general it will, however there will be some differences in details. And as you probably know, the details is the place where God is
    CMOS camera settings.
    When it comes to CMOS cameras there are a few more settings to fiddle around, that were not present in oldie goldie CCD cameras.
    GAIN – this one is probably the easiest to understand, yet there are some problems to select desired setting. Gain is more less the same as ISO setting in digital cameras. When we increase the gain, all the signal from sensor is higher, so the image will be brigher. But also visible noise will increase and dynamic range will decrease.
    OFFSET (or sometimes BIAS) – this is constant value added to signal read from sensor. The idea behind is to avoid pixels that have value of zero and could cause artifacts in the image.
    USB traffic – this one determines the bandwith of USB port that will be reserved for the camera. To ensure highest possible transfer rate USB traffic should be set as high as possible (or low – see later in the post), but increasing it too much may effect in unstable work.
    So how to set all these new parameters?
    USB trafficis probably the easiest to adjust. We need to set it to the highest value where system works stable. Of course it does not need to be very granular setting, you can choose for example 60 or 70, but does not make much sense to choose between 64 and 63. When we already have the optimal value it is worth to expose one bias frame, then lower USB traffic by 1/3 and expose one more bias frame and compare. If both have similar noise distribution then we can stay at our USB traffic setting.
    Also in some cases (especially when we use camera with USB2.0 port) increasing USB traffic may cause slower transfer and longer frame download. Then we need to decrease USB traffic, so the frame is downloaded to computer faster. It also has been noted, that in some versions of CMOS camera drivers lowering USB traffic settings effects in higher transfer. In this case we need to do changes in reverse mode – set this setting as low as possible, so the transfer will be fastest, but system works stable.
    Offset setting is also quite straightforward to set. For each gain setting we need to adjust offset, so there will be clear gap in bias frame histogram to the left of histogram peak. And there will be no pixels with zero value in the whole frame. For high gain settings it is sometimes impossible to eliminate all zero value pixels even at maximum offset – then we need to leave offset set at maximum and that is all we can do.
    And now gain setting that is sometimes most problematic to select. For planetary imaging when we operate in milisecond range exposure times we set gain to high value. In this case we can use image preview to see if selected gain is good for us – for example if the planet image is not oversaturated. For so called “lucky imaging” when exposure time varies from fraction of second to a few seconds we can set gain to values between unity gain 1) and unity gain multiplied by 2-3. And for long exposure astroimaging, where exposure time is 1 minute or more – well there are two approaches to adjusting gain:
    • we set gain to minimum value and expose frames as for CCD cameras
    • we set gain to “some value” between minimum and unity gain and expose frames with time shorter than for CCD cameras
    In theory the second way allows us to achieve maximum possible signal to noise ratio and also maximum possible dynamic range in the final image. This is due to two facts. First – increasing gain also reduces in the same time read noise (but (!) it also reduces effective pixel capacity, so effective dynamic range). At low gains this increase is significant, for higher gains this increase is minimal. And second – collecting the same amount of material with short exposed subframes we will have larger number of frames. Since the analog to digital converter (ADC) resolution of CMOS cameras is usually limited to 12 or 14 bits that in theory would give us better dynamic range in final stacked image.
    In real life these advantages are not so obvious. First – in long exposure astrophotography read noise is small fraction of total noise in the image. And since the read noise in CMOS cameras is very low anyway, there is no visible benefit when it will be lowered. And second – when we split material to larger number of short frames the dynamic range increase in the final image will not be substantial. This is again to the same fact as before – dynamic range of the camera (so the ratio between pixel depth and read noise) sensor can be larger than ADC resolution, but dynamic range of the registered image (so the ratio between pixel depth and total noise in the frame) will be definitely lower than 12 bits (for long exposure astroimaging). And, as I mentioned before, increasing gain reduces effective pixel capacity. So dynamic range will be limited, and also more stars may be oversaturated.
    There is one more possible advantage of using shorter exposed frames at increased gain – quantiziation error. If the camera sensor pixel has depth 16000e, and ADC resolution is 12 bit, so 4096ADU levels, then the read value of 100ADU 2) may indicate 399 or 400 or 398 or 401 registered electrons and that will be additional source of noise. Quantization error may play a little role for single frames, where background is very dark – like in narrowband imaging. But quantization error is not a systematic error, but random error, so when we stack subframes to final image this error will also be decreasing. In general – the less noise sensor registers (dark sky, slow instrument, filters, short exposures) the more read noise and quantization noise will affect the final image. In extreme cases it may happen, that for example under exceptionally dark skies with color camera (RGB filters) and f/10 telescope exposure times 3-5 minutes would require some gain increase to achieve optimum outcome. Or the single exposure time may be increased.
    There is one sweet spot for camera settings that would give minimum effect of read and quantization noise to final stacked image. It lays somewhere between zero and unity gain for gain setting, and somewhere between 30 and 300 seconds for exposure time, and will also vary with the number of single frames we collect. But the benefit achieved with this sweet spot can be easily compensated by just collecting a few percent more of subframes with gain just set to zero. In my opinion it makes life much easier (consider collecting also calibration frames for each combination of settings).
    That’s for the theory. Some time ago I made small comparison and exposed the same target with two different settings. First attempt was at gain=0 and I exposed 10 frames 5 minutes each (with luminance filter). Second attempt was at gain=10 and I exposed 50 frames 1 minutes each. After calibrating and stacking I was not able to see and probe any difference between final images. I mean – any difference that would be larger than just an uncertainity error. That’s why I stay at gain=0 for long exposure astroimaing till the moment I will find out there is a better way
    In the next part I will write about calibrating frames at CMOS cameras. So far there is one important thing to remember: each combination of temperature, gain, offset and USB traffic usually requires separate set of calibration frames. So if you want to play with all settings – reserve much time for it.
    Clear skies!

    1) unity gain is a gain where one recorded electron is trasformed to 1ADU in analog to digital converter. At this setting pixel effective depth is reduced to ADC resolution (so for 12 bit ADC it will be 4096)
    2) one need to remember that ADC resolution in CMOS cameras is usually 12 or 14 bit, but the signal transferred to computer is expanded to 16 bit anyway. So when 12bit ADC reads 100ADU it is stored in computer as 1600ADU

    !reply!

    !thread_magic! !report!

  • TA的每日心情
    慵懒
    7 dd
  • 签到天数: 3 天

    [LV.2]偶尔看看I

    1

    viewthread_userinfo_threads

    3

    viewthread_userinfo_posts

    85

    viewthread_userinfo_credits

    初级会员

    Rank: 2

    viewthread_userinfo_credits
    85
     !thread_author!| !poston!: 2018-2-12 16:30:42
    | !thread_show_all!
    http://astrojolo.com/gears/cmos-calibration-challenge/
    After some time with deep sky imaging with CMOS cameras I need to say that calibration for this kind of imaging is even more important, than in CCD cameras. The reason for this is amp glow that comes up for long exposure astrophotography. For cooled CCD cameras one can live without bias calibration, also without flats, can remove hot pixels with dithering. But amp glow in CMOS cameras can be removed effectively only with dark calibration.

    If you worked before with CCD cameras then as a first iteration you may calibrate images as usual – using bias, dark and flat frames. But the most important information for CMOS calibration is, that you must have calibration frames for each combination of:
      ● temperature
      ● gain and offset settings
      ● USB traffic settings
    So it is good to choose and adjust some optimum settings preset that we will use for some time and we don’t need to create calibration frames too often. Calibrating CMOS images in classic way works just fine if our light frames have corresponding calibration frames. We should not rely on dark scaling for CMOS calibration, amp glow does not scale good. So if we change exposure time or temperature or gain for light frames and there is no corresponding dark frames in calibration library we must make them and use them. That is why it is good idea to have also dark flats, especially if we need to expose flat frames for several seconds (like for narrowbands).
    But when we are sure we do not use scaling there is no actual point in having bias frames. And here comes another calibration approach, that do not use bias frames. In this scenario for calibration long exposure images we use:
      ● bunch of dark frames (they already contain bias)
      ● some flat frames (to get rid of vignette and dust donuts)
      ● bunch of dark flat frames with settings the same as for flats

    What about the actual numbers? The more the better, but in general we should have no less than 10 dark frames, 20-40 will be good (that is the number I use). For some cases when we have many light frames (like I had a few times almost 100 light frames, and each was 3 minutes) then it may have sense to increase the number of corresponding dark frames. Some guides advise to have as many darks as we have light frames, and that will work of course as well. Making dark frames can be pretty time consuming, and that’s why it is better to find optimum settings and use them as much as possible, so we do not need to create new calibration set once we decide that our settings are wrong and we need to adjust them. For flats 10-20 frames should be enough.
    And then if our processing software can automate the process (like MaxIm DL) then all we need to do is to put all frames to the library and let the program do the work. If we need to calibrate manually, then we need to: create master darks from dark frames, calibrate flat frames with corresponding master dark, create master flatfrom calibrated flats and then calibrate light frames with master dark and master flat.
    I used classic way of calibration for over the year and then tried this “biasless” approach and must say it works pretty well. One may ask “why to get rid of biases – they cost almost no time!”. Well, they cost disk space, because we should have many of them. And they also introduce some small amount of noise, very small, but still. But the main advantage I noticed is that I have no more problems with wrong flats that do not remove vignette correctly, especially for narrowband imaging under light polluted sky.  
    Clear skies!

    该用户从未签到

    2

    viewthread_userinfo_threads

    36

    viewthread_userinfo_posts

    158

    viewthread_userinfo_credits

    初级会员

    Rank: 2

    viewthread_userinfo_credits
    158
    !poston!: 2018-2-12 21:13:25
    | !thread_show_all!
    post_reply_quote

    原来CMOS要这么校准。。。学习了。。
    来自微站

    该用户从未签到

    2

    viewthread_userinfo_threads

    36

    viewthread_userinfo_posts

    158

    viewthread_userinfo_credits

    初级会员

    Rank: 2

    viewthread_userinfo_credits
    158
    !poston!: 7 dd
    | !thread_show_all!
    这网站干货好多,作者太厉害了,一直在更新很多国内找不到的资料,感谢楼主分享
    来自微站
  • TA的每日心情
    开心
    2018-2-10 21:40
  • 签到天数: 1 天

    [LV.1]初来乍到

    153

    viewthread_userinfo_threads

    645

    viewthread_userinfo_posts

    8910k

    viewthread_userinfo_credits

    管理员

    Rank: 9Rank: 9Rank: 9

    viewthread_userinfo_credits
    892599

    最佳新人活跃会员热心会员推广达人宣传达人灌水之王突出贡献优秀版主荣誉管理论坛元老

    !poston!: 7 dd
    | !thread_show_all!
    多谢分享~认真学习下
    !reply!

    !thread_magic! !report!

    !post_credits_rule!

    QQ|手机版|小黑屋|开源天文论坛     开源天文论坛 TinyAstro 讨论组

    !time_now! , !processed_in! 0.101576 !seconds!, 35 !queries! .

    Powered by Discuz! X3.3

    Release 20170120, © 2001-2018 Comsenz Inc.

    MultiLingual version, Rev. 555, © 2009-2018 codersclub.org

    !fastreply! !scrolltop! !return_list!