Understanding Sensor Analysis in Sharpcap

[Stub Mandrel]

I've managed to get some comparative results for my Touptek Mono:

And ZWO ASI120MC:

These both use the SAME chip, and I would expect the mono version to be more sensitive, but in Sharpcap it appears LESS sensitive.

I THINK these figures explain why - the ZWO has much higher gain giving the apparently 'brighter' image, but at about 0.1 e/ADU this is at the expense of dynamic range and not actually giving me more data.

The mono cam has a sudden drop in read noise  at 200 gain - I assume for long exposures this will be the best setting balancing noise against dynamic range.

Note both cameras were very warm (about 27 degrees) for this test.

 

Can anyone help me better understand these figures?

 

 

 

Edited July 28, 2019 by Stub Mandrel

[vlaiv]

I don't think you have anything in these values that would describe sensitivity of sensor.

I'm not sure how the measurements were made, but e/ADU part needs flat panel to be established. If you did use flat panel, then it should be measured value, if not, it's probably read from driver instead.

Let's examine values and what they mean.

1. Gain - it is usually arbitrary number assigned in drivers to represent e/ADU value (actually inverse - as people are used to idea higher gain means lower e/ADU, or higher ADU/e - which would be reciprocal). ZWO drivers are following rather good convention and they use 0.1db units of gain, so 200 gain is actually gain of 20db. We will get back to that when examining relative gain and relative gain in db.

2. e/ADU - this one is straight forward. Light comes in photons and sensor pixels are recording photons by capturing electrons in potential well. When a photon hits pixel there is certain probability that electron will be captured (photon ejects electron from photo sensitive material and it's being captured by potential well). This probability is quantum efficiency of a sensor, so sensor with 75% QE has 75% chance of converting photon to electron. After exposure there will be number of electrons in each pixel's potential well. When value is read out it is converted to number called ADU (that is short for Analog-Digital Unit), or what we know as "pixel value" - which is a number, a integer number recorded in fixed number of bits. This number is calculated like this - number of electrons is divided with e/ADU and it gives numeric value in ADU which is then rounded and recorded in fixed number of bits.

3. Read noise is standard deviation of very short exposure pixel values when bias is removed. In this case it is expressed in electrons. It is measured by taking bias subs, then stacking them in such way that it removes bias signal (easy way to do it is stack first half of subs to one stack and other half of subs to second stack and then subtracting those two stacks - what remains is pure read noise scaled down by square root of total number of subs stacked in both stacks). This way you get value in ADU units, so you use e/ADU to convert back to electrons. Read noise in electrons can be compared between gain values - not one in ADUs because intensity of ADUs will depend on gain setting.

4. Full well capacity is essentially how many electrons can any one pixel capture. In this case, true full well capacity is not given, but rather effective full well capacity, which is consequence of number of bits used to record each ADU value. That is 12 bits in this case or values in range 0-4095. So this column actually holds values that you get when you multiply 4096 with e/ADU (converting from max ADU back to electron count) - as that is maximum number of electrons that you can effectively record with these sensors.

5. Relative gain

and

6. Relative gain in db

are just a way to represent how e/ADU changes as you increase "gain" setting. First value of reference gain is 1 - that is "starting point". Each next value is just simply how many times e/ADU is smaller than this base value (divide any e/ADU with first - reference e/ADU and you will get relative gain). Relative gain in db is just previous column expressed in db (logarithmic) scale. For db scale look here:

https://en.wikipedia.org/wiki/Decibel

7. Dynamic range is calculated by dividing full well capacity with read noise and taking logarithm with base two of that. It sort of represents "number of bits in binary system needed to record useful signal level above read noise that is capable of being recorded by pixel" - note this is my explanation for it and not some general definition. I don't really find any use of this number in AP although it is often quoted. I think it is more useful in regular photography with single exposure as it gives you idea of how much histogram manipulation you can do when you have raw image (like adjust exposure stops up/down and such - higher dynamic range - more adjustment possible).

 

 

 

 

 

[Stub Mandrel]

9 hours ago, vlaiv said:

I'm not sure how the measurements were made, but e/ADU part needs flat panel to be established. If you did use flat panel, then it should be measured value, if not, it's probably read from driver instead.

Diffuse and rather low daylight.

See if I'm getting this right?

As (in principle) the only difference between the sensors is the bayer filter the QE should be the same and to a first approximation, the mono cam will collect about three times as many photons per pixel per unit time.

Any other differences come from the electronics.

So let's assume that at a particular level of white light you get 100e on the mono cam and 33e on the colour.

At their maximum gains this gives 213 ADU and 305 ADU respectively, so the colour will appear over half a stop  brighter if these are scaled directly to display levels, even though the mono image has three time the 'real' qauntisation depth?

 

[vlaiv]

22 minutes ago, Stub Mandrel said:

Diffuse and rather low daylight.

See if I'm getting this right?

As (in principle) the only difference between the sensors is the bayer filter the QE should be the same and to a first approximation, the mono cam will collect about three times as many photons per pixel per unit time.

Any other differences come from the electronics.

So let's assume that at a particular level of white light you get 100e on the mono cam and 33e on the colour.

At their maximum gains this gives 213 ADU and 305 ADU respectively, so the colour will appear over half a stop  brighter if these are scaled directly to display levels, even though the mono image has three time the 'real' qauntisation depth?

 

Almost

Thing with Bayer matrix is that it does not divide 400-700 range into exact thirds. It's actually a bit more complicated as this QE curve for ASI120MC shows:

From this graph you can see that all three colors have some sensitivity over whole range 400-700nm.

Bayer matrix lowers QE compared to mono about 10% or so. It can be seen for example in this graph of QE:

But you are right about what will intensity look like if you take same exposure given max gain settings - converted to ADU, even if you take 1/3 of electrons for green pixel vs mono variant if you apply above e/ADUs for max gain you will end up with roughly 50% higher ADU value and it will look brighter on screen if scaled to same range.

Of course, point of sensitivity of camera is not how bright things look on screen as you can adjust brightness (do different range scaling, or apply non linear histogram transform). Point of sensitivity is achieved SNR in given time with light source of set intensity (and other things being equal - like sampling rate and aperture).

 

 

[Stub Mandrel]

5 minutes ago, vlaiv said:

Of course, point of sensitivity of camera is not how bright things look on screen as you can adjust brightness (do different range scaling, or apply non linear histogram transform). Point of sensitivity is achieved SNR in given time with light source of set intensity (and other things being equal - like sampling rate and aperture).

I think my worry has been is that with apparently 'similar' settings - maximum gain (to allow shortest exposure to fight seeing) and the display settings (brightness, gamma all on default) - the mono cam + a colour filter has appeared much less bright for a given exposure than the colour one.

Now I know this isn't actually a problem and I can compensate by turning up the display brightness to compensate (effectively mimicking the extra gain) and rely on a combination of Sharpcap's Histogram 'brain' to ensure I keep the planet part of my data out of the 'red zone' while keeping the exposure as short as possible.

It also give me the confidence to try again on Venus with my extreme UV-pass filter.