M51 with a ZWO ASI224

[vlaiv]

1 minute ago, cuivenion said:

Hi, I thought that would be the case, but as I wasn't sure I just played it safe, thanks for the info. The last point has gone over my head a bit. I've not come across quantisation noise. Looks like I've got some reading to do. You've obviously got a deeper understanding of this than me, you couldn't finish off that equation showing the correct gains could you please?

Quantization noise is rather simple to understand, but difficult to model as it does not have any sort of "normal" distribution.

Sensor records integer number of electrons per pixel (quantum nature of light and particles). File format that is used to record image also stores integers. If you have unity Gain - meaning e/ADU of 1 - number of electrons get stored correctly. If on the other case you choose non integer conversion factor you introduce noise that has nothing to do with regular noise sources. Here is an example:

You record 2e. Your conversion factor is set to 1.3. Value that should be written to the file should be 2x1.3 = 2.6 but file supports only integer values (it is not up to file really, but ADC on sensor that produces 12bit integer values with this camera), so it records 3ADU (closest integer value to 2.6, in reality it might round down instead of using "normal" rounding).

But since you have 3 ADU and you used gain of 1.3 what is the actual number of electrons you captured? Is it 2.307692..... (3/1.3) or was it 2e? So just by using non integer gain you introduced 0.3e noise on 2e signal.

On the matter of gain formula - it is rather simple. ZWO uses DB scale to denote ADU conversion factor, so Gain 135 is 13.5db gain over the least gain setting. So how much is lowest gain setting then?

sqrt(10^1.35) = ~4.73 and here it is on graph:

Ok, so if unity is 135 (or 13.5db or 1.35 Bell), we want gains that are multiples of 2. Why multiples of two? In binary representation there is no loss when using powers of two do multiply / divide (similar to decimal system where multiplying with 10 and dividing with 10 only moves decimal dot) - so guaranteed quantization noise free (for higher gains, for lower gains you still have quantization noise because nothing is written after decimal dot).

DB system is logarithmic system like magnitude. So if you multiply power / amplitude you add in db-s. 6db gain is roughly x2.

(check out https://en.wikipedia.org/wiki/Decibel there is a table of values - look under amplitude column).  Since gain with ZWO is in units of 0.1db - 6db is then +60 gain.

Btw, you should not worry if gain is not strictly multiple of 2 but only close to it, because closer you are to 2 - quantization noise is introduced for higher values (because of rounding) - and with higher values SNR will be higher (because signal is high already).

 

 

[cuivenion]

34 minutes ago, vlaiv said:

Quantization noise is rather simple to understand, but difficult to model as it does not have any sort of "normal" distribution.

Sensor records integer number of electrons per pixel (quantum nature of light and particles). File format that is used to record image also stores integers. If you have unity Gain - meaning e/ADU of 1 - number of electrons get stored correctly. If on the other case you choose non integer conversion factor you introduce noise that has nothing to do with regular noise sources. Here is an example:

You record 2e. Your conversion factor is set to 1.3. Value that should be written to the file should be 2x1.3 = 2.6 but file supports only integer values (it is not up to file really, but ADC on sensor that produces 12bit integer values with this camera), so it records 3ADU (closest integer value to 2.6, in reality it might round down instead of using "normal" rounding).

But since you have 3 ADU and you used gain of 1.3 what is the actual number of electrons you captured? Is it 2.307692..... (3/1.3) or was it 2e? So just by using non integer gain you introduced 0.3e noise on 2e signal.

On the matter of gain formula - it is rather simple. ZWO uses DB scale to denote ADU conversion factor, so Gain 135 is 13.5db gain over the least gain setting. So how much is lowest gain setting then?

sqrt(10^1.35) = ~4.73 and here it is on graph:

Ok, so if unity is 135 (or 13.5db or 1.35 Bell), we want gains that are multiples of 2. Why multiples of two? In binary representation there is no loss when using powers of two do multiply / divide (similar to decimal system where multiplying with 10 and dividing with 10 only moves decimal dot) - so guaranteed quantization noise free (for higher gains, for lower gains you still have quantization noise because nothing is written after decimal dot).

DB system is logarithmic system like magnitude. So if you multiply power / amplitude you add in db-s. 6db gain is roughly x2.

(check out https://en.wikipedia.org/wiki/Decibel there is a table of values - look under amplitude column).  Since gain with ZWO is in units of 0.1db - 6db is then +60 gain.

Btw, you should not worry if gain is not strictly multiple of 2 but only close to it, because closer you are to 2 - quantization noise is introduced for higher values (because of rounding) - and with higher values SNR will be higher (because signal is high already).

 

 

Hi, I'm not going to pretend I completely understood that. But basically you're saying if you choose a gain that is 6db,12db,18db, 24db etc (195, 255, 315, 375) above unity gain then I'll avoid quantization noise?

[vlaiv]

1 minute ago, cuivenion said:

Hi, I'm not going to pretend I completely understood that. But basically you're saying if you choose a gain that is 6db,12db,18db, 24db etc (195, 255, 315, 375) above unity gain then I'll avoid quantization noise?

Yes, unity gain included (it also avoids quantization noise).

[cuivenion]

4 minutes ago, vlaiv said:

Yes, unity gain included (it also avoids quantization noise).

Great thanks for the info.

 

[Demonperformer]

5 hours ago, Demonperformer said:

That is MEGA-seriously tempting ....

Oscar Wilde once said* “The only way to get rid of a temptation is to yield to it” ... button pressed ...

*The Picture of Dorian Gray, 1890

[AstroNtinos]

Wow very nice image I thought asi224 was only for planetary imaging but does very nice work for deep sky.