When is Low Conversion Gain Useful?

[PeterC65]

Over the summer, I've been experimenting with a new (to me) Altair GPCam2 327C, using my existing 72mm F6 APO scope. After much research I've concluded that the camera gain needs to be set to the point when it just switches to High Conversion Gain, as this gives minimum read noise while still maintaining dynamic range and full well depth. This has got me wondering why cameras ever operate in Low Conversion Gain mode.

In HCG mode the read noise is lower and the dynamic range higher for a given gain, and there is no difference to the full well depth. So why would anyone operate a camera in LCG mode?

On the face of it, I'd prefer my camera to only operate in HCG mode so that I always had the lower read noise and higher dynamic range. But I assume I'm missing something?

 

[vlaiv]

1 hour ago, PeterC65 said:

and there is no difference to the full well depth.

How did you figure that out?

From what I can see for Altair camera, there is significant drop in FWC:

On a separate note, when low gain is applied, following happens:

- one utilizes full well capacity to the maximum (actual full well capacity - not value clamped by ADC). That can be beneficial in some applications

- If ADC bit count is lower than needed - there will be quantization error involved. Certain level of read noise is needed to perform dither on the data and make that quantization error less noticeable / looking more like regular noise. For that reason it is ok and even necessary to have higher read noise levels on lower gain settings ("above" unity gain).

 

[symmetal]

HCG was introduced to give a higher quality picture in low light conditions primarily for security or industrial use cameras.

LCG is really just HCG turned off and is the way cameras operated before HCG was available. In normal lighting conditions LCG is preferable as you get a larger well depth and camera read noise is not an issue.

Astrophotograpy has found that HCG has a distinct advantage, due to the very low light conditions the camera is operated in.

Alan

[wimvb]

I only use the hcg mode for narrowband imaging. I do all LRGB imaging at 0 gain, and offset 8. That is with an ASI294MM. Works good enough for me. I expose for the stars (ie, no saturated stars except for the very brightest), and extract the nebulosity in post pricessing. At 0 gain, the camera has its true full well depth, and highest dynamic range.

[symmetal]

Sony haven't released info on how the HCG and LCG actually works, at least I couldn't find anything, and manufacturers just give similar rough indications and no specifics. It does appear that three separate settings are available on the later sony sensors with HCG which affect gain.

HCG on/off. If off it's in LCG mode.

HCG gain multiplier

'Normal' camera gain

Zwo have tried to make it simple with having a fixed HCG gain multiplier and turning HCG on when the normal gain is set to 100 or above. 😃

Some manufacturers have the HCG on/off separate to normal gain, giving more control but needs more consideration to avoid conflicting settings. 🤔

Atik have possibly made all three available in the Horizon but in preset steps with confusing names which doesn't really tell you what mode it's in. 😬

The newest Sony sensors have a Dual-Conversion Gain mode which has two ADCs available for each pixel, so each pixel is converted twice simultaneously, one with a low gain and the other with a higher gain. The two ADC outputs are then combined to give a HDR, high dynamic range, signal output. Whether this is a HCG ADC combined with a LCG ADC, or two different 'normal' ADC gain settings, or a combination, I don't know.

Alan

[CraigT82]

5 hours ago, symmetal said:

The newest Sony sensors have a Dual-Conversion Gain mode which has two ADCs available for each pixel, so each pixel is converted twice simultaneously, one with a low gain and the other with a higher gain. The two ADC outputs are then combined to give a HDR, high dynamic range, signal output. Whether this is a HCG ADC combined with a LCG ADC, or two different 'normal' ADC gain settings, or a combination, I don't know

This is the thrust of Sony’s ‘Starvis 2’ (Clear HDR) technology. The sensor brackets an image simultaneously as you say in order to increase DR, but I’m not sure if that feature would be utilised at all by the Astro camera manufacturers?  I wonder if the thing with ZWO having different HGC switch point to Player 1 is something to do with this Starvis 2 tech.

The IMX662 and 678 also have the Starvis 2 tech and so will be interesting to compare how ZWO and other manufacturers use these chips too (once the other manufactures have caught up and released some cameras).

The Starvis 2 chips also have a different, ‘vertical’ photo well architecture which allows a greater charge build up, explaining why the full well capacity of these chips are much greater then their non Starvis 2 predecessors.  This is definitely something that the Astro cam producers can use! 
 

[PeterC65]

14 hours ago, vlaiv said:

How did you figure that out?

From what I can see for Altair camera, there is significant drop in FWC:

 

The full well depth curves are continuous, meaning that the switch from LCG to HCG has no effect on full well depth. Compare this with the read noise and dynamic range where there is a step change at the switching point.

As @symmetal has mentioned, the LCG / HCG control and the normal gain control are separate controls on the sensor. Some camera manufactures choose to keep them separate; some choose to combine them. For those who keep them separate, SharpCap will decide when LCG / HCG switching occurs (unless you use the ASCOM driver). For those who combine them, different manufacturers select different switching points (the ASI585Mc switches at 252, the Uranus-C switches at 180, both use the IMX585 sensor).

I asked the original question because the LCG / HCG switching point is a choice, sometimes made by the camera manufacturer, and if so, made differently by different manufacturers, sometimes made by software, and sometimes made by the user.

As I understand it:

• Full well depth increases as gain reduces but is unaffected by LCG / HCG switching. So if you want a deeper full well you go for lower gain.
• Read noise is always lower in HCG mode. This may not be relevant at lower gains when quantisation noise dominates but ...
• Dynamic range is always higher in HCG mode.

So given the choice, which camera manufacturers have, and we sometimes have, why wouldn't you always select HCG mode (for all gain settings) and maximise the dynamic range? This is what I was asking in the original post. Is there a situation when HCG isn't a good thing?

[CraigT82]

Does anyone know exactly why the DR jumps when HGC switches on? 

[vlaiv]

34 minutes ago, PeterC65 said:

This is what I was asking in the original post. Is there a situation when HCG isn't a good thing?

To be honest, I don't really get all LCG, HCG (don't know which is which) thing, but I'll try to answer original question based on some QHY cameras which have both modes on.

First lets address some important things.

Dynamic range.

This term is in my view completely useless as it is used. What does it represent? Maximum signal value divided with read noise? Why is that important?

If we want to address daytime or night time surveillance use cases - why is read noise important bit? What would be more sensible thing to be called dynamic range? I'd say ratio of highest and lowest signals that can be distinguished. For low signal to be distinguished we need something like SNR5 or some other measure - but whatever measure we introduce, shot noise will be dominant and hence actual strength of that signal will play important part - not read noise.

In above case if we have FWC of 12000 and FWC of 4000 - it is clear which one will have larger dynamic range even if read noises are different.

If we adopt SNR5 as standard and have read noise 3 and 1 - we will have ~27.71e and ~25.96e as signals, so dynamic range will be 12000/27.71 = 433 and in other case it will be 4000/25.96 = 154

In astrophotograpy - dynamic range completely does not make sense as in either of above definitions as we stack to improve SNR so final SNR and any sort of dynamic range will depend on total integration time and number of stacked subs.

Now, why have two modes?

Well, if we have only a single mode - like green line above, then what is the point in having 12K FWC when we can only utilize something like 4500 with 12bit ADC. Or to phrase it differently - why add only 12bit ADC when 14bit ADC will solve the issue of FWC.

So we have compromise in terms of price / complexity of sensor.

Second thing is having low read noise at low gain setting.

At low gain setting - system gain is ~3 in purple readout mode. With this mode we are reading complete full well of 12K and putting it into 0-4095 range of 12bits.

If we observe what will happen with very low signal:

0,1,2 electrons will be read out as 0ADU

3,4,5 electrons will be read out as 1ADU

and so on

Signal will have certain level of "posterization" to values. Just take ADU and convert it back to electrons - 0ADU will give 0e, 1ADU will give 3e, 2ADU will give 6e and so on.

We have introduced this "stepping" error which is very predictable and not random at all. Stacking won't help get rid of such error. This error must be mixed with another error to make it more natural. This is where read noise steps in and makes things fuzzy enough. This is called noise shaping:

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

1e of read noise is not enough to shape 3e of posterization and read noise needs to be higher than that.

This is why we have higher read noise on higher e/ADU values (among other things - it is handy for this purpose).

In the end - to address your question, why would anyone use purple read out mode over green (even in astronomy) - well if one wants to simultaneously detect two signals that have large intensity difference - that wouldn't otherwise fit in green readout mode and thing being recorded is transient so we can't use stacking of some sorts - but only single exposure.

In general, in astrophotography - not very useful, so just use green mode - or do what @wimvb does and use full well capacity (just mind exposure length as you are then in high read noise domain).

 

[vlaiv]

3 minutes ago, CraigT82 said:

Does anyone know exactly why the DR jumps when HGC switches on? 

DR as defined is just ratio of highest signal recorded to read noise.

As ratio of two numbers - you can make it big by making numerator bigger or reducing denominator.

In case of HGC (if that is low read noise version) - we reduce denominator.

However - DR is simply useless as measure of anything really, so not sure why it is used.

I'd be much obliged if anyone explained usefulness of DR to me, maybe I'm missing something.

[PeterC65]

1 hour ago, vlaiv said:

I'd be much obliged if anyone explained usefulness of DR to me, maybe I'm missing something.

Dynamic range, measured as the ratio of the maximum possible signal divided by the read noise, may not be fully representative. But dynamic range in the broader sense of the term, meaning the range of distinctly measured signal, for the smallest to the largest, is surely useful in astrophotography? Perhaps it would be better measured as the maximum possible signal divided by the total noise (from whatever source), since it is the noise that limits the smallest possible signal that can be distinctly measured. Stacking will increase dynamic range by this measure.

1 hour ago, vlaiv said:

Well, if we have only a single mode - like green line above, then what is the point in having 12K FWC when we can only utilize something like 4500 with 12bit ADC. Or to phrase it differently - why add only 12bit ADC when 14bit ADC will solve the issue of FWC.

I'm not sure what the green and purple lines in your graphs represent @vlaiv. Are they showing LCG mode and HCG mode?

Surely it is beneficial to have a larger full well depth even if the ADC resolution gives rise to quantisation errors when reading its contents? The full well depth and quantisation error are just different things.

1 hour ago, vlaiv said:

 

Signal will have certain level of "posterization" to values. Just take ADU and convert it back to electrons - 0ADU will give 0e, 1ADU will give 3e, 2ADU will give 6e and so on.

We have introduced this "stepping" error which is very predictable and not random at all. Stacking won't help get rid of such error. This error must be mixed with another error to make it more natural. This is where read noise steps in and makes things fuzzy enough. This is called noise shaping:

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

1e of read noise is not enough to shape 3e of posterization and read noise needs to be higher than that.

This is why we have higher read noise on higher e/ADU values (among other things - it is handy for this purpose).

Leaving aside any reasoning based on dynamic range, I agree that the switch from HCG to LCG should happen at the point when read noise is needed to overcome quantisation noise. It ought to be possible to calculate this point. Taking the ASI585MC as an example, the switch happens at a gain of 252 when the read noise falls from 4.2e to 1.5e, but at this gain the e-gain is 0.8e/ADU, so the extra read noise is not required to mask quantisation errors. At a gain of, say, 100, when the e-gain is 3.8e/ADU, then the extra read noise may be needed.

It is the spec for the ASI585MC camera that really prompted this post as I think it is unnecessarily high.

 

[PeterC65]

Despite all of the above discussion, I still can't see a situation when HCG isn't a good thing and therefore why it isn't enabled all of the time.

 

[vlaiv]

3 minutes ago, PeterC65 said:

Dynamic range, measured as the ratio of the maximum possible signal divided by the read noise, may not be fully representative. But dynamic range in the broader sense of the term, meaning the range of distinctly measured signal, for the smallest to the largest, is surely useful in astrophotography? Perhaps it would be better measured as the maximum possible signal divided by the total noise (from whatever source), since it is the noise that limits the smallest possible signal that can be distinctly measured. Stacking will increase dynamic range by this measure.

We already have measure of strongest signal - that is full well capacity. Weakest signal that can be recorded - depends on host of things. Read noise is probably one with the least impact.

For astrophotography, such measure is completely useless.

You can control complete process with:

a) exposure length

b) total integration time

Say you have 6K full well capacity.

Want to record source that emits 6K per minute - but you want to avoid saturation? Expose for less than a minute (say half a minute).

Want to record large maximum FWC of say 300K - add up 100 of such subs and you will accumulate 300K of signal (if each sub is 3k max).

In any case - stacking solves issue of weakest signal. Want to record something really weak signal? Stack enough subs to push its SNR above detection threshold.

9 minutes ago, PeterC65 said:

I'm not sure what the green and purple lines in your graphs represent @vlaiv. Are they showing LCG mode and HCG mode?

They are showing some sort of "read mode 0" and "read mode 1". QHY often just numbers read out modes without all that LCG/HCG or whatever they are called.

I've seen some of their camera models that have more than 2 readout modes - as much as 4 or maybe 5.

10 minutes ago, PeterC65 said:

Surely it is beneficial to have a larger full well depth even if the ADC resolution gives rise to quantisation errors when reading its contents? The full well depth and quantisation error are just different things.

I'm not sure it is. For long time 16 bit ADC was the norm and people valued large FWC.

When I started using ASI1600 - I happily accepted 12bit ADC and 4000 FWC.

By control of exposure length and number of exposures - any source can be recorded. 60K FWC is not much of advantage really over say 4K.

One might argue that with large FWC - stars won't saturate. My counter argument is that - for any FWC there is enough of difference in star brightness that some stars will saturate and best way to deal with that is not to increase FWC but rather to use exposures of different lengths based on what one images. Even using multiple exposure lengths of same field and then combining signal is going to work well - and in fact, it will solve issue of saturated stars - regardless of FWC and how bright star might end up in FOV. It just works for every scenario.

 

[vlaiv]

10 minutes ago, PeterC65 said:

Despite all of the above discussion, I still can't see a situation when HCG isn't a good thing and therefore why it isn't enabled all of the time.

 

It's not a good thing if you want to exploit full - full well capacity (for any reason) that is available to you.

If you have 12bit camera and lowest gain that you can set with HCG mode has e/ADU of say 1.1 - max number of recorded electrons will be 4096 * 1.1 = ~4500. It does not matter if your camera has 14K FWC - that won't be accessible to you.

While I think that FWC is not important in stacking - there are certainly cases where one might want to have large FWC - like if stacking is not available due to transient nature of phenomena.

[PeterC65]

I should have mentioned that I'm approaching this from an EEVA rather than an AP perspective. Exposure time needs to be short and post processing minimal (just live stacking), so that the image is close to real time. For AP, when long exposures are possible and lots of post processing, the compromises will be different.

@vlaiv your QHY read modes 0 and 1 don't look like the LCG and HCG of other cameras, at least not with regard to full well capacity and e-gain.

 

[andrew s]

It might be worth reviewing the basics noise regimes.  At zero to low signals read noise (including random telegraph noise, quantisation noise etc. ) dominates and is independent of the signal. This is followed by signal shot noise and is proportional to the square root of the signal. Then photon response non uniformity (PRNU) dominates as it is proportional to the signal level. Finally, saturation. 

Various strategies can be used to reduce the effects in the various regimes.

Personally,  irrespective of this I set my cameras for the best signal to noise ratio they can deliver usually just as the high gain mode cuts in and  then change tbe exposure accordingly. I also create and use a high s/n ratio master PRNU flat. 

Regards Andrew 

[PeterC65]

2 hours ago, andrew s said:

Personally,  irrespective of this I set my cameras for the best signal to noise ratio they can deliver usually just as the high gain mode cuts in and  then change tbe exposure accordingly. I also create and use a high s/n ratio master PRNU flat.

This goes to the heart of my question. Most people seem to set their camera gain to just above the HCG switching point. So wouldn't it be better if this point was at the minimum gain possible so as to make use of the maximum full well depth, maximum dynamic range, and maximum range of gain?

A setting of 252 for the ASI585MC for example blows away half of the available gain range (and associated full well depth / dynamic range).

 

[vlaiv]

43 minutes ago, PeterC65 said:

This goes to the heart of my question. Most people seem to set their camera gain to just above the HCG switching point. So wouldn't it be better if this point was at the minimum gain possible so as to make use of the maximum full well depth, maximum dynamic range, and maximum range of gain?

A setting of 252 for the ASI585MC for example blows away half of the available gain range (and associated full well depth / dynamic range).

 

If you are coming at this from EEVA perspective then answer is simple - use very high gain setting.

With short exposures you don't really care about FWC - you won't be able to saturate pixels with interesting signal (non star) in few seconds of each exposure and high gain equals low read noise.

Since you'll be using short exposures, you really want to minimize read noise.

Only difference there is as far as SNR is concerned between one long exposure and stack of shorter exposures of equal total length is in read noise. With long exposures we minimize this impact with other noise sources - like LP noise or thermal noise or shot noise, but with short exposures - you don't have that and it is beneficial to minimize read noise.

That is the same thing planetary imagers do when utilizing lucky imaging technique with very short exposures - choose setting with the least amount of read noise (one of the reasons why ASI224 is so popular - look at read noise figures).

[pipnina]

9 hours ago, vlaiv said:

DR as defined is just ratio of highest signal recorded to read noise.

As ratio of two numbers - you can make it big by making numerator bigger or reducing denominator.

In case of HGC (if that is low read noise version) - we reduce denominator.

However - DR is simply useless as measure of anything really, so not sure why it is used.

I'd be much obliged if anyone explained usefulness of DR to me, maybe I'm missing something.

I think DR only helps us astronomers by reducing star blow-out and helping us avoid HDR techniques on high contrast targets such as M42, or imaging planets with their moons at the same exposure time.

I think it is worth saying however that, a good sensor is mostly linear (some more than others... but the IMX 571 for example is 99%+ linear as per some sharpcap screenshots I've seen.). This means one stop of dynamic range represents something close to a doubling of ADU. Which means for a 12 bit camera you genuinely cannot represent more than 12 stops of dynamic range, as each binary bit you add only allows us to double the maximum ADU value. Considering that our astrocams add offset which pushes the noise floor to above 0, this means we lose several stops already as you can wipe out 3 stops in the ADU range of 0 to 8 (1-2, 2-4, 4-8 are all 1 stop apart). Leaving only 9 stops between 8 and 4096 ADU. Thus going from 12 to 16 bit ADC not only means you can avoid posterization but actually means you improve dynamic range too.

Maybe there's a flaw in this somewhere but it's how I've currently rationalised it all in my head...

[vlaiv]

19 minutes ago, pipnina said:

I think DR only helps us astronomers by reducing star blow-out and helping us avoid HDR techniques on high contrast targets such as M42, or imaging planets with their moons at the same exposure time.

Not sure about that.

You can't escape HDR techniques in astrophotography as difference in brightness can be quite substantial - even for targets that we don't usually consider high contrast.

Take any of the brighter galaxies. Core is likely to be ~mag16 while outer reaches / spiral arms are known to go below mag26. That is 10 magnitudes of difference or more, between bright and faint parts or x10000 brightness difference. If we have SNR5 for faint parts (which means signal being about 25-30 electrons or more depending on LP) - that will make bright parts be 300000e.

No regular amateur astronomy camera can record 300000e.

In the end, we need to compress those x10000 of difference in only 8bits of intensity levels that regular images show, so we indeed must use HDR techniques. Stacking is first, non linear stretch is second ...

25 minutes ago, pipnina said:

This means one stop of dynamic range represents something close to a doubling of ADU. Which means for a 12 bit camera you genuinely cannot represent more than 12 stops of dynamic range, as each binary bit you add only allows us to double the maximum ADU value.

DR is well defined and it is log base 2 of ratio between highest signal and noise level. If you double max ADU - you will increase ratio by factor of x2 and log base 2 will increase by 1.

However - that does not mean that 12bit camera can have only up to 12bit dynamic range - it can have more than that.

Remember, it is signal level divided by read noise, regardless of bit count used to represent data. 12 bits can only hold 0-4095 values, but that does not mean all 12bit cameras are limited to capturing only 4096e worth of signal. Many have much larger FWC - like 20K or similar.

If you look at DR graph for ASI224 for example:

Al those circled data points are above 12 stops of dynamic range although camera has 12bit ADC. This is because it has 19.2K FWC (and that is more than 14 bits for max signal).

30 minutes ago, pipnina said:

Thus going from 12 to 16 bit ADC not only means you can avoid posterization but actually means you improve dynamic range too.

You really don't need 16bit camera to avoid posterization. That is myth.

All cameras in question capture more intensity levels than we are used to see on our screens (8bit), and we never see posterization if we create smooth gradient.

In fact, computer screen that I'm using right now has 6bit panel and I haven't seen posterization (6bit per channel for total of 18bits per color).

 

[pipnina]

Not entirely sure I follow the dynamic range explanation as I'm still struggling to understand how a dynamic range in electrons (FW / RN) can be expressed properly when converted to ADU, since 14 doublings of brightness can't fit in 12 doublings of 2 as a base unit. Unless the electron read noise at that high a DR is merely expressed as less than one ADU on average? Not sure if that is the right conclusion to draw.

14 minutes ago, vlaiv said:

In fact, computer screen that I'm using right now has 6bit panel and I haven't seen posterization (6bit per channel for total of 18bits per color).

Displays actually use a neat trick to get around this. When a 6-bit per pixel (BPP) screen is asked to display a brightness of 129/256, it can't send a signal of 129 to the pixel, but it can send 64/128 and 65/128 which are equally far apart from that value of 129/256. So it will choose either the higher one or lower one and flip between them every frame if looking at a static image, and the neighbouring pixels that have the same brightness are also all randomly chosen to be above/below the unsettable value of 129/256. This dithering of the values simulates an 8-bit colour depth to our eyes.

Posterization can be seen on true 8-bit panels as well, most often in video games or CGI or gradients from graphics programs. This is because there is no natural dithering in the form of noise that we'd get from a photograph for example. So we see gradients as big solid steps up. In video games it's most visible in smoke effects in dark scenes I find.

I also made a picture to demonstrate posterization when displaying a solid gradient in an 8-bit colour space, while showing a software-dithered gradient below it. It's best viewed on a high monitor brightness and has to be viewed at 100% zoom (not lower, but I think higher is fine) to not break the illusion of a better colour depth.

Opening this image in a new browser tab or a photo viewer and then looking at 100% of higher shows how lacking even true colour 8-bit displays are in some situations. Dithering is a necessary process in computer graphics to get a clean looking gradient. Sadly only HDR content and pro artist monitors use 10 or 12 BPP displays.

 

[vlaiv]

6 minutes ago, pipnina said:

Not entirely sure I follow the dynamic range explanation as I'm still struggling to understand how a dynamic range in electrons (FW / RN) can be expressed properly when converted to ADU, since 14 doublings of brightness can't fit in 12 doublings of 2 as a base unit. Unless the electron read noise at that high a DR is merely expressed as less than one ADU on average? Not sure if that is the right conclusion to draw.

Very simple really.

Say you have 12bit ADC and you have 20K FWC and maybe 2e of read noise.

DR is simply log_base_2 (20000 / 2) = log_base_2 (10000) = ~13.2877

It has nothing to do with ADU at that point.

Conversion to and from ADU is by using e/ADU value (gain). In this particular case, say that e/ADU is ~5.

You want to measure dynamic range on your sub.

You take your sub that has values in range of 0-4096 and you convert it back to electrons by multiplying with e/ADU value so you end up with 0-~20000 values

You measure max signal in electrons, you measure standard deviation of bias sub (after bias signal is removed) - and you get max electron count and value of read noise - you divide the two and take log base 2 and you get DR

What ADC is giving you is just "resolution" to write down measured number.

In above case - where e/ADU is 5 - you get to write down "to every fifth electron", meaning that electron values 0-4 will be represented by value 0 in 12bit numbers, 5-9 will be 1ADU, electron values 10-14 will be 2ADU and so on.

When you have value of say 47ADU - you don't really know how much electrons you captured, except that it was somewhere between 47*5 and 47*5+4 or 235 and 239 electrons for that pixel.

That is quantization error and that is bad thing, but if you have some read noise - it gets hidden away (noise shaping), but it has no effect on DR as it is defined.

Why then DR falls as we increase gain? Again very simple thing really

Say that we have gain set to 0.5e/ADU

You again take a sub, your sub will have 0-4095 values and when you multiply every pixel with 0.5 to get electrons - you will get only 0-2047 values out of it for electrons. Any signal above 2047e will saturate pixel at that gain - it will be too bright to accurately record.

You then do the same as above - 2047/read_noise and take log base 2

You are bound to have lower DR just by virtue of lowering max signal that you can record. You can maintain same level of DR only if read noise is reduced proportionally - but it never is.

 

[vlaiv]

I did a little search on google to really see about this dynamic range - and, sure enough, wiki article gives completely different definition of dynamic range - much more in line with what I would expect:

Quote

Dynamic range (abbreviated DR, DNR,^[1] or DYR^[2]) is the ratio between the largest and smallest values that a certain quantity can assume. It is often used in the context of signals, like sound and light. It is measured either as a ratio or as a base-10 (decibel) or base-2 (doublings, bits or stops) logarithmic value of the difference between the smallest and largest signal values

and specific to photography:

Quote

Photographers use dynamic range to describe the luminance range of a scene being photographed, or the limits of luminance range that a given digital camera or film can capture,^[52] or the opacity range of developed film images, or the reflectance range of images on photographic papers.

source:

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

No mention of read noise as it is really not important in this context (and neither is regular dynamic range in context of stacking as stacking two images - adds one bit of DR, stacking 4 images adds 2 bits and so on - one can have arbitrary dynamic range given enough time - which we already know, more time we spend imaging -better image we get ).