New Mono CMOS camera from Starlight Xpress based on IMX304

[andrew s]

Here is the reply from Terry Platt of SX a direct cut and past. It is a closed Yahoo group so you would have to join to see it.

"Hi Andrew,

I’m not sure where they get their numbers from, but they are clearly incorrect. The first fact to note is that Sony do not list the read noise for any of their sensors, so the chip manufacturer isn’t a source. The second fact is that we measured the noise at around 2.9 and 3.1 electrons on two cameras. The third fact is that even ZWO list this chip as giving 3.2 at unity gain (unity gain is at 0dB) – here is their graph:

I think that your correspondents are assuming that unity gain means maximum gain, where the read noise does fall to near 1 electron – but that isn’t what we are quoting in the specification. Max gain greatly reduces the full well depth and is not desirable for many purposes.

All the best,
Terry"

The link to the graph was missing so I have asked Terry to post it again.

 

[Adam J]

33 minutes ago, andrew s said:

Here is the reply from Terry Platt of SX a direct cut and past. It is a closed Yahoo group so you would have to join to see it.

"Hi Andrew,

I’m not sure where they get their numbers from, but they are clearly incorrect. The first fact to note is that Sony do not list the read noise for any of their sensors, so the chip manufacturer isn’t a source. The second fact is that we measured the noise at around 2.9 and 3.1 electrons on two cameras. The third fact is that even ZWO list this chip as giving 3.2 at unity gain (unity gain is at 0dB) – here is their graph:

I think that your correspondents are assuming that unity gain means maximum gain, where the read noise does fall to near 1 electron – but that isn’t what we are quoting in the specification. Max gain greatly reduces the full well depth and is not desirable for many purposes.

All the best,
Terry"

The link to the graph was missing so I have asked Terry to post it again.

 

Actually that explains everything. They are defining unity gain as 0 dB gain. All the other manufacturers define unity gain as the gain at which 1e is encoded to 1-bit of information on the 12-bit a/d and that happens at 11 dB of gain for the IMX290 not 0 dB which the other manufacturers simply call zero gain. The upshot of that is that the SX figures are vertually identical to the other manufacturers figures once you account for the difference in definition of unity gain. This is very good news.

When you look at the zwo graph you will see that the gain scale is not in dB but 0.01dB per gain increment. Unity gain is clearly not defined as zero dB on that graph but as gain 110 (11dB)

They would be best advised to use the long established definition as if people don't realise they are using a different definition of unity they will appear to be at a disadvantage when in fact it's identical.

For the record I did not think unity means max gain. 

I now regard the inconsistancy as resolved. 

Adam

 

Edited June 8, 2019 by Adam J

[andrew s]

@Adam J I am glad that is now resolved. We can all rest easy 😊

Regards Andrew 

[Ryan_86]

Thanks for your info guys. The csx-304 may be the camera for me in readiness for winter. 

I am keeping a close eye on all your comments👍🏼

Clear skies🌌

[fireballxl5]

It seems to me that SX are just using correct engineering definition of "unity" gain, i.e. a gain of 1 is 0dB. The others need to correctly define the units being used in their data sheets. 

It reminds me of amusing anecdote from work, when a report was issued with the data presented in linear form. The results were deemed unacceptable (signal levels too high) to someone on the distribution and so the author simply re-presented the data in dBs, making the values lower. There was no further comment 😏

Edited June 9, 2019 by fireballxl5
minor typo

[andrew s]

3 minutes ago, fireballxl5 said:

It seems to me that SX are just using correct engineering definition of "unity" gain, i.e. a gain of 1 is 0dB. The others need to correctly define the units being used in their data sheets. 

It reminds me of amusing anecdote from work, when a report was issued with the data presented in linear form. The results were deemed unacceptable (signal levels too high) to someone on the distribution and so the author simply re-presented the data in dBs, making the values lower. There was no further comment 😏

While I agree it has long been the case that CCD gain ( actually inverse gain) has been given in e per adu. This is the term used by James R Janesick who pioneered CCD use in astronomy.  See "Scientific Charged Coupled Devices" by JRJ.

Regards Andrew 

[Adam J]

6 hours ago, fireballxl5 said:

It seems to me that SX are just using correct engineering definition of "unity" gain, i.e. a gain of 1 is 0dB. The others need to correctly define the units being used in their data sheets. 

It reminds me of amusing anecdote from work, when a report was issued with the data presented in linear form. The results were deemed unacceptable (signal levels too high) to someone on the distribution and so the author simply re-presented the data in dBs, making the values lower. There was no further comment 😏

The others including the likes of ATIK, Orion, QHY, Altair, ASI do define the units being used in their data sheets I'll post example when not on my phone. They all use the same definition of unity gain. But who is technically correct is not the issue. The issue is that unlike myself and you and some others here I am willing to bet most people don't think in dB at all. As such it will likely confuse the majority of their customers sufficiently such that they see one noise level and the other and they just pick the lower value, or as above they wonder if the anti-amp glow is increasing read noise, that would be a real shame for a good company such as SX. Prior to the response from SX it even confused everyone in this thread full of people who understand dB myself included. Sometimes you should just go with the flow 🙂 

Edited June 9, 2019 by Adam J

[vlaiv]

Haven't read the whole topic (will do tomorrow), but in my view both unity gain and gain in dB how most manufacturers use it is quite right way to do it.

Gain in dB is relative measure - it needs a reference point, in same way that magnitudes (stellar) need reference point. Choosing unity gain to be reference point might make sense, but it is likely to confuse most people as we would end up with both positive and negative gain (in the same way we have positive and negative mags when using Vega as reference point). It makes more sense to keep gain positive and choose reference point to be in terms of other characteristics of sensor - quite reasonably manufacturers choose gain value at which ADC matches full well capacity. This is a good baseline value - 0dB gain in this case will let you gather as much light per pixel as sensor allows (is designed to do) - no "amplification", hence 0 value.

Unity gain as term then makes complete sense - it is gain in dB units at which there is 1:1 correspondence (unity conversion factor)  between e and ADU.

[andrew s]

18 hours ago, andrew s said:

While I agree it has long been the case that CCD gain ( actually inverse gain) has been given in e per adu. This is the term used by James R Janesick who pioneered CCD use in astronomy.  See "Scientific Charged Coupled Devices" by JRJ.

Regards Andrew 

I put this to SX and here is the reply from Terry

"Yes, electrons per ADU is universally used to define the output of a CCD after A-D conversion, but it’s an ‘after the process’ gain, not the gain of the variable amplifier that CMOS chips include. In CMOS, there is a ‘gain factor’ applied before A-D conversion and this is defined in terms of 0 – N decibels (typically 0 – 40dB). This amplifier gain can be set in software, so we define its gain as unity when giving the base value of readout noise. It isn’t the same thing as the electrons per ADU that is often quoted in specifications. However, I agree that we need to be a bit more specific in our data sheet.

 

Regards,

Terry"

Regards Andrew

[Thommy]

On 07/06/2019 at 22:51, Adam J said:

What application do you have in mind for that camera / what scope are you thinking of matching it to? It's potentially a very sensitive camera. A replacement for icx825 based cameras like the atik 414ex, potentially as sensitive with a larger sensor and similar pixel size. 

I currently use a SX MX7 but with a retrofitted Sony 829 chip in combo with a MN190. The large pixels (not square!) gives me a resolution of around 1.9 "/px which is just too low. CSX 249 would increase the resolution to 1.3 "/px and increase the field size a bit as well. That´s fine with me as I normally image galaxies spanning no more than 15-20'. Actually I consider the new SX Trius 825 or 694 Pro or as well, should I decide to continue with CCD's. Anyway, the discussion in this thread brings some interesting information...

[Thommy]

CSX-249 would be sensitive due to the pixels size. With regard to dynamic behavior it is not optimal. With a full well depth of 33k and, say 2 e read noise (should be lower  according to preliminary specifications), it has a dynamic range of 1:16500 which is very good, BUT since dataformat is 12 bit, you only have 4096 stops. For this camera a data format of 14 bit would be more appropriate to take advantage of its dynamic range. For CSX-304 the dynamic range is 1:5500 which is consistent with 12 bit.

[Adam J]

Still not a single image taken by this camera (CSX-304) online as far as I can tell. 

Adam

[pete_l]

On 13/03/2019 at 16:14, andyboy1970 said:

he said that the Asi1600 cameras have dark frames subtracted on camera so the data that arrives on your laptop isn't actually RAW data but already has dark frame subtracted to limit the amp glow.

An easy thing to (dis)prove. Take a 30s exposure. If the image is delivered directly after the exposure finishes, no dark frame was subtracted. If you have to wait another 30s, you can deduce that it did subtract a dark.
Of course, this presumes that the camera did take a true 30s exposure, not a 15s one followed by a 15s dark. But again that is easy to verify by comparing a 30s exposure from the same chip in another manufacturer's camera with the ASI one.

I think I can sniff a bovine, here.

[JamesF]

25 minutes ago, pete_l said:

An easy thing to (dis)prove. Take a 30s exposure. If the image is delivered directly after the exposure finishes, no dark frame was subtracted. If you have to wait another 30s, you can deduce that it did subtract a dark.
Of course, this presumes that the camera did take a true 30s exposure, not a 15s one followed by a 15s dark. But again that is easy to verify by comparing a 30s exposure from the same chip in another manufacturer's camera with the ASI one.

I think I can sniff a bovine, here.

I think if it worked either of those ways then people doing, say, 600 second exposures would have been generating a fair bit of fuss.

James

[Stub Mandrel]

On 27/01/2020 at 18:12, pete_l said:

On 13/03/2019 at 15:14, andyboy1970 said:

he said that the Asi1600 cameras have dark frames subtracted on camera so the data that arrives on your laptop isn't actually RAW data but already has dark frame subtracted to limit the amp glow.

An easy thing to (dis)prove. Take a 30s exposure. If the image is delivered directly after the exposure finishes, no dark frame was subtracted. If you have to wait another 30s, you can deduce that it did subtract a dark.
Of course, this presumes that the camera did take a true 30s exposure, not a 15s one followed by a 15s dark. But again that is easy to verify by comparing a 30s exposure from the same chip in another manufacturer's camera with the ASI one.

I think I can sniff a bovine, here.

Definitely wrong

Best way to disprove is to look at a 5 minute sub and compare to one with darks subtracted - the very faint amp glow disappears.

[Adam J]

https://www.gxccd.com/art?id=579&cat=1&lang=409

Ok so it looks like Moravian Instruments has just brought out a camera with the same sensor.

and if you scroll right to the bottom of that link unlike SX they actually have example images!

Also at $1500 that is a much more realistic price than the SX version.

Adam