Sensor size, complexity, pixel size and cost?

[hjulenissen]

 

[Erik Kaffehr]

• It may also be that camera electronics can have difficulty to handle larger amount of data within limited power budget.

I'm not a sensor expert but I do know about A/D converters. The higher the resolution (pixel count), the longer it takes for the A/D to convert all those analog pixel readings into their digital values, which affects shooting rate and power dissipation.

Would not a bsi sensor with a higher number of embedded adcs offset that (at a higher sensor complexity/power cost)?

Multiple A/Ds in parallel would definitely fix the A/D throughput problem. But embedded in the sensor? I'm not aware of camera sensors with embedded A/Ds. I can think of a lot of circuit problems that would create and not many advantages.

Practically every sensor in every camera you buy today has embedded column ADCs. Though they weren't the originators of the technology, Sony introduced it to the still camera market with the IMX021 in the Nikon D300 and Sony A700. Every Samsung sensor used it. Panasonic used the architecture from the GH2 onwards, Canon has used it from the 1D X II and most of their cameras now use it.

It's actually much more likely that Canon's image sensors —

No, it isn't much more likely. By and large, I do like to do a bit of prior research before making a post.

including the ones in the 1D X II and 5D4 — do not have on-chip ADCs and still use an external A/D conversion circuitry.

https://www.dpreview.com/forums/post/61749398

That's a 5Ds, not a 1D XII or 5D4. The 1D X II was, so far as I know the first Canon camera to have column ADCs and was widely reported to be so at the time.

The picture is actually of a 5D4 PCB...

OK, lost track of which image you were referring to.

As to your "doing a bit of prior research", it seems to have escaped your attention that that PCB sports a chip manufactured by Analog Devices, Inc. — have you ever heard of that company ?

Indeed, AD, makers of all kinds of chip, including such things as LDVS transceivers:

https://www.analog.com/en/parametricsearch/11462

which would typically be used to interface the output from a digital output sensors (which generally interface to LVDS) to a processor which doesn't have an LVDS interface (not so many do, built-in)

These LVDS transceivers are functionally simple devices, and come in small, in-line packages with a limited number of leads — around 8 to 20, they are just serial protocol conversion chips, after all — and their package looks completely different from the ADI chip found in the 5DIV.

The ADI chip on the 5DIV PCB is a square — presumably BGA — chip that bears a striking physical resemblance to the ADI ADC chips customarily used in Canon's DSLRs.

Some of the LDVS buffer chips come in BGA packages, and it possibly is not just a simple buffer. It's possibly a deserialiser, or other decoding function. AD makes all sorts of stuff. It doesn't match any of their AFEs.

Possibly the best way to find out what it is is to ask Roger Cicala, who does post here to look at the original, higher res version of the image and see whether the chip can be identified. Then we'll know for sure, and if I'm wrong, I will immediately accept it. Until then, your assumption that the chip must be an AFE doesn't hold much water. Can you even identify a candidate part?

Logic doesn't seem to be your forte. There's no need for me to identify a candidate part — I've already told you that, as Canon likely buys millions of these chips, the part is quite likely a custom part specific for Canon, with specs not found anywhere in ADI's publcly available information.

Furthermore, you're implying that Canon's engineers were clever enough to add LVDS transceivers to their image sensor, but not to their latest DIGIC chip. Canon's engineers probbaly din't realize that their image sensors were going to be interfaced with their DIGIC chips. That sounds soooo plausible.

What do you think the DIGIC chips are? They are off-the-shelf application processor chips. Canon does not design them, Texas Instruments do. It's quite possible that for a quite esoteric interface, it was worth using an additional interface chip.

The fact that Canon likely uses / licenses TI's ARM CPU and SoC-related intellectual property doesn't automatically mean that DIGICs are "off-the-shelf" chips. Given Canon's purchasing power and in-house design capabilities, it's quite possible that Canon actually design a significant part of the functionality of the ASIC they name "DIGIC", designing and adding to it various functionalities they might need for their cameras — e.g. I/O interfaces connected to the image sensor, memory cards, rear LCD, motor control etc.

The resulting ASIC, incorporating Canon's and TI's IP, would then be manufactured at a foundry facility.

Here's a page that hints that Canon — which is a multi-billion high-technology company, after all — might well have in-house EDA tools and could be using VHDLs to design their own logic chips:

"[..] Ensuring Efficient Development of Large-Scale System LSI

Canon develops its own system LSIs, ..."

http://www.canon.com/technology/now/element/cp.html

The other problem that you have, is if you believe the unidentified chip do be an AFE chip, that AD does not have one that handles 8 channels (which is why, for instance, the 5D III required four 2 channel chips)

https://www.analog.com/en/parametricsearch/11269

Given that Canon produces millions of DSLRs per year, it's quite likely that ADI is willing to produce custom ADC / analog front-end chips, whose reference numbers and specs won't be found on ADI's public product list.

It's possible, but there is the interesting fact that Canon's been happy to use off the shelf AD devices before,

Like these four AFE chips from the 1DX II

AD80246 as also used in 1200D, Rebel T5, Canon 5D mark III, Canon 600D; Canon, Canon Rebel T3i; Canon 6D, Canon G1x.

and generally, one would expect a custom chip to be labelled 'Canon', if Canon retained the IP. If Canon didn't, it would go in the AD catalog.

It doesn't seem to occur to you that the AD80246 is not an "off-the-shelf" part. The fact that a Nikon D850 might contain an image sensor marked Sony, IMX-blah-blah doesn't mean that the D850 uses an "off-the-shelf" part, for example.

There's no publicly available information for the AD80246 anywhere, which would tend to indicate that it's a custom development for one specific customer; the part is not publicized, and does not seem to be avaialble for purchase by customers other than Canon.

Also worth noting is the signal conditioning components right on the analog inputs, which seem to absent on the 1D XII. Notice also on the 1D X, they are on the back of the sensor assembly, not on the main PCB. I wonder why they changed that?

Huh ? Which "signal conditioning components" ? How could you tell ? Where have you seen a picture of an 1DXII sensor assembly, for example ?

The "reports" one reads on the Internet that the 1D X II has on-chip ADCs are just based on hearsay of uninformed speculation and unfounded assumptions — e.g. that one can tell (by the color ?) of the peripheral areas visible on a low-res picture of a sensor package that they are column ADCs. I find such assertions quite ridiculous, to be honest.

You can find it what you like. It's one of a number of indicators:

- the performance characterisation of the chip.

There are other possible avenues for DR improvement.

https://www.dpreview.com/forums/post/61785976

I wasn't talking about DR improvement, I was talking about the read noise characteristic. Different thing.

You seem not to be even aware that noise degrades DR, and that DR can thus be improved by reducing noise, including of the "read noise" variety...

- the layout of the chip interface, which is atypical of analog output chips, which typically have connections each end. Column ADC chips generally have power supplies distributed round all four sides, the reason being that if you don't, the column ADCs power consumption disrupts the rather delicate power supply arrangements to the pixel array

- the oversize top and bottom borders, required to accommodate the ADCs

- the visible ADC structures on the chip

Oversize top and bottom borders do not imply that they must be ADC structures, as such borders were already present in image sensors contemporaneous with the first generation Canon 1D X:

https://cweb.canon.jp/eos/lineup/1dx/feature-highquality.html

Lessly oversize,

The 1 DX chip is classic Canon, classic analog output, no complex power supply, connections concentrated at the ends, smaller borders top and bottom. The 1 DX II chip is a break in years of practice, with power supply distribution top and bottom, (to what, I wonder) what looks very like a digital IO on the left hand side. Actually quite similar to a Sony, but more tidily arranged.

and without the visible structures running along just where you'd expect the busses for a column ADC to run.

The electrical contact layout might as well have been influenced by the chip carrier's electrical and thermal characteristics, and tighter path length constraints between the silicon and the package's pins.

The 1DX has a 18MP sensor and a 12fps frame rate, and the sensor must thus have a 220 megapixels/sec output bandwidth.

The 1DX II has a 20MP sensor, but can also record 4K video at 60FPS. As a 4K frame contains about 8 million pixels, the 1DX II's sensor must have an output bandwidth of about 480 megapixels/sec. The much higher bandwidth requirement could have driven the chip designers to bring some electrical contacts closer to the middle of the chip, so that the output signals have a shorter path to travel.

I, on the other hand, do not have to rely on chip border width,

I'm not relying on any one thing, I produce a number of things together, and say that on the balance of probability Canon has produced column ADCs for this chip and the subsequent ones which behave similarly. See one duck, it might be a goose. See a couple, you get more sure they're ducks see five, you become pretty sure they're ducks.

and can simply point towards an ADI chip on the 5DIV's PCB that bears a striking resemblance with ADI's ADC chips found in other Canon cameras.

Very little resemblance, in fact, apart from the AD logo, and AD makes a huge variety of chips. All you have is that there is a chip labelled AD in the path from sensor to processor. You have no idea what it is.

"Very little resemblance" ? REALLY ?


Unidentified Analog Devices chip on 5DIV PCB (left) vs. AD80246 ADC/AFE found on 5DIII's sensor board (right)

- a number of Canon patents showing that they have been developing the technology.

It would be a bit silly to draw conclusions from one of those things, almost as silly as drawing a conclusion from a single unidentified AD chip on a PCB, but when you put them together, it's likely, in the absence of contrary evidence, that column ADCs is the correct diagnosis. After all, why would Canon not have produced column ADCs? Every other sensor manufacturer has.

Canon have certainly been working on the technology, but nothing indicates that they turned it into mass production reality yet.

Nothing indicates that they haven't. Why do you believe that they would be so laggardly when every other manufacturer and their dogs have column ADCs

There might also be "blocking" patents — e.g. related to noise processing / digital CDS — already owned by competitors with much larger semiconductor / image sensor businesses, and who have done extensive R&D and actual manufacturing and patenting in the relevant analog and digital circuit technologies well before Canon.

So these are 'blocking patents' after the technology they're trying to block has been developed by someone else? That doesn't work. In any case, as I said, Canon has certainly produced demonstrator chips with column ADC, and they did it before the release of the 1D XII. Funny that.

Sorry to have been unclear. When I suggested that "blocking" patents might exist, I of course meant that Canon's competitors might be owning these patents, which would then "block" Canon from using some technology.

Your argument is thus basically of the level of people invoking "akums razor" [sic].

Also see https://www.dpreview.com/forums/post/61755597

The only argument that you have is a chip that could be many things. You think that you have discovered something big, but all you've discovered is a chip with AD on it. Identify what it is, I'd be more impressed.

And one other thing, when it was released, Canon said it had ADC on the sensor chip:

First of all, there’s the sensor, which includes a number of firsts for Canon. It’s a 20.2-million-pixel full frame CMOS sensor that’s the firm’s first to use on-chip analogue to digital conversion. Canon says that by shortening the signal pathways used when reading the signal off the sensor, this should result in lower noise, with the main benefit being increased dynamic range and cleaner shadows at low ISOs.

https://www.trustedreviews.com/reviews/canon-eos-1d-x-mark-ii

Quoting people who are just parroting stuff they've read on some Internet forum or rumor site, and have no capability and willingness to actually verify by themselves whether, say, a signal carried by a flex cable is analog or digital adds absolutely zero credibility to your argument.

The modders at centralds.net, by the way, say "We guess that the CMOS cable delivers only digital signals."

Let me enlighten you: "guess" means, they do not actually know.

So far, you have one piece of evidence. An unknown chip form a manufacturer that makes many. That is the reason that you need to identify the chip to make your case, until you do that, the rest is bluster. I now have seven pieces of evidence, and could doubtless gather more. Non is but itself conclusive, but neither is your counter evidence, so where we stand now, my seven is beating your one. You have to find something else to make the case.

Given the intelelctual standards you've demonstrated when qualifying something as "evidence", I pity the person who'll have to stand trial before a jury containing the likes of you...

So, you're signalling the end of a rational discussion, and on to the ad-hominem stuff in the absence of anything else to support your case. Time, I think, to call a halt.

As soon as you start to qualify as "evidence" something that isn't — e.g. postulating that an oversize border on the sensor chip is necessarily an ADC circuitry, or citing random pronouncements found on the Internet, or postulating unplausible roles like LVDS conversion for a particular chip, etc. — then, the discussion belongs more to the domain of fantasy than rationality.

Hi,

A closely related question, do you also think earth is flat?

There is a a lot of hearsay evidence to the contrary.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Xasan]

Hi,

It would be interesting to hear from folks knowledgeable about the cost aspects of modern sensor designs.

My understanding is that increasing the sensor size over proportionally expensive.

But, I would assume that increasing the number of pixels would not be very expensive.

The big dilemma is this: to put serious money into R&D to improve the current generation of sensors, or to put it into R&D aimed at newer technologies.

Incremental sensor improvements resulted in Nikon Z and Canon R underachieving with respect to image quality.

The hope is that as soon as new technologies are introduced pixel count can be addressed.

Clearly, increasing the complexity will increase the probability of component failure, but I also think that bad pixels can be mapped out. The complexity of the non pixel components should probably increase with the square root of pixel components. Doubling the number of pixels increases number of columns with sqrt(2).

The trend is to delegate more and more functions to the pixel level.

It seems that APS-C is pretty much stuck at 24 MP and 24x36 is moving slowly upwards from 36 MP.

• There can be quite a few reasons for the slow development:
• Optimum balance between noise and resolution may have been reached.
• It may be that we get into diminishing returns with regard to resolution.
• It may also be that camera electronics can have difficulty to handle larger amount of data within limited power budget.
• Cost obviously is a factor that plays an important role.

On the other hand, we get into a situation when medium format uses smaller pixel size than 24x36 mm, something like 3.8 microns on the 100 MP 44x33 mm sensor.

Any comments from sensor experts?

Best regards

Erik

Hi,

I wouldn't say that either the 5Ds or the Z7 underachieve in image quality.

EOS R, not 5Ds.

Since you've mentioned 5Ds, EOS R is perceived to be better than EOS R (higher res). D850 is perceived to be better than Z 7 (no banding, AF).

But of course you are right, all of those are absolutely formidable cameras. Sadly, it doesn't mean all and everything, and never was.

To be clear, saying "underachieve" I mean the comparative perception of image quality by the opinion makers (compared to what they think is state of art in the niche+), and thus the opinion of the substantial part of the market. Tony N., 3 negative Z 7 reviews, total 700,000 views, that's not something to ignore.

Absolute technical numbers don't drive marketing of camera bodies, and marketing decides where the moneys go.

The issue is not more pixels or smaller pixels. The issue is new pixels.

Reviewers criticizing the current crop of new cameras are putting impediments to incremental upgrades. In a sense it is useful.

But, that may also depend on your definition of image quality. It is not meaningful to discuss image quality before we have a definition of the term.

Best regards

ErikHi,

I am a landscape photographer, essentially shooting base ISO on tripod.

In my use case the 5DsR offers the best detail of all 24x36 cameras, but may lack DR at base ISO. I don't own a 5DsR, I do my shooting on a Sony A7rII.

The 5DIV uses a sensor technology much more like the one Sony uses.

I have a close friend who shoots Canon, but our way of shooting is very different, so even if we shoot the same things at the same occasions, our images will be very different.
The way I shoot, I would still feel that I am more limited by sensor than limited by lens.
So, I would appreciate to have a sensor that actually resolves all the information coming from the lens...

Best regards

Erik

We are once again at the point where the architecture for small and large pixels and sensors needs to be different, but we will reach better convergence in not so distant future.

 

[Xasan]

you need to identify the chip to make your case

Identifying the chip doesn't make the case. External ADCs may coexist with internal ones, to be used for video, certain raw modes, rgb measurements in real time, other needs.

A more useful data may be this:

 

[bobn2]

you need to identify the chip to make your case

Identifying the chip doesn't make the case. External ADCs may coexist with internal ones, to be used for video, certain raw modes, rgb measurements in real time, other needs.

Possible, but I wouldn't actually think very likely. If he can identify it as an ADC chip I would think that's curtains for my argument. On the other hand, part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

But, there is plenty of scope for a 'front-end-processing' circuit. I'm working with a Sony sensor at the moment, and the 8 line serialised LVDS data is not handled by any off-the-shelf applications processor - nor is it something you can just feed into the processor and expect it to decode sensibly. If you're not in volume, about the only option is to use an FPGA to deserialise it and repackage it into something the processor can handle. So, it doesn't surprise me that there is a chip there, nor does it surprise me that it's an AD part, they do all kinds of video conversion chips, deserialisers and the like.

A more useful data may be this:

--
Ride easy, William.
Bob

 

[Xasan]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

 

[Antisthenes]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

As English obviously isn't your first language, perhaps you don't realize that there's a significant ambiguity as to who's meant by the "he" in this sentence of yours: "Contrary to what he says Canon does use front-end".

Or maybe you don't know what "Contrary" means in English.

Anyway, let me also point out that "front end", a term that happens to be used by Analog Devices, Inc. to describe their camera-oriented ADCs, would be rendered as "フロントエンド" in Japanese katakana:

https://www.dpreview.com/forums/post/61755597

 

[Antisthenes]

you need to identify the chip to make your case

Identifying the chip doesn't make the case. External ADCs may coexist with internal ones, to be used for video, certain raw modes, rgb measurements in real time, other needs.

Possible, but I wouldn't actually think very likely. If he can identify it as an ADC chip I would think that's curtains for my argument. On the other hand, part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Logic doesn't seem to be your forte.

Canon might have significant control of the DIGIC's functionalities, but you apparently don't realize that the DIGIC chip is, essentially, a "logic" chip — i.e. a digital chip.

Analog chips, in general, require design methodologies and semiconductor manufacturing processes that differ from digital logic. High-speed ADCs are thus the kind of circuit where companies like Analog Devices, Inc. (ADI) have a significant accumulated experience and competitive advantage, and attract clients like Canon.

The issue is simple: you argue that the oversize borders e.g. of the 1DXII's image sensor chip must be ADC circuitry, and that the Analog Devices chip found on the 5DIV's PCB must thus be some kind of digital device, e.g. a LVDS converter. I argue that aspects like oversize chip borders aren't significant, and explanations about requiring a digital protocol conversion between Canon's own chips aren't very plausible, and that a chip that has a striking physical resemblance to the ADC chips used in other DSLRs could be, well, an ADC, and that it's thus quite imprudent to proclaim that Canon cameras have moved to on-sensor ADCs and don't use off-chip ADCs anymore.

Incidentally, reducing the physical length of digital output lines can improve a system's noise figure:

https://www.dpreview.com/forums/post/61785976

There's thus a justified reason to reduce as much as possible the physical length of the connection between the ADC and the consumer of the digital data.

One could thus also easily imagine, as an evolution of that trend to reduce connection length — and considering modern chip-scale packaging know-how driven e.g. by the huge smartphone market — a stacked 3D integration e.g. in package-on-package form of an ADC die supplied by ADI, a RAM die from a memory supplier and a DIGIC die coming from a foundry, in a single physical package.

 

[bobn2]

Logic (with the exception of the now obsolete ECL) drives the transistors into saturation. Analog doesn't (if it did, it wouldn't be analog). Low voltage and current based signalling by its very nature uses unsaturated drivers, and is typically based on mixed signal processes. That is why a mixed signal specialist, such as Analog Devices, has such a large portfolio of LVDS and similar interface chips. It is also why processors implemented using pure logic processes often don't have such interfaces integrated.

 

[Detail Man]

Logic (with the exception of the now obsolete ECL) drives the transistors into saturation.

True enough, my knowledgeable friend. It's cool that bipolar junction transistors in linear mode can do a faster job of transmitting "quanta" than (saturating) logic gates - where the total current draw (with the higher quiescent currents involved) can be comparatively lower than the total current draw of CMOS construction gates at switching frequencies > 20 MHz.

 

[bobn2]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

He will argue about the semantics of 'front-end'. The Canon correspondent you have is from their support office, I'm not sure that what she said will bear too much dissection.

Still, on the topic of front ends, I found this tear-down of a Nikon D800:

http://www.fixyourcamera.org/nikon-d800/

I don't think that anyone would deny this has a column-ADC sensor (actually, the Sony IMX094). The data interface to the sensor is via the white connector, top centre-left of the board. Notice that it interfaces to the main processor (the chip labelled 'Expeed') via a 'front-end' chip (the other one labelled 'Nikon').

--
Ride easy, William.
Bob

 

[Erik Kaffehr]

• It may also be that camera electronics can have difficulty to handle larger amount of data within limited power budget.

I'm not a sensor expert but I do know about A/D converters. The higher the resolution (pixel count), the longer it takes for the A/D to convert all those analog pixel readings into their digital values, which affects shooting rate and power dissipation.

Hi,

Most recent sensors have ADCs on chip mostly doing column wise conversion.

So doubling megapixels increases the number of ADCs by 41% each ADC in turns needs to be 41% more conversions.

With off sensor ADCs very high conversion rates are needed. Say the device has 6 ADCs and 24 MP, shooting 5 FPS. In that case, each ADC needs handle 5*24E6/6 = 20E6 conversions per second.

With same MP count we would have 6000 converters, so the conversion rate would 5*24e6/6000 = 2E4. These converters are probably ramp type converters.

I would think that some professional Nikon models have off sensor conversion while Nikons using Sony sensors use on sensor conversion. Canon seems to have switched from off sensor to on sensor conversion based on noise characteristics .

Best regards

Erik

 

[bobn2]

Logic (with the exception of the now obsolete ECL) drives the transistors into saturation.

True enough, my knowledgeable friend. It's cool that bipolar junction transistors in linear mode can do a faster job of transmitting "quanta" than (saturating) logic gates - where the total current draw (with the higher quiescent currents involved) can be comparatively lower than the total current draw of CMOS construction gates at switching frequencies > 20 MHz.

I'm not sure they are necessarily bipolar transistors, a whole load of mixed signal is exclusively MOS. What's interesting is the particular nature of the data connection, which appears to be multiple twisted pairs rather than the normal flex,





which in turn means pretty much certainly it isn't analog. Synchronising parallel analog channels operating at those speeds on a cable like that would be a nightmare (and the reason that on the 1D X, digitisation happened on the sensor card). It simply wouldn't survive the hacked cable extension that the modders in the link I posted earlier did. Still, there is the question of what kind of interface would need a cable like that, and one answer is TMDS. Analog Devices has a slew of TMDS interface products.

--
Ride easy, William.
Bob

 

[John Sheehy]

I would think that some professional Nikon models have off sensor conversion while Nikons using Sony sensors use on sensor conversion. Canon seems to have switched from off sensor to on sensor conversion based on noise characteristics .

Be careful with that type of analysis; the number of pixels plays a huge role in base-ISO PDR (or DxO "Print" DR). Look at their ranking in terms of pixel count, and there is a very strong correlation.

For any given type of sensor readout, the more pixels, generally speaking, the higher the base-ISO DR, because most of "pixel-level" post-gain read noise has nothing to do with the pixels themselves, and everything to do with the signal path beyond the amplifiers.

 

[Xasan]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

As English obviously isn't your first language,

So? Digital cameras are not your line, but you have no problem discussing digital cameras.

perhaps you don't realize that there's a significant ambiguity as to who's meant by the "he" in this sentence of yours: "Contrary to what he says Canon does use front-end".

Judging by your nervous response my English is good enough for you to understand exactly what I meant.

Or maybe you don't know what "Contrary" means in English.

Oh, I do.

Anyway, let me also point out that "front end", a term that happens to be used by Analog Devices, Inc. to describe their camera-oriented ADCs, would be rendered as "フロントエンド" in Japanese katakana

I know. Front-end means other things as well. I doubt that it is important how AD uses the term as the diagram is drawn by Canon. If Canon's response is not clear to you little can I do.

AD also uses "front-end" in a variety of manners, you can come across "Integrated Mixed-Signal Front End", "FET Input Analog Front End With ADC Driver", which means that to AD front-end is also something more than just a "camera-oriented ADC". To say it is used by AD "to describe their camera-oriented ADCs", as you did, is incorrect. You either don't know, or you are trying to make a point whatever it takes.

 

[Erik Kaffehr]

I would think that some professional Nikon models have off sensor conversion while Nikons using Sony sensors use on sensor conversion. Canon seems to have switched from off sensor to on sensor conversion based on noise characteristics .

Be careful with that type of analysis; the number of pixels plays a huge role in base-ISO PDR (or DxO "Print" DR). Look at their ranking in terms of pixel count, and there is a very strong correlation.

For any given type of sensor readout, the more pixels, generally speaking, the higher the base-ISO DR, because most of "pixel-level" post-gain read noise has nothing to do with the pixels themselves, and everything to do with the signal path beyond the amplifiers.

Hi John,

Sorry to say, I absolutely disagree.

Let's compare Canon 5DIII and Canon 5DIV, with the later model having 30 MP while the older having 23 MP:




You see that Canon 5DIV is nearly ISO-less DR drops evenly with ISO, the 5DIII has pretty constant DR up to 400 ISO. The ISO-less behaviour is pretty typical of on sensor column converters.




Comparing Sony's gen II models the 42 MP A7rII has best DR at base ISO and the 12 MP A7sII has lowest DR at base ISO.

Comparing Nikon's latest generation:






The Nikon D850 at 46 MP has much higher DR at base ISO and drops linearly with exposure, The increase around 400 ISO is the DR-PIX tech that Sony licensed from Aptina. The D5 has much lower DR at base ISO, but it is near constant to say 800 ISO, again D850 behaviour is characteristic for on sensor column converters while D5 is typical for off sensor ADCs with variable analogue gain.

Best regards

Erik

--
Erik Kaffehr
Website: http://echophoto.dnsalias.net
Magic uses to disappear in controlled experiments…
Gallery: http://echophoto.smugmug.com
Articles: http://echophoto.dnsalias.net/ekr/index.php/photoarticles

 

[Antisthenes]

Logic (with the exception of the now obsolete ECL) drives the transistors into saturation.

True enough, my knowledgeable friend. It's cool that bipolar junction transistors in linear mode can do a faster job of transmitting "quanta" than (saturating) logic gates - where the total current draw (with the higher quiescent currents involved) can be comparatively lower than the total current draw of CMOS construction gates at switching frequencies > 20 MHz.

I'm not sure they are necessarily bipolar transistors, a whole load of mixed signal is exclusively MOS. What's interesting is the particular nature of the data connection, which appears to be multiple twisted pairs rather than the normal flex,



which in turn means pretty much certainly it isn't analog.

The 5DIV can take 30MP pictures at 7fps. The pixel rate from the sensor must thus be at least 210 megapixels/sec. At 14 bits per pixel, the raw bit rate would thus be about 3 gigabits/sec. If the flex cable connecting the sensor board to the main board consisted of, say, 12 digital lanes, then each lane would have to transmit about 250 megabits/sec. The analog bandwidth required to transmit without too much distortion the sharp voltage transitions of a 250Mb/s digital bitstream would reach well into the GHz range, where transmission line issues like characteristic impedance and line topology matter.

Could the 5DIV's blue flex cable, with its 90-degree folding and kinks, carry a 3 Gigabits/s data stream using e.g. TMDS ?

Here's a relevant layout rule:

"[..] Signals within a TMDS pair shall not have any 90 degree corners. Corners shall be chamfered."
https://ez.analog.com/video/w/documents/687/hdmi-layout-guideline

Hmm. it thus seems that, for high-speed digital transmission lines, sharp 90-degree corners and turns are frowned upon.

Now, let's look at what actual digital transmission channels, coming either from ADCs (Canon cameras) or from image sensors with digital outputs (Nikon and Sony cameras) look like:

Canon 1DX:


Canon 1DX digital pixel transmisson channel: two smooth, white cable bundles with no kinks or tight 90º turns

Canon 5D3:


Canon 5D3 digital pixel transmisson channel: smooth cable bundle with no kinks or tight 90º turns

Canon 6D:


Canon 6D digital pixel transmisson channel: straight flex cable, no kinks or 90-degree folds

Nikon D600:


Nikon D600 digital pixel transmisson channel: straight flex cable, no kinks or 90-degree folds

Nikon D800:


Nikon D800 digital pixel transmisson channel: straight flex cable, no kinks or 90-degree folds

Sony A7R2 :


Sony A7R2 digital pixel transmisson channel: straight flex cable, no kinks or 90-degree folds

So, on the one hand, we have this 5DIV's blue flex cable, with its kinks and tight 90-degree folds, which you postulate must be carrying a multi-gigabit digital signal requiring GHz+ bandwidths.

On the other hand, looking at actual digital transmission cables from Canon, Nikon and Sony cameras, we see that none of them has folds, kinks and tight 90-degree turns, and are thus completely different from the 5DIV's cable topology.

I wonder why. Could it be that the engineers at Canon, Nikon and Sony, when they design the physical transmission channels that carry high-speed digital data, actually take into account well-known guidelines about the topology of the signal paths ? Like, say, avoiding tight 90º folds or turns ?

Could it be that the 5DIV flex cable thus carries a signal with a much lower bandwidth requirement than a GHz+ digital channel ? Say, an analog signal requiring about 20MHz, between the sensor board and an ADC chip on the main circuit board ? The Analog Devices chip on the 5DIV's main circuit board looks remarkably similar to the Analog Devices ADC chips found in other Canon DSLRs, after all.

Hmm.

Synchronising parallel analog channels operating at those speeds on a cable like that would be a nightmare (and the reason that on the 1D X, digitisation happened on the sensor card). It simply wouldn't survive the hacked cable extension that the modders in the link I posted earlier did.

Amusing assertion.

A 30MP Canon 5DIV's sensor at 7fps would spew out 210 megapixels/s.
Transmitted over a flex cable consisting of, say, 12 analog channels, each channel would have to transmit about 18 megapixels/s. As an analog, video-like baseband signal, the required bandwidth per channel would thus be slightly less than 20 megapixels/s, or around 20MHz of analog baseband bandwidth. This is a far cry from the 1GHz+ analog bandwidth that would be required if 210 million pixels encoded at 14 bits were transmitted in digital format.

In coaxial cable, the propagation speed of a signal would be about 200,000 km/s.

At 20 MPix/s, each pixel would occupy about 10 meters of cable.

This means that, when transmitting pixels over 12 parallel analog lanes, if their line length difference is less than 10 meters, then, the inter-lane timing skew at the receiver's end would be less than 1 pixel.

Cutting a bundle of micro-coax cables so that their individual lengths are similar, to within a few millimeters, is trivial.

Synchronising parallel analog channels in a near-perfect fashion, for baseband transmissions requiring about 20MHz of bandwidth, is thus a non-issue.

So, to recap, you're now blithely postulating that:

1. Canon's engineers were clever enough to implement a communication protocol — LVDS / TMDS / SLVS / whatever — on their image sensor, but too dumb to implement it in their DIGIC chip, perhaps because they didn't realize that the image sensor would likely be communicating with the DIGIC chip.
2. The 5DIV's flex cable, despite having kinks and bends and 90-degree folds that are frowned upon for high-speed digital data transmission channels, is necessarily a high-speed LVDS / TMDS / SLVS / whatever digital data transmission channel.
3. The centralds.net modders are too clumsy and wouldn't even be able to adjust the lengths of the micro-coax cables in their cable bundles so that there's less than, say, a 10-meter length difference between the individual cables.

What next ? Are you going to assert that the slightly different blue hue of the oversize sensor chip edges in the pictures of the 1DX and 1DX II sensors establishes beyond doubt that the 1DX sensor is analog, while the 1DX II has on-chip ADC circuitry ?

 

[Antisthenes]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

He will argue about the semantics of 'front-end'. The Canon correspondent you have is from their support office, I'm not sure that what she said will bear too much dissection.

Still, on the topic of front ends, I found this tear-down of a Nikon D800:

http://www.fixyourcamera.org/nikon-d800/

I don't think that anyone would deny this has a column-ADC sensor (actually, the Sony IMX094). The data interface to the sensor is via the white connector, top centre-left of the board. Notice that it interfaces to the main processor (the chip labelled 'Expeed') via a 'front-end' chip (the other one labelled 'Nikon').

Let's see:

Nikon D7200: Toshiba sensor
Nikon D5: Toshiba sensor
Nikon D800: Sony Sensor
Nikon D850: Sony Sensor

Maybe Nikon likes to keep their sensor options open, and doesn't want to hard-wire the Expeed chip for one specific sensor interface — e.g. LVDS or SLVS — because different suppliers' image sensors might require different communication protocols, embedded / separate clocks, different lane widths...

Of course, keeping their interfacing options flexible would mean that protocol conversion chips could sometimes be required to mediate the traffic between a particular image sensor and the Expeed chip.

As for Canon, is their situation identical to Nikon's ? Do Canon's ILCs use sensors that are not designed and manufactured by Canon themselves ? Is there a plausible reason why Canon's own image sensors and DIGIC chips would implement different, incompatible digital communication protocols, and therefore require a digital protocol conversion chip sourced e.g. from a third-party like Analog Devices, Inc. ?

 

[Antisthenes]

part of what he was denying was that Canon would use a 'front-end processing circuit' when they have, apparently full control of the DIGIC.

Contrary to what he says Canon does use front-end, it is right there on the diagram they supplied (the one in my previous post).

As English obviously isn't your first language,

So? Digital cameras are not your line, but you have no problem discussing digital cameras.

perhaps you don't realize that there's a significant ambiguity as to who's meant by the "he" in this sentence of yours: "Contrary to what he says Canon does use front-end".

Judging by your nervous response my English is good enough for you to understand exactly what I meant.

I'm not nervous. I was just intrigued, because I have never argued that Canon doesn't use a front end chip. To state it clearly for the reading comprehension-challenged, I've actually been arguing all along that Canon might still be using an analog front end chip from Analog Devices.

Or maybe you don't know what "Contrary" means in English.

Oh, I do.

Anyway, let me also point out that "front end", a term that happens to be used by Analog Devices, Inc. to describe their camera-oriented ADCs, would be rendered as "フロントエンド" in Japanese katakana

I know. Front-end means other things as well. I doubt that it is important how AD uses the term as the diagram is drawn by Canon. If Canon's response is not clear to you little can I do.

That reply from a Japanese Canon support person is but a jumble of abysmal syntax smothering nebulous concepts. If you think that meaningful information about the topic is conveyed by such a botched reply, then, you're clearly delusional as to the level of your understanding of the subject matter and of the English language.

AD also uses "front-end" in a variety of manners, you can come across "Integrated Mixed-Signal Front End", "FET Input Analog Front End With ADC Driver", which means that to AD front-end is also something more than just a "camera-oriented ADC". To say it is used by AD "to describe their camera-oriented ADCs", as you did, is incorrect. You either don't know, or you are trying to make a point whatever it takes.

It apparently doesn't even occur to you — probably because you don't even know what "mixed-signal" means — that all these instances of "front end" happen to refer to devices that process analog signals...

 

[Antisthenes]

So, you're signalling the end of a rational discussion, and on to the ad-hominem stuff in the absence of anything else to support your case. Time, I think, to call a halt.

As soon as you start to qualify as "evidence" something that isn't — e.g. postulating that an oversize border on the sensor chip is necessarily an ADC circuitry, or citing random pronouncements found on the Internet, or postulating unplausible roles like LVDS conversion for a particular chip, etc. — then, the discussion belongs more to the domain of fantasy than rationality.

Hi,

A closely related question, do you also think earth is flat?

There is a a lot of hearsay evidence to the contrary.

I have this suspicion that flat-earthers are people resembling you, with cognitive abilities so deficient that they cannot even tell the difference between primary sources and hearsay, between established facts and unsubstantiated, ignorant speculation...