The EOS 1D X Sensor Demystified...

[Canon Rumors]

<div name="googleone_share_1" style="position:relative;z-index:5;float: right; margin: 70px 0 0 0;"><g:plusone size="tall" count="1" href="http://www.canonrumors.com/2011/10/eos-1d-x-sensor-broken-down/"></g:plusone></div><div id="fb_share_1" style="float: right; margin: 0 -50px 0 10px;"><a name="fb_share" type="box_count" share_url="http://www.canonrumors.com/2011/10/eos-1d-x-sensor-broken-down/" href="http://www.facebook.com/sharer.php">Share</a></div><div><script src="http://static.ak.fbcdn.net/connect.php/js/FB.Share" type="text/javascript"></script></div><div class="tweetmeme_button" style="float: right; margin-left: 10px;"><a class="tm_button" rel="&style=normal&b=2" href="http://www.canonrumors.com/2011/10/eos-1d-x-sensor-broken-down/"></a></div>
<p><strong>What’s in a sensor?

</strong>[<a href="http://image-sensors-world.blogspot.com/2011/10/canon-full-frame-35mm-sensor-news.html">ISW</a>] has given a nice and concise breakdown of the technology inside the new full frame sensor for the EOS 1D X. Below is a direct quote of the breakdown.</p>
<ul>
<li>New photodiode construction has resulted in an improved photoelectric conversion rate that gives increased light sensitivity.</li>
<li>Improved transistors inside the pixels are said to make SNR higher</li>
<li>The first time that gapless microlenses have been employed on a Canon full-frame sensor.</li>
<li>14fps speed is achieved by a 16-channel analog output with two-vertical-pixel simultaneous readout. The 16 outputs are muxed in 4 ADCs siting on a separate image processor chip Digic 5+. It is around 1.4 times faster than the previous generation EOS-1D Mark IV and said to be a first for a 35mm full-frame digital sensor. At ISO 32,000 or higher the frame rate is reduced to 10fps.</li>
</ul>
<p><strong>Source <a href="http://image-sensors-world.blogspot.com/2011/10/canon-full-frame-35mm-sensor-news.html">Image Sensor World</a></strong></p>
<p><strong></strong><strong><span style="color: #ff0000;">c</span>r</strong>

<strong> </strong></p>

 

[torger]

Re: EOS 1D X Sensor Broken Down

My first question -- will there be line (pattern) noise?

My second question -- why is frame rate reduced at the ultra high ISOs?

 

[Woody]

According to the original CPN article here:
http://cpn.canon-europe.com/content/education/technical/eos_1d_x_explained.do
"newly designed CMOS circuits have also been designed to capture a wider dynamic range, meaning more detail is retained in both shadows and highlights in comparison to previous cameras."

Sounds like Canon is finally listening to what many folks say about the circuits on the sensors: these circuits limit the dynamic range that can be achieved on Canon sensors.

Now, whether it translates into real gains for dynamic range remains to be seen. I hope it all works out.

 

[Mt Spokane Photography]

We can only guess at why frame rate at high ISO's is reduced, but there is on-chip NR going on, even with RAW, so it might just be taking more time to read out due to more complex NR.

 

[Loswr]

Re: EOS 1D X Sensor Broken Down

My second question -- why is frame rate reduced at the ultra high ISOs?

Most likely it's simply a data transfer issue. I'd assume that 18 MP at 12 fps is pretty near the max throughput. On a current 18 MP sensor (7D, in this case), an example RAW file (same scene, data from TDP) has the following file sizes:

• ISO 100 - 24.7 MB
• ISO 3200 - 29.3 MB
• ISO 12800 - 34.1 MB

Since data quantity goes up with increasing ISO (same trend on all Canon bodies), at some point the frame rate needs to slow down to compensate.

 

[outsider]

The RAW format is a lossless compressed format.
The file size gets higher at higher ISOs because there is less to compress (due to higher nosise).

The data coming off the image sensor is likely coming off at the same rate regardless of ISO setting
(doubt the RAW compression happens on the imaging chip), so I would not think that's the bottle neck.

So maybe this is a limitation on the write rate to CF cards?

 

[Sampo]

The RAW format is a lossless compressed format.
The file size gets higher at higher ISOs because there is less to compress (due to higher nosise).

No, it's because CR2 format doesn't use adaptive Huffman, but predefined tables that are optimized for common case value distribution.

If adaptive Huffman or arithmetic coding (like h.264 CABAC) was used, high ISO raw images would be significantly smaller. Unfortunately, processing cost in terms of power consumption and silicon area needed would likely be higher, especially with more complex coding schemes.

Maybe future RAW formats will have PNG or lossless JPEG style spatial predictor functions. Combine that with adaptive arithmetic coding, and file size savings would likely be significant, even halved.

But may that doesn't really make sense - file size is not really a big issue anymore. Current scheme is very reliable - flip one bit in current CR2 format, and you can recover rest of the image with just one pixel error given software that can resync to Huffman stream. More complex coding could mean larger blocks of the image become corrupted without sophisticated error recovery and correction. Otherwise one bit flip could render whole image unusable.

Besides, if file size was really an issue, Canon would probably stop embedding a thumbnail AND a full size JPEG image in every RAW file! I prefer reliability over file size any day or night.

---

Correction: Having taken a look at reverse engineered CR2 implementation [1], I have to say I was wrong. CR2 RAW compression is actually based on modified Lossless JPEG [2]. Main differences are in data ordering, and of course that CR2 contains Bayer-filter values, not actual color components in any color space.

Sorry if I misled anyone! I think I mixed CR2 with some other camera manufacturer old RAW file format that used simple Huffman coding.

[1]: David J. Coffin, http://cybercom.net/%7Edcoffin/dcraw/
[2]: ITU CCITT recommendation T.81 - http://www.itu.int/rec/T-REC-T.81/e

 

[KitH]

Re: EOS 1D X Sensor Broken Down

Most likely it's simply a data transfer issue. I'd assume that 18 MP at 12 fps is pretty near the max throughput. On a current 18 MP sensor (7D, in this case), an example RAW file (same scene, data from TDP) has the following file sizes:

• ISO 100 - 24.7 MB
• ISO 3200 - 29.3 MB
• ISO 12800 - 34.1 MB

Since data quantity goes up with increasing ISO (same trend on all Canon bodies), at some point the frame rate needs to slow down to compensate.

Is that the data quantity goes up as a function of increasing noise (or decreasing signal to noise ratio)?

that is, are there smaller areas of pixels that can all be treated together as if they're the same, before some noise gets in the way and asks the encoding to record a new different value in the data. Let's assume the same happens back again afterwards (or else it's not noise, it's part of the picture) so there's two new data points from the passage of a particular "packet" of noise - and this effect aggregates to pump up the file size?

I just took a look at those numbers, there's a bit of a trend going on. Need more data points, but the shape I'd expect is a runaway file size explosion with increasing noise (until the picture is random hiss, where it levels out), hence the data throughput constraint is the limiting factor on ISO. More throughput capacity on the chipsets gives more ISO up to a point where the first thing to break is the frames per second speed.

 

[traveller]

The RAW format is a lossless compressed format.
The file size gets higher at higher ISOs because there is less to compress (due to higher nosise).

No, it's because CR2 format doesn't use adaptive Huffman, but predefined tables that are optimized for common case value distribution.

If adaptive Huffman or arithmetic coding (like h.264 CABAC) was used, high ISO raw images would be significantly smaller. Unfortunately, processing cost in terms of power consumption and silicon area needed would likely be higher, especially with more complex coding schemes.

Maybe future RAW formats will have PNG or lossless JPEG style spatial predictor functions. Combine that with adaptive arithmetic coding, and file size savings would likely be significant, even halved.

But may that doesn't really make sense - file size is not really a big issue anymore. Current scheme is very reliable - flip one bit in current CR2 format, and you can recover rest of the image with just one pixel error given software that can resync to Huffman stream. More complex coding could mean larger blocks of the image become corrupted without sophisticated error recovery and correction. Otherwise one bit flip could render whole image unusable.

Besides, if file size was really an issue, Canon would probably stop embedding a thumbnail AND a full size JPEG image in every RAW file! I prefer reliability over file size any day or night.

So much for "Demystifying the EOS 1DX Sensor", all this techno-babble has me even more mystified! Hopefully when someone gets hold of the actual camera we'll see some real life tests that show us what it can offer to photographers in practice...

 

[pierlux]

The RAW format is a lossless compressed format.
The file size gets higher at higher ISOs because there is less to compress (due to higher nosise).

No, it's because CR2 format doesn't use adaptive Huffman, but predefined tables that are optimized for common case value distribution.

If adaptive Huffman or arithmetic coding (like h.264 CABAC) was used, high ISO raw images would be significantly smaller. Unfortunately, processing cost in terms of power consumption and silicon area needed would likely be higher, especially with more complex coding schemes.

Maybe future RAW formats will have PNG or lossless JPEG style spatial predictor functions. Combine that with adaptive arithmetic coding, and file size savings would likely be significant, even halved.

But may that doesn't really make sense - file size is not really a big issue anymore. Current scheme is very reliable - flip one bit in current CR2 format, and you can recover rest of the image with just one pixel error given software that can resync to Huffman stream. More complex coding could mean larger blocks of the image become corrupted without sophisticated error recovery and correction. Otherwise one bit flip could render whole image unusable.

Besides, if file size was really an issue, Canon would probably stop embedding a thumbnail AND a full size JPEG image in every RAW file! I prefer reliability over file size any day or night.

Amen!
...and... Wow! Sampo, first post here and you're mythical to me already! Post more often, please!

 

[Sampo]

...

Amen!
...and... Wow! Sampo, first post here and you're mythical to me already! Post more often, please!

Thanks, but I was actually wrong. Refer to my updated post for details. http://www.canonrumors.com/forum/index.php/topic,1949.0.html#msg37508

 

[cpsico]

According to dxo labs the 1dsIII has about 12 stops of dynamic range, the Nikon d3x has an amazing 13.7 both at iso 100. If its near 14 stops like the Nikon flagship with low noise that would be amazing for wedding and landscape photographers.

 

[Joseph]

Re: EOS 1D X Sensor Broken Down

Most likely it's simply a data transfer issue. I'd assume that 18 MP at 12 fps is pretty near the max throughput. On a current 18 MP sensor (7D, in this case), an example RAW file (same scene, data from TDP) has the following file sizes:

• ISO 100 - 24.7 MB
• ISO 3200 - 29.3 MB
• ISO 12800 - 34.1 MB

Since data quantity goes up with increasing ISO (same trend on all Canon bodies), at some point the frame rate needs to slow down to compensate.

You should think " file size " has no affect though , since the images pass to the buffer first - the buffer would slow you down from shooting - before a shutter would.

I have mentioned this in the past , but what the hell is stoping you ( CANON ) from putting in high speed , low latency buffers - several manufacturers have memory running beyond 6ghz " 25GB/sec + " , they are very small in size , and have low power consumption. Maybe cost effectiveness , maybe the Digic processors have an unsupported bit rate , or a bit rate too low to transfer any faster ?

Can anyone find any specs on the Digic 4 , or 5's bus ( 16bit/32bit/64bit ect.) ?
Do Canon Sensors have a bit/bus rate ?
Can anyone find out the memory type Canon uses for beffering (SG , DDRx , SD / ECC or Non ECC ect.)?
Can anyone find info on the buffer amounts in Canon DSLR's ?

Sorry if there is a forum somewhere else for this - this article just made me think about it again.

 

[distant.star]

Yes, I too found the "demystifying" mystifying.

More like baffling with BS.





So much for "Demystifying the EOS 1DX Sensor", all this techno-babble has me even more mystified! Hopefully when someone gets hold of the actual camera we'll see some real life tests that show us what it can offer to photographers in practice...
[/quote]

 

[Woody]

According to dxo labs the 1dsIII has about 12 stops of dynamic range, the Nikon d3x has an amazing 13.7 both at iso 100. If its near 14 stops like the Nikon flagship with low noise that would be amazing for wedding and landscape photographers.

I believe even Nikon's 16 MP APS-C cameras like the D7000 achieve 13.9 eV of dynamic range (so much for all the myths about larger pixel size giving wider dynamic range), beating out their very own venerated 12 MP FF D3s which has 12 eV of dynamic range. Canon's Archille's heel really lies in their sensor electronics. We'll wait for real world tests to see if they have overcome this.

 

[Meh]

Larger pixels giving more Dynamic Range is not a myth just an oversimplified statement because there is more to it than just pixel size. It is true when those other things are equal. DR is the ratio of the largest signal to the lowest signal that can be measured . That ratio can be expressed as a simple ratio, stops/EV, or dB.

The maximum signal is the most electrons that can be stored in the photosite (one electron is release for each photon absorbed). This is also referred to as saturation value or full-well capacity. Note that it's the size of the photosite that matters... the area of each pixel that is sensitive to light which in the past I believe has been as low as 30% but in recent years was up to about 50%. This is called the fill-factor

The lowest signal is taken as the noise floor.. which is the read noise at whatever ISO your using. Most CMOS sensors up until the most recent generation actually had higher read noise at ISO 100 but the Sony sensor used in the D7000 greatly improved on this and the read noise is fairly constant with ISO settings.

The Sony also has much higher full-well capacity even though the pixels are smaller. This is most likely because they were able to increase the fill-factor... i.e. a higher percentage of the pixel area is used for the photosite. In other words, even though the pixels are smaller the photosites are larger (higher max signal). Combined with the lower read-noise at ISO 100 (lower noise floor) the ratio of largest to smaller signal is much larger giving the very high 13.9 stops of DR.

Hopefully, we'll see these advances in the 1DX sensor as well but there are many other tradeoffs in designing a sensor to optimize for speed, heat, minimizing blooming (current leakage), etc. that the Sony sensor may have sacrificed... I don't know that's the case just stating the possibility.

 

[torger]

According to dxo labs the 1dsIII has about 12 stops of dynamic range, the Nikon d3x has an amazing 13.7 both at iso 100. If its near 14 stops like the Nikon flagship with low noise that would be amazing for wedding and landscape photographers.

I believe even Nikon's 16 MP APS-C cameras like the D7000 achieve 13.9 eV of dynamic range (so much for all the myths about larger pixel size giving wider dynamic range), beating out their very own venerated 12 MP FF D3s which has 12 eV of dynamic range. Canon's Archille's heel really lies in their sensor electronics. We'll wait for real world tests to see if they have overcome this.

Dxomark's dynamic range measurement is actually quite useless and confuses many. What they measure is the mathematical/engineering type of dynamic range, how small a signal can be until it is equal size as the noise. That is not the same as saying that shadows 13.9 stops down are usable to the photographers, when the signal is equally large as the noise there is not a usable picture! The measurement does not say anything about how much dynamic range that is useful in a picture, well only that it is considerably less than 13.9 stops. Typical usable range in a good digital camera is around 7 stops if I remember correctly -- the reason why it so much smaller than the measured dynamic range is that you must have much more signal than noise in your shadows for a picture to be pleasing.

It is better to look at their SNR 18% measurement, which shows how much noise there is in parts that are not insanely dark, and there you can see for example that the 5D mk2 has less noise than D7000, despite that its "dynamic range" is only 11.9.

But then there's another factor -- pattern noise -- which Canon happens to have problem with, meaning that you can see a pattern in the noise. This is not measured by DxOmark, but does reduce the useful dynamic range, because most people really don't want to push the shadows such that you start seeing noise pattern. Noise that is totally random is much more pleasing to the eye.

And there's yet another factor -- auto exposure -- if you let the camera expose the picture itself it may play it safe and leave space at the top to avoid clip highlights, or it may push the histogram as far to the right as it can. If I remember correctly the D7000 autoexposure exposes much farther to the right than for example 7D. This means that D7000 will seem to have more dynamic range than it actually has. When you expose manually you can max out the sensor as you like.

In all, useful dynamic range is subjective and need to be tested, it is not easily measured.

 

[BillHardie]

I am surprised and more than a little disappointed that the new D1X sensor does not seem to provide an extended dynamic range. Surely this is something that would be a winner for Canon!