Ad
Ad
Ad
Pages: « 1 2 3 [4] 5 »   Bottom of Page
Print
Author Topic: 16 bit dslr  (Read 7465 times)
hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


Re:
« Reply #60 on: January 28, 2014, 01:32:48 PM »
ReplyReply

Engineering or Physiology, discrete values are the essence of "digital".
As an engineer (I cannot really speak for physiology), I'd disagree. Discrete values are not (in my humble opinion) the _essence_ of "digital". As this discussion has shown, the physical signal can have discrete and continous aspects. The physical theory can lead us to this or that conclusion (depending on how many physics classes one attended). The essence of digital lies in the flexible interpretation/mangling of the signal that is transmitted or received. This is what enables packetized networks to "jump" across umpteen links and large distances with your jpegs presented in all their glory (while an "analog" transmission would probably have all kinds of visible degradations)

-h
Logged
allegretto
Sr. Member
****
Offline Offline

Posts: 371


« Reply #61 on: January 28, 2014, 11:32:31 PM »
ReplyReply

In Physiology we refer to "digital" as information that is composed of discrete values which are binary. It turns out that your nervous system is exactly that. A neuron's responses can be modulated by inhibitory or excitatory factors which alter the thresholds for their gates to open, but they still exhibit the given values (depending upon the tissue type) it just takes more or less stimulus to trigger due to the modulators.

If that seems analog to you then I have no argument, call it as you wish. Interestingly, modulation is such a complex factor in the CNS that it is a whole "Science" by itself. The issue from before was quite different and there is no need to re-hash.

xpat - cool here!
Logged
david distefano
Full Member
***
Offline Offline

Posts: 112


« Reply #62 on: January 30, 2014, 08:52:25 PM »
ReplyReply

as the op of this conversation i got lost about page 3. all i was really asking was the possibility of a 16 bit sensor on a dslr a better option than say the rumored nikon d4x at 54mp. i also read on this conversation that 16 bit really isn't 16. some of you have said that 2 to 3 bits is, if i read correctly, noise. do you also lose the same amount on a 14 bit sensor such as the d800/e?
Logged
Telecaster
Sr. Member
****
Offline Offline

Posts: 768



« Reply #63 on: January 30, 2014, 09:22:58 PM »
ReplyReply

My take, minus the thread digressions: more photosites have benefits, and so does greater usable dynamic range. I'm not convinced that with current sensors 16-bit A/D conversion offers any real-world benefit. But near-future tech is IMO likely to redraw the battlefield, so to speak. So hang tight.   Smiley

-Dave-
Logged
LKaven
Sr. Member
****
Offline Offline

Posts: 788


« Reply #64 on: January 30, 2014, 09:51:05 PM »
ReplyReply

as the op of this conversation i got lost about page 3. all i was really asking was the possibility of a 16 bit sensor on a dslr a better option than say the rumored nikon d4x at 54mp. i also read on this conversation that 16 bit really isn't 16. some of you have said that 2 to 3 bits is, if i read correctly, noise. do you also lose the same amount on a 14 bit sensor such as the d800/e?

The width of the A-D converter, 14 or 16 bit, has mostly to do with who is making the converter.  The Sony chips do conversion on the sensor, and 14 bits are exactly as many as are needed.  Sourcing A-D converters on the open market, you will likely find more 16-bit converters, since 16 bits is a convenient 2-byte quantity.  

MFDB makers have been sourcing these commodity 16-bit converters all along.  The words "16 bits" are then inserted in the specs and marketing literature as-if to suggest that the camera produces 16-bits of dynamic range per pixel.  This meme is hard to kill.

The D800/e makes good use of its 14 bits in per pixel response.  Most CCDs, due to read noise, yield only about 12 and a fraction bits per pixel.  It's when you downsample 80M of these pixels to about 12M to print, that you gain some bit precision in the process.  
« Last Edit: January 31, 2014, 02:34:14 PM by LKaven » Logged

hjulenissen
Sr. Member
****
Offline Offline

Posts: 1666


« Reply #65 on: January 31, 2014, 12:46:46 AM »
ReplyReply

all i was really asking was the possibility of a 16 bit sensor on a dslr a better option than say the rumored nikon d4x at 54mp.
That would be an apple to oranges comparision.
Quote
i also read on this conversation that 16 bit really isn't 16. some of you have said that 2 to 3 bits is, if i read correctly, noise. do you also lose the same amount on a 14 bit sensor such as the d800/e?
This is a bit like judging the top speed of a car by seeing how far the speedometer will go. You can be pretty sure that a car won't go any faster than the highest rating on its speedometer, but cannot be certain that it will be able to reach the highest rating.

-h
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3467


« Reply #66 on: January 31, 2014, 03:47:39 AM »
ReplyReply

as the op of this conversation i got lost about page 3. all i was really asking was the possibility of a 16 bit sensor on a dslr a better option than say the rumored nikon d4x at 54mp. i also read on this conversation that 16 bit really isn't 16. some of you have said that 2 to 3 bits is, if i read correctly, noise.

Hi David,

That is correct, and the noise is coming from several sources.

Suppose that a sensel could collect the charge of 60000 converted photons then, by the random arrival rate of photons, photon shot noise will be sqrt(60000) = 244.95 photons (0.408%) at the high exposure end. That's not much, but will become noticeable when we start reducing the level of exposure. Cutting the exposure in half (or doubling the ISO ) will leave a maximum of 30000 photons with a shot noise of sqrt(30000) = 173.21 photons (0.577%).

By the time we have reached an exposure level of 14 stops below maximum (= deep shadows) there will be only 3.66 photons of exposure recorded on average, with a photon shot noise of 1.91 (52.26%). At 15 stops below maximum the signal will become 1.83 photons with a noise of 1.35 (73.9%). That is only caused by the random nature of light particles as they build up the exposure over continuous time. Note: discrete particles, continuous time, hence light is not a digital but an analog signal. It becomes digital after quantization.

You can imagine that any additional electronic noise from circuits that process (e.g. read-out and dark current) that already noisy signal, and slight amplification differences between the per sensel transistors (PRNU) will immediately wreak havoc on the already marginal signal/noise ratio. It will be very difficult to exceed 14 bits of real signal.

Only by reducing that additional (read, PRNU, etc.) noise (e.g. by super cooling and very high component quality), or increasing the storage capacity (well depth) and exposure time, can we approach between 15 and 16-bit signal accuracy per sensel. With the shrinking sensel sizes, I think that 14 to 14.5-bit signal accuracy will be the practical limit, for which 16-bit components would be required (14-bit components will be too close to the best signal level with no room to spare).

Quote
do you also lose the same amount on a 14 bit sensor such as the d800/e?

The D800/D800E can approach 14-bits of actual signal with a well depth of some 45000 electrons, but it clips part of the Read noise before writing the Raw data. So the maximum signal is something like 15.46-bit, minus 1.4-2 bits of noise.

Do note that this is all before demosaicing, gamma, and tone curve adjustment, which may amplify or reduce the visibility of noise in the final image.

Cheers,
Bart
« Last Edit: January 31, 2014, 05:21:57 AM by BartvanderWolf » Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7257


WWW
« Reply #67 on: January 31, 2014, 01:53:17 PM »
ReplyReply

Hi,

The number of bits is just a question of the number of electron charges a pixel (called FWC)  can hold and the noise reading those pixels. If you divide the FWC with readout noise you get the dynamic range of the sensor as large number. You need a number of bits to represent that number, that would be log(N) / log(2).

So let's say FWC is 60000 electrons, readout noise is 15 electron charges, than the dynamic range would be 60000/15 = 4000. Log(4000) / log(2) = 11.97 so 12 bits would be adequate for that sensor.

Now, let's reduce readout noise 3 electron charges, so we would have 60000 / 3 = 20000. Log (20000) / log(2) = 14.3, so that sensor would need 15 bit's.

Increasing MP makes the sensors smaller half the area -> half the FWC. If you increase MP the DR per pixel will be less, so the mythical 54 MP camera would perhaps have 30000 in FWC and 3 electron charges of readout noise. 30000 / 3 = 10000, log (10000) / log (2) = 13.3 so it would do just fine with 14 bits.

Going the other direction, making pixels larger could increase DR per pixel. Fat pixel could perhaps hold 150000 electron charges, but if those sensors have high readout noise they still don't need that many bits.

So 14 bits are quite safe now.

Best regards
Erik


as the op of this conversation i got lost about page 3. all i was really asking was the possibility of a 16 bit sensor on a dslr a better option than say the rumored nikon d4x at 54mp. i also read on this conversation that 16 bit really isn't 16. some of you have said that 2 to 3 bits is, if i read correctly, noise. do you also lose the same amount on a 14 bit sensor such as the d800/e?
Logged

BartvanderWolf
Sr. Member
****
Online Online

Posts: 3467


« Reply #68 on: January 31, 2014, 03:28:20 PM »
ReplyReply

So 14 bits are quite safe now.

Hi Erik,

I'd say that with almost 14-bit DR for the D800 we're just safe per sensel, with virtually no real room to spare.

However, the trend is towards smaller sensels, and that's when we'll have to consider DR per unit area, with multiple sensels per such unit area. We'll ultimately get to something like 1 micron sensel pitch ('well depth' of maybe 1000-1500 electrons each) without the need for an OLPF because diffraction and residual lens aberrations will limit the contrast beyond the Nyquist frequency enough to avoid aliasing.

Cheers,
Bart
Logged
LKaven
Sr. Member
****
Offline Offline

Posts: 788


« Reply #69 on: January 31, 2014, 04:02:16 PM »
ReplyReply

Hi Erik,

I'd say that with almost 14-bit DR for the D800 we're just safe per sensel, with virtually no real room to spare.

However, the trend is towards smaller sensels, and that's when we'll have to consider DR per unit area, with multiple sensels per such unit area. We'll ultimately get to something like 1 micron sensel pitch ('well depth' of maybe 1000-1500 electrons each) without the need for an OLPF because diffraction and residual lens aberrations will limit the contrast beyond the Nyquist frequency enough to avoid aliasing.

Meanwhile, worries about Photoshop's 15-bit limit are encroaching.  I'm sure way back when they never thought we'd have real 16-bit image data, and that they could steal a bit for some reason.  But it looks like the day of reckoning may have come.
Logged

bjanes
Sr. Member
****
Offline Offline

Posts: 2756



« Reply #70 on: January 31, 2014, 05:57:44 PM »
ReplyReply

Hi Erik,

I'd say that with almost 14-bit DR for the D800 we're just safe per sensel, with virtually no real room to spare.

However, the trend is towards smaller sensels, and that's when we'll have to consider DR per unit area, with multiple sensels per such unit area. We'll ultimately get to something like 1 micron sensel pitch ('well depth' of maybe 1000-1500 electrons each) without the need for an OLPF because diffraction and residual lens aberrations will limit the contrast beyond the Nyquist frequency enough to avoid aliasing.

I'm glad to have 14 bit files with the D800e, which can encode the full DR of the sensor, which is about 13 EV. However, the useful photographic DR is less than the engineering ER of 13.3 EV, depending on what noise floor is used for practical photographic DR. One can derive DR at other noise floors from the DXO data as Emil explains here.

Using this method, I derived the DR for noise floors of 0, 6, 12, and 18 db.



The calculations are tabulated here. The interpolation method is that in the referenced Wikipedia article on log interpolation.



The dynamic range derived from Imatest are shown here.



So 14 bits gives a bit of safety for the D800e and likely would be sufficient for the new PhaseOne IQ250. The 16 bit files for the older PhaseOne sensors are merely marketing hype, or more charitably, related to the use of off the shelf 16 bit ADCs used for those CCD cameras that have only about 12 stops of DR.

Bill
Logged
LKaven
Sr. Member
****
Offline Offline

Posts: 788


« Reply #71 on: January 31, 2014, 07:31:44 PM »
ReplyReply

The Emil Martinec files, which should be digested at this point, contain some of the best writing anywhere on the internet on the subject.  Haven't seen Emil around for quite a while.  I'm sure he's busy working out cutting edge string theory instead.
Logged

Fine_Art
Sr. Member
****
Offline Offline

Posts: 1063


« Reply #72 on: January 31, 2014, 07:36:52 PM »
ReplyReply

Good to know Bill, there are probably a lot of people that got sucked in by believing they were missing something from 14 to 16 bits. Probably the only real 16bit (of real data) CCD cameras available to consumers are the tiny cooled chips for astro-photo. $1000 for a 640x480 peltier cooled chip, maybe greyscale. Kept 30oC below ambient temp, they may have the noise floor low enough.
Logged
david distefano
Full Member
***
Offline Offline

Posts: 112


« Reply #73 on: January 31, 2014, 10:34:30 PM »
ReplyReply

[Good to know Bill, there are probably a lot of people that got sucked in by believing they were missing something from 14 to 16 bits.]

this has been interesting. i for one believed i was losing out on shades of color since i sold my mfdb and have been using the d800. ok i have this question, (since i am still contemplating a mfdb to use with my arca swiss) if nikon or canon or sony come out with their 54mp cameras and paired with the zeiss otus lenses and printing to a maximum size of 24x30 what am i really going to gain with a mfdb. (excluding the 60 and 80mp backs, too much money.) a p45 used is going for about $7,000 which is about my limit unless lotto hits. i shoot images for the pure enjoyment not as a business that would be able to write off the equipment.
Logged
Vladimirovich
Sr. Member
****
Offline Offline

Posts: 1279


« Reply #74 on: January 31, 2014, 10:40:31 PM »
ReplyReply

I'm sure he's busy working out cutting edge string theory instead.
that thing rumored to have 26bits !!!
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7257


WWW
« Reply #75 on: February 01, 2014, 12:06:29 AM »
ReplyReply

Hi Bart,

I don't think so. My guess is that we are going to pixel sizes down to 3 microns on "real cameras". It is possible to make smaller pixels, but I would guess that lenses set practical limits. Small pixels are possible but the approach of a larger sensor with reasonable size pixels seem more rational to me.

Best regards
Erik




Hi Erik,

I'd say that with almost 14-bit DR for the D800 we're just safe per sensel, with virtually no real room to spare.

However, the trend is towards smaller sensels, and that's when we'll have to consider DR per unit area, with multiple sensels per such unit area. We'll ultimately get to something like 1 micron sensel pitch ('well depth' of maybe 1000-1500 electrons each) without the need for an OLPF because diffraction and residual lens aberrations will limit the contrast beyond the Nyquist frequency enough to avoid aliasing.

Cheers,
Bart
Logged

LKaven
Sr. Member
****
Offline Offline

Posts: 788


« Reply #76 on: February 01, 2014, 01:44:54 AM »
ReplyReply

I don't think so. My guess is that we are going to pixel sizes down to 3 microns on "real cameras". It is possible to make smaller pixels, but I would guess that lenses set practical limits. Small pixels are possible but the approach of a larger sensor with reasonable size pixels seem more rational to me.

Fossum sees APS-c sensors topping out at about 100MP or so.  I haven't computed the pixel size for that, but he's on record as saying that pixels will continue down to the 900nm range.  At under 2um the APS-c sensors will benefit from BSI. 

Of course then there's Fossum's idea for a JOT sensor.  One photon per pixel, sampled at 250-1000 times per second across the entire sensor surface.  Bitslice image planes will get integrated through a sigma (summation) unit, possibly after getting shifted to account for subject movement.
Logged

BartvanderWolf
Sr. Member
****
Online Online

Posts: 3467


« Reply #77 on: February 01, 2014, 04:05:57 AM »
ReplyReply

Fossum sees APS-c sensors topping out at about 100MP or so.  I haven't computed the pixel size for that, but he's on record as saying that pixels will continue down to the 900nm range.  At under 2um the APS-c sensors will benefit from BSI.

Hi Luke,

Indeed. For those not familiar with Dr. Eric Fossum, he's the inventor of the CMOS image sensor.

Quote
Of course then there's Fossum's idea for a JOT sensor.  One photon per pixel, sampled at 250-1000 times per second across the entire sensor surface.  Bitslice image planes will get integrated through a sigma (summation) unit, possibly after getting shifted to account for subject movement.

Yes, there are several innovative ideas being tossed around. Here is a nice collection of PDF papers on the various issues and solutions. Some show examples of existing 1.1 micron pitch sensors with a well capacity of some 2700 electrons.

Cheers,
Bart
« Last Edit: February 01, 2014, 04:08:03 AM by BartvanderWolf » Logged
thierrylegros396
Sr. Member
****
Offline Offline

Posts: 636


« Reply #78 on: February 01, 2014, 04:41:28 AM »
ReplyReply

First audio Compact Disc players (Philips 1984 if I remember) used 14 bits DA converters.
It was just the minimum required to ensure the noise will be masked in ambient noise (about 84dB SNR).

Then 16 bits converters appears (about 96dB SNR).
Then 18 bits (about 108dB), and after 20 (120dB) and so on till today.
Same for AD converters for recording the source.

What to say about that?!

It seems that 16dB is more that enough in a practical point of view, altough golden ears prefer slightly more.

I don't know if it's the same for optical sources and medias (Screen, but certainly not paper), but it's possible.

Thierry
Logged
ErikKaffehr
Sr. Member
****
Offline Offline

Posts: 7257


WWW
« Reply #79 on: February 01, 2014, 04:54:56 AM »
ReplyReply

Hi,

Audio CD-s recorded at 16 bits. But photography is based on light and light has a noise of it's own. So 16 bit is now 12-14 bit of signal and 2-4 bits of noise.

Just to demonstrate it a bit, check Phase One web site.

IQ 250 has 14 EV dynamic range and 14 bit processing. The IQ 260 and 280 have "16-bit opticolor" and 13EV DR (that is 13 bits) and the IQ-250 has 14 EV DR (14 bits).

So the older backs have 16 bit (13 bit signal + 3 bit noise) while IQ 250 has 14 bit (14 bit signal + 0 bit noise). Just to make life easy, you can take the 14 bit from the IQ 250 and convert it to 16 bit.

Here is the 'C' code:

sixten_bit_signal = 14_bit_signal << 2 + rand() % 4;

Best regards
Erik




First audio Compact Disc players (Philips 1984 if I remember) used 14 bits DA converters.
It was just the minimum required to ensure the noise will be masked in ambient noise (about 84dB SNR).

Then 16 bits converters appears (about 96dB SNR).
Then 18 bits (about 108dB), and after 20 (120dB) and so on till today.
Same for AD converters for recording the source.

What to say about that?!

It seems that 16dB is more that enough in a practical point of view, altough golden ears prefer slightly more.

I don't know if it's the same for optical sources and medias (Screen, but certainly not paper), but it's possible.

Thierry
Logged

Pages: « 1 2 3 [4] 5 »   Top of Page
Print
Jump to:  

Ad
Ad
Ad