GFX100 Sensor details Published - 16-bit digital output

Started Oct 31, 2018 | Discussions
Chris Dodkin
MOD Chris Dodkin Forum Pro • Posts: 12,674
GFX100 Sensor details Published - 16-bit digital output

Imaging website FRAMOS has published a more detailed overview of the new Sony sensor destined for the Fujifilm GFX100

Product information "IMX461AQR-C"

The IMX461AQR-C is a diagonal 55 mm (Type 3.4) CMOS active pixel type image sensor with a square pixel array and 102 M effective pixels. This IC incorporates maximum 36 dB PGA circuit and 16-bit A/D converter. 16-bit digital output makes it possible to readout the signals of 102 M effective pixels at high-speed of 2.7 frame/s in still picture mode. In addition, vertical subsampling binning and 3-horizontal pixel weighted binning realize high-speed 12-bit digital output for shooting moving pictures.

https://www.framos.com/us/imx461aqr-c-21053

IMX461AQR-C

Manufacturer: Sony

Type: Area SensorsSensor

type: CMOS

Shutter type: Rolling shutter

Pins / Balls: 544

Pixel size x [µm]:3.76

Pixel size y [µm]:3.76

Bit Depth: 12

Protective Film: yes

-- hide signature --

Your time is limited, so don't waste it arguing about camera features - go out and capture memories

 Chris Dodkin's gear list:Chris Dodkin's gear list
Fujifilm X100F Fujifilm X100V Fujifilm X-T2 Fujifilm GFX 50S Fujifilm GFX 50R +43 more
Rick Knepper
Rick Knepper Forum Pro • Posts: 17,870
Re: GFX100 Sensor details Published - 16-bit digital output

Chris Dodkin wrote:

Imaging website FRAMOS has published a more detailed overview of the new Sony sensor destined for the Fujifilm GFX100

Product information "IMX461AQR-C"

The IMX461AQR-C is a diagonal 55 mm (Type 3.4) CMOS active pixel type image sensor with a square pixel array and 102 M effective pixels. This IC incorporates maximum 36 dB PGA circuit and 16-bit A/D converter. 16-bit digital output makes it possible to readout the signals of 102 M effective pixels at high-speed of 2.7 frame/s in still picture mode. In addition, vertical subsampling binning and 3-horizontal pixel weighted binning realize high-speed 12-bit digital output for shooting moving pictures.

https://www.framos.com/us/imx461aqr-c-21053

IMX461AQR-C

Manufacturer: Sony

Type: Area SensorsSensor

type: CMOS

Shutter type: Rolling shutter

Pins / Balls: 544

Pixel size x [µm]:3.76

Pixel size y [µm]:3.76

Bit Depth: 12

Protective Film: yes

Chris, the top paragraph states the ADC is 16 bit but the specs at the bottom state Bit Depth is 12. Can you help me understand?

-- hide signature --

Once you've done fifty, anything less is iffy.

 Rick Knepper's gear list:Rick Knepper's gear list
Pentax 645Z Canon EOS 5DS R Fujifilm GFX 50S Sony a7R IV Pentax smc D FA 645 55mm F2.8 AL (IF) SDM AW +11 more
Chris Dodkin
OP MOD Chris Dodkin Forum Pro • Posts: 12,674
Re: GFX100 Sensor details Published - 16-bit digital output

12 bit is for video mode apparently - allows faster frame rate

-- hide signature --

Your time is limited, so don't waste it arguing about camera features - go out and capture memories

 Chris Dodkin's gear list:Chris Dodkin's gear list
Fujifilm X100F Fujifilm X100V Fujifilm X-T2 Fujifilm GFX 50S Fujifilm GFX 50R +43 more
NAwlins Contrarian Veteran Member • Posts: 4,814
Bit depth - current supposition and some history
1

Imaging website FRAMOS has published a more detailed overview of the new Sony sensor ...

Product information "IMX461AQR-C"

The IMX461AQR-C ... incorporates ... 16-bit A/D converter. 16-bit digital output makes it possible to readout the signals of 102 M effective pixels at high-speed of 2.7 frame/s in still picture mode. In addition, vertical subsampling binning and 3-horizontal pixel weighted binning realize high-speed 12-bit digital output for shooting moving pictures.

* * *

Bit Depth: 12

Chris, the top paragraph states the ADC is 16 bit but the specs at the bottom state Bit Depth is 12. Can you help me understand?

I agree that the quoted text does appear to imply, or at least is most consistent with, 'normal' output being 16-bit depth and "high speed" output being 12-bit depth. Related to that, some of the stuff about binning seems geared to getting manageable video out of the ultra-high-resolution sensor.

However, just to point out, I don't think we should assume anything, because in the not-ancient history of analog-to-digital and digital-to-analog conversion, some curiosities have appeared, maybe for good reasons and maybe as marketing nonsense. The compact disc digital audio medium appeared in 1982, with 16-bit audio encoding. By about 1988, some of the player manufacturers started touting that their players had 18-bit or even 20-bit digital-to-analog converters. The stated rationales included that converter accuracy fell with the least significant bits, and therefore that a higher-bit converter not using its 2 or 4 least-significant bits was more accurate; and/or that some sort of dither or somethings was introduced as the least-significant bits for somehow-improved sound. One can imagine that the same basic benefit from the reverse process might be true, or at least claimed, for analog-to-digital conversion.

Also, many older medium format digital backs had 16-bit output--but you can argue about how 'real' the two or three (or four) least-significant bits were. If the analog-to-digital converter gets a sufficiently noisy signal from the imaging sensor, past a certain point it doesn't matter how many bits the converter is capable of outputting, or how many it does output; the least-significant bits are essentially all noise.

My personal guess is that--insofar as AFAIK Sony integrates analog-to-digital conversion with its current photo sensors--Sony would not build in, much less tout, this capability unless the two extra bits contain significant 'real' image data. But if you want to be more cynical about that, I can't say the history of such things proves you wrong.

 NAwlins Contrarian's gear list:NAwlins Contrarian's gear list
Nikon Coolpix S30 Canon PowerShot S120 Sony Alpha DSLR-A580 Tamron SP AF 90mm F/2.8 Di Macro Tamron SP 70-300mm F/4-5.6 Di USD +5 more
Chris Dodkin
OP MOD Chris Dodkin Forum Pro • Posts: 12,674
Re: Bit depth - current supposition and some history

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

-- hide signature --

Your time is limited, so don't waste it arguing about camera features - go out and capture memories

 Chris Dodkin's gear list:Chris Dodkin's gear list
Fujifilm X100F Fujifilm X100V Fujifilm X-T2 Fujifilm GFX 50S Fujifilm GFX 50R +43 more
Erik Kaffehr
Erik Kaffehr Veteran Member • Posts: 4,205
Re: Bit depth - current supposition and some history
2

Chris Dodkin wrote:

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

Hi,

Sony specs say 14 bit readout.

But, Phase One already has the IQ4150 on preorder. The IQ4150 is specified to have 16 bits, albeit at very slow readout.

The ADCs are ramp type, AFAIK, and that means the precision can be changed by making the stepping on the ramp slower. So, it may be technically possible to extract 16 bits but the camera may slow down to 1 FPS or so.

The GFX 100 sensor is going to have the same pixel dimensions as the 150MP sensor.

If more than 14 bit readout is needed depends on the "engineering DR" of the pixel, which is full well capacity / readout noise. You can calculate it in EV like DR = log(FWC/readout noise) / log(2). Both FWC and readout noise are measured in e-.

If DR is below 14 EV, at the pixel level, no more than 14 bits are needed to fully represent data.

DxO-mark shows engineering DR at the pixel level if you switch to 'screen' mode. They did not measure the GFX 50S but:

  • Pentax 645Z has a DR of 13.39 EV
  • Hasselblad X1D has a DR of 13.39 EV

Both cameras use the same sensor design as the GFX50S

The GFX 100S has half the pixel area compared to the GFX 50S. That would probably reduce DR a bit, sensor development would reduce redaout noise on the other hand. So I would guess that 14 bits would be enough to represent the output of the sensor.

Fuji could add a readout mode for 16-bits, but the only benefit would be to slow down the camera by a factor of four.

Sixteen bits are important to Phase One, for marketing reasons.

Best regards

Erik

-- hide signature --

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

KlasG New Member • Posts: 21
Re: Bit depth - current supposition and some history

Erik Kaffehr wrote:

Chris Dodkin wrote:

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

Hi,

Sony specs say 14 bit readout.

But, Phase One already has the IQ4150 on preorder. The IQ4150 is specified to have 16 bits, albeit at very slow readout.

The ADCs are ramp type, AFAIK, and that means the precision can be changed by making the stepping on the ramp slower. So, it may be technically possible to extract 16 bits but the camera may slow down to 1 FPS or so.

The GFX 100 sensor is going to have the same pixel dimensions as the 150MP sensor.

If more than 14 bit readout is needed depends on the "engineering DR" of the pixel, which is full well capacity / readout noise. You can calculate it in EV like DR = log(FWC/readout noise) / log(2). Both FWC and readout noise are measured in e-.

If DR is below 14 EV, at the pixel level, no more than 14 bits are needed to fully represent data.

DxO-mark shows engineering DR at the pixel level if you switch to 'screen' mode. They did not measure the GFX 50S but:

  • Pentax 645Z has a DR of 13.39 EV
  • Hasselblad X1D has a DR of 13.39 EV

Both cameras use the same sensor design as the GFX50S

The GFX 100S has half the pixel area compared to the GFX 50S. That would probably reduce DR a bit, sensor development would reduce redaout noise on the other hand. So I would guess that 14 bits would be enough to represent the output of the sensor.

Fuji could add a readout mode for 16-bits, but the only benefit would be to slow down the camera by a factor of four.

Sixteen bits are important to Phase One, for marketing reasons.

Best regards

Erik

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

Chris Dodkin
OP MOD Chris Dodkin Forum Pro • Posts: 12,674
New 16 Bit Sensors

The additional data from FRAMOS shows both the new X-T3 and the proposed GFX100 sensors as having 16-bit A/D and 16-bit digital output - we were not discussing the GFX50 sensor.

X-T3 Sensor Specs

The IMX571BQR-C is a diagonal 28.3 mm (Type 1.8) CMOS active pixel type image sensor with a square pixel array and 26.11 M effective pixels. This sensor incorporates maximum 36 dB PGA circuit and 16-bit A/D converter. 16-bit digital output makes it possible to readout the signals of 26.11 M effective pixels at high-speed of 6.84 frame/s in still picture mode. In addition, it realizes 12-bit digital output for high-speed 4K moving picture by window readout mode. This sensor is designed for use in consumer cameras.

GFX100 Sensor Specs

The IMX461AQR-C is a diagonal 55 mm (Type 3.4) CMOS active pixel type image sensor with a square pixel array and 102 M effective pixels. This IC incorporates maximum 36 dB PGA circuit and 16-bit A/D converter. 16-bit digital output makes it possible to readout the signals of 102 M effective pixels at high-speed of 2.7 frame/s in still picture mode. In addition, vertical subsampling binning and 3-horizontal pixel weighted binning realize high-speed 12-bit digital output for shooting moving pictures.

-- hide signature --

Your time is limited, so don't waste it arguing about camera features - go out and capture memories

 Chris Dodkin's gear list:Chris Dodkin's gear list
Fujifilm X100F Fujifilm X100V Fujifilm X-T2 Fujifilm GFX 50S Fujifilm GFX 50R +43 more
NAwlins Contrarian Veteran Member • Posts: 4,814
Re: Bit depth - current supposition and some history
3

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

The number of bits limits how many steps into which you can divide up the representation of levels of light. Your math is correct on the number of steps. But you can use that number of steps to represent any range / size of range you want, and for example an 8-bit JPEG does that with gamma encoding.

But I think all of the major current digital cameras / digital camera sensors use linear analog-to-digital converters. What that means is that as much light as the sensor can hold just before it clips is represented by every bit being a 1, so 65535 with 16-bit representation, 16383 with 14-bit representation, etc. And then, here's the key, half as much light (-1 EV from maximum) is represented by half that number, and so on. So basically, with linear analog-to-digital conversion, the dynamic range capable of being represented is limited by the number of bits.

Of course, at some point the analog values get low enough that all the sensor is fed is noise. Suddenly switching to 16-, 20-, or 24-bit analog-to-digital converters isn't going to let you capture greater dynamic range because the physical properties of the sensor just can't capture it and convey it to the converter.

 NAwlins Contrarian's gear list:NAwlins Contrarian's gear list
Nikon Coolpix S30 Canon PowerShot S120 Sony Alpha DSLR-A580 Tamron SP AF 90mm F/2.8 Di Macro Tamron SP 70-300mm F/4-5.6 Di USD +5 more
Erik Kaffehr
Erik Kaffehr Veteran Member • Posts: 4,205
Re: Bit depth - current supposition and some history

KlasG wrote:

Erik Kaffehr wrote:

Chris Dodkin wrote:

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

Hi,

Sony specs say 14 bit readout.

But, Phase One already has the IQ4150 on preorder. The IQ4150 is specified to have 16 bits, albeit at very slow readout.

The ADCs are ramp type, AFAIK, and that means the precision can be changed by making the stepping on the ramp slower. So, it may be technically possible to extract 16 bits but the camera may slow down to 1 FPS or so.

The GFX 100 sensor is going to have the same pixel dimensions as the 150MP sensor.

If more than 14 bit readout is needed depends on the "engineering DR" of the pixel, which is full well capacity / readout noise. You can calculate it in EV like DR = log(FWC/readout noise) / log(2). Both FWC and readout noise are measured in e-.

If DR is below 14 EV, at the pixel level, no more than 14 bits are needed to fully represent data.

DxO-mark shows engineering DR at the pixel level if you switch to 'screen' mode. They did not measure the GFX 50S but:

  • Pentax 645Z has a DR of 13.39 EV
  • Hasselblad X1D has a DR of 13.39 EV

Both cameras use the same sensor design as the GFX50S

The GFX 100S has half the pixel area compared to the GFX 50S. That would probably reduce DR a bit, sensor development would reduce redaout noise on the other hand. So I would guess that 14 bits would be enough to represent the output of the sensor.

Fuji could add a readout mode for 16-bits, but the only benefit would be to slow down the camera by a factor of four.

Sixteen bits are important to Phase One, for marketing reasons.

Best regards

Erik

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

Hi,

The issue is that the data is unprecise. That lack of precision depends mostly on the random arrival of photons but also on the readout noise.

The dynamic range is the ratio between the maximum signal and the noise floor.

Let us say that a pixel can hold 64000 electron charges, representing 64000 photons.

If that pixel has say 13.5 EV dynamic range the noise floor will be 64000 /(2 ^ 13.5) -> 5.5 e-. We need 16 bits to represent 64000, but the last 2-3 bits will just contain noise. So, we can just throw away those last two bits.

Bright patch values on ColorChecker

The above figure shows raw histogram of an uniform bright patch on a color checker passport. You set the the values vary pretty wildly. The average value for say the G1 channel is 29988, but it varies wildly, so if you call it 29984 or 29988 doesn't matter. The data is inprecise.

Now, if you look at a pretty dark patch:

Dark patch on the ColorChecker

This patch is taken on another ColorChecker in a very dark part of the image. The histogram still represents a single color, here the average G1 channel has the average value 168.9 but the values still vary a lot. So, if the value is 165 or 170 doesn't matter at all. The data are inprecise.

What you also may that the histogram is pretty spiky and the spikes are 4 bins apart. That really mean that this file just contains 14 bits of data, that has been expanded to 16 bits.

I would suggest reading Jack Hogans explanation of the issues involved:

https://www.strollswithmydog.com/how-many-bits-to-fully-encode-my-image/

Best regards

Erik

-- hide signature --

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

JimKasson
JimKasson Forum Pro • Posts: 29,225
Re: Bit depth - current supposition and some history

NAwlins Contrarian wrote:

However, just to point out, I don't think we should assume anything, because in the not-ancient history of analog-to-digital and digital-to-analog conversion, some curiosities have appeared, maybe for good reasons and maybe as marketing nonsense. The compact disc digital audio medium appeared in 1982, with 16-bit audio encoding. By about 1988, some of the player manufacturers started touting that their players had 18-bit or even 20-bit digital-to-analog converters. The stated rationales included that converter accuracy fell with the least significant bits, and therefore that a higher-bit converter not using its 2 or 4 least-significant bits was more accurate; and/or that some sort of dither or somethings was introduced as the least-significant bits for somehow-improved sound. One can imagine that the same basic benefit from the reverse process might be true, or at least claimed, for analog-to-digital conversion.

In audio and imaging applications, absolute accuracy is not very important. Linearity and monotonicity are.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Sony a7 III Nikon Z7 Fujifilm GFX 100 Sony a7R IV +2 more
JimKasson
JimKasson Forum Pro • Posts: 29,225
Re: Bit depth - current supposition and some history

Erik Kaffehr wrote:

Chris Dodkin wrote:

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

Hi,

Sony specs say 14 bit readout.

But, Phase One already has the IQ4150 on preorder. The IQ4150 is specified to have 16 bits, albeit at very slow readout.

The ADCs are ramp type, AFAIK, and that means the precision can be changed by making the stepping on the ramp slower. So, it may be technically possible to extract 16 bits but the camera may slow down to 1 FPS or so.

The GFX 100 sensor is going to have the same pixel dimensions as the 150MP sensor.

If more than 14 bit readout is needed depends on the "engineering DR" of the pixel, which is full well capacity / readout noise. You can calculate it in EV like DR = log(FWC/readout noise) / log(2). Both FWC and readout noise are measured in e-.

If DR is below 14 EV, at the pixel level, no more than 14 bits are needed to fully represent data.

DxO-mark shows engineering DR at the pixel level if you switch to 'screen' mode. They did not measure the GFX 50S but:

  • Pentax 645Z has a DR of 13.39 EV
  • Hasselblad X1D has a DR of 13.39 EV

Both cameras use the same sensor design as the GFX50S

The GFX 100S has half the pixel area compared to the GFX 50S. That would probably reduce DR a bit, sensor development would reduce redaout noise on the other hand. So I would guess that 14 bits would be enough to represent the output of the sensor.

Fuji could add a readout mode for 16-bits, but the only benefit would be to slow down the camera by a factor of four.

Sixteen bits are important to Phase One, for marketing reasons.

You took the words right out of my mouth, Erik.

Jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Sony a7 III Nikon Z7 Fujifilm GFX 100 Sony a7R IV +2 more
JimKasson
JimKasson Forum Pro • Posts: 29,225
Re: Bit depth - current supposition and some history

NAwlins Contrarian wrote:

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

The number of bits limits how many steps into which you can divide up the representation of levels of light. Your math is correct on the number of steps. But you can use that number of steps to represent any range / size of range you want, and for example an 8-bit JPEG does that with gamma encoding.

But I think all of the major current digital cameras / digital camera sensors use linear analog-to-digital converters. What that means is that as much light as the sensor can hold just before it clips is represented by every bit being a 1, so 65535 with 16-bit representation, 16383 with 14-bit representation, etc. And then, here's the key, half as much light (-1 EV from maximum) is represented by half that number, and so on. So basically, with linear analog-to-digital conversion, the dynamic range capable of being represented is limited by the number of bits.

Of course, at some point the analog values get low enough that all the sensor is fed is noise. Suddenly switching to 16-, 20-, or 24-bit analog-to-digital converters isn't going to let you capture greater dynamic range because the physical properties of the sensor just can't capture it and convey it to the converter.

Nice explanation.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Sony a7 III Nikon Z7 Fujifilm GFX 100 Sony a7R IV +2 more
KlasG New Member • Posts: 21
Re: Bit depth - current supposition and some history

Erik Kaffehr wrote:

KlasG wrote:

Erik Kaffehr wrote:

Chris Dodkin wrote:

Previous sensors for Fuji have stated 14 bit - but this one, and the new sensor in the X-T3, both of which are paired with Fuji's later processor chip, appear to have 16 bit specs.

Currently the X-T3 only states 14 bit output though - in line with the comments you made.

Hi,

Sony specs say 14 bit readout.

But, Phase One already has the IQ4150 on preorder. The IQ4150 is specified to have 16 bits, albeit at very slow readout.

The ADCs are ramp type, AFAIK, and that means the precision can be changed by making the stepping on the ramp slower. So, it may be technically possible to extract 16 bits but the camera may slow down to 1 FPS or so.

The GFX 100 sensor is going to have the same pixel dimensions as the 150MP sensor.

If more than 14 bit readout is needed depends on the "engineering DR" of the pixel, which is full well capacity / readout noise. You can calculate it in EV like DR = log(FWC/readout noise) / log(2). Both FWC and readout noise are measured in e-.

If DR is below 14 EV, at the pixel level, no more than 14 bits are needed to fully represent data.

DxO-mark shows engineering DR at the pixel level if you switch to 'screen' mode. They did not measure the GFX 50S but:

  • Pentax 645Z has a DR of 13.39 EV
  • Hasselblad X1D has a DR of 13.39 EV

Both cameras use the same sensor design as the GFX50S

The GFX 100S has half the pixel area compared to the GFX 50S. That would probably reduce DR a bit, sensor development would reduce redaout noise on the other hand. So I would guess that 14 bits would be enough to represent the output of the sensor.

Fuji could add a readout mode for 16-bits, but the only benefit would be to slow down the camera by a factor of four.

Sixteen bits are important to Phase One, for marketing reasons.

Best regards

Erik

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

Hi,

The issue is that the data is unprecise. That lack of precision depends mostly on the random arrival of photons but also on the readout noise.

The dynamic range is the ratio between the maximum signal and the noise floor.

Let us say that a pixel can hold 64000 electron charges, representing 64000 photons.

If that pixel has say 13.5 EV dynamic range the noise floor will be 64000 /(2 ^ 13.5) -> 5.5 e-. We need 16 bits to represent 64000, but the last 2-3 bits will just contain noise. So, we can just throw away those last two bits.

Bright patch values on ColorChecker

The above figure shows raw histogram of an uniform bright patch on a color checker passport. You set the the values vary pretty wildly. The average value for say the G1 channel is 29988, but it varies wildly, so if you call it 29984 or 29988 doesn't matter. The data is inprecise.

Now, if you look at a pretty dark patch:

Dark patch on the ColorChecker

This patch is taken on another ColorChecker in a very dark part of the image. The histogram still represents a single color, here the average G1 channel has the average value 168.9 but the values still vary a lot. So, if the value is 165 or 170 doesn't matter at all. The data are inprecise.

What you also may that the histogram is pretty spiky and the spikes are 4 bins apart. That really mean that this file just contains 14 bits of data, that has been expanded to 16 bits.

I would suggest reading Jack Hogans explanation of the issues involved:

https://www.strollswithmydog.com/how-many-bits-to-fully-encode-my-image/

Best regards

Erik

Super, thanks!

KlasG New Member • Posts: 21
Re: Bit depth - current supposition and some history

NAwlins Contrarian wrote:

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

The number of bits limits how many steps into which you can divide up the representation of levels of light. Your math is correct on the number of steps. But you can use that number of steps to represent any range / size of range you want, and for example an 8-bit JPEG does that with gamma encoding.

But I think all of the major current digital cameras / digital camera sensors use linear analog-to-digital converters. What that means is that as much light as the sensor can hold just before it clips is represented by every bit being a 1, so 65535 with 16-bit representation, 16383 with 14-bit representation, etc. And then, here's the key, half as much light (-1 EV from maximum) is represented by half that number, and so on. So basically, with linear analog-to-digital conversion, the dynamic range capable of being represented is limited by the number of bits.

Of course, at some point the analog values get low enough that all the sensor is fed is noise. Suddenly switching to 16-, 20-, or 24-bit analog-to-digital converters isn't going to let you capture greater dynamic range because the physical properties of the sensor just can't capture it and convey it to the converter.

Thanks for all the good explanations!

So with 14-bits the brightest 1 EV will have 8192 possible steps?

With 16-bits the brightest 1 EB will have 32768 possible steps?

Is this a difference that can be noticed? Or is the extra bit depth only valuable in the lowest dynamic range?

JimKasson
JimKasson Forum Pro • Posts: 29,225
Re: Bit depth - current supposition and some history
2

KlasG wrote:

NAwlins Contrarian wrote:

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

The number of bits limits how many steps into which you can divide up the representation of levels of light. Your math is correct on the number of steps. But you can use that number of steps to represent any range / size of range you want, and for example an 8-bit JPEG does that with gamma encoding.

But I think all of the major current digital cameras / digital camera sensors use linear analog-to-digital converters. What that means is that as much light as the sensor can hold just before it clips is represented by every bit being a 1, so 65535 with 16-bit representation, 16383 with 14-bit representation, etc. And then, here's the key, half as much light (-1 EV from maximum) is represented by half that number, and so on. So basically, with linear analog-to-digital conversion, the dynamic range capable of being represented is limited by the number of bits.

Of course, at some point the analog values get low enough that all the sensor is fed is noise. Suddenly switching to 16-, 20-, or 24-bit analog-to-digital converters isn't going to let you capture greater dynamic range because the physical properties of the sensor just can't capture it and convey it to the converter.

Thanks for all the good explanations!

So with 14-bits the brightest 1 EV will have 8192 possible steps?

Yes

With 16-bits the brightest 1 EB will have 32768 possible steps?

Is this a difference that can be noticed?

No, and it will be entirely swamped out by photon noise and pixel response non uniformity.

Or is the extra bit depth only valuable in the lowest dynamic range?

Pretty much. And there only if the read noise is low enough.

Jim

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Sony a7 III Nikon Z7 Fujifilm GFX 100 Sony a7R IV +2 more
NAwlins Contrarian Veteran Member • Posts: 4,814
Re: Bit depth - current supposition and some history

I am confused, is it 1 bit per 1 EV?

I thought that increasing the bit depth allowed for more precise readout.

For example, if we had a bit depth of 4, only 16 values could be assigned resulting in similar readings being rounded into the same value.

For 14-bit, we get 16384 possible values.

For 16-bit we get 65536 possible values, allowing finer graduations.

Could someone explain how this relates to dynamic range?

The number of bits limits how many steps into which you can divide up the representation of levels of light. Your math is correct on the number of steps. But you can use that number of steps to represent any range / size of range you want, and for example an 8-bit JPEG does that with gamma encoding.

But I think all of the major current digital cameras / digital camera sensors use linear analog-to-digital converters. What that means is that as much light as the sensor can hold just before it clips is represented by every bit being a 1, so 65535 with 16-bit representation, 16383 with 14-bit representation, etc. And then, here's the key, half as much light (-1 EV from maximum) is represented by half that number, and so on. So basically, with linear analog-to-digital conversion, the dynamic range capable of being represented is limited by the number of bits.

Of course, at some point the analog values get low enough that all the sensor is fed is noise. Suddenly switching to 16-, 20-, or 24-bit analog-to-digital converters isn't going to let you capture greater dynamic range because the physical properties of the sensor just can't capture it and convey it to the converter.

Nice explanation.

Thanks. Except after the fact I spotted a typo: I meant to say, in the last paragraph, that at some point the analog values get low enough that all the analog-to-digital converter is fed is noise. GIGO and all that.

 NAwlins Contrarian's gear list:NAwlins Contrarian's gear list
Nikon Coolpix S30 Canon PowerShot S120 Sony Alpha DSLR-A580 Tamron SP AF 90mm F/2.8 Di Macro Tamron SP 70-300mm F/4-5.6 Di USD +5 more
NAwlins Contrarian Veteran Member • Posts: 4,814
Gamma encoding to the rescue

So with 14-bits the brightest 1 EV will have 8192 possible steps?

With 16-bits the brightest 1 EB will have 32768 possible steps?

Is this a difference that can be noticed? Or is the extra bit depth only valuable in the lowest dynamic range?

Gamma encoding of JPEGs etc. relies on there being no need to encode the brightest 1 EV in 32768 steps (or even 128 steps in 8-bit representations) and shifts some of that data / representational capacity to the shadows. Think about it, with 8-bit encoding, a linear representation would mean that relative to full brightness (255), -8 EV is 1 and -9 EV is 0, so if e.g. the brightest non-specular highlight is +4 EV from midtone, anything below -4 EV from midtone would clip to black, and you'd have very visible banding in the shadows. The step from -1 to -2 EV relative to midtone would only have 4 intermediate steps, and I think people would notice banding with steps averaging 0.25 EV. The step from -2 to -3 EV relative to midtone would only have 2 intermediate steps, and people would definitely notice banding with steps averaging 0.5 EV.

But with gamma encoding, we can say, e.g., that for our 8-bit JPEG or whatever, the brightest highlights only really need (say) 10 steps instead of 128, which allows us to give the shadows a lot more steps, and representation of some image details further into the shadows.

More information:

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

 NAwlins Contrarian's gear list:NAwlins Contrarian's gear list
Nikon Coolpix S30 Canon PowerShot S120 Sony Alpha DSLR-A580 Tamron SP AF 90mm F/2.8 Di Macro Tamron SP 70-300mm F/4-5.6 Di USD +5 more
StoryInPictures Regular Member • Posts: 233
Re: Bit depth - current supposition and some history

JimKasson wrote:

Erik Kaffehr wrote:

Sixteen bits are important to Phase One, for marketing reasons.

You took the words right out of my mouth, Erik.

Jim

So is this a nice way of saying there are 16 bits but two are too noisy to use?

JimKasson
JimKasson Forum Pro • Posts: 29,225
Re: Bit depth - current supposition and some history
1

StoryInPictures wrote:

JimKasson wrote:

Erik Kaffehr wrote:

Sixteen bits are important to Phase One, for marketing reasons.

You took the words right out of my mouth, Erik.

Jim

So is this a nice way of saying there are 16 bits but two are too noisy to use?

With P1, so, far, 16 bits is overkill. You could make a case for 15 in some of the newest cameras, though.

-- hide signature --

Posted as a regular forum member.
https://blog.kasson.com

 JimKasson's gear list:JimKasson's gear list
Nikon D5 Sony a7 III Nikon Z7 Fujifilm GFX 100 Sony a7R IV +2 more
Keyboard shortcuts:
FForum MMy threads