Jump to content

Nikon is starting to tick me off


bob_bill

Recommended Posts

"By making photosites smaller, you can improve dynamic range.."

 

- No, you can't. You can make tonal transitions appear smoother through dithering more and smaller photosites, but dynamic range is limited by how many bits are used to digitise the linear analogue signal.

 

In linear space, 14 bits maps exactly to 14 stops, no more. In fact all A/D converter specifications I've seen exclude the least-significant bit from their accuracy. E.g 2% absolute, +/- 1LSb.

 

So in reality a 14 bit converter limits the dynamic range to 13 stops (+ some statistical 'guesswork'). This is easy to see mathematically if you convert the maximum number that can be represented by 14 bits (16,383) to a sensor brightness range.

 

Using a 16 bit converter increases this by a theoretical 2 stops to 65,535:1. Alternatively, a log curve could be applied to the analogue signal before digital conversion. Applying a gamma curve or LUT in digital space doesn't work to increase dynamic range; only to prevent waste of digital discrimination in the lighter tones that would occur from retaining a linear representation.

 

Multiple photosite signals cannot be manipulated after digitisation to increase their brightness range.

 

Besides, 16 bit A/D converters are pretty much industry standard in every other application - except digital cameras it seems.

 

Andrew. A square bellows buckles inwards when deformed to shift the lens. There's a limit to the amount of shift before the bellows intercepts the light path. Tapered bellows are also more flexible and squash flatter, since the folds fall inside one another.

Edited by rodeo_joe|1
Link to comment
Share on other sites

  • Replies 85
  • Created
  • Last Reply

Top Posters In This Topic

If we assume adjacent pixels have independent noise, by combining each block of 2x2 pixels from a 45MP camera into one pixel of a 11MP final image, the per pixel SNR increases by a factor of two. You thus gain some dynamic range and you can store the final image into a 16-bit file. In practical applications you rarely actually use every pixel as a separate image element; resampling to a lower resolution for printing or display is typical. The high resolution cameras have a slight advantage because the ADC noise plays a smaller part in the shadow noise in the final presentation size (simply because each pixel corresponds to a smaller area of the print) than it would in a file from a lower resolution camera. In practice it is typical that this extra shadow quality is evident in low ISO images.
Link to comment
Share on other sites

It's not about the noise floor. While noise limits the depth of shadow detail that can be captured, the dynamic range limit is purely down to how wide a brightness ratio the ADC can encompass.

 

Let's plug some numbers in to clarify:

Say it takes 0.001 Lux-seconds exposure to store enough energy in a single photosite to trigger a theoretically perfect ADC to output binary 1. It then follows that 0.002 Lux-seconds are needed for binary 2, and 0.004 Lux-seconds for binary 4, etc.

 

With a 14 bit ADC the maximum sensor brightness that can possibly be captured is 16.383 Lux-seconds. Whereas a 16 bit ADC could cope with 65.535 Lux-seconds - assuming the sensor is also capable of holding that energy level.

 

Therefore the combination of sensor and ADC bit-depth are directly responsible for the 'dynamic range' a camera is capable of capturing. Not the noise floor, and not the number of pixels.

 

Incidentally I've experimented with 'pre-flashing' a digital sensor using the multiple exposure facility. It does work to lift shadow detail and thereby compress the sensor brightness range into something that's easily handled. So there's another technique that could be used to allow better use of available ADC bit depth. The 'pre-flashing' could in fact be an artificially introduced charge in the photosites.

 

AFAIK. None of those techniques - analogue logarithmic amplification, or 'pre-flashing' - have been used in a digital camera to improve dynamic range or SNR.

Edited by rodeo_joe|1
Link to comment
Share on other sites

I'm purely a software person, Joe, so I'm a bit confused.

 

My understanding is that the sensitivity of the bottom-most level in the sensor sites is already approaching a single photon. Averaging over multiple sensor sites gives you a value that represents a small number of photons in a large area, so that's the dark level benefit - but you'd get the same from large sensor sites.

 

My argument about increasing pixel count was about the maximum well depth - the brightest value, or record of maximum number of photons, hitting a sensor site. If you record, say, four high resolution pixels, each of those records up to 14 bits worth of photons hitting the sites (with a scale factor, but if we're at roughly the "capture single photons" stage, there are limits on what that can be of what's being recorded is to be vaguely linear). If you sum those 14-bit values, you get a sensible 16-bit number.

 

A single large sensor site covering the same area gets hit by four times as many photons. If you still go through a 14-bit DAC, either you saturate, or you have to limit the sensitivity at the low-light end to compensate.

 

From an information perspective, you're comparing 14 bits (16, as you say, 384 levels) to 4×16384=65536 or 16 bits. The issue is, I argue, that the sensor site isn't recording illuminance (incident luminous flux per standard unit area) - it's recording the incident luminous flux just hitting that sensor site. (Actually integrated over time, if anyone's annoyed by my abuse of units, but the same was true of the per-area measure.)

 

Which, incidentally, comes back to my favourite argument about sensor size equivalence and the relationship between ISO and sensor size. It's all interconnected (even though a colleague keeps yelling "unnecessary dependency" at me when I see the world this way).

 

Of course, of it were that simple, there would be no difference between high and low res sensors at high ISO; there's more at play. (The A7RIII sensor is very good at high ISO, but I think we believe Sony do some raw preprocessing which confuses DxO...)

 

Is the idea of "pre-flashing" to lift the sensitivity of the sensor to the level where photons start registering (like a half-level offset, digitally)? I was under the impression that modern sensors are sensitive enough that this wasn't needed, but that's very much in the "nasty little analogue electronics" black box that I try not to have to know about. :-)

Link to comment
Share on other sites

While there are multiple definitions for dynamic range, it is typical that it is defined as the ratio between the largest signal that can be recorded and the smallest signal that is discernible from noise. Different sites that test dynamic range use slightly different criteria, i.e. SNR = 20 or SNR = 1 can be used to determine the limit where image detail is just discernible from noise. For example, photonstophotos use SNR = 20 and dxomark use SNR = 1 (in normalized 8 MP images).
Link to comment
Share on other sites

As I've banged on about before, my opinion is that trying to capture a real sensor brightness range of over 12 stops is a pretty futile exercise. However, the benefits of a higher ADC bit depth aren't just in theoretical 'dynamic range'. The refinement of colour discrimination in shadow areas can't be denied. At the very least we would have 3 bits defining the lowest shadow tone, and not just one.

 

"Is the idea of "pre-flashing" to lift the sensitivity of the sensor to the level where photons start registering.."

 

-Basically, yes. In our theoretically perfect 14 bit world; adding 0.002 Lux seconds of light (or equivalent electrical charge) lifts the shadow detail above the noise floor. While the highlights of 16.383 Lux-seconds become an insignificantly larger 16.385 Lux-seconds. The shadow luminosity is affected significantly, while the highlights change hardly at all.

 

As I said, this can be done experimentally using a double exposure. But, as I discovered, extreme care is needed in order to regulate the amount, colour quality and evenness of the flashing exposure.

 

Going back to the large tapered bellows: That too would help with increasing sensor brightness range. A larger dark chamber can absorb more stray light than a smaller one, given the same surface reflectivity. Another plus point for a mirrorless body with less restriction on the dark chamber size and construction

Link to comment
Share on other sites

I'm not objecting to the desire for 16-bit raw, but I'm standing by my "smaller pixels as an alternative" theory. You don't get more detail if one level represents one photon (ish). If you have a larger sensor site so that relationship isn't true, then yes, more bits help - although Nikon have never felt inclined to optimise dynamic range in their low-pixel cameras.

 

Unless there's a base level below which the ADC won't trigger, I'm not sure I see the benefit of priming the sensor. People have reported that Nikon already adjust the raw levels to remove the noise floor; priming the sensor may avoid this effect (which I don't approve of anyway). Still, if raw records linear intensity, you've still got noise, just with a higher number. I don't see how that correction can't be done digitally.

 

Are smaller bellows not better able to block unwanted light paths? I agree that mirrorless should allow this to be more controlled, but I'd say it's more to do with not having a large uncontrolled mirror box in the system.

 

(Sorry if I sound argumentative - just trying to understand!)

Link to comment
Share on other sites

"Are smaller bellows not better able to block unwanted light paths?"

 

- That's not been my experience.

If the lens has a wide image circle, then stray light hits a small bellows and bounces around to create non-image-forming flare. (The peaks of the bellows folds are usually quite reflective BTW.)

Now imagine a bellows that opens out enough not to cut into the cone of light from the lens. That image circle will hit around or behind the sensor, and reflect into a larger cavity than the small bellows could provide. There'll still be flare, but it can be better controlled.

 

And if you're trying to provide lens movements, a small bellows that cuts into the light cone is the last thing you want!

Edited by rodeo_joe|1
Link to comment
Share on other sites

I'll trust you, Joe - I have very little experience with bellows. I was kind off assuming they worked like the light baffles inside lenses (as well as, obviously, being flexible). I'm not sure how a larger cavity helps (the light will still hit something); bouncing from behind the sensor may well help a bit, although there are a lot of mechanical bits in a camera that are probably harder to black out than I'd have assumed bellows would be. Light tends to exit lenses to some extent even at relatively extreme angles, the question is how much, and that depends greatly on the lens design and its internal baffling.
Link to comment
Share on other sites

"I was kind off assuming they worked like the light baffles inside lenses.."

 

- The main difference is that light baffles are usually metal and can be made extremely thin. Therefore the lens-axis-parallel surfaces of a metal baffle have a very small area and low angle reflection is minimised.

 

OTOH, a bellows has quite a lot of axis-parallel surface, at each internal fold, and its reflectivity is also compromised by the blacking material having to be flexible.

 

Sorry about the cumbersome description 'lens-axis-parallel surface' BTW. I couldn't think of a more succinct way to describe the top/sides/bottom of a lens baffle or bellows fold.

 

I haven't worked out the reflective maths of dark chamber size. It just seems intuitive that a larger surface area at a greater distance will reflect less light (or rather absorb more) than a smaller area close by.

  • Like 1
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...