Resolution required of colour channels given modern diode based systems?

[DMills]

Hi all,
I have just been hacking on something and noticed an interesting issue....

The DAC in play has 8 bit resolution for each of the colour channels, and it looks just fine feeding a correctly set up PCAOM in front of a gas rig.

However I recently built a cheapus maximus little projector for use at home for building shows and noticed something that I do not see with the gas rig.
This little projector is a hundred mW of green, a snifter more red and 445 to taste, but what I find is that by the time I have dialled in the DAC to bring the thing to threshold for the green (Crap driver doesn't make threshold until about a volt on the mod line, I know, I know....), and applied a suitable gamma correction to give as close to a linear apparent brightness as I can get then 8 bits is just not enough to give a smooth fade at the bottom end of the curve.

This screws with colour fidelity any time you use a gamma a long way from 1.0 (which is often) as the gamma mapping function ceases to be surjective.

Does anyone have good data as to how many steps you need on the output side of the gamma conversion to give a smooth perceived fade at low levels (Or at high levels if your rig needs gamma the other way)?

The problem is actually more interesting then it first appears as if each source in the projector has different gamma then the steps will occur at different points in the fade so a smooth full white to black fade gets blocks of odd colours as it goes down.

I have a suspicion that given a sane range of gamma values we probably need about 10 or 12 bits of resolution to get below the point at which fades seem to have discreet steps.

Any thoughts?

Regards, Dan.

[weartronics]

Does anyone have good data as to how many steps you need on the output side of the gamma conversion to give a smooth perceived fade at low levels (Or at high levels if your rig needs gamma the other way)?

8 bits per-channel is plenty. This is what the vast majority of computer systems use, with a gamma correction of 1/2.2 prior to 8-bit quantization. Of course higher resolution systems exist, but it is difficult to discern the difference, even for trained eyes.

The problem is actually more interesting then it first appears as if each source in the projector has different gamma then the steps will occur at different points in the fade so a smooth full white to black fade gets blocks of odd colours as it goes down.

If each color has a different response, any correction must be per-channel.

I have a suspicion that given a sane range of gamma values we probably need about 10 or 12 bits of resolution to get below the point at which fades seem to have discreet steps.

I doubt that your problem stems from the gamma correction. I suspect it's due to compensating for the driver offset/scale in software. It means that regardless of the available resolution, you're not even using it! I know FlexMod and Die4Drive have provision for offset/scale in hardware to use the full DAC range, and I suspect other drivers are similar. If your driver does not have this feature, you could implement it with a single op-amp on the color signal.

If you want to implement the driver offset/scale in software, you might consider one of the C-Media based sound card DACs with 16-bit color channels. It wastes some bits, but these bits are very cheap.

If you don't want to change your DAC, you might also consider Eye-magic IRIS safe (or a similar system of your own design). This is an off-the-shelf digital solution with automatically-calibrating closed-loop feedback for intensity control with gamma correction. I have never seen or used one, but in principle it sounds like it would even compensate for discontinuities in DPSS response due to mode-hopping and temperature fluctuations.

Best regards,

weartronics

[DMills]

8 bits is just fine on the INPUT side of the gamma calculation, even 6 bits is probably more or less ok, but the effect of the exponential in the gamma calculation is that as gamma moves away from 1.0 the available 2^8 steps become compressed at one end of the scale or the other as seen from a intensity linear input.

Sure the driver on this POS green is utter junk, and I should throw something better together but I doubt that anything simple is going to linearise it to the point that this problem goes away.

Now it is undoubtedly a non issue for beam shows, and probably not much of an issue with a better laser, but with this thing having a gamma of about 4 by eye, it is a bit boring with THIS particular junky laser.

Regards, Dan.

[mixedgas]

Start with the instrumentation:
http://www.homecinema-fr.com/colorimetre/index_en.php

Steve

[DMills]

That sensor is now unobtainable (No longer manufactured), which is boring, but it turns out that Agilent do a small board intended for LED backlight drift compensation that might make for a very interesting optical feedback sensor for doing something similar to the Iris product.

Regards, Dan.

[buffo]

you might also consider Eye-magic IRIS safe (or a similar system of your own design).

DZ investigated producing a hobbyist version of this exact product about 9 months ago. He spent several weeks experimenting with various feedback mechanisms coupled to a fast photodiode, and the conclusion was that yes, this is a viable solution to perfect color response, provided your lasers can respond fast enough.

Unfortunately, for all of the DPSS lasers that DZ experimented with, the *only* way to make them respond fast enough was to run them at a significant fraction of their rated maximum power (somewhere between 15 and 25%).

In fairness, however, remember that when DPSS technology was first invented, they said that they could never be modulated. Fast forward to today, and clearly that isn't the case. But even so, we all know that the cheapest direct-injection red diode has far better modulation response (not to mention power linearity!) than even the most expensive DPSS laser.

Adam