Visual effects of DAC Bit Depth & Resolution- can you see it?

[simonb]

Dear Forum,

I have been thinking about the pair of 800 series Digital to Analogue Controllers (DACs) that my lasershow controller will be using to drive the X & Y galvos (via galvo amps of course). They are typical commercial/industrial D.I.L devices. Not bad, but representative maybe of some of the low to mid range DIY DAC designs.

But is 8 bits adequate?

I mean 2^8 gives 256 discrete steps in X & Y. But is this enough? Can you see these discrete steps?

If you project a laser tunnel; a cone from 2 sine waves 90 degrees out of phase, can you see the 'spokes'? I have seen many such tunnel effects where there are definite radial lines of increased brightness. I had assumed these were deliberate anchor or emphasis points, but maybe that is the visual effect of not enough steps. But then again I have seen (smaller numbers) of tunnels that are completely fluid, with nothing apart from that beautiful effect of smoke/haze in the laser beam!

After all, the control software is generating digital sine waves from some mathematical lines of code in the software, rather than pure analogue sine waves from a quality signal generator/oscillator. Not enough bit depth & your sine-waves would resemble stepped waveforms, like a pair of staircases back to back. But then surely the sample rate also contributes equally (mentioned later below).

Surely if you can see the steps on the small screen of an oscilloscope (in XY mode), it will also be visible in your laser effect when projected? Which effects would show the most pronounced effects? I had thought a diagonal line/flat scan at 45 degrees would be worst, because with not enough steps would result in a staircase sort of result?

But then again, that's what all that active filtration & anti-quantisation opamp circuitry after the DAC does I guess, including for example:-

(i) Removing any effect of the system clock signal,
(ii) Smoothing out adjacent steps with the time constant of capacitance,
(iii) Reducing any quantisation noise, jitter between adjacent levels/steps,
(iv) Reducing any aliasing (if present, or if it would have any effect?).

But how well? How much can active signal processing post DAC compensate for inadequate bit depth?

Which made me wonder about the main question, which is: How much visual benefit would you get from increased bit depth, like those below:-

Bit Depth Number of Discrete Levels / Approximate Levels. (No of Discrete Levels = 2 ^ Bit Depth)
-----------------------------------------------------------------
8 bits = 256 levels, ~ 256 levels.
10 bits = 1024 levels, ~ 1 thousand levels.
12 bits = 4096 levels, ~ 4 thousand levels.
14 bits = 16,384 levels, ~ 16 thousand levels.
16 bits = 65,536 levels, ~ 65 thousand levels.
18 bits = 262,144 levels, ~ 262 thousand levels.
20 bits = 1,048,576 levels, ~ 1 million levels.
24 bits = 16,777,216 levels, ~ 16 million levels.
28 bits = 268,435,456 levels, ~ 268 million levels.
32 bits = 4,294,967,296 levels, ~ 4 billion levels.

If I had to guess where this Trade-Off occurs, I would say between 12 to 14 bits, possibly 10 to 14 bits.

I know from Audio files/formats, that most people can tell the difference between an 8 bit file & a 16 bit CD quality file, & some may be able to tell the difference between 16 bits & 24 bits even.

I know this depends much on the sample rates too (~44kHz for CD, up to 196kHz in cases of very high quality sound cards (actually is that kHz or k samples-per-second?)), as well as the music choice. But can you really hear an improvement above 24bits?

Similarly with a lasershow DAC, how aparent would these higher bit depths actually be in a normal typical indoor lasershow? Huge outdoor shows over greater distance could certainly show inadequacy.

Also what about the actual sample rates too. If you consider a graph, bit depth/number of bits as the vertical axis, then sample rate is the horizontal axis. Ridiculously slow would be obvious &possibly dangerous, but what point does sample rate cease to offer any visual benefit?

I look forward to your assistance/opinions.

My thanks,

Simon B.

P.s. Led Zepplins Stairway to Heaven sounds awful at 8bits with a very Low sample rate.

-----------------------------------------------------------------
Sent from my iPad2:-
Simon.D.Bond MEng, BSc(Hons), IEng, MRAeS.
-----------------------------------------------------------------

[Solarfire]

The problem with laser show scanning via galvanometers is not the DAC but rather the point output of the laser show software. Unlike the digital to analog audio conversion where the digital signal is timely constant, in laser shows the digital output is determined by the points of which a graphic is made up of. So a DAC with higher resolution won’t do any good due to the lack of sufficient interpolation between 2 points that are being output. Also in show scanning in part one relies on the inertia (very limited frequency response) of the galvanometers, which also gives a smoothing effect. In order to get best results it’s essential that points be optimally placed in a graphic frame at a given output point rate.

[mixedgas]

There used to be a semi-scientific best on the Pangolin website on minimum # of bits for a good image. 10 is about minimum. 12 is almost too much for the eye until you start manipulating the images.

The problem is when you rotate or scale a image at 8 bits. Chew on that for a while. You end up with a effective 4 or 6 bit image.

Many systems use either a 16 bit DAC or use a scaling DAC driving another DAC's reference port for this very reason.


Although the ILDA test pattern does not look bad when I ran it on 8 bit hardware at full size by truncating off the LSBs. I've done shows with 8 bit systems with 16 levels of scaling (AMIGA 500 and ZAP!). Its not bad for beams, or very carefully designed graphics, but compared to modern systems, you can see the jaggies on moves. The images have a "Cartoonish" flavor of all their own at 8 bits.

Another thing to consider is at 30K, you can see it in the graphics when X and Y do not update at the same time, so DACS must be "DOUBLE BUFFERED" in hardware, or all update on chip, or have zippity quick data input rates.

Steve