CurtPalme.com Home Theater Forum

[emdawgz1]

[perisoft]

Ah hah, this will be fun.

Just thinking off the top of my head, and given my penchant for software processing, here goes.

First: Rear projection. But not a lenticular screen - full viewing angle via a completely matte surface. That, of course, means you need insane amounts of brightness, but we're not counting cost.

Second: Not too huge, but pretty big. Maybe 15-20 feet horizontal screen size. If I want one bigger and money was no object I'd just do film.

So how do I get the brightness?

909s, of course - blended, of course. But I'd use an array of them blended, probably 4 by 2, and stack each unit, so sixteen projectors.

Of course, making them all blend is impossible, right? Certainly not doable with the onboard geometry controls, and what about drift? So, behind the audience I'd put a scanning digital camera - or an array of high resolution SLRs - which would constantly image the screen surface and compare the output with the input. So, the data output from the source material (DVD, bluray, whatever) would be streamed to a very fast computer which would split the video streams to each of the sixteen 909s.

Because the effective resolution of the 909s blended four wide and two high would be something along the lines of 8000x4000, we can afford to lose a lot of resolution in the software convergence and blend procedure and still end up with 4000x2000 pixels effective.

And because we have a *real* input source telling us EXACTLY where each pixel we're trying to project really ends up, we can constantly alter the blend and convergence patterns to ensure that the image is perfect even if the projectors drift - because it's done "ex post facto" there's no 'real' knowledge of each projector's output necessary; it just happens automatically.

The initial setup procedure would involve first building a spatial map and then building a luminance map:

-Fire each pixel on each tube of each projector and watch where it ends up on the screen. Now we have a lookup table which we can invert to tell a projector where to light up a pixel to make a pixel light up at a given physical geometry on the screen surface.

-Map the overlaps between the projectors - obviously since it's a blend, some spots will get hit by four projectors and some by two, and probably some by eight.

-Build a blend map given that data. I'm not 100% sure how I'd do this yet.

-Now, go back the other way and run through the LUT inverted, as described, and check the luminance output of each pixel. Probably you'd do this at several levels to build a very precise gamma map.

Now, to display a pixel, you run through two LUTs - one for the gamma/blend and one for the position - and end up with a **PERFECT** image. Absolutely perfect.

As the projectors run, of course, the cameras continue to look at the output and adjust every pixel every few minutes to ensure that drifts and changes in output levels don't affect the image.

You could do it with smaller blends, too, of course, but I'm not sure at what point it makes sense. Dropping to a quadzilla would certainly make the processing infrastructure a LOT easier, but this was an exercise in set-your-hair-on-fire utter perfection!

And that's what I'd do. I haven't thought about the sound yet. I bet you could do it for a quarter million - achase, if you want to have a go just give me a holler.

[Moose]

[perisoft]

Yeah, that's what I meant - eight pairs of stacked projectors, in a blend. And a big, honking AC unit to keep them cool!

(And now that I think about it, it would make sense to do a triple stack. Then run each projector at 2/3rds max. If a projector dies, the software will crank up the other two to match within a minute or two, and then you can swap it out with another one (just bolt it in place and plug it in) and do a reconfiguration of the lookup tables. Voila!

[cmjohnson]

Since we're in fantasy land now, I'm going to revise my design:

An array of 16x9 Marquee 9500LC Ultras, fully modded to the hilt, in rear projection mode, with a backup projector for EACH one,
always ready to go at a moment's notice. That's 280 Marquees in the installation! Each is projecting on the SMALLEST screen that
it can be perfectly focused to while also using the entire usable phosphor surface. All edges blended, of course.

Convergence handled by a prototype "flash acon" system which has better than 2 pixel resolution per displayed pixel, consisting
of some VERY expensive digital cameras and customized software and test patterns. The system even auto-converges every
few minutes while you watch a movie, as it looks at the input video, calculates what it should look like on the screen to the acon
camera, and makes on-the-fly adjustments in sub-pixel increments as often as it finds a discrepancy. Colorimetry and color
space conversion is handled by it as well. Of course, the system is fully capable of auto-adjusting every individual Marquee,
and what's better....can fully adjust them all at ONCE.

Processor power for all this is provided by a Cray XT4 supercomputer array, in a configuration that's "adequate" for the demands
of the full system. Surplus computing capacity is dedicated to a "software Teranex" solution for sub-pixel image processing across the
entire array. With scalability up to 120,000 processor cores, this shouldn't be a problem.

In its spare time, it will also be used to drive the most awesome flight simulation system ever built into a home theater.


CJ

[Gary Murrell]

[AnalogRocks]

All Gino has to do is turn up the lights in his theater room.

[AnalogRocks]

All Gino has to do is turn up the lights in his theater room.

[JustGreg]

I'd hire Curt, Kal, draganm, and Clarence. It would kick the sh*t out of anybodys.
I don't necessarily care how it all would fit together and what names were burnt in or stuck on the parts n pieces....but I know it would kick some serious ass!

[perisoft]