Blending, again.

135" 16x9 screen, two Electrohome Marquee 8000s, one nVidia 6800, and a bit of Avisynth:

A few more, and one for scale.

Hi Tim,

The screen is 135" in the diagonal, so it's 120" wide. The picture below shows the un-blended overlap; I think it's about 35-40% per projector on the screen. The resolution per projector is 1208x1080 and the 2412x1080 resultant screen real-estate is overlapped by 512 pixels for a net 1920x1080 image. The rasters are as large as they can be in the vertical direction, and the top of the rasters are about as large as they can be in the horizontal.

Edit: Added picture.

Hi Gino,

Those HQF lenses should settle the question, fairly definitively, as to whether a 2.35 blend with CRT projectors is possible with true transparency as it regards the blend-zone. If those don't cut it, all that's left is those Nikon lenses from the Sony HDIH series, right? If you don't mind me asking, what size image were you throwing, per projector, with the GT-17s?

Anyhow, I hope you're enjoying tuning out what appears, already, to be a top-shelf display.

Scorch and Cliff,

Thanks. Cliff, you are in for a treat. Seeing the image that this little junker yields is pretty amazing, and you're about to go for a ride in the Enzo.

Andy,

Nice to hear from you. Don't get rid of that Quadro 4000 yet, it just might come in handy.

Hi Tim. Just MPEG-2, mostly, and what can be played on your garden variety HTPC.

Hi there, Nick.

I have to defer on that question, because the tubes in one of the projectors are quite worn. The array goes deficient in blue, on one side, at the high IREs. It is at these high IREs, specifically in the blue component of the image, that the blend irregularities are most apparent.

However, the blend zone is transparent in all of the material I've seen since applying this admittedly kludgey compensation. Previously, the vertical band of increased brightness in the middle of the screen when using nVidia's blending (Andy's "skunk stripe") had the effect of destroying the depth of the image and serving as a constant reminder that two projectors were employed to create the image. Now, however, the image looks continuous, if you will, across the width. Apart from an almost imperceptible drop in sharpness at the center 10% of the screen (the result of less-than-ideal convergence) I would say the image is indistinguishable from that of a single projector, at the very least with regard to colorimetry. In this regard, I would say that the nVidia approach is now at par with the VLC solution.

If anyone wants me to take any pictures, now is the time to ask. Lefty is getting a new set of tubes tomorrow, and there is a chance that the projector might not make it through the procedure. But barring that, I will be able to give a more confident answer once the new tubes are in.

Hi RaptorHT,

The HTPC employs a Core2Duo processor running at 3Ghz, with 1 Gigabyte of RAM and an PCIe variant of the 6800U. This allows the computer to run the nVidia blend software (2% processor usage according to the Windows Task Manager), the video playback software (~40% of one core using ~20MB/s Mpeg-2@HL, using the DScaler5/IVTC audio and video decoders) and the Avisynth script (~50% of one core, using 2 instances of Overlay in a multithreaded setup). I don't see any stuttering with either MPEG-2 or with high bitrate H.264 variants. So the computer handles HD by software decoding, and isn't suffering from any stuttering, even with the blend apparatus in place. I haven't tried it yet with any HD acquisition cards, yet, though I don't think that acquisition will present a greater processing load to the system than decoding compressed material. I'm not sure that the PCIe bus will handle simultaneously porting uncompressed HD-SDI/component into the CPU and out to graphics card, but that's my next project.

Hi Gino,

63"x76", if I recall, is in the middle of the intended range of best-focus for the GT-17s. If the HQF lenses do yield better edge focus at that size, it would seem that you will have given a compelling data point regarding the focusing ability of the two lenses. So you're using a 10% overlap? I guess that explains needing the razor sharp lenses.

I've been interested, too, in the light output advantages of using the full raster. TSE mentioned, a while ago, that light output is a function of beam current. So, assuming no blooming is happening, if a smaller raster has the same current applied to the raster over the course of drawing the frame, will it have the same light output as the larger raster? I'll pull out my light meter and try and to get a grip on what's going on. At any rate, it seems that a larger raster will wear slower than a smaller one, and that less overlap per projector will yield more light.

Don, Cliff's meet sounds like it's going to be both an eye opener and a great time. I'd think, though, that with top-end 9"er the primary benefit to blending would be a legitimately cinematic 2.35 presentation, as 16x9 1080p isn't a problem. And what a fantastic idea, building with room for expansion. I'm taking notes...

Tim, thanks for the heads up. I had to replace all three tubes, though, so I ended up resetting the projector before I put the tubes in. This sets everything to 50 except for the contrast, which goes to 35, right? The Marquees are pretty amazing machines - even without a horizontal linearity control it only takes me about 2 hours to get the machine into a workeable state. They're really a great baseline for blending.

Hi Ken and Dokworm, you guys are spot on; the video card is a nVidia 6800 Ultra running Quadro 4000 drivers, and the compensation for the light vertical strip is an overlay. Thanks for all of the help with this.

Terry, I'm working on getting some pictures together to try and illustrate what's been done, so hopefully anyone who tries the nVidia and/or Avisynth route can avoid some of the pitfalls I got stuck on.

In brief, the method described here relies on the idea that a blend of two projectors should create a coherent image with regard to luminosity and colorimetry across the horizontal dimension of the image. If one were to split and overlap an image, with 60% on both projectors, the resultant image would have a band with twice the luminosity in the middle 20%. Simply halving the luminosity of both projectors on the innermost 20% of the blend would, assuming perfect colorimetry, yield an image that has the correct luminosity over the resultant image. In practice, with CRT projectors, this yield an image with three distinct color characteristics; one representative of the left projector, one of the right, and one that is the sum of both halves. Our eyes are sensitive to color change to such a level that no degree of manual calibration of the two projectors will yield a unified image with a simple halving of the intensity. To "blend" the images, therefore, takes a more complicated luma attenuation scheme, horizontally, across the overlapped region. Paul Bourke illustrates these blending concepts in "Edge Blending with Commodity Projectors," (http://local.wasp.uwa.edu.au/~pbourke/colour/edgeblend/)

nVidia, in their Quadro drivers, supplies an application called "nvKeystone" that effects such a blend, over and on top of a hardware function that splits and overlaps the image onto two outputs of a dual-head graphics card. The parameters for this split, overlap and blend combination can be modified using the "advanced" tab of the "display" control panel in Windows, once the appropriate driver set has been installed. I assume that the mathematical basis for this software implementation is sound; but in practice, with CRT front projectors, the package does not yield a visually unified image. This may be the result of CRT projectors having a different gamma response than is expected by the algorithm, or may suggest that the projectors are set up differently with regard to full white/black levels than is assumed. nVidia doesn't supply any guidelines with their software, so all that can be said is that the result is close, but certainly does not yield a watchable image.

Without getting too far into the nVidia control panel itself, there is generally a zone, vertically, of increased brightness at the horizontal midpoint of the screen flanked laterally by two areas characterized by a slight deficit in brightness.

So without knowing what the expected projection device characteristics that nVidia expects are, I simply left the projectors set at setting that were as close to one another as possible and yielded a good image for a home theater environment, and then went to work on attenuating the blend.