[WIP] Dual out to single shutterglass signal conversion

[iondrive]

Hi all, I've got some interesting ideas for you...

The ultimate goal of this project is to take a dual projector 3d output signal and get a single projector shutterglass output from it. Of course this means you could also use a CRT monitor. This would be a good project for people who want to use iZ3D's driver with shutterglasses without losing sync by using iZ3D's dual projector output selection instead of the shutterglass output. You'll see what I mean if you keep reading.

I don't know if I'll actually do this but someone else could use the ideas for their own version.

Reasons for doing it this way:
One problem with a passive polarization setup is that you need a special screen to preserve the polarization. If you don't want a special screen and just want to use your wall instead, then you should probably consider shutterglasses with a DLP projector.


Version 1: get a 3d signal to your dual projectors and block the light with shuttering LCD glass.

OK, I don't expect anyone to do this but it's good to explain it in order to get started with the ideas of the final version. The reason to not do it this way is because it will end up pretty dark and I expect not that good.
Note that you should use DLP projectors instead of LCD for various reasons.

For this version, just setup your 2 projectors and block each one with LCD glass that is connected to the shutterglass signal. This should block each beam alternatively in sync with your glasses so that when your left eye is clear, only the left projector image is hitting the wall. When your right eye is clear, only the right projector image is hitting the wall. This should work but is awkward because you need bigger LCD glass to block the projectors and not overheat and not leak much light. Of course the two projectors' refresh rate should be EXACTLY the same and also EXACTLY in sync with each other. I've tested this and it IS possible to perfectly sync two monitors in either Clone or Dual Monitor modes although it's possible some cards/setups could have problems. I guess you should try to verify this capability before you embark on this project although most people won't have the hardware to test it. I can talk more about testing later. You can use E-D Activator to turn on your glasses if your dongle is connected to the #1 monitor. iZ3D must be used in Dual monitor mode and not Clone mode if I am correct. For now, on to Version 2.


Version 2: like Version 1 but don't use LCD shutters on the projectors.

OK, so for this version, take out the LCD shutters on the projectors and insert a box before each projector's VGA-in cable. This project uses analog signals but maybe someone could adapt it to digital. The box has circuitry that, when activated, blacks out alternating frames of video for every other refresh. This mimics the function of the LCD glass that was removed but does not have the downsides like dimming or overheating. The blanking generated by the two boxes must be 180 degrees out of phase and can be triggered by the shutterglass signal or else just activated with a switch and you would have a 50/50 chance of being out of phase. Cycle the switch if it's out of phase or else build in a parallax inversion switch or use Ctl-F8 with iZ3d's driver. On to the final version.


Version 3: take signals for the two projectors and combine them into one output for one DLP projector or CRT monitor.

Now design another circuit that can switch VGA input signals with each refresh so that you get one output. Send this to one DLP projector and it should work. I guess you could skip the blanking circuit and just make the switching circuit. Now you can use iZ3D's driver with shutterglasses and never lose sync. Hurray. You might need to use Ctl-F8 to do field inversion occasionally between cutscenes or something but not if you used the shutterglass signal to help trigger the video signal swapping properly. I suppose I should have just skipped to version 3. OK, sorry. Anyway, I think this is an awesome idea and wouldn't be surprised if others had thought of it before me but I haven't seen it anywhere so I thought I'ld post it myself. Please point me to it if you know of any other threads that would help with this project.


Hopefully this is a project that will appeal to people like RageDemon and QuadropoeniX. (hint hint)

Later.

[iondrive]

OK, found the "[WIP] Active S-3D dual projector system" thread,

For Version 1, I was wrong about the projectors needing to be exactly in sync but I think it should still be better that way unless there's some other timing issue like shutterglass transition speed. Other Versions still need perfect sync.

Version 4: shutterglass signal out instead of in
I guess you could design the box differently so that it generates it's own shutterglass signal and then you wouldn't need the E-Dim dongle. Let's call that version 4. Wired glasses could plug into the box or else it needs a wireless emitter jack. OK so now I can research do-it-yourself vga pass-through devices.

later.

[iondrive]

Hi all,

Well it's been a while and I think it's time for another post.

The Chip I tried so far:
So I've actually worked on this and I've learned that the magic word is multiplexer. For those who don't know, multiplexing is just switching between a bunch of inputs to one output. So far, the chip I tried has it's problems so I'm going to have to try another setup. The chip was the CD4053 and I chose it because I could get it locally but the problem is some ghosting and signal degradation. It turns out that many demultiplexers can also multiplex and that's the case with this one. It's kind of generic and has 3 separate sets of 2-channel 'plexers on one chip. It's like 3 single-pole--double-throw switches (three sets of 2-inputs and one output), and in this case, one each for red, green and blue analog color signals (0V - 0.7V). It was a good place to start and here's the datasheet but you can skip this one since there are better chips for this app.

http://www.datasheetcatalog.com/datashe ... 053B.shtml" onclick="window.open(this.href);return false;

Usage:
Signal "A" selects between inputs ax and ay, and so on.
For Version 2 of this project (the blanking circuit), you would use 2 of these chips, one for each projector and send the RGB data to the chips' input pins and send all other signals bypassing the chips straight to the PJs. Then RGB outputs go to the PJ and the shutterglass signal from the E-Dim dongle's 3-pin mini-din (top hole) goes to the chip's "A" "B" and "C" pins. For one projector, the RGBs go to ax, bx, and cx, and for the other PJ, the RGBs go to the other chip's ay, by, and cy. Ground all other abc inputs to give you black color data for your blanking function. Even though I say don't use this chip since you will see some ghosting during the blank phase, I'm explaining this since the setup will be similar for other chips. The key here is "ground the other rgb/abc inputs to give you black color data for your blanking function". You may be able to use a different signal than the E-Dim dongle such as the DDC pin 12 on the vga cable but I'm describing the way I did it. It worked with my test CRT setup except for the ghosting.

For Version 3, just use one chip and send both sets of RGB signals to the chip but it gets more complicated than that. From my tests, I've learned that you also need to send both Hsync signals to another switch for this to work right. That means you still need 2 chips since there are only 3 switches per chip. Also, you still use the E-Dim dongle for the shutterglass signal but don't use the signal from the wired glasses jack since it has the wrong freq and voltage. You have to use the signal meant for the IR emitter. This setup works but there are still problems.

The 3-CRT testing setup:
If you're going to try crazy stuff like this, I suggest this setup for testing. Your computer has a card with 2 outputs. I'm testing on a win98se with a Geforce FX 5200 since my XP computer is still down. I have 2 CRT's connected with a VGA Y-connector on each cable. This gives me easy access to data lines since the unplugged Y-end is female and I can just stick wires in holes there. RGB data is on pins 1, 2 and 3 while pins 6, 7, and 8 are return lines that should be grounded. In practice, grounding only pin 6 was enough. For the 3rd CRT (the shuttering display), get a male to female VGA converter so you can stick wires in holes there too. Basically this CRT needs 6 things: RGB, Hsync, Vsync, and ground. You can skip DDC on this one, just make sure that all CRTs can do the same res and freqs. The 3rd CRT gets RGB and Hsync from your chips and you can just send Vsync from display #1 to CRT3 bypassing the chips. Before turning on CRT3, I like to make sure that CRT1 and 2 are in sync. That can be tricky and I'll leave that for another post. I hope someone can help me with that since windows likes to not listen to me when I tell it to change freqs. There's a prog call Refreshlock that I might try someday.

Problems:
I think I'm running out of time for tonight and problems take some time to describe. They involve switching to clone mode or dual mode and then having weird freqs on your monitors. It might be a driver problem.

On to better chips:
Eventually I found this other chip, the MAX4885, and it looks really good for this app since it's specially designed for vga signal switching. I got a free sample of this one and found out that it's only 3mm square and has 32 wire-spots that need to be soldered to. How am I going to do that? Oh well, live and learn. I should have looked closer at the specs. Feel free to LOL. It's funny. Get the datasheet here if you're curious:

http://www.maxim-ic.com/quick_view2.cfm/qv_pk/5186" onclick="window.open(this.href);return false;

But all is not lost. This lead me to the Maxim website and they have other chips I might be able to get. Also, there is a test evaluation kit available for the 4885 with VGA jacks and everything. The problem is that the best deal I've found is $140 for 4 Eval Kits. Maybe I can figure something else out. If you want to browse Maxim chips, their website is pretty good:

Goto http://www.maxim-ic.com/products/switches/" onclick="window.open(this.href);return false; and select video multiplexer apps (active and passive) in the product tree. The "active (buffered)" webpage looks like it has some good chips but don't get excited because it's often hard to order small numbers of these things. Most websites want you to order alot of them or else a minimum order of at least $200 or something. In the case of the MAX4885, I could get a free sample though so that was very nice. Too bad it's so tiny. Anyway...

Finally,
hopefully the final version of this will only need one monitor but at this point in time I think it will need two because of some DDC issues. For a one monitor setup, I think I need a way to fake a virtual CRT on one port. I know about using resistors to simulate a dummy VGA CRT but then I can only get 640x480 resolution on the second monitor while I would like at least 1024x768. I tried using a KVM with one monitor and both VGA ports going in to the KVM but I get the same result as with the resistor method. If anyone knows how to fake a 1024x768 monitor on a second port, that would be helpful.

By the way, this project could also be called a "Passive to Active 3d Signal Converter" and you can Google "passive to active" and 3d and find some things like http://www.christiedigital.co.uk/EMEAEN ... verter.htm" onclick="window.open(this.href);return false;. There's another too (the XPO.2PA) but these things are hard to find prices for and one I found was 12 thousand dollars!!! Sheesh!

I'll post more in a month or two.

--- iondrive ---

PS: I'll include drawings once I get this setup working better.

PPS: If you decide to try some other multiplexing chip, make sure it's for analog signals. Chips for multiplexing digital signals won't work for this project unless you want to try making a digital version.

[Likay]

I think some of the problems is caused by unsync of the actual videocard outputs. The syncpulse does not appear at the same time neither on my 7900 or my 8800... By changing profile under the nvidiadriver thus forcing a reboot of the gpu sometimes can cause better sync but it's all very random.
Unless you manage to get a perfect sync between the channels you definitely need something that stores left and right eye view images in a memory and alternatively "shutters" between them. This is of course expensive though.

So: Try to verify if you're able to get the vertical sync syncronized from both ouputs before continuing.

[iondrive]

Hi Likay,

I've read the word "genlock" in regards to this kind of thing and so googled it and learned something:

http://en.wikipedia.org/wiki/Genlock" onclick="window.open(this.href);return false;

Then I wondered which nvidia cards had a genlock feature and found:

http://www.nvidia.com/object/IO_10793.html" onclick="window.open(this.href);return false;

but I still want to just use the cards I have already.

The advice you gave me is good and although my XP system is still down, I feel pretty confident that I can H-sync the outputs there since I can do it on my test system. I had to fight with it a little but I still hope that I can find some good procedure to make both signals sync every time. I wonder if the nvidia driver programmers could do that. Usually it takes a few tweaks before the sigs are synced and in one case it needed a reboot to get synced but I think that once you've got it, it will still be synced in future boots. I'm sure some games could ruin the sync but they might not do that if they're set to the same res and rate as your desktop. Also cutscenes could be a problem too. I'll just have to try it and see. Obviously this software side of the project needs more work. As I said earlier, maybe some 3rd party software like Refreshlock will do the trick for me. It's a struggle but I'm still hopefull.

You know what's funny, the passive to active converters I've looked at still need genlocked signals. So these things that are thousands of dollars might be doing things cheaply after all. It's easy to miss in my previous post since I didn't shout it, but what I've already tried did work. I had my 3rd monitor alternating eye-views perfectly from the left and right views going to the other 2 monitors. I even used the shutterglasses with it to verify it. I just can't shout about it yet because there are so many loose ends. I have to get this working better with XP. Then I can say hurray.

Tech details: I've got an oscilloscope now, nothing fancy, not digital, but I can still get a good look at some of these signals and when the video signals are in sync, they are off by about 2-4 microseconds. That's H-sync, V-sync, and color data too. I tried the shift screen left/right function in nvidia's display control window but that only shifts the H-sync pulse and not the color data timing. That's why you need to switch in H-sync data to your 3d display too. Two microseconds are negligible for V-sync but not H-sync. Another solution would be to have some kind of adjustable delay line but I'm not keen on that idea. I think it's a little awkward for analog signals, easier for digital.

Anyway, thanks for posting Likay. It's nice to have someone respond.

Later.

PS: frame-locking is also interesting
http://www.nvidia.com/object/IO_10794.html" onclick="window.open(this.href);return false;

It kind of looks like Quadro cards have this. There's some kind of tweak that lets you use some Geforce cards as Quadro cards. Hmmm, I'll check that someday if I need to.

Update: the Quadro G-sync 2 is another card that you use with either a Quadro FX 4800 or 5800.

[mickeyjaw]

Software genlock for windows:

http://graphics.ethz.ch/~mwaschbu/winsgl/" onclick="window.open(this.href);return false;

It is designed to genlock separate PCs in a cluster, rather than two outputs on one graphics card, but it is open source so maybe it can be adapted?

[mickeyjaw]

Iondrive:

With respect to your post in the other thread, i believe i have figured a way to make this work on one gfx card. WinSGL, like SoftGenlock only handles the primary GFX card output. What you need to do is use an e-dimensional dongle on the secondary output. Put it in page-flip mode using the E-D activator, and run a cable from the VESA DIN back to the parallel port instead of connecting an external clock. This signal will let WinSGL know when the vertical retrace is occurring on the seconary output, which it otherwise knows nothing about. You can then use WinSGL in slave mode on the primary output, and it will sync the primary output to whatever the secondary is doing.

[iondrive]

ohhhhhh, nice!!!

I'll still be working on the hardware side of things until I get to a certain point, then I'll tackle the software side after I recover my XP system. This weekend was spent on mplayer. But I'm glad you got me doing that since it helped me retackle interlaced mode on windows with mplayer which I failed at before because I didn't know about the directx:noaccel option. That's what did it for me.

Question: can mplayer on Linux span two monitors? If so, then maybe this project can work under Linux for movies and other.

By the way, another reason I don't use Linux for 3d is because it's normally connected to an LCD monitor.
Also by the way, I use Gentoo.

bye now.

LINUX-HEADS UNITE!!!

[mickeyjaw]

Question: can mplayer on Linux span two monitors? If so, then maybe this project can work under Linux for movies and other.

Yes indeed it can, and the -xineramascreen option is the one to go for. I have 2 monitors,
-xineramascreen 0 -fstype none -fs gives me fullscreen on monitor 0,
-xineramascreen 1 -fstype none -fs gives fullscreen monitor 1,
-xineramascreen -2 -fstype none -fs gives me fullscreen across both monitors.

Yes that is right -2 for span, -1 is fullscreen on current screen (not that useful really).

For some reason it only works if the -xineramascreen is at the start of the filter chain before anything else, and only if you explicitly set the fstype option somwhere too, but it does indeed work. I guess you are thinking of using fil/tile to produce left/right dual outputs yes? If so, let me know how you get along. Also looking forward to your tests with softgenlock, keep up the good work

BTW, I use Ubuntu 99% of the time for what it's worth, with an XP partition for the odd gaming session.

P.S. Can you not just persuade IZ3D to produce an Over/Under driver with adjustable blank line insertion and use a sync doubler to get your shutter signal instead?

[iondrive]

Hi Mickey,

I asked iZ3D for gap control and I've read that it's in the new driver version but have not tested it yet. I would like that and it would be good with my CRT but I really want to use my DLP porjector that maxs out at 85 Hz so I want frame-sequential for that.

Yes, since mplayer can span, that means that people can use it for dual-projector passive polarization mode. I later realized that this project is not normally needed for linux since softgenlock can do the shuttering. It does it a little differently right? It spans across a double-wide virtual desktop on one monitor and then flips that, right?
I have some more ideas to wonder about with mplayer. Basically, can you mirror only one image in a side-by-side mode? If you know how to do that, please put it in the mplayer thread.

Do you think you can use softgenlock on Linux in the way you suggested for me in windows?

I'm so glad you're posting on this forum.

later.

[mickeyjaw]

I think it should still be possible to use a sync doubler with iz3d then, if you have a 1024x768@85hz PJ you would have to force a mode with 1024x(1536+number of blank lines)@42.5hz then turn on sync doubling and the PJ should sync to it as 1024x768@85hz no? Maybe better to use an x-force dongle if you can get one as i've heard they have a hardware switch to turn them on/off, or maybe you can set a hotkey combo. Point is the PJ wont sync to the 1024x1536@42.5hz resolution so you'll be blind until you turn the sync doubler on. You really should try this if you haven't yet.

With respect to softgenlock on linux, as well as syncronising multiple gfx cards like WinSGL, it can also provide a kind of software page flipping. You set the computer up such as you would have a double width virtual desktop eg 2048x768, with the screen res at 1024x768@85hz. You move the mouse off the edge of the screen, the screen pans. The gfx card stores a 2048x768 frame buffer, but only draws 1024x768 of it. The gfx card has a register set telling it where the viewport to draw starts and ends in memory. When you pan around, xorg just changes these registers, and softgenlock does the same. During each VBLANK, softgenlock twiddles these registers so that the left half is drawn on one frame and the right side on the next. You then just need to draw fullscreen left/right and softgenlock will pageflip it for you (in theory)

Softgenlock sucks, however, as it requires a 2.4 kernel with RTAI patches to work 100% reliably. There are three ways to compile it: realtime kernel module, normal kernel module and userspace program. I could only get the userspace program to compile on Ubuntu 8.04, the realtime module needs a 2.4 series with RTAI, the last kernel the normal module was known to compile on was 2.6.3. With a realtime kernel and module, the swapping thread can always be sure to wake up during the vertical retrace to move the viewport. With a userspace program, it still swaps left/right if your cpu hits 100% sometimes. Also it needs to run as root to get the direct hardware access, and sometimes crashes taking the system down with it. This is not good when I host various services on the same box.

Should still be possible to use softgenlock to sync the primary output to the secondary on the same card as well. AFAIK, keeping video timings in step is much easier than software page flipping as you only need to tweak the timings every few frames or so, and not exactly on the vblank. Tempted to test, but I wouldn't know how to check if my signals were perfectly in time or not?

[iondrive]

Hi Mick,

I was just thinking of the 42.5hz idea myself too. I have some X-force glasses/dongles but I'm not too crazy about that idea. I'm not sure if I can get windows to do that freq but I think I might try it someday when I'm curious. I would have to disable DDC and then maybe I might get that option but I'm not sure if I have a monitor that can go that low although I realize I don't technically need one. Yes, my projector can take 1024x768 input although native res is 800x600.

I reviewed the mplayer tile command and now understand your gap control technique. I'll post a more detailed explanation about it in the mplayer thread soon. I think there's a small chance it might have a downside with regards to A/V sync because of tfields but otherwise it looks less confusing than my expand usage.

I like the softgenlock idea for pageflipping a doublewide virtual desktop but too bad it has problems. I wonder if windows could do something like that but I won't be heading in that direction.


OK, new issue: How to check if your monitors are in sync:

Let's skip the "use an oscilloscope" answer since most people won't have that.

Using shutterglasses to check sync, you need these things:
2 dongles and glasses and an old CRT television showing a tv signal. You probably need two DVI-to-analog converters if your card has two DVI outputs instead of two analog outputs. Wired glasses might make things easier since you're using two pairs simultaneously.
EDIT: if you're using two CRTs, then you can just use one dongle and glasses and no TV. If you're using one CRT and one LCD, then the dongle must be on the LCD and the LCD must be the primary display.

Start clone mode and set all settings the same for each monitor. Choose 60Hz for starters to make things a little simpler and don't forget to set timing modes the same too: General Timing Formula (GTF), Discrete Monitor Timing (DMT), Coordinated Video Timing (CVT), or Fixed Aspect Ratio Timing (FART). I don't know yet if any of these are better than the others for sync purposes but anyway, you find that subwindow by right-clicking on the corresponding monitor in the nView window and selecting "adjust display". After that, use ED-Activator and both dongles will get triggered to start shuttering. This is why you use clone mode for starters instead of dual mode. In dual mode ED-Activator only activates the primary monitor's dongle although I suspect mickeyjaw might know a way to trigger the secondary one by using a GIF image. Now look at the TV through the glasses with each eye looking through a different pair. If they're in-sync, then each eye will see the same dark band on your tv or rarely, no band at all. That's all there is to it. You may ask, "why not use your CRT monitor instead of a TV?". The answer is because this way is more generic and will work even if one monitor is an LCD and the other is a DLP. If you're using two CRTs, then you can get away with just one dongle and glasses. Just get them shuttering and look at the other CRT. If it looks normal, then you're in sync. I guess I should have started with that description. oops. edit. Alright, that wan't too bad. Moving on...

Regarding the WinSGL input signal from video port 2, I would have to use the Vsync signal from pin 14 instead of the output of the 3-pin mini-DIN because that signal is 1/2 the freq of the refresh rate. I look at it as the right-eye shutterglass signal because it's 5V when the right-eye is black and 0V when the right-eye is clear. The wired signal is even weirder. I'll be starting a shutterglass signal thread soon to talk about the deatails of various shutterglass signals that I've learned about. I just don't know what to write about next.

bye

PS: I think if I make my own dongle for this project, I should include a circuit that lights an LED to indicate when the monitors are in sync. The convenience of such a thing would probably be worth the effort and I have an idea about how to do that. I think it should work but it might need some kind of fine tuning to work better.

[mickeyjaw]

Ah yes, the GIF images. I will post them here as the s***y wiki I use can't host attachments.
Edit: Junked didiwiki, installed mediawiki now so the GIFs are hosted there as well now see http://linux3d.magicbox.org.uk

Basically, all the E-D activator does is draw a series of coloured bars across the top of the screen. The dongle watches the top lines of each frame for these bars. All you have to do to set the mode on the dongle is ensure that the colour bars are drawn full screen width starting at row 1. I just open them in a image viewer and position the window by hand just off the top of the screen (title bar off top of screen so screen starts with colour bar). When I get round to it (don't hold your breath) I will write a simple pygame/sdl launcher app to draw a fullscreen window with the appropriate bars, launch mplayer with the right options, then draw the off bar again before quitting.

Hopefully, the filenames should be self explanatory...

vsyncdoublingon.gif

pageflipon.gif

interleavedreverseon.gif

interleavednormalon.gif

off.gif

Edit: DOH!!!! I think I have fried one of my dongles by pushing it too hard. I tried putting it into interleaved mode at 2048x1536@85hz, and may have also accidentally enabled sync doubling at 1280x1024@120hz instead of line interleaving. Now the picture jumps around all over the place and the monitor loses sync at anything above 1024x768@100. I will have to go rummaging in my boxes of junk for my other dongle, I know I have a brand new one still in the box somewhere as I just took the glasses and put the rest 'safely' away...

[iondrive]

Alright, thanks for the GIFs. I'm sure they'll work since I tried something myself. I used the "print-screen" key during ED-Activator usage and pasted into Paint and it worked to activate the dongle on the secondary screen in dual monitor mode so that's good. I'll be able to activate either one in dual mode now. Yay. Even under Linux. Awesome.

What do you think the "normal" max values are for the various modes?
I forgot what the max was that I've done but I'm pretty sure I've done interlaced at 1600x1200x60Hz.
For normal shuttering, I think it was 140Hz.

PS: if you have a newer ED dongle, then you could also include the GIFs for the 2 LCD modes but I'm not really asking you to add them since they are not important to me and I can get them myself if I need to. Just to inform: Ive tested these and LCD mode 1 is interlaced and LCD mode 2 is normal shutterglass mode. The difference between LCD mode and normal modes is that the shutterglass signal is delayed by 3/4 of one refresh cycle period. That's refresh cycle, not shutterglass lens cycle. I don't know how/why they came to that decision but it doesn't work good on my LCD monitor so it's kind of useless to me. It makes the lower half of the screen OK but the top half has much ghosting. I think they could have made a better decision and I will work on that as a side project since it's so closely related to my other projects. I have not started that forum topic yet. bye now.