How the hell does DLP-Link keep sync?

[cybereality]

There seems to be some confusion on how the DLP-Link glasses used on DLP 3D projectors actually work. I've asked people here and also on other forums (like AVS) and it seems nobody knows the answer. What is common knowledge is that the DLP display shows an encoded white flash on the screen in order to have the glasses switch eyes. So that part makes sense, in terms of the glasses keeping sync with the refresh rate of the display. However, it does not explain how the display knows which frame is for the right eye and which is for the left. This would obviously have to come from the content source (ie a PC). Previously this was done using the VESA DDC line on a VGA cable which worked great for legacy shutter-glasses. Another method was blue-line encoding, where the bottom scanline of each image had a blue-line (different size for left or right image) and thus the display (or dongle) could tell which eye was which. However, it does not appear that DLP-Link is using any of these legacy methods. So whats the deal here? How does it work?

I would be very interested to hear from any driver developers (like the iz3D guys) that have played with the hardware and understand how it works. Also, if there were any engineers from TI (ideally, since its their spec) or from the various manufacturers who have implemented DLP-Link in their products (ie Viewsonic, Optoma, etc.) I would love to hear from them too. There has got to be a simple answer to this, but it seems like no one knows what is going on. This is the exact reason why stereo 3d is so daunting to mainstream users and why it remains a niche product. You practically need a PhD in Computer Science just to get your 3D equipment to work. I've been in to 3D for years and I am on this forum religiously and every single day someone makes a post or asks a question that I have to do some serious research just to try to answer. And this DLP-Link sync question has me stumped. It really shouldn't have to be this complicated. Someone please help.

[tritosine5G]

I hate this story...

So , DLP projectors have something called the color wheel. This generates the colors themselves. In theory you should have R G B colors . But the clever techies at TI tought out something else, to maximise the lumens, lets use white segment as well.

So the RGBW color sequential "classroom" DLP was born with highly competitive features (lumen).

Now this is history , we have RGBWY ( red , green, blue, white , yellow) colorwheels, this is called brilliantcolor™ now, and supposed to give better shades. At 60hz, 2500+ contrast 2000+ lumen , all is beatiful.

Now enter 120hz 3d: the nice RGBWY color sequence is no longer in working order, WY is sacrificed. We only have RGB , hence 1/3 the lumen output.

What happens to WY ( white yellow?) sequences. Easy, dimmed. In the IR case, both are dimmed.

Now comes DLP Link: if DLP link is engaged, the white segment is used to send a flash thru the screen, and the optical detectors in the glasses pick that white flash up. You dont see it because the shutter dim's the flash.

So its a sad loss because brilliantcolor is no longer in use.

[Fredz]

If that's so hard to find an answer about how the sources are synchronised to DLP projectors, I think it's simply because there is no synchronization at all. The fact that some DLP projectors include a "3D Sync Invert" option is a good argument for this theory.

As you said, the only way to have synchronisation would be to do some sort of signaling in the source, be it in the images themselves (BLC/WLC) or in the hardware responsible for the output (VGA DDC).

Maybe there have been some agreements between graphics chips manufacturers (VGA, DVI, HDMI)/graphics card manufacturers (NVIDIA, AMD/ATI, Intel, etc.) and TI to use a convention for the order of the images (first image in 120Hz is always the left one for example) but I seriously doubt it. We would have heard about that after all this time.

IMO, if TI is so secretive about its technology it's because it does have flaws, namely the fact that they don't have any solution to the source/display synchronisation.

I'd still be curious to hear anyone experience about sync loss (or lack thereof) using a DLP projector and only DLP link glasses (ie. no 3D Vision at all). But I'm afraid there are not a lot of people in this situation, hence the reason why we never hear about these kinds of problems.

[taz291819]

If you think about it, it really doesn't have anything to do with DLP-Link. The display doesn't know which frame is first, hence the reason a "reverse" function is included in the display (and in some glasses).

This same problem is inherit to IR-based shutter glasses also, from my testing anyway. DLP-Link (like IR-based emitters) is simply used to sync the display to the glasses.

The crazy part is, after reading on AVSForum, is that owners of DLP FP's experience the "swap" problem when going from full screen to windowed mode, it's basically a 50-50 crapshoot. My DLP RP doesn't experience that problem at all. I don't understand how they would act differently, since they're using the same chip.

[cybereality]

@tritosine: Once again you completely ignore the topic of discussion and go off on some weird tangent. Not sure what to tell you, bro.

The crazy part is, after reading on AVSForum, is that owners of DLP FP's experience the "swap" problem when going from full screen to windowed mode, it's basically a 50-50 crapshoot. My DLP RP doesn't experience that problem at all. I don't understand how they would act differently, since they're using the same chip.

Well DLP RP doesn't have the problem because it uses checkerboard as a transmission format. So the left/right images are encoded directly into the format (ie the odd pixels on the odd rows are always the left eye, even pixels on even rows are the right eye, etc.). Its pretty easy then for the TV to extract both images and send a sync signal to the VESA port. If you are using a FP which takes a paged-flipped input then it would all depend on the exact millisecond you started the game, or hit play in the Stereoscopic Player. Unless there is some super-secret encoding going on that nobody knows about. Are we going to have to wait like 5 years for an ex-TI engineer to release the specification on Wikileaks?

[tritosine5G]

arghhh

i told you exactly how it works.
Its a white flash embedded between frames, when the white flash occurs, every viewers glasses see the same bright white light all across the xpand cinema, the photo detector picks the signal up (or @ your home, won't matter) This won't directly drive the glasses, what drives the glasses is a PLL circuit I presume so. L->R swap won't happen, and you won't see the white flash either because the glasses filter it , clever huh.
It isnt hard but looks like i have all pieces of the puzzle and you you guys dont have many alltogether

dlp link is for ultra fast displays

oh Im busy because my novint falcon works with borderlands as well as mass effect2, also my silverscreen turned out to be good, wanna use my quad esl57 now as well, I wanna vacuum-ize (torus-ize) this silver fabric to see how it looks etc.

Damn.

[Fredz]

@tritosine: Once again you completely ignore the topic of discussion and go off on some weird tangent. Not sure what to tell you, bro.

This is music to my ears...

Unless there is some super-secret encoding going on that nobody knows about. Are we going to have to wait like 5 years for an ex-TI engineer to release the specification on Wikileaks?

Reading the compatible third-party projector list on the Optoma 3D-XL site, they mention that with the Vivitek projector the 3D must be enable/disabled till the sync is correct because it doesn't have a 3D resync OSD option. This seem to confirm that there is no solution to this problem without using a 3D resync button. We'll have to wait for its availability in January to have some more informations with other projectors though.

[cybereality]

i told you exactly how it works.
Its a white flash embedded between frames, when the white flash occurs, every viewers glasses see the same bright white light all across the xpand cinema

Yes, I know about that. That only explains how the glasses shutter rate syncs with the display refresh rate. It doesn't explain how the display knows which frame is for the left eye and which frame is for the right eye. That is the piece of the puzzle that is missing. The display could just as easily be outputting pseudo-stereo, perfectly synced with the glasses via the white flashes, but your eyes would be seeing the wrong thing.

Reading the compatible third-party projector list on the Optoma 3D-XL site, they mention that with the Vivitek projector the 3D must be enable/disabled till the sync is correct because it doesn't have a 3D resync OSD option. This seem to confirm that there is no solution to this problem without using a 3D resync button. We'll have to wait for its availability in January to have some more informations with other projectors though.

Yeah, I saw that too. It does imply that maybe the other projectors on the list need to be manually synced via the "invert 3d" option. However, it is also possible that Vivitek just didn't want to pay TI however many thousands of dollars it costs to get their super-above-top-secret DLP-Link spec. And this wouldn't be a first. For example, Viewsonic released "Nvidia 3D Vision compatible" projectors before actually getting certified. Its possible those other projectors have something implemented we don't know about.

Also, this is just about the most concrete info I have seen on DLP-Link. And its coming straight from a very respected member of the stereoscopic community: Peter Wimmer, maker of the Stereoscopic Player. He seems to have knowledge that DLP-Link is a hoax:

Here is what I know about DLP-Link:

* There is no L/R signal on the VGA/DVI cable.
* There is a L/R signal between the projector and the glasses.

This means, if a projector would support blue line code or a similar technology,
it could provide the glasses with correct L/R information, assuming the
application sends such codes. To my knowledge, there are no such projectors yet,
so you have 50% change that L/R is reversed. In the Stereoscopic Player you can
fix this by pressing the F7 key (swaps L/R).

Btw, if you cannot find the Blue Line Code option in the Stereoscopic Player:
The player doesn't support it anymore because of patent issues.

Peter

http://movies.groups.yahoo.com/group/3dtv/message/26855" onclick="window.open(this.href);return false;

Really, if this is true, I would consider TI and other CEM's promotion of "3D Ready" front projectors to be fraudulent. In fact, I think it could be the basis for a class-action lawsuit. Who's with me? I am not trying to get rich here. I am willing to settle out of court for a life-time supply of 3D equipment.

[cybereality]

Just found this topic on AVS, someone actually got a reply from Optoma confiming the issue:

Hello Mathew,



You are correct that the sync needs to be corrected if the content changes (either starting, stopping, or paused). There is no fix for this other than making the correct menu option when this happens.



Best regards,



Joshua Allen

Optoma Technology, Inc.

3178 Laurelview Court

Fremont CA 94538

Technical Support Rep/Customer Service

Email: JoshuaA@optoma.com

PH: 510.897.8568

Fax: 510.897.8601



From: mathew_orman@yahoo.com [mailto:mathew_orman@yahoo.com]
Sent: Tuesday, December 07, 2010 1:34 PM
To: Optoma - Customer Services
Subject: Optoma Technical Support

Dear Customer,

Your inquiry has been received. Our customer representative will reply within 24-48 hours. If your question is more urgent or you would like to follow-up, please call us at 888-887-5001 during normal weekday business hours.

Best Regards,
Optoma Support Team
------------------------------------------------
E-mail:mathew_orman@yahoo.com
First Name:Mathew
Last Name:Orman
Phone Number:
Address: xx xxxx Drive
City: Nashua
State:NH
Zip:030xx
Country: USA
Product Model NumberLP-Link projector
Product Serial Number:none
Purchased Date:none
Problems:The DLP-Link has no sync with 3D content and one must manually set the correct eye order every time movie starts, movie is paused or if there is a glitch. What is the problem and how can I correct it. Please advice. Sincerely, Mathew Orman

http://www.avsforum.com/avs-vb/showthread.php?t=1296612" onclick="window.open(this.href);return false;



This is also supported by this YouTube comment by Optoma UK:

Regarding DLP link glasses, you will still need to use the projector’s sync-invert option to “optimise” your 3D experience
OptomaEMEA 13 hours ago

http://www.youtube.com/watch?v=5jXQyq9PaqY" onclick="window.open(this.href);return false;

[Likay]

I'm just inflicking: To keep sync between the channels only a white flash during one of the views (just for example the right image) is needed. In the glasses the necessary frequencydoubling can be done with a pll-circuit+2bit counter (in electronics a very simple circuit). This is how i always assumed how dlp-link worked and the principle will keep a perfect sync and id of the channels.
If it's done this way and projectors/glasses still swaps the eyes it must be that the manufacturers haven't done their lessons...
It would be stupid doing 3d-hardware without ability to keep the id of left and right...

[Fredz]

A stereo driver developer should be able to count the frames when using a stereo mode. Using this information he could choose to always send say the left image first. Maybe the DLP-Link specs say something about the image orders so that the source and the glasses are always in sync.

Just suppositions on my part at this time, I would need to do some tests with a DLP projector to test this since the software part (page flipping + frame counting) is already feasible with the appropriate software (namely Genlock on Linux).

[Likay]

Yeah. Software pageflipping is something else and so far only nvidia seems really successful. One obvious part of it s that they can "cheat" access to all "lowlevel" data needed.
Iz3d had problems with software pageflipping ever since they started to support shutters and have difficulties with eyeviewswapping. They did implement some features for sync like bluelinecoding and in earlier drivers even a totally configurable id for the line (white, blue, specific patterns whatever...) but not too many hardwares have support for such.

[Fredz]

I wasn't talking about software page-flipping but about true hardware page-flipping, by swapping the address of the CRTC without having to do a copy of buffers in video memory.

It's quite easy to do on Linux with NVIDIA graphic cards without having access to their driver. I really don't understand why everybody say that NVIDIA has an advantage in this case, I've just been using hardware page-flipping that has been known since the MS-Dos days through VGA standard registers programming. I guess it could work on other VGA compatible graphic cards too, like AMD/ATI or Intel ones.

I've already posted some code to do this for any resolution on Linux (see Compiling and using Genlock on Linux). It would be quite easy to create a counter that is incremented at each vertical retrace to send the left image for even or odd values of the counter in order to keep the sync with the DLP projector.

The only difficulty is to adapt the code to Windows, which isn't easy because their driver model is atrocious to say the least compared to Linux. I'm running Windows at this moment so I'll eventually try to adapt the code for this platform using WinSGL as a base.

The same is true for DDC signals, although they're not using standard registers in this case. The guy behind NVclock managed to identify and use the registers involved for EDID and I've already used that code to pilot DDC glasses under Linux with 7XXX and 8XXX NVIDIA graphics card : DDC glasses activation for NVIDIA graphics cards under Linux.

[Likay]

I wasn't talking about software page-flipping but about true hardware page-flipping

My mistake.

[tritosine5G]

Now I experimented too, lets try 100hz RGBWY on H5360 plus IR.

Occasional eye swap, ugly nvidia stereo scaling, otherwise it works.

Im into lawsuits, petition pages, because who would want dark 120hz RGB when you can have flickerfree bright 100hz RGBWY.

I also need better glasses or a polarization modulator for 100hz.

[cybereality]

A stereo driver developer should be able to count the frames when using a stereo mode. Using this information he could choose to always send say the left image first. Maybe the DLP-Link specs say something about the image orders so that the source and the glasses are always in sync.

Yes, if the driver developer could do this the content would at least stay in sync with the display refresh, however it doesn't guarantee the glasses are in phase with the content. That would require some constraint as to the image order: ie left image is always odd frames, for example. And, correct me if I'm wrong, but does DVI/HDMI even have inherent frame numbering? When does it start counting from? When you turn on the computer? When you turn on the display?

[tritosine5G]

In my 100hz experiment , I was able to configure VSYNC totally.

This means I edited the game's config file to give me smoothed VSYNC : 50 or 100hz.
(smooth to either 51 or 101 which is available)
Then I loaded a small map, where I had 100hz easily. Smooth. Then I tried a fight scene 50/100 vsync swapping worked perfectly. Only that goddamn nvidia scaling and nvidia glasses between me, and huge luminance.

Eye swapping happened between opening 2 aplications, image viewer. But not ingame. Im going AMD gfx for sure.

[Fredz]

Yes, if the driver developer could do this the content would at least stay in sync with the display refresh, however it doesn't guarantee the glasses are in phase with the content. That would require some constraint as to the image order: ie left image is always odd frames, for example.

In fact with the technique used in Genlock it's quite simple to do since the two images are in the same buffer, ie. a virtual screen twice the width of the display resolution. So it's just a matter of choosing the right address for the CRTC register (0 or WIDTH) for odd or even vertical retraces.

It remains to be confirmed if this technique (virtual screen) can be easily implemented on MS Windows systems though. I've written a DirectX test app to do that but I can't confirm if it does really work before being able to access CRTC address registers (which need to be in a device driver). I managed to compile WinSGL some minutes ago, so I'll probably be able to test this soon.

And, correct me if I'm wrong, but does DVI/HDMI even have inherent frame numbering?

There is probably no inherent frame numbering with VGA either, although there is a counter available in the NVIDIA /proc entry on Linux that seem to correspond to the number of vertical retraces since the machine was switched on. But as I said adding a counter for each vertical refresh in the driver could be used to do just that, without using proprietary drivers specific informations.

I don't think this vertical retrace detection has anything to do with the type of connector either (be it VGA, DVI or HDMI), else you wouldn't be able to sync to vblank on anything different than a display connected on the VGA port.

When does it start counting from? When you turn on the computer? When you turn on the display?

The counting would start each time a 120 Hz mode is selected, but for this to work it should be confirmed that the DLP-Link flashes are always sent in the same order when the projector is switched to a 120 Hz display mode.

If this is the case but the order is different for each projector, an option could be used in the driver to specify the images order for a specific projector. That would only be needed when the driver or the projector is first installed, using for example a stereo test application displaying the LEFT and RIGHT words on the screen and asking the user if the result is correct.

This technique should prevent any eye inversion during a whole session (ie. computer on -> computer off). This is feasible since the driver could keep the counter updated even if the stereo is not enabled but the 120 Hz mode is still active.

To handle missed vertical refreshes a second counter could be added to the driver to measure the duration between two consecutive frames. If the duration is greater than one frame (ie. > 8.33ms for 120Hz), the ordering could be corrected accordingly (number of missed frames = duration modulo 8.33).

The extinction of the projector should be handled though as you pointed, but I've really no idea about how to do that for now.

[cybereality]

Ok, so lets just assume that DLP projectors always show the left image first (ie on odd vertical retrace) and the right image on even retraces. This count would start from when the computer first initialized the 120Hz mode. Here is the question though: when a full-screen DirectX application loads, does this transition to full-screen re-initialize the display's mode? Such that the 3D driver (aka Nvidia driver) could make sure it started counting from when the game started (and thus when the 3D driver took control of the display output). In this case, 3d games would always be in sync. 3D Blu-Ray applications also use the Nvidia driver so they would have the same access to do this. Not sure if any of that is actually happening, but who knows?

[Fredz]

Here is the question though: when a full-screen DirectX application loads, does this transition to full-screen re-initialize the display's mode? Such that the 3D driver (aka Nvidia driver) could make sure it started counting from when the game started (and thus when the 3D driver took control of the display output).

The graphics mode changes must be intercepted in order to reset the counter, that could be done by using a resident application which intercepts DirectX calls or by using a library interposer (that's what VRizer does on Linux for example).

Its task would be to detect when an application requests a full screen mode, and then modify the frequency to 120Hz (or another value depending on defined parameters) and reset the frame counter. It should also be used to detect hotkeys for stereo parameters modification and to identify the calling application to apply appropriate default settings from the registry or the configuration files.

The only thing that's not clear is if the DLP-Link mechanism is reseted in this case like I said.

3D Blu-Ray applications also use the Nvidia driver so they would have the same access to do this. Not sure if any of that is actually happening, but who knows?

I think it would work the same way for any fullscreen application, be it a game, a video player or anything else as long as a full screen mode is requested.

[Fredz]

Yeah. Software pageflipping is something else and so far only nvidia seems really successful. One obvious part of it s that they can "cheat" access to all "lowlevel" data needed.

I just tested page-flipping on Windows XP with a GeForce 7600GT and it does indeed work, just like in Linux.

If you want to test it, you'll need the UserPort application from Thomas Franzon (I was too lazy to implement a kernel device driver myself) and this little one from me. You first need to copy the UserPort.sys file in the C:\WINDOWS\system32\drivers, launch the UserPort.exe application and click on the Start button. Then you can run pageflip.exe, enter a display width and click the Flip ! button.

If you use a virtual screen you'll see its hidden part, but you can still test it in a non-virtual screen by entering a lower value than your screen width. You'd better not move the mouse after the first flip because the mouse cursor doesn't move like the screen so it's going to be quite difficult to come back to the previous state in this case.

Since this page-flipping seems to work correctly on Windows, I'm even more confident that it can be automatically done on each vertical retrace. But it's going to be a little bit more complicated to implement I'm afraid. I'll see...

[cybereality]

Found this email response from TI posted over at AVS:

Thank you for your interest in DLP(r) Link.

You are correct that DLP 3D Ready projectors do not support the DLP HDTV
checkerboard input as the display devices used in our projectors and our
HDTVs are made differently. The DLP 3D Ready projectors do, however,
support 120 Hz or 60 Hz inputs as well as 120 Hz output frequencies.

The native format for the DLP 3D Ready projectors is frame sequential 3D
(as opposed to checkerboard, side-by-side, or over/under). The benefit
of using this format is that there is no compression of the image as in
the other 3D formats, but the drawback is that the projector must know
how to decode which frame is used for the left eye and which frame is
used for the right eye. There are many proprietary encoding and decoding
formats that will allow the frame sequential source material to insert
metadata into the content to mark each frame and have the display device
decode it correctly, but unfortunately, there is no mainstream standard
of how to do this with current 3D content. Current DLP 3D Ready
projectors, therefore, do not automatically support this type of
automatic left/right detection. Most, if not all, DLP 3D Ready
projectors have a menu control to invert the L/R sequence if a pseudo
stereoscopic condition exists. The same control also exists on most 3D
software and 3D compatible PC graphics cards.

In recent months, this 3D standard situation has changed with the
adoption of the HDMI v1.4a standard. This standard defines 7 different
3D formats that all display devices must support. Since it is a digital
interface, the content is in a fixed format and it is very easy for the
display device to know which frame is used for left eye and which is
used for right eye. Our more recent home theater 1080p DLP projectors
that support HDMI v1.4a (including BluRay) formats do not have the
problem of pseudo stereoscopic image display. As more and more of the
DLP 3D Ready projectors adopt support for the HDMI v1.4a standard, the
pseudo stereoscopic issue will no longer be an issue.

Sincerely,

DLP(r) Call Center
Texas Instruments
888-357-2984
http://www.dlp.com" onclick="window.open(this.href);return false;

http://www.avsforum.com/avs-vb/showthre ... st19687245" onclick="window.open(this.href);return false;

[Fredz]

Thanks for the info ! So none of the non-HDMI 1.4 DLP projectors can be synchronized to the source because of a lack of standard.

Now we've got the definitive answer on this subject at last, too bad it's not the one we were expecting...

[billqs]

Well that sucks! It's a complete copout from TI. TI could certainly have set a standard for left eye/right eye identification had they wanted to, after all they created the DLP Link (r) "standard" themselves. I'm glad newer projectors will be forced to the 1.4a HDMI standard, but that leaves the older projectors out in the cold.

Do we know how prevalent wrong eye assignage is with DLP Link?

[Fredz]

I guess it's a matter of 50% chance...

[snarfbot]

i always wondered why the projectors were so finicky with color settings and whatnot when in 3d-mode. hell i wondered why they had a "3d-mode" to begin with, it should just accept a 120hz signal and use it without any gimmicks. anyway that explains a great deal, mtbs3d should have an article that goes in depth about the various methods of 3d-displays, with pros/cons and the technological limitations.

anyway doesnt seem like such a big deal, flip the sync once in game, and your good to go after that right?

[billqs]

I don't think it's a deal breaker for me, but given we had several easy to implement solutions any of which TI could have used, it would have been nice if they had gone with one.

As long as you can invert and be good to go, it is probably ok... but nothing close to ideal.

[tritosine5G]

All I can say is , between 60hz 2D and 120hz 2D , the difference is less than the difference between 2 screens.

For example , 120hz white vs 120hz silver, silver 120hz looks, pops better ,than 60hz 2d, and its as bright.

Yet to try 120hz silver front projection vs. 120hz black retroreflective , I expect similar step up in quality white -> silver was, Its goin to get 2-6x brighter again, and ambient light rejection going to improve again.

An interesting thing to do with retroreflective and DLP is to mount the projector close to your head , inside a hush box.

Now this starts to get interesting: use 20mm wooden MDF to make the hush box. Make it somewhat large. Open up the projector, put the enclosure away. Look for the light dump/ heatsink. Replace it , and fans with something big.

That's it , contrast should be improved to ~ 5000:1 ( on -off contrast ), and the thing dead silent

[relaxman]

Hi
I have an Acer P1500 FullHD 3D projector, and use it with T/B and SBS footages. With DLP link glasses left-right order is always good. Always!
So the pj. knows left-right order of course (because the halved video format), but how knows the glasses the correct order?
DLP flash is visible only every 2nd frame? Or are there 2 separate flash, one shorter and one longer?
I like to know it

[cybereality]

@relaxman: I'm not 100% sure how the white flash is encoded, but I believe it's done in way that marks which eye should be open for each frame. My guess is that the flash length is slightly different for left and right eyes. The syncing between the projector display and the glasses was never a problem. Since the beginning, that has worked. The issue talked about in this thread is the syncing between the source material (like a PC) and the display. It's mostly a moot point now that HDMI 1.4a is popular and a common standard, so we have many ways to accept a 3D signal that actually work.

[relaxman]

Ok, i'm just interested in this part (pj and glasses)
thx