Quote:
Originally Posted by
fluffysheapÂ
Thank you for your responses! I've looked at Marc Repnow's page, but I didn't find much useful information there, most of it is about the physical setup of the tachistoscope, or else is behind a paywall. I did see your thread on [H] but apparently did not see his, I'll look for it.
From what you've written, you seem to be aware of his research already -- you must have seen at least a few of the posts (StrobeMaster == Marc Repnow) because he's the one who mentioned that LightBoost does half a frame of buffering.
Quote:
The problem for me is that the existing Lightboost monitors have some horrible image quality.
Even if you cherrypick to an ASUS VG278H, bump its contrast ratio up slightly, and calibrate with an i1 Pro (to correct for gamma bleaching effect), and get approximately 750:1 contrast ratio? This is greatly improved from many stock out-of-the-box LightBoost picture. Your challenge is a homemade scanning backlight with perfect uniformity is far more challenging to get good image quality than hacking an existing LightBoost monitor. I think you can do it, but could be challenging!
Please take pictures of your progress; Blur Busters Blog would love to give publicity to your monitor hacking work!
Quote:
With an IPS/PVA panel it seems there are two choices. One, build a scanning backlight array, or two, use the existing edgelight (or a brighter version of it) with a low refresh rate and some huge blanking interval. Instead of 120Hz, perhaps 8ms update + 0.33ms blank, use 85Hz, 8ms update + 3.75ms blank/illuminate. This is not really much different from what the Lightboosts do now, by partially buffering the frame - effectively, they are building their own blanking interval into the monitor electronics.
That is a sensible approach. However, the
existing 120Hz IPS overclocks have a lot of streaking between refreshes -- that means you have pixel persistence that is creating motion blur that exceeds one refresh length. Due to greater than 10ms+ real-world GtG pixel persistence, there's never fully-refreshed frames to time a strobe backlight through. Even 85Hz will still be too high to hide much pixel persistence. I'd highly recommend doing a strobe backlight at only 60 Hz, but that is a lot of flicker. I am thinking more of 8ms update + 8ms blanking interval, or a two-pass refresh (8ms update in dark + 8ms repeat-refresh update (plus backlight strobe near the end of the second update)). The pixel persistence turd can only be polished so much.
Quote:
The trouble is that this would give a good image at the top of the screen, but a pretty bad one at the bottom, because the bottom will still be changing while the light is on. But you wouldn't have to change anything except to add a way to strobe the edgelight.
LightBoost monitors have this problem too; the bottom edge of the monitor have more trailing artifacts than the rest of the screen.
One thing that LightBoost monitors actually do is start the next refresh before turning off the strobe. Due to pixel persistence means the pixels haven't noticeably begun transitioning in the first few tenths-of-milliseconds of applying a voltage to an LCD pixel.
So for an ~85Hz mode, utilizing 8ms refreshes and nearly 4ms blanking interval (this is not enough time for IPS, but this is just an example only), and a 1ms strobe, your strobe timing cycle could be similar to this:
T+0ms -- top-down LCD refresh begins (pixels won't yet have strong visibility until maybe 1 to 2ms later)
T+1ms -- backlight turns on (refresh from previous refresh still displayed)
T+2ms -- backlight turns off
T+8ms -- top-down LCD refresh finishes; start of extended blanking interval (wait-out the pixel persistence)
T+11.7ms -- next refresh begins.
(NOTE: This is a hypothetical example only; to demonstrate creative phasing of strobe flash)
Yes, this is a compromise. Push the strobe timing into the refresh, rather than during the blanking interval.
LightBoost LCD's actually does a somewhat off-phase strobing roughly like this, although it's not as drastically delayed phase timing like the above, since it's a TN rather than an IPS. LightBoost turns on the backlight during the blanking interval, but turns off the backlight slightly after the next refresh has already begun. But for IPS, you'll need to push the phasing of your backlight strobe further. You can adjust the phasing of the timing of the backlight flash, so that the center 3/4ths of your LCD looks fairly clean, with some nasty (but forgivable) ghosting along the top and bottom edges of the screen. You simply adjust the phasing of the strobing, to occur earlier or later, until the clearest motion is along a centre band of your screen. There's a slow fade into a ghosting double-image along the vertical axis of the screen, like those seen at the very bottom edge and top edge of a LightBoost LCD.
I sincerely doubt that 3.7ms is enough time for most of IPS LCD pixel transitions to finish, but at least you'll have a small band that's relatively clear motion (it may only be about 1/3 to 1/2 screen height). Truly, TN pixel transitions are faster.
But you'll still have less motion blur than without a strobe; you'll just have an unavoidable trailing double-image effect (fainter along a center band across your screen, stronger at top/bottom edges of screen).
Your problem is finding an IPS LCD that meets these requirements:
(1) Overclockable IPS LCD; to allow you to do fast refreshes (refreshing the panel in 8ms), like QNIX Q2710, Catleap 2B, Overlord X270OC
(2) Electronics that accept an artificially large vertical blanking interval (e.g. 4ms blanking interval at 85Hz)
(3) LCD panel scanout occurs at the same rate as the signal (so you preserve large blanking interval intentionally injected into your signal via nVidia Custom Resolution Utility or ToastyX CRU)
(4) Minimum streaking in the IPS LCD. Let's assume 5ms IPS, with about 10ms real-world.
Personally I suspect it's easier to get better image quality on a TN panel with a modified LightBoost backlight.
- No overclock artifacts; many IPS overclocks start having artifacts/streaking/compromises when you push them to 120Hz
- Using a full RGB LED edgelight
Quote:
Going even further to 60Hz, 8ms update + 8.66ms blank (6ms settle + 2.66ms illumination), would allow the bottom of the screen time to settle before illuminating, but you would have only about a 15% duty cycle, and the existing edge light probably would not be up to it.
You could add a boost circuit to flash the backlight about 3x brighter. This is usually safe for most LED's. It will wear the LED's down faster but the tests have shown that not much lifetime is lost, if done carefully. LightBoost monitors actually do this; in the best models of monitors, there is a boost circuit that can be modified to allow 100cd/m2 during LightBoost=10% (which is dark 85% of the time) -- that's >600cd/m2 brightness if it was shining continuously. This is a 2x boost factor, since the monitor is normally rated at 350cd/m2 in non-LightBoost mode. See
CREE boosted LED pulse current driving specifications and warnings. You can push about 5x current briefly through LED's to get about 3.5x brightness. Conservatively, you can probably safely get about 2x-2.5x brightness without much LED lifetime loss.
Also, don't forget to play with out-of-phase strobing (e.g. pushing the timing of the strobing slightly into the start of the next refresh), so that you can push the center-band of motion clarity to the middle band of the screen, with the top-edge of the screen slightly ghosting into the next refresh, and bottom-edge slightly ghosting into the previous refresh, as a compromise.
Quote:
So it seems to me that the best approach really is a scanning backlight.
If you can do reasonable optics, you could go for it. I'd love to see you succeed. But with a home-made scanning backlight, you can open yourself to risk of uneven backlight and other problems that cause a backlight quality that looks much worse than a TN LightBoost LCD with a well-engineered edgelight. Consider this factor. Also, the ghosting during strobes at 85Hz (using 8ms+4ms cycle, with strobe pushed ~1-2ms into the next refresh) may actually still look superior to a scanning backlight, because of light diffusion within a scanning backlight. The best local dimming displays can only amplify the contrast ratio by approximately 10x (give or take) in real-world checkerboard contrast ratio measurements, which suggests backlight diffusion between on-segments and off-segments will be a major limiting factor. As a stylized example: 2000:1 LCD becomes 20,000:1 measured contrast ratio with local dimming. Let's say, a homemade backlight array might be twice as worse at controlling diffusion (e.g. 5x brightness difference between between on-segments and off-segments of a scanning backlight, for a blank-black screen). You might do better, you might do worse, but let's say the 10x becomes 5x with a homebrew approach. That means your blacks will be a 20% grey whenever any backlight is illuminated somewhere else on your screen. This will create a permanent double-image crosstalk effect throughout the whole screen. You will have to deal with both problems: double-image ghosting caused by scanning backlight diffusion, AND double-image ghosting caused by pixel persistence leaking between refreshes. If you do full-strobe backlight flashes, you only have to worry about double-image ghosting caused by pixel persistence leaking between refreshes. Even with the tight timings, the lack of backlight diffusion may actually win out! I have a big suspicion that you'll find better motion blur by using a strobe backlight rather than a scanning backlight, even for an IPS LCD. But maybe you want to test both approaches on two displays? Test to see which homebrew approaches win out?
Quote:
This is barely different from what scanning backlight TVs do now, except using a real high-speed input signal instead of frame interpolation or whatever other silliness they do on a TV.
Some of them actually have full strobe modes (e.g. Sony Motionflow Impulse) that flashes the whole panel all at once, rather than simply scanning it. The strobe method is also more 3D-glasses friendly since the shutter can open more briefly (to capture the strobe), reducing 3D crosstalk. (That's why nVidia 3D LightBoost was invented, but some HDTV's actually also use a similar technique).
Quote:
A manufacturer doing this could even do it with edge lighting, if they could find edge lights bright enough. I would probably be willing to pay $1000 for a large monitor that did this from the factory if the results were decent (which is about what it would cost to build it myself, anyway). Homebrew, it wouldn't necessarily have to be at full 120hz - maybe 85Hz using the big blanking interval with a four section backlight. But I bet you could manage the full 120Hz if you tried harder. If you had 8 sections instead of 4, and the panel refreshed every 8ms, and took 6ms to settle, you'd spend 1ms on the signal, 6ms settling, and have 1ms left for illumination at 12% duty cycle, which is enough.
If you can bump your budget to $1300, then one idea is modding a SEIKI 4K HDTV. It supports 120Hz native signal input (at 1920x1080), and it apparently has less motion blur than overclocked 120Hz IPS LCD's, which may make it a more successful panel to modify its existing edgelight to strobe. It also happens to be refresh-rate multisync, so you can drive it at 1080p@85Hz. Tests would be needed to see if it accepts artificially large blanking intervals. If so, then you could hack its existing edgelight to strobe. And you're getting a 4K display to boot (though you won't have a useful strobe mode during 4K, though!) -- but converting it to an essentially impulse-driven 1080p display that ends up having less motion blur than a plasma (but not as good as CRT).
Quote:
You have talked about needing complicated optics for this, but I am not sure. I bet just putting white or mirrored partitions in between the LED strips would be enough, only to reduce crosstalk between the strips. Fancy parabolic mirrors and whatnot will improve efficiency, I don't disagree, but I don't think it's necessary. There are plenty of TVs with LED array backlights and they don't have such things. Even if the partitions were black instead of white, they would probably not significantly impact the overall brightness of the screen if you only had three of them (for a four-section backlight).
Perhaps you can use straight angled mirror partitions between ribbon rows (like V-shaped ribs between LED ribbons), that might produce adequate efficiencies since you only need thin narrow strips of mirrors. But you will still be handicapped by backlight diffusion (As seen by limited real-world contrast ratio amplification for local-dimming backlights, suggesting you'll probably get the neighbourhood of 20% backlight diffusion).
Quote:
Maybe the right choice is to just wait and hope the manufacturers realize that every high-refresh monitor should have a strobed or scanning backlight. And it doesn't need to be NVidia exclusive - Lightboost is the whole end to end 3D setup with the emitter, glasses and all that. The VG248QE is selling like hotcakes and half the market has to go to huge contortions just to use it at all. Anybody can use a plain strobed or scanning backlight.
If the manufacturers did it, that would be nice. I definitely hope it does happen over the next year or two, it's worth milking maximum benefits of LCD while we're waiting for other better technologies to come out (whatever they may be -- OLED, or blue-phase LCD's, or currently unobtainium technology) as many LCD's are capable of much better motion clarity than they currently are, with the creative help of active backlight operation (strobing).
Quote:
Scanning backlights also cause less lag than strobed backlights. A strobe backlight adds half a frame of lag, a scanning backlight adds lag only equal to the settle time of the pixels, plus a rounding error. (Granted, on IPS/PVA the settle time is probably more than half a frame anyway, but on TN it would be less)
True, input lag can be a consideration. And, IPS settle time can actually be more than a full frame (e.g. majority of pixel transitions greater than the length of a refresh), even when measuring to just 90%-complete-transited level. But you've got a tough decision to make because you are going to be hamstrung by backlight diffusion within scanning backlights.
If you choose an overclockable IPS monitor, or SEIKI 4K, your choices seem to be (in order of simplicity):
-- Modify existing edgelight to strobe. (Best chance of retaining image quality)
-- Rip out existing edgelight & aim a superior edgelight where the old edgelight was. (To access extra brightness, better CRI, ability to replace LED's if worn out or blown out by over-boosting or experimentation accidents, etc.)
-- Create a scanning backlight. (High risks of not having as good uniformity as original edgelight/diffuser). They are also more inefficient than edgelights.
I have long advocated the scanning backlight, but I have come to appreciate the amazing engineering due diligence of existing strobe backlights, and have come to believe more in the strobe backlight approach than the scanning backlight approach. You just only need to tweak things to make the pixel persistence window wide enough (long enough blanking interval for pixel-settling), that the ghosting of a full-strobe backlight is reduced to be less than the ghosting otherwise caused by scanning backlight diffusion. A 5ms IPS LCD may be 10ms real-world, but the pixel transitions are actually often already more than 80% finished after 5ms. You weigh that remaining 20% grey caused by that, versus the 20% grey caused by scanning backlight diffusion. For a 11.7ms refresh cycle (8ms refresh and 3.7ms blanking interval), you theoretically have enough safety margin to have less than 20% ghosting (i.e. remnants of previous refresh) intensity along the middle 3/4ths band of your screen.
You must research and think through this carefully, versus the backlight diffusion. You may wish to test a LCD's potential for scanning backlight diffusion by lighting up a row of backlight LED's along one part of the screen (Displaying a black color), and doing light sensor measurement of the on-segment area and the off-segment area. You will find it virtually impossible to get a good contrast ratio. If you are lucky, you can be extremely efficient and manage to pull off just a 5% or 10% leakage, but when you try to do this, the LED rows start to become visible. Attempting to use diffusers to eliminate the seams of uneven lighting between LED rows, worsened the leakage between off-segments and on-segments, leading to the 20%. This is why I no longer really like scanning backlights as much as strobe backlights. Strobe backlights are actually unbounded in their efficiency -- once you manage to make the pixel transitions mostly complete or fully complete (e.g. remnant pixel persistence pushed to below human eye visible levels, like in some good LightBoost LCD's), motion blur can be as sharp as your ability to shorten the strobe flash. e.g. MPRT measurement of 0.1ms and even 0.01ms is possible on TN LCD's when using manageable refresh rates (e.g. 60Hz or 120Hz), assuming you used a strobe backlight sufficiently bright enough. In this situation, the MPRT exactly matches the strobe length -- that's how amazingly efficient a strobe backlight is, all you need is sufficient brightness in the flash to compensate for the briefness of the flash. Scanning backlights can't gain MPRT's matching the strobe length of a scanning backlight segment, due to the diffusion problem; you start hitting a wall. Your game is really, to find a way to open the pixel persistence window large enough that a strobe backlight becomes "good enough" to overcome the inefficiency of a scanning backlight. Then again, I might be wrong, and you might not be able to open the window large enough -- and you're stuck with having to do the complexity of a scanning backlight -- but I wouldn't give up on a strobe backlight yet. Yes, both of the two approaches have its risks.
Polishing turds can be so much fun, eh?
But LCD's are here to stay -- for a long time -- so we might as well have fun with improving motion blur via backlight control. Amazing improvements are achievable with modern strobe backlights on fast LCD's.
Edited by Mark Rejhon - 5/15/13 at 10:41pm
