LCD motion blur: Eye-tracking now dominant cause of motion blur (not pixel persistence)

Hello,

Reading several academic scientific papers -- it is easy to come to the conclusion that vast majority of motion blur on _recent_ LCD panels is caused by eye tracking motion rather than pixel persistence.

On very new LCD's, most pixel persistence is now only a tiny fraction of a refresh. Motion blur on LCD displays occurs due to several factors, including pixel persistence and eye tracking. Recently, with today’s faster LCD’s, pixel persistence now only has a minor factor in motion blur. Pixel persistence ceases to be a dominating factor during the first few milliseconds (e.g. 2 milliseconds) of a single refresh (e.g. 16 milliseconds for 60Hz refresh). Although there's some residual persistence after the first few milliseconds, it is now mostly gone well before half a refresh cycle, and thus contributes far less than half of the motion blur found in today's modern LCD displays. In many new 3D 120Hz displays, nearly all the residual persistence is now gone between refreshes, since they must clearly show alternate frames for 3D active shutter glasses.

LCD's are sample-and-hold displays: They illuminate each refresh continuously & statically for the whole refresh. Continuous eye movement during tracking moving objects, causes the static image of each LCD refresh to be blurred across your retina, before the next refresh steps the image forward in the next frame. Your eyes are continuously moving, even during the middle of a continuously-illuminated LCD refresh. A human eye, even with saccades and imperfect tracking motion, is still in continuous movement when tracking moving objects on a LCD display. The sample-and-hold nature of LCD is observed can be seen in a high-speed camera video of LCD vs CRT refreshing). The image stepping effect, during each refresh, of a sample-and-hold display, produces a motion blur effect that's still noticed in fast motion even at high framerates. This is eye-tracking based motion blur, and is explained in several academic papers below. This motion blur factor persists even well beyond 120Hz, which is why even 120Hz LCD’s still have more motion blur than even a regular 60Hz CRT.

In summary: LCD's are sample-and-hold displays, so they illuminate each refresh continuously & statically for the whole refresh, while CRT flickers during their refresh.

Thus, the dominant factor of motion blur is now caused by eye tracking motion (for newer LCD's).

The distinction between eye-tracking based motion blur, and pixel persistence based motion blur, is explained in several scientific/academic papers, including:

Dynamic-Scanning Backlighting Makes LCD TV Come Alive
http://www.informationdisplay.org/issues/2005/10/art6/art6.htm

LCD motion-blur analysis, perception, and reduction using synchronized backlight flashing
http://adsabs.harvard.edu/abs/2006SPIE.6057..213F

Perceptually-motivated Real-time Temporal Upsampling of 3D Content for High-refresh-rate Displays
http://www.mpi-inf.mpg.de/resources/3DTemporalUpsampling/3DTemporalUpsampling.pdf
(See section 3, and Figure 1 -- it also covers non-interpolation related factors)

There are still many people who disagree about the existence of such a distinction, despite scientific proof, and manufacturers already know about the distinction.

[Edited to add] In many older LCD technologies, pixel persistence is the major motion blur barrier and cannot be bypassed. However, recently the scales have tipped -- this is no longer true anymore for modern LCD panels. Pixel persistence disadvantage of LCD (influences motion blur) is different from the continuous-backlight motion disadvantage of LCD (influences eye tracking motion blur). These are two separate disadvantageous traits of LCD from a motion quality perspective. It is important to understand the difference between the two. What is harder to understand is which factor dominates. Recent research has discovered that the dominating factor has tipped -- pixel persistence is no longer the dominant cause.

Comments?
Edited by Mark Rejhon - 2/22/13 at 11:53am

About pixel persistence now being a less dominant factor on modern LCD displays:

Here is a screenshot of pixel persistence during fast motion on a modern LCD computer monitor (Acer VG236H 120 Hz monitor, using prad's PixPerAn benchmark software). When run at 60 Hz, the panel is fast enough to nearly completely eliminate ghosting before the end of the refresh cycle. (Screenshot Credit: NCX, reviewing the monitor on WeLoveGamesToo forum)



In this image from NCX's review, observe the "Trace Free" modes -- this is a response-time-acceleration technology, allowing the vast majority of residual pixel persistence disappears before the next refresh begins. This is an older 120Hz computer monitor (from year 2010), but still has excellent reviews. Several newer 120 Hz displays are reported to do even better than this now (with less artifacts, such as coronas);

Ghosting now is largely gone before the start of the next refresh cycle. However, motion blur persists due to the store-and-hold nature of the LCD display, and continuous eye tracking movement while static refreshes "step" image movements forward. The ghost-free moment near the end of a refresh, before the next refresh, also happens to be the 'perfect' window of opportunity for a scanning backlight to strobe through, for CRT-quality motion on an LCD display.

Some of you already know that I'm currently working on a custom-made Arduino-driven scanning backlight for a LCD computer monitor, and this is one of the potential donor computer monitors I'll be modifying for my BlurBusters.com blog. (I've now posted a new FAQ about scanning backlights).
Edited by Mark Rejhon - 2/22/13 at 3:25pm

You are correct -- this is true for your display and other older LCD technologies. In many older LCD technologies, pixel persistence is the major motion blur barrier and cannot be bypassed. However, recently the scales have tipped -- this is no longer true anymore for modern LCD panels. Pixel persistence disadvantage of LCD (influences motion blur) is different from the continuous-backlight motion disadvantage of LCD (influences eye tracking motion blur). These are two separate disadvantageous traits of LCD from a motion quality perspective. It is important to understand the difference between the two. What is harder to understand is which factor dominates. Recent research has discovered that the dominating factor has tipped. Also, let me copy an answer to one of my questions in my FAQ from my Scanning Backlight FAQ.

Q: How can an LCD display get perfectly sharp motion like on a CRT?

CRT displays have long been known to have no visible motion blur. A scanning backlight can also be designed to simulate CRT scanning as closely as possible from a motion quality perspective. In order for this to be possible on LCD displays using a high-performance scanning backlight (without using motion interpolation), three main pre-requisites are necessary:

1. Pixel response complete by the end of the frame refresh cycle.
2. Very short strobes of backlight. Dark at least 90% of the time.
3. Very bright backlight to compensate for very short strobes.

Pixel response complete by the end of the frame refresh cycle.
Even if pixel response is only 2 milliseconds out of a refresh cycle (e.g. 1/60 second, which is about 16.67ms), there is still residual ghosting that often leaks over several refresh cycles. When this happens, pixel persistence is still a limiting factor for motion blur.


(Credit: DisplayMate Multimedia with Motion Bitmaps Edition)

Fortunately, some new LCD panels have finally become fast enough to eliminate noticeable ghost trails within one refresh cycle (about 16.67 milliseonds at 60 Hz).


(Credit: NCX's review of Asus VG236H monitor)

The emergence of 3D 120 Hz means that LCD panel manufacturers have an incentive to make them refresh more quickly, to allow alternate-frame shutter glass operation. LCD display makers have been working hard to clean up as much traces of pixel response (ghosting) as possible before the next refresh. As a result, minor ghosting on new LCD panels no longer leaks noticeably into the next refresh cycle. The scanning backlight can then be strobed during a ghost-free moment during the LCD refresh, once per refresh. Thus, pixel response is no longer the absolute barrier for motion blur reduction. (The next FAQ question explains this topic in greater detail.)

Very short strobes of backlight. Dark at least 90% of the time.
Very short strobes are needed, similar phosphor decay in a CRT display, as seen in this high-speed video of CRT scanning. Different CRT displays have different phosphor decay times. Common CRT computer monitors typically have approximately 1 to 2ms of phosphor decay. While phosphor illumination is near-instant, the phosphor brightness fades more gradually over the subsequent 1 to 2 millisecond time period.

As seen in this high-speed video of CRT scanning, about one-tenth of a CRT display is brightly illuminated at one time during a 1/60th second refresh. In order to do the same with an LCD display, a scanning backlight needs to brightly illuminate only one-tenth (or less) of the display at a time, which means backlight flashes lasting similar to the length of phosphor decay. That is between 1 to 2 milliseconds out of a 16.67ms refresh cycle, about 10% of a refresh cycle.

As a result, this requires a high-performance scanning backlight to be dark at least 90% of the time, in order to match the motion quality of a sharpness of a consumer CRT computer monitor. The shorter the illumination is, the better the motion quality is (Science & References).

Very bright backlight to compensate for very short strobes.
Very short strobes will normally result in a darker image. The backlight needs to be very bright (at least 10 times brighter than normal) to compensate for the very short strobes. As observed in the high-speed video of CRT scanning next to an LCD display, phosphors on CRT illuminate extremely brightly for a very short time period -- at least 10 times brighter than a typical LCD backlight.

Existing HDTV's that use scanning backlights, do not have a backlight 10 times brighter than normal. These scanning backlights do not have short strobe cycles, so they do not reduce motion blur as much as a CRT display. In addition, these displays often have a dimmer image when the scanning backlight mode is enabled. Several models also combine motion interpolation, which is unsuitable for computer and gaming use due to input lag. Finally, the extreme amount of extra brightness can be extremely expensive to engineer into a backlight.

Fortunately, the prices of LEDs have fallen dramatically. LEDs are well suited for scanning backlights due to their fast switching speed and brightness. LEDs are now available at costs less than 25 cents per watt at factory cost, and prices are continuing to fall. LEDs is also now available in mass-manufactured ribbon format, which may lead to cheap assembly and manufacture of ultra-bright backlights. It is now becoming increasingly possible to engineer an overkill of a backlight necessary for CRT-quality perfect motion on an LCD display, without a drastic increase in the cost of the display.

All three pre-requisites have now been largely solved for high-performance scanning backlights. This means it is now possible for LCD panels to have the same perfectly fluid motion that CRT displays have. The next move is to put the technology pieces together!

---

For other questions in my FAQ, there are more questions in my Scanning Backlight FAQ.
Edited by Mark Rejhon - 2/22/13 at 3:28pm

Another item from the Scanning Backlight FAQ is ...

Q: How do you bypass pixel persistence as a motion blur barrier? (with a scanning backlight)

A: Thanks to tests on new LCD panels, it has now recently become possible to bypass the pixel persistence barrier. In some displays, the vast majority of pixel persistence is less than a single frame of a refresh. This provides an excellent window of opportunity for a massive motion-blur reduction from a scanning backlight. It is possible to wait for pixels to finish refreshing, and then strobe the backlight after the pixels have largely finished refreshing, but before the next refresh. A single refresh at 60Hz takes about 16.67 milliseconds, and pixel persistence has now become far less than this.

Example LCD Refresh cycle
T+0ms - Refresh begins (unseen in dark)
T+2ms - Most of the refresh is finished (unseen in dark)
T+14ms - Residual ghosting is practically gone (unseen in dark)
T+15ms - Strobe backlight brightly for 0.5 milliseconds (seen by human eye)
T+16.67ms - Next refresh begins (unseen in dark)

The human eye sees the 0.5 millisecond strobe portion of a refresh that is visible, instead of the pixel persistence portion of the refresh that is now made invisible by a turned-off backlight. Thus, the human eye no longer sees motion blur caused by pixel persistence limitations, provided the display is able to finish refreshing before the next refresh cycle.

Photographic Example -- Click Here
Some displays use response-time acceleration technologies such as ASUS Trace Free, which helps eliminate perceptible ghosting completely before the next refresh. This image contains screenshots of an animated object from PixPerAn benchmark software provided from NCX's review of the Asus VG236H monitor on the WeCraveGames.com forum.

[image] -- Observe that higher Trace Free modes on the 120Hz Asus display, completely eliminate perceptible ghosting by the end of the current refresh cycle, before the next refresh cycle, for a specific area of the display. (Note: These are fast-shutter camera photographs of motion on an LCD display, so the images are noisy only because of the fast exposure setting.)

A scanning backlight can do a very quick, brief, bright "pulse of light" to illuminate a  fully-refreshed portion of the LCD. This now allows motion blur reduction to approach CRT quality for perfectly crystal-clear fast motion during video games.

---

More questions are answered in my Scanning Backlight FAQ.
Edited by Mark Rejhon - 2/22/13 at 3:29pm

If by "occlusion", you mean a synchronization issue between the scanning backlight and shutter glasses:

Occlusion would not cause blurring, but would cause parts of the picture to be invisible (e.g. shutter glasses open while the backlight is turned off). There are times when 3D shutter glasses are closed for both eyes at the same time; this is because it needs to wait for the LCD to refresh to the next frame for the other eye (prevent ghosting/crosstalk) before opening the other shutter in your 3D shutter glasses. Can you explain why occlusion from glasses would cause blurring? It is simply only a synchronization issue.

Occlusion would have nothing to do with presenting the wrong image to the wrong eye, so occlusion would not contribute motion blur, since the shutter glasses would already be calibrated to show only the correct eye image. It is simply adjusting the timing of the scanning backlight (or full-strobed backlight) to illuminate the whole backlight for one frame, while the shutter glasses is open for one eye.

As long as the shutter glasses run 'in sync' with the scanning backlight, black frame insertion, or other strobed technology, it works. Existing scanning backlights already do this (e.g. Elite LCD HDTV) already does this, supporting scanning backlight operations works with 3D. So, it's confirmed by these existing displays: occlusion do NOT cause motion blur.
Also, for light hitting the human eye -- certain kinds of high-performance scanning backlight designs can be made equivalent to CRT scanning -- and CRT's is compatible with 3D, and 3D does not add any motion blur for CRT.

For more information about how scanning backlights work, see Scanning Backlight FAQ and also Science & References.
Edited by Mark Rejhon - 2/22/13 at 3:30pm

To more accurately measure phosphor decay time using a camera, you need to use a fast shutter speed. The faster the better, but 1/1000sec will do. At some point, the amount of phoshors stops shrinking with each shrink of exposure duration -- now you're capturing the limits of phosphor decay because you're starting to capturing more illumination from actual decay rather than just illumination from the scanning in progress over a longer exposure. I've observed many medium-persistence CRT computer displays, it's approximately 1 to 2 milliseconds (illuminates near instantly with the electron gun, but fades out over a millisecond or two).

In answer to the two issues you brought up...

Measuring CRT Phosphor Decay using a Camera -- You can still quite approximate phosphor decay even if you're somewhat limited by exposure time, if you're careful to count scan lines. For example, an NTSC TV has 525 scan lines vertically for 1/30 second. That's 262.5 scan lines over 1/60 of a second for one interlaced "field" of a frame. Taking a photo of the television using a camera with 1/1000sec shutter (hopefully precision-calibrated in shutter accuracy), the television set only has time to "scan" a little over 15 scan lines (262 * 60 / 1000 = 15.75) (Incidentially, this number is familiar! An NTSC TV is 15.75 kilohertz in horizontal "scan rate"!) But in a camera photo taken with 1/1000sec, you'll see many more than 15 scan lines -- that's the phosphor decay. TV's vary in phosphor persistence and depending on color, but if you've got 1 millisecond phosphor decay, your 1/1000sec photograph of an NTSC TV will photograph about 30 illuminated scan lines. (15 scan lines from the actual scan, and 15 extra scan lines from phosphor decay). You need enough resolution in your photo to count the scanlines, then you can accurately calculate phosphor decay based on known shutter speed and subtracting the scanned scan lines from the phosphor decay scan lines. For a reasonable approximation, you can use illuminated screen height. One 1/60sec refresh is 16.67 miliseconds, so if about 1/10th of the screen height is bright in a 1/1000sec photograph, phosphor decay is certainly pretty close to 1.6 milliseconds (1/10th of a refresh). About 5% of the screen height would be the photo exposure time (subtracted), and subtract another 3 - 10% due to VSYNC, but this isn't a very significant change -- it's still solidly within "1 to 2 milliseconds" at least for the majority of the phosphor decay cycle. The high-speed video of CRT scanning was taken with a 1000fps camera, which means the shutter speed is no slower than 1/1000sec. In the slowest possible shutter speed for this specific high speed video, if it had zero phosphor decay (instant fade), then only 15 scan lines would be visible in this video (15 scanlines equals 5% the height of a single NTSC field refresh, because 15 is about 5% of 262). However, in this video, clearly more than 15 scanlines is visible -- in fact, there's CRT ghosting visible for the whole refresh. CRT phosphor decay isn't an exact science since it can take a long time for it to completely stop emitting photons -- but it stops emitting usable brightness after apporoximately 2 or so milliseconds, at for this specific YouTube video. So as you can see, you can measure CRT phosphor decay using mathematics based on your camera's shutter speed, and excluding the scanned scanlines (scans occuring while shutter was open) from the decaying scanlines (scans that occured before the camera shutter opened).

Flicker reduction for scanning backlight There's many ways to design a scanning backlight to have less objectionable flicker.
-- Scanning the backlight instead of strobing the whole backlight at once, means a more constant rate of light hitting your eyeballs, which is subconsciously more comfortable than sudden surges of photons. Also, there's another good extra reason the backlight is scanned -- LCD refresh is already done top-to-bottom, so you want to strobe the backlight top-to-bottom in phase with the LCD refresh, so now you're adding top-to-down flicker, and it just happens to be similar to scanning of a CRT, thus the name "scanning backlight". The scanning effect has a slightly higher flicker fusion threshold than the full-strobe effect.
-- Flicker can be reduced even further by running at 120Hz or 240Hz refresh (preferably native rather than interpolated), it's even less objectionable. For computer monitors, 75 or 85 Hz can work, and reduce graphics reqiurements (GPU) when matching framerate and refresh rate.
-- Phosphor decay can be emulated simply by adding small capacitors to each scanning backlight segment. This adds a fade-out cycle to the scanning backlight -- you've added a decay effect to the scanning backlight that mimics phosphor decay.
-- More segments also can help as well, but beyond a certain point, it makes little difference.
There are those people who are extremely sensitive to CRT flicker -- those people will be the ones most sensitive to various scanning backlight tweaks. For a many-segment scanning backlight with decay effect added to each segment, it starts to look more indistinguishable from a CRT under a high-speed camera; and the closer you get to the advantages (and disadvantages!) of a CRT. Then again, you try to get as many advantages as possible, with as few disadvantages as possible. (e.g. use a higher refresh rate to eliminate flicker). Fortunately, scanning backlights are easily put into regular backlight mode by the press of a button or feature.
Edited by Mark Rejhon - 10/16/12 at 9:23am

There's a flaw in your test: Moving eyes left and right does not simulate a strobing backlight. You'd need to turn off the room lights and use an xenon strobe light (1/1000sec flashes or shorter). You have to repeat the same test again, but in total darkness -- using a flashing xenon strobe lamp. You will observe sharp speakers (no blurry speakers) in all situations, no matter how fast you move your eyes in any direction, and even when waving hand in front of your eyes. This is exactly the same effect that reduces motion blur -- the short flashes and the long darkness between flashes -- this is a very important aspect of a scanning backlight. When waving your hand while the xenon strobe is flashing -- Sometimes your hand will go in front of your eyes while the xenon strobe flashes (meaning you see nothing), other times, the xenon strobe flash will occur when your face is not blocked by your hand (meaning you see a sharp speaker). That's the synchronization aspect I'm talking about. If you can precisely time your hand wave to go fast between the flashes, your hand won't block a flash from being seen by your eyes.

See? It's just a simple matter of synchronization -- between the timing of the strobes and the timing of the shutter glasses.
Also, it's already proven in existing HDTV scanning backlights, that 3D+occlusion, do NOT cause motion blur.
Therefore, scanning backlights are fully compatible with 3D shutter glasses, if properly synchronized.

Your occlusion test is invalid without a totally-dark room and an xenon strobe light. (Xenon strobe lamps are about $25 each at party stores, especially common at halloween time.). Also, it's preferable that a more scientific method is done -- there are several academic papers for useful reading here in Science & References.

--

Note: The scanning backlight design I'm creating is targeted towards computer monitors at the moment. The technology exists in expensive LCD HDTV's already -- There's 3D compatible scanning backlights commercially already in Elite LCD HDTV, and the Samsung's with "Clear Motion Rate 960" such as model UN55ES8000. For television material, it is important to note that the benefit of scanning backlights only really shine during live action panned fast (e.g. fast pans during hockey/football, red bull air races, nascar, ski racing, etc) -- the same kind of materials that CRT displays excel at -- and live-material 3D broadcasts of those material is not very common. (Scanning backlights do not generally benefit movies) You need to test with video material that show glaring motion blur deficiencies on LCD but otherwise looks perfectly free of motion blur on CRT. That's that kind of material -- fast motion sports and videogames -- which you really need to benchmark motion blur on (in order to see benefits beyond "120Hz equivalence"). My scanning backlight design is targeted at computer monitors, and videogaming at 60fps+ and beyond.
Edited by Mark Rejhon - 2/22/13 at 3:31pm

Another thing that's useful to know. 3D _active_ shutter glasses can contribute to eye-tracking-based motion blur. This is unrelated to scanning backlights and occlusion. With active shutter glasses, you're getting 120 samples per second, spaced 1/120sec apart - 60fps per each eye, shown during alternate 1/120th second pariods. The even samples go to one eye and the odd samples go to the other eye. But video sources are often 60 frames (or fields) per second rather than 120 frames per second.

For original capture of nearly all 3D video-based material, both the left/right eyes are captured simultaneously. But for alternate-frame shutter glasses, you're presenting the frames temporallly differently than they were captured. You are now temporally displacing the left eye frame versus right eye frame. This can produce a motion blur disadvantage caused by eye tracking. This is because you're now getting 120 samples per second (either left eye or the right eye), but linear motion (e.g. panning) doesn't step forward smoothly at 120 movements per second, but only 60 movements per second (tantamount to a frame repetition, except for a different eye). So you're getting "60fps judder on a 120Hz display" with video material with 3D active shutter glasses. Thus, you can perceive motion blur with active shutter glasses that cannot be fully overcome with a scanning backlight, because of this "frame repetition" effect. This may be the related issue being thought of here.

Going to 120fps solves this problem. This can be done using either motion interpolation, or via computer games (a computer outputting at native 120Hz, with a very fast GPU, does not have the "limited-to-60fps" video source availability problem). For active (alternate-frame) shutter glasses viewing, the behaviour of the display is tantamount to a full 120Hz display, requiring 120fps for fully fluid operation, to avoid motion blur caused by eye-tracking through a frame repetition (the blurry-effect on edges seen in frame-repetition judder, during pans and fast movements). Video games capable of running at 120fps (in full 3D mode) would be immune to this effect, being able to generate all samples necessary for perfectly sharp motion with active shutter glasses.

I've now updated my Scanning Backlight FAQ under the question
"Q: What type of video material benefit the most from a scanning backlight?"
Edited by Mark Rejhon - 2/22/13 at 3:31pm

There's no such thing as "the left and right frame don't fit together stereoscopically" from this context -- the human vision system doesn't work on a discrete-basis like that. "Persistence of vision" behaves differently -- for maximum "soap opera" effect (often desired in high-framerate materials such as video games and sports, but not always for movies or intentionally-low-framerate materials), it is better for the frame presentation to mimic real-life.

Allow me to explain why it is more proper/natural at 120fps
The easiest way to explain this: It's like real world -- if you wear 120 Hz shutter glasses in real life but watch real life (Example: sit in front of the TV and enable the shutter glasses. Then look away from the TV set to watch other real-life objects such as a person walking across the room near the TV). The person is moving continuously. Your shutter glasses is blocking one eye at a time. So you never have both of your eyes see the moving real-life object in exactly the same position at exactly the same instant in time. But your eyes are successfully able to see this, and the movement of real-life objects in the same room is still more fluid than movements on the TV.

Similiar effects can be seen when looking between slats of a tall picket fence from a moving vehicle or bicycle, etc -- the movement allows you to see more of a scene through the "slits" between the fence slats. Your left eye and right eye are combining the view through persistence of vision, and your left eye is getting different angle views at various different times than your right eye view, yet you're able to combine a scene, and if you're moving fast enough, you also get 3D depth perception as you whoosh by.

Eyes are always continuously tracking. Watching a moving object as the shutter switches from left to right, if the object didn't move (in 3D space in 3D mode), your human vision system gets the same effect of a repeated frame (Even though it's of a different angle), your left eye saw the same scene from one angle, then 1/120th of a second later, your right eye sees the same scene (of exact same instant) from a slightly offset angle. But your eyes are continuously tracking in 3D, even over a 1/120th second interval, and if there's no movement in that 1/120 second, then your vision system perceives the telltale judder of a frame repeat. That's exactly what happens.

So, to simulate real-life at the maximum fluidity with 3D shutter glasses, one could theoretically do:
- Capture left-eye frame at even 1/120th second intervals (T+0/120second, T+2/120, T+4/120, T+6/120)
- Capture right-eye frame at odd 1/120th second intervals (T+1/120second, T+3/120, T+5/120, T+7/120)
This would work properly only with alternate-frame shutter glasses (where each eye take turns), but would be unmatched (on time-basis) for polarized systems (where both eyes sees their image simultaneously). You wouldn't need to capture 120fps pairs, just capture at 120fps one frame at a time - left eye, right eye, left eye, right eye.
(Or render, of course -- as in 3D video games, but this is not done in actual practice since games always render both pairs for the same in-game instant.)
This would make it mimic the real-life scenario I explained more, because of the 1/120th second stepping.
Thus, getting the "soap opera effect".

(Extra thought: This is theoretically already compatible with existing 3D systems, you simply use existing workflows and 3D mastering workflows, except tweak the timing of the right-camera shutter forward by 1/120th of a second relative to the left eye, for 60fps material. So left eye running at 60fps at even-numbered 1/120th second intervals, and right eye running at 60fps at odd-numbered 1/120th second intervals, to keep capture time exactly relative to presentation time. (i.e. keep everything "perfectly aligned" in capture-scene-time versus shutter-presentation-time, to be consistent with real life). But of course, you run into problems when presenting on polarized systems, so this isn't something that is ever done in practice. The polarized system, in this situation, would have the unfortunate effect of simultanous presentation of left/right eye frames taken at different times. Just something that's possible to do.)

Another way is simply capture both left/right eye pairs at 120fps, and present only one eye of each frame pair. (e.g. left eye of one pair, then right eye of the next pair, and then left eye again of the subsequent pair, and so on). This is quite much more inefficient, obviously, since half of the frames are never presented. This would result in exactly the same "soap opera effect" as the above method explained. But this more-wasteful method would certainly maximize compatibility with multiple types of active and polarized systems (Both 60Hz polarized and theoretical future 120Hz-native polarized).

It's all not being done in actual practice by filmers and videographers (at this time, to my knowledge, due to the incompatibilities introduced to non-shutter-type 3D systems), except when doing motion interpolation, e.g. 120fps motion interpolation (supported by *some* 3D displays), then you're getting the 'real-life-smooth' feel of 120fps @ 120Hz, the soap-opera-effect, through 3D shutter glasses.

This technique is doable with PC based 3D too, games can always render both pairs of frames 120 times a second, even if only one of the two is ever seen by the human eye during high-framerate 120fps moments (until framerate slowdowns gives time for the other eye, of the same 3D frame pair, to be shown)

It's still 60fps per eye, but video captured for right eye's 60fps would have been photographed by the camera 1/120th second different relative to left eye. When you begin to think this way (in order to mimic real-life, like looking away from the TV set and watching a person walk across a room, while you're wearing 120Hz 3D glasses), it suddenly starts to make a hell lot more sense; though the knowledge about vision physics is not quite this fine-tuned by many people... If you're still confused, I will attempt to try to find some references, but one good reference is this one:
"Perceptually-motivated Real-time Temporal Upsampling of 3D Content for High-refresh-rate Displays"
It may not explain all bases, but it certainly scientifically explains a lot of this.
Edited by Mark Rejhon - 10/20/12 at 6:14pm

I assume you are you asking "Is eye-tracking the dominant reason of motion blur on plasma?". That question is not as obvious as for LCD (it's sometimes yes, and sometimes no. Cheap plasmas that distribute subfield refreshes willy-nilly throughout the entire normal image refresh, behave essentially as a sample-and-hold display to the human eyes, and thus would have the same motion blur. The plasmas that have more motion blur than others -- THAT additional motion blur on the lower-quality plasmas, is nearly entirely caused by eye-tracking. The only way to reduce motion blur for the same material on a plasma, is either through interpolation (rarely used) or through clustering the subfield refreshes to the point where the plasma behaves more like an impulse display.

See Panasonic Germany's own YouTube about the 2500Hz FFD drive (Europe) in their Panasonic VT50:
Look at the diagram -- that the Panasonic employee is pointing at, they cluster their subfield refreshes into a timespan of only 0.4 milliseconds -- (1/2500second) - this produces an equivalence to 2500Hz. That makes this plasma behave more like an impulse driven display. [edit -- corrected an error]

Cheaper plasmas don't cluster subfield refreshes the way Panasonic says they do with their VT50, and thus those plasmas will have more motion blur. Phosphor persistence is an insignificant cause of motion blur (just like CRT). Plasmas often don't have motion interpolation like LCD does. So, when comparing plasmas that have no motion interpolation, then -- 100% of the perceiveable motion blur differences on plasma displays (that have no motion interpolation), is caused by eye-tracking based motion blur. So you compare two plasmas -- neither has interpolation -- and one has more motion blur than the other? That's eye-tracking based motion blur.

(P.S. My scanning backlight aims to do exactly the same objective -- queeze the illumination into one similarly short time period.)
Eye-tracking-based motion blur affects most displays unless they have ultra-short single strobes per refresh (e.g. 1ms, like CRT) to the point where the fastest motion the human eye can track (on full framerate material with sharp frames) has no noticeable additional motion blur.

Regarding eye-tracking motion blur:
3LCD -- same consideration as LCD. If pixel response time is less than half a refresh, then eye-tracking blur is dominant.
LCoS -- same consideration as LCD. If pixel response time is less than half a refresh, then eye-tracking blur is dominant.
DLP -- depends. Can be as bad as LCD, but there's plenty of opportunity for impulse control (firmware creativity)
Plasmas -- depends. Can be as bad as LCD or as good as CRT

DLP, projectors, plasmas, etc, are affected to lesser or greater extent by eye-tracking-based motion blur, depending on whether they're sample-and-hold displays versus impulse-driven displays, and how short the impulses are. The shorter the impulse, the less eye-tracking-based motion blur there is.

Also your question was changed to "Is eye-tracking part of the cause of motion blur on any other display?" -- then the answer is an unequivocial YES for nearly all displays. Only displays that have strobes that are too fast to allow eye-tracking motion blur to be perceived by human eyes. (e.g. like a 1/2,500sec photograph of fast-action sports -- at 1/2500sec, you typically totally freeze all motion without motion blur.)
Edited by Mark Rejhon - 11/11/12 at 1:50pm

Yes, DLP is actually impulse driven BUT at the pixel level, and at very high frequencies. So you're getting many impulses per refresh. The entire screen is never completely black at any time (no black frame effect), and there's no sequential strobing done (no CRT style effect). The distributed impulses at hundreds of Hz are so rapid, and so distributed throughout a refresh, with no regards to motion blur reduction -- so it causes DLP to average out to a sample-and-hold display.

However, you can design DLP to compress the impulses into one cluster of impulses per refresh, to reduce motion blur on a DLP. Some microdisplays [corrected] such as Sony SXRD and Christie DLP, have a black frame insertion feature, to clulster the pixel impulses so that there's enough black frame between the impulses. That turns the DLP more into a impulse-driven display from a human motion-blur perception perspective.

So just like plasma, you get more motion blur on some DLP's and less motion blur on other DLP's.
Edited by Mark Rejhon - 11/12/12 at 5:16pm

Yep.

Flicker reduction is why several high end LED-backlit LCD displays combine interpolation (to 240Hz), combined with intentionally timed backlight impulses -- to get the rest of the way to their respective "960" ratings (e.g. Samsung CMR 960, Sony Motionflow XR 960). That said, there's more complexities in turning an LCD into an impulse-driven display -- subtle factors such as backlight diffusion between on-segments and off-segments of scanning backlight, pixel persistence limitations, etc.
Fortunately, I'm catching an opportunity in history where a DIY hobbyist guy (with sufficient engineering _and_ human vision knowledge) can pull off a zero-motion-blur gaming LCD for under $1000 of modifications to an existing 23" monitor (without any panel modifications needed; just backlight). Panel technology has finally caught up, thanks to 3D-compatible panels eliminating 99%+ of inter-frame pixel persistence by the end of the refresh. However, manufacturers haven't dared spend the expense of 150 watts per square feet of backlight, required for CRT-equalling impulses. Once it's done properly in accordance to scientific principles -- viola -- zero motion blur gaming LCD. And conveniently run at native 120Hz computer refresh, of course, to solve the 60Hz flicker problem without adding any interpolation.

P.S. Feel free to post comments to posts at my BlurBusters.com Blog.
Edited by Mark Rejhon - 2/22/13 at 3:31pm

What I said was an intentional simplification.
I apologise for the phrase of "willy nilly" when I meant to describe a "longer duty cycle" (which we, obviously, agree on) -- it's still impressive technical engineering achievement of the invention of the plasma display: In plain English -- essentially a massively scaled-up version of a very old-fashioned green vaccuum-fluorescent alarm clocks or oven range clocks (the blue-green glow clocks of the 80's) but with over 2 million sets of pixels (R,G,B) of precisely pulsed illumination to generate the complete rainbow of colors. I appreciate there's an engineering challenging to get gas-filled cells (of a plasma screen) to generate a complete-looking rainbow of colors.

The technicalities you say, are correct. That said, compare a 1998-era or even an entry level 2005-era plasma -- to today's Panasonic or Elite plasmas -- and those older plasma distribute the subfield refreshes over a wider region of a refresh, than a narrower region of a refresh. (The early models seemed to approach 100% duty cycle to my eyes -- not in continuous pixel illumination, but in terms of spreading the subfield flickers through the entire refresh, even if pixels were dark most of the time. That is not intrisinically impossible.) ... So that's why I said "willy nilly" (scientifically incorrect phrase, since all plasmas ever invented, have intentional engineering in their subfield refreshes, no matter how rudimentary). The older/cheaper plasmas (in comparison to today's best plasmas) don't compress the subfield refreshes over as short a time period as current ones -- basically, a longer duty cycle, as you were saying.
Perhaps you're right, 5ms _certainly_ would be the limiting factor. I was referring to phosphor on CRT. I can definitely see a difference in increased motion blur at that level, and even down to 2ms -- it needs to be 1ms (actual light fall-off measurement) less to become mostly a non-factor.
Pretty true. I've never seen a plasma have more motion blur than a 60Hz LCD (no motion enhancements being done).

That said, motion blur on different plasmas do vary... That said, it's still 35-40% of sample and hold (if factoring in another +5ms to +8ms of phosphor decay in very old/cheap plasmas, then we're easily talking 60%-90% of sample and hold, aren't we? I can easily tell 60Hz strobes when the duty cycle is 50% or less (end-to-end flicker cycle including phosphor decay), and I'm unable to easily tell much flicker effect on some really old plasmas.) Again, when I say "duty cycle", I don't mean the amount of time a pixel is continuously illuminated, but the time from the first subfield refresh to the last subfield refresh of the same signal refresh -- even if it's not currently being done anymore, physics do not limit electronics from deciding to briefly pulse a pixel 4 times 4 milliseconds apart (that's far more than a human-eye perceived "35-40% duty cycle" from the _perspective_ of eye-tracking-based motion blur. Perhaps plasmas don't do that anymore for the last decade -- but, regardless, in current modern plasmas, even a 35% duty cycle is quite a huge percentage of a frame, enough to be tantamount to behaving like a sample-and-hold display to the human eyes for a lot of fast motion material -- I look at many plasmas, they're still far worse than CRT. Of course, the best ones are stunning with very clear motion, even if not as good as a very good CRT.

______

[scientific-question]
While we're talking about the science of phosphor decay...
...I'm curious -- What measurement of phosphor decay do you use? As phosphor has no scientific definitive decay limit (zero photons) -- which is why we can still see a CRT or even plasma faintly glow in the dark long after it has been turned off and even unplugged) -- it's a continuous logarithmic-like decay that goes on and on and on. For human eye perception during normal average picture level, phosphor decay becomes a mostly non-factor -- when it's decayed somewhere in the high 90%'s range -- and that's where someone quotes a measurement. In the high-speed video of CRT, most of the phosphor has decayed by about 2ms (1/8th of a 16ms frame) but there's still some glow going all the way to the next refresh (16+ms later). What defines the universal standardized method of phosphor decay measurement, for the purposes of display meant for human vision? One party can use a 80% cutoff, while another party can use a 99% cutoff, so the same phosphor may be quoted at a 1.2ms decay under one measurement method, and a 3.5ms decay under a different measurement method. Is there a standardized decay measurement cutoff value? Is it defined in any FCC, NTSC, SMPTE document or specification?
[/scientific-question]
Edited by Mark Rejhon - 2/22/13 at 3:32pm

Your diagrams are accurate. They are excellent in pointing this out. Seeing this, you are definitely right that the ramp-up pulses would not contribute very significantly at first. Now I can go by the same definition of "duty cycle".
That would explain why many older plasmas did not noticeably flicker -- 8-10ms is more than half of 1/60th of a second. The 90% decay makes sense as a cutoff measurement. It'd be nice to see a citation of a specification or standard. Also, the bright glow that still occurs in phosphor for the first few hundred microseconds after the electron gun passes (no further injection of energy) -- from what I read, that's still scientifically called phosphorescence too, isn't it? So it's the cutoff point that defines where it doesn't matter much to discussion.

Yes, I see the color separation effects from the asymmetry of phosphor decay.

Good to see a plasma expert here -- I may want to ask you for permission to reuse these images on the scanningbacklight.com website. Perhaps you can peer-review some images/graphs I will create of scanning backlight light output.

[Edit to add, in regards to my scanning/strobed backlight] -- The graphs for full panel-at-once strobe would be extremely similar to the CRT one (and potentially be better than CRT in motion blur, given the right panel that had less % remnant pixel persistence at the end of its refresh, versus the amount of green-ghosting that medium-persistence computer monitor CRT has) -- I'd do this on a fast-scanned 3D compatible panel which is ideal for this purpose -- they fast-scan their refresh and get rid of as much persistence as possible, so that the 3D glasses shutters can open sooner for the opposite eye -- these LCD panels are full-backlight-strobe compatible, can then stop worrying about doing complicated backlight scan patterns with this type of panel. For this, there is no backlight diffusion problem, but a potential flicker problem because it's a very brief full-panel strobe (unless it was at 120Hz+) - tantamount to a photo flash equivalent (1/2000 to 1/4000sec) of a backlight freezing the individual frames of motion, completely preventing visible eye-tracking-based motion blur from occuring. There would only be about 1% faint afterimage (remnant LCD persistence during illumination timing - same amount of remnant seen during crosstlak in 3D shutter glasses, potentially less, if I strobe a little bit later into the refresh than shutter glasses normally open at). This ~1% remnant pixel persistence is more for GTG, while it's closer to 0% for full-black and full-white since those full stops are much faster on these panels. Bidirecitonal response-time-acceleration technology is useful, as long as I strobe at the right timing (not before or after the pixel rebounds, or I get more inaccurate color during my strobes) - they have to fine-tune that for 3D shutter glasses operation, anyway. Timing of the strobes will be essential for best image quality, so it'll be an adjustment. Either way, I may need to use a photodiode and an oscilloscope, on some PixPerAn screens, to pull scientifically accurate graphs. I will have scanning modes, but I'm also beginning to realize that backlight diffusion is going to be a major limiting factor during scanning modes, so these strobe modes will be extremely useful for achieving the "zero motion blur LCD" goal. Regardless, the Arduino circuit I have, with some high-power MOSFET amplifiers, will enable 16-channel operation of a segmented backlight, in any independently addressable combination (sequentially illuminated, or entire backlight strobed at once).
Edited by Mark Rejhon - 11/12/12 at 7:19pm

Good catch. I meant "Some microdisplays...."

Theoretically, specially motion-compensated subfields could greatly reduce the increased dithering noise during fast motion -- I wonder how Panasonic's FFD handles the ramp up to the peak output, perhaps they have discovered a way to pulse (dither) creatively along the motion vector to keep the dithering consistent while the eye is tracking -- as a result, greatly reducing noise during motion. But that would contradict against the need to ramp up the same gas cells to peak. (xrox, any comment?) I even see this too (green phosphor ghosting during moving white objects on black background -- in first-person shooter video games)
Edited by Mark Rejhon - 11/12/12 at 7:00pm