Artificial reason of ghosting in <1ms OLED. Why it was done like that?

[SeekingKnowledge]

Hello BlurBusters! (Chief, I hope you will check this as I never saw someone talk about this).

First of all I've already read this https://blurbusters.com/faq/oled-motion-blur/ and my question is not about this:

We know that <1ms OLED still have motion blur because of high MPRT (16.6ms on 60Hz). However when I run ufo ghosting test on Meizu Pro 6 (AMOLED screen) it has very big ghosting (ghosting, not blur because of MPRT). It surprised me a lot, because accordingly to RTINGS measurements OLED has ~0.1-0.5ms of color transitions. I've wrote my own grey color changer (from 0 to 255 and other shades), run it on Meizu and recorded with 960fps camera (video attached).

On this video we see rolling scan (16.6ms because it's 60Hz) and this scan changes color instantaneously (camera has 1 frame per 1 ms of real life so transitions have <1ms). Still it doesn't explain why there is ghosting. But here is strange thing.

If you will check frame by frame (or at 0.25x speed) transition from 0 to 255 (4th square on top on the left side) you will see that scanout switch pixel not from 0 to 255, but from 0 to ~127 - wait 16.6ms for new scanout - ~127 to ~230 - wait 16.6ms for new scanout - ~230 to real 255 (pure white) and so here is reason of ghosting. With <1ms response time screen do not change 0 to 255, but stretches this transition in time and since it's done once per scanout - we have real response time of 3xscanout time ~ 50ms. (It does same thing for other big transitions, but 0-255 is easy to demostrate)

I hope you understood all this info (check picture for details please). My questions are: is it known? does this technique has some name? If so, then why it's needed? Why we split 1 transition across multiple scanouts and increase response time artificially?

My theory is that true 0.1ms response time on low Hz screen (like 60Hz) will introduce stutters that people do not like. Imagine you change your IPS phone with >20ms responce time to 0.1ms OLED which has no ghosting at all and so it amplifies stutters (RTINGS talk about this issue when they review OLED TVs, but since TV shows 24fps video content it's more clear why it's important) I don't understand why it's used on 60Hz phones. Or maybe I wrong and OLED tech is not designed to have so fast changes in brightness (like transition from off state to full bright 255 can reduce lifespan and so we better to split this transition to multiple transitions so pixel will not be overloaded). But from what I see in RTINGS OLED review - TVs do not use it, they change from 0 to 255 instantly - https://i.rtings.com/images/reviews/tv/ ... -large.jpg On the other hand I don't have other OLED phones so I can't check existence of this trick on other phones.

So overall what is true? If this tech has special name then I will be happy to google it, but so far I can't find answer.

Video: https://drive.google.com/open?id=1I73CE ... q2LtWJr6qi
If you don't want download this, here is example how 0-255 transition was splitted to multiple phases. It doesn't fade between scanouts like LCD. It changes color shade instantly when scanline triggers it.

OLED response.jpg (586.08 KiB) Viewed 8294 times

Thanks

[Chief Blur Buster]

You got all of your information and homework down correctly, and obviously aware of Why Does Some OLEDs Have Motion Blur?

This is common on OLEDs -- even on low-persistence OLEDs -- including my Oculus Rift (example: Load Robo Recall -- the bright white interval screen with black text, turn head around and the black text will ghost on white backgrounds)

I think OLED GtG is measured by the GtG 90% metric, as seen in GtG vs MPRT: Pixel Response FAQ.



So the 0.1ms-0.5ms is honest at the GtG90% metric, however, the 10% and 90% electronic-measuring-equipment cutoff points neglects to mention the fact that below-10% and above-90% is still human visible:



OLED GtG is really fast but OLED GtG is not 100% perfect in the first refresh cycle.

The below-10% and above-90% is very relevant in this modern era of the Vicious Cycle Effect where bigger displays, higher resolutions, wider FOV, closer viewing distances, reduced motion blur, all coincide simultaneously to make human-visible limitations more visible.

These days, we are now seeing human visibile differences where 0.25ms(MPRT100%) versus 0.5ms(MPRT100%) versus 1.0ms(MPRT100%) for very fast panning speeds (e.g. fine text at 4000 pixels/second on a low-persistence 4K display).

The OLED GtG quirk is also found on iPhone X too when viewing www.testufo.com/eyetracking -- the ghost squares appear trailing behind the squares.

[Chief Blur Buster]

Now, why this happens -- is probably because the OLED active matrix transistors aren't completely transitioning to its final switched state in the first scanout voltage pulse.

So there's some incomplete transistor switch, since it's like trying to adjust a water faucet perfectly for a perfect flow rate, since subpixel greyscale switching in LCD/OLED is effectively defacto analog even in a digital display.

It's the amount of electricity being adjusted like a water faucet through the active matrix transistor driving the subpixel. Metaphorically like a dimmer circuit. Imagine millions of analog-knob dimmer switches! One for each subpixel. And you only have a brief instant (since scanout voltage is refreshed one pixel row at a time) to set the correct setting. The modern LCD and OLED are really engineering achievements in digital-driving of analog-greyscale pixels.

In an LCD, the transistor drives the amount of power imparting momentum to rotating molecules of a liquid crystal being used as a polarized shutter (diagram). In an OLED, it's the direct power illumintation of a light emitting pixel.

You can do it differently (e.g. LED JumboTrons, DLP projectors) and instead do high-speed digital PWM to generate the colors. But in an LCD/OLED, it's a more analog adjustment of electricity (like a dimmer) through a pixel. This has many advantages, there's no temporal dithering effects and artifacts, but you are beholden to GtG imperfections. Like what you're seeing -- for both LCD and OLED.

You can use an ultrabrief (fast) camera shutter, and such screens (LCD and OLED) can be temporally exact (as long as they're not strobed and as long as they're not using multi-refresh-cycle temporal dithering to compensate for bit depth limitations of a single Hz -- which is more effectively just an A/D converter bitdepth limitation, not a direct pixel limitation). Whereas, if you try to ultra-briefly photograph a DLP, you get a distorted photograph. This is because LCD and OLED pixels aren't digitally flashed to generate color, these subpixels actually analog-controlled to the amount of desired brightness for that pixel. That is challenging to do fast AND accurately. But you have none of the temporal-dithering artifacts you get with DLP and other technologies in this refresh rate race to retina refresh rates.

(Note: There can be some refresh-cycle-level temporal dithering - aka FRC (not subrefresh like for DLP). That's simply because of limited color depth per refresh cycle, e.g. a bit depth limitation in the D/A conversion at the individual refresh cycle, e.g. 6-bit per Hz, as the control of the subpixels is where digital become analog again)

____

The measurement numbers you are seeing -- the GtG gets vastly more than 90% of the way in the first scanout, but gets new voltage-kicks in subsequent scanout getting the pixel closer to its final value.

I've seen this GtG-stairstepping behaviour manifest itself on some LCD panels, if you run a photodiode oscilloscope on the slow VA colors (that takes multiple refresh cycles), you'll see the GtG-stairstepping effect there too, as the GtG gets new "kicks" from the repeated scanout pulses of consecutive refresh cycles.

Useful relevant links are:
- Pixels are metaphorical GtG soccer balls to be kicked
This is for LCD, but explains how brief voltage kicks are during scanout, and affects OLED to an extent too
- The Complexity of GtG
This helps explains how complicated it is to measure GtG on LCDs. Most of it doesn't apply to OLED, but some of it does, GtG measurement matrix tables as an example, as a method of averaging GtG and meeting the 90% metric

These posts may not fully explain OLED behaviours, but will help gain some useful insight about how incomplete GtG kicks are during scanouts, and how GtG can be re-kicked during subsequent scanouts (faint GtG stairstepping effect in a multi-refresh-cycle photodiode oscilloscope measurement) -- this applies to both LCD and OLED.

[SeekingKnowledge]

I think OLED GtG is measured by the GtG 90% metric, as seen in GtG vs MPRT: Pixel Response FAQ.

Thanks for reply. I will read all these links (some of them I already checked), but it seems that we talk about different things which makes me think that this tech has no name and is unknown by far.

You talk about fact that LCD changes shades slowly and we measure 10-90 GtG and pixel still need some time to reach from 90 to 100. But in video we see absolutely different picture. It's not like on LCD. I probably made picture with bad explanation. Video (on 0.25x speed for example) shows that transition from 0 to 255 was smth like programatically splitted to 3 steps. 0 to 127 (done instantly)-> wait for scanout 16.6ms -> 127 to 230 (done instantly) -> wait for scanout 16.6ms -> 230 to 255 (done instantly). Check my plot:

OLED diff 2.png (56.06 KiB) Viewed 8231 times

This trick turned true 0.1ms screen into blurry ~36ms response time soap. For me it looks like artificial change to:
1) keep the blur so user will not dislike phone (fast OLED can't fade like LCD do, also devs can't simulate fading because they have only 16,6ms scanout, so they split big transitions to few phases across scanouts)
2) not fry pixel, but from rtings reviews we see that every OLED TV simply instantly jumps from 0 to 255 for about ~0.2ms, it doesn't split this color transition to long phases like phone do and there is no problems with this approach.

[Chief Blur Buster]

We are indeed talking about multiple separate things -- however, when I refer to the behavior also existing on other displays, I meant GtG jumps to literally 95%-99%, rather than halfway as you indicate. That would indeed be a weird behavior.

OLED GtG stairstepping example
Get an iPhone X and run www.testufo.com/eyetracking -- you'll see it jumps to roughly GtG99% on first scanout, then GtG100% on second scanout. Some OLEDs definitely only jump to approximately ~GtG99% (IIRC, somewhere between 98% and darn-near-100%) on first scanout pass, and only finally reach 100% on the second scanout pass. But 99% is so close to 100% that it may be above the cutoff thresholds for considering GtG 100%

LCD GtG stairstepping example
Get a VA panel monitor, turn off overdrive, and use a photodiode oscilloscope on the VA panel on dim transitions such as This 0x20-vs-0x60 Flicker, and you'll see curves with multiple humps (stairsteps) because the VA curves fast and slows down, curves fast again and slows down, curves fast and slows down. There will typically be about 2 to 3 faint re-accelerations of the curves at the scanout passes for GtG color combos that overlaps multiple refresh cycles.

This can also happen if you turn off overdrive on a high-Hz TN and IPS panel too. The GtG stairstepping effect also shows up in TestUFO ghosting, this is an actual pursuit camera photograph of LCD GtG stairstepping (and also seen with human eye too while tracking eyes on the moving object):



(The arrows correspond exactly with the scanout passes, showing the GtG-curve-reaccelerations)

GtG stairstepping is fainter on LCD than for OLED, and it's more of a squiggly curve rather than sharp stairsteps (I have to run a photodiode on this and show a graph example).

All my examples (LCD or OLED) are jumps to well past GtG50% in the first scanout (curvy or sharp squarewave) and are simply law-of-physics related.

Obviously what you're describing is different, but I was describing stairstepping effects (whether squarewave stepping or curve-reaccelerations) in other contexts.

Now, I get it -- GtG 50%? What you're describing is much more unusual. A jump only to GtG 50%? Wow. There's no reason for OLED to be that slow, unless there's a reason. I wonder why. I wonder if that GtG speed is a "battery management technique" -- since sudden average picture level brightness changes can cause difficulty for a power subsystem (Battery) to keep up with those sudden power changes. Though capacitors should solve that. I would be interested to know what rationale for an OLED to go only GtG50% on first scanout pass...

[flood]

i haven't read through this carefully but i observe something similar in my samsung galaxy chromebook which has an oled display. here are some testufo pictures (the sync track is quite hard to see but it is aligned quite well in these two pictures)

[Chief Blur Buster]

i haven't read through this carefully but i observe something similar in my samsung galaxy chromebook which has an oled display. here are some testufo pictures (the sync track is quite hard to see but it is aligned quite well in these two pictures)

Yes, good observation about the OLED GtG incompleteness -- I see faint white dots trailing to the left of the saucer part.

flood, I bet if you viewed www.testufo.com/eyetracking, you'll see the ghost-squares behind the squares too. I see that on approximately half of the OLED displays I see. One TV I tried was perfectly flawless, while the iPhone X had the ghost-square effect of an approximate ~GtG99% completion in first scanout pass.

<Pursuit Camera Commentary>
Excellent long-exposure 1/15sec handwave pursuits at 60Hz! You got a very steady hand. That's VERY challenging to do accurately, I often intentionally shorten my exposure to only 2-3 tickmarks for lower 60 Hz refresh rates in sample-and-hold operation, due to difficulty of aligning 4 tickmarks in hand-wave pursuits.

Handwave smartphone pursuit camera are much easier at 1/20sec or 1/30sec exposures for a 60Hz sample-and-hold display, since the reduced handwave-mistrack error massively outweigh the reduced exposure-stacking (to simulate human vision averaging behavior) for the vast majority of non-temporally-dithered sample-and-hold displays.

The multi-exposure stacking (and exact integer multiple of refreshes, to compensate for shutter-not-VBI-aligned) is way more important for pursuit camera on temporally dithered displays such as DLP and plasma.
</Pursuit Camera Commentary>

[flood]

yup, see gray ghosts behind the black squares

interesting, my pixel 3a (also oled) looks nearly identical in this test, even the intensity of the ghost.

i had to take maybe 50 some photos to get those two . the fact that it's a 13in display makes it even harder.
anyway i have to say, the idea of the sync track is brilliant