Understanding Persistence versus Motion Blur

Advanced display talk, display hackers, advanced game programmers, scientists, display researchers, display manufacturers, vision researchers & Advanced Display Articles on Blur Busters. The masters on Blur Busters.
Post Reply
User avatar
Chief Blur Buster
Site Admin
Posts: 11647
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Understanding Persistence versus Motion Blur

Post by Chief Blur Buster » 15 Jan 2014, 13:20

RealNC wrote:There's still something I don't fully understand in regards to strobing and persistence. These methods are intended to make persistent displays behave like CRTs. Why is FPS important? Back in the CRT days, games had the same problems we have today; they often would fall below the vertical frequency of the monitor. Playing a game at 30FPS would look just as sharp on my 120Hz CRT as a 60 or 120FPS game.

So why is high FPS important for blur-free display on LCDs, if it wasn't important on CRTs?
Strobed LCD behaves similiar as CRT.
You don't need high FPS for a strobed LCD, except to reduce flicker (as strobing is more noticeable than phosphor decay).

However, high FPS is often desirable on flickerfree displays
You have to understand persistence from the perspective of frame visibility length.
Higher framerate reduces persistence on flickerfree displays.
Higher framerate does not necessarily reduce persistence on strobed displays (if you don't change strobe length).

Persistence == frame visibility time == sample-and-hold (For the purpose of these discussions here, these mean the same thing)

Frames on a flickerfree display tends to have a persistence equal to frame length. (assuming GtG is not a major factor)
60Hz flickerfree = 1/60sec persistence = 16.7ms motion blur at 60fps @ 60Hz
120Hz flickerfree = 1/120sec persistence = 8.3ms motion blur at 120fps @ 120Hz
240Hz flickerfree = 1/240sec persistence = 4.16ms motion blur at 60fps @ 60Hz

Frames on a strobed display has a persistence equal to strobe length.
Strobing at 1/60sec flashes = 1/60sec persistence = 16.7ms motion blur (at any refresh rate)
Strobing at 1/120sec flashes = 1/120sec persistence = 8.3ms motion blur (at any refresh rate)
Strobing at 1/240sec flashes = 1/240sec persistence = 4.15ms motion blur (at any refresh rate)
Strobing at 1/1000sec flashes = 1/1000sec persistence = 1.0ms motion blur (at any refresh rate)
(Obviously, refresh intervals that are longer than the strobe, or you can't fit the strobes :D )

This is why flicker displays (CRT, plasma, LightBoost, black frame insertion) can produce less motion blur at lower refresh rates, than flickerfree displays. Nearly all flickerfree displays (including common LCD displays) are all high-persistence.

1ms of persistence translates to 1 pixel of motion blur during 1000 pixels/second motion
(This result has been so reliably repeatable in testing of recent monitors, I now call this Blur Busters Law)
Assumes: Strobed, instant transition, squarewave transitions, and framerate == refreshrate (== stroberate if strobed). Slow transitions, repeat strobes, strobe crosstalk, and phosphor decay fuzzies this up, but newer displays have become faster and more squarewave, and far more accurately resembles this equation. Several monitors I have sitting here, XL2720Z, VG248QE, XL2411T, VG278H, (and to an extent) FG2421, all follow this "Blur Busters Law" remarkably closely in tests.

Persistence (in X milliseonds), translates to X pixels of motion blurring during 1000 pixels/second framerate==refreshrate motion. So if you have 4.1ms of persistence in a panning test pattern at 1000 pixels/second, you will get 4.16 pixels of motion blurring. For 500 pixels/second, you get 2.08 pixels of motion blur, and for 2000 pixels/second you get 8.3 pixels of motion blur. Pursuit camera photographed motion blur trail lengths can easily be seen at PHOTOS: 60Hz vs 120Hz vs LightBoost as well as PHOTOS: LightBoost 10% vs 50% vs 100%.

Most motion blur on modern LCDs are not because of GtG transitions. The motion blur is caused simply because the display is high-persistence, and today's flickefree technologies are always high-persistence (unless you use motion interpolation). It is caused by eye tracking motion blur (aka persistence), as illustrated in the following motion test found at http://www.testufo.com/eyetracking

View the below animation on an LCD, with strobing off/LightBoost turned off:
1. Look at the stationary top UFO. 2. Next, look at the moving bottom UFO.


(Make sure the above is running at full frame rate, in a stutterfree browser such as Chrome with Aero mode enabled, or animation won't be accurate)

The motion blur you are seeing above in the moving UFO, is not caused by LCD GtG.
The motion blur you are seeing above in the moving UFO, is caused by high persistence on flicker free displays.

Modern LCDs such as those found in 120Hz and 144Hz gaming monitors, are rate at 1ms to 2ms pixel transition speed. This is a tiny fraction of a refresh cycle (1/60sec = 16.7ms, and even 1/120sec = 8.3ms). But we still see lots of motion blur compared to a CRT with a 2ms phosphor, because of the persistence effect explained in the animation at http://www.testufo.com/eyetracking

As you track eyes on moving objects, your eyes are in different positions at the beginning of a refresh than at the end of a refresh. Flickerfree displays means that the static frames are blurred across your retinas. So high persistence (long frame visibility time) means there is more opportunity for the frame to be blurred across your vision. That's why you still see motion blur on 1ms and 2ms LCDs, and this is why strobing hugely helps them (make them behave more like CRT persistence).

There's no motion blur during 60Hz strobing on an LCD either. But LightBoost is like a fixed-frequency CRT limited to refresh only at 100-120Hz since NVIDIA chose that rate for LightBoost for the flicker fusion threshold during 3D vision (60Hz/60Hz per eyes), and LigthBoost was originally designed for 3D Vision as well, not just to eliminate motion blur. That's the main reason for a higher strobe rate on LightBoost, even though 60Hz LightBoost works perfectly (emulated via software-based black frame insertion, http://www.testufo.com/blackframes when viewing this page on a LightBoost display, since it ends up suppressing every other strobe ...)

Makes better sense now?
NEXT POST: Detailed explanation of http://www.testufo.com/blackframes#count=3 to explain the relationship of persistence length / duty cycle affecting motion blur trail length.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

Image
Forum Rules wrote:  1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
  2. Please report rule violations If you see a post that violates forum rules, then report the post.
  3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!

User avatar
Chief Blur Buster
Site Admin
Posts: 11647
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Re: Understanding Persistence: Strobed & non-strobed, CRT vs

Post by Chief Blur Buster » 15 Jan 2014, 14:03

Long-time Blur Busters readers will recognize this chart I created many months ago:

Image

The milliseconds is equal to persistence, and the motion blur trail lengths are proportional to each other -- e.g. you see a >10x longer motion blur trail at 60Hz, than you see during LightBoost=10%. e.g. Where you may see about ~16 pixel of motion blurring during 60Hz in panning tests such as http://www.testufo.com/photo (very close to 1000 pixels/second by default), you would now see only ~1 pixel of motion blurring during LightBoost=10%.

From a motion blur perspective (excluding other effects such as decay and stroboscopic effects, wagonwheel effects) it doesn't matter how you create this persistence:

-- Shorten persistence, by adding more frames. (Reduces frame visibility time to 1/Hz) -- flickefree method
This applies to flickerfree displays and sample-and-hold displays, including LCD and standard AMOLED

-- Shorten persistence, by adding black period between frames. (decrease frame visibility time) -- strobe method
This applies to light-modulated displays, including CRT, plasma, strobed LCD, rolling-scan OLED, black frame insertion

This really creates same amount of tracking-based motion blur (assuming framerate==refreshrate==stroberate, and also that stroberate is high enough to exceed flicker fusion threshold). That said, there are a lot more complicated factors at play such as plasma subfields, pixel-level PWM in DLP, phosphor decay effects, and other factors in short-persistence techniques, that makes it hard to predict motion blur behavior. There are also other side effects such as stroboscopic behavior and wagonwheel artifacts, more visible at lower strobe rates, but tracking-based motion blur is identical between a perfectly-1ms-persistence 60Hz display and a perfectly-1ms-persistence 1000Hz display (during framerate == refreshrate == stroberate). However, strobe-backlight LCD has very predictable persistence, because LED backlights/edgelights turn ON/OFF very rapidly -- pure phosphorless LEDs are also used to transmit data in optical fiber -- and even phosphor decay (~0.1ms) in white LED is currently insignificant relative to current strobe lengths used in current strobe backlight displays.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

Image
Forum Rules wrote:  1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
  2. Please report rule violations If you see a post that violates forum rules, then report the post.
  3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!

User avatar
Chief Blur Buster
Site Admin
Posts: 11647
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Re: Understanding Persistence: Strobed & non-strobed, CRT vs

Post by Chief Blur Buster » 27 Mar 2014, 16:42

Here's a new diagram comparing persistence of different displays:

Image

Good Educational Persistence Animations
Persistence relating to frame rates (30fps has double persistence as 60fps) -- http://www.testufo.com
Good animation demo of motion blur from persistence -- http://www.testufo.com/eyetracking
Good animation demo of black period to lower persistence -- http://www.testufo.com/blackframes
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

Image
Forum Rules wrote:  1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
  2. Please report rule violations If you see a post that violates forum rules, then report the post.
  3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!

ScepticMatt
Posts: 36
Joined: 16 Feb 2014, 14:42

So what persistence should I choose

Post by ScepticMatt » 29 Mar 2014, 16:57

Very nice. Let me try to add to that, I hope you don't mind.

1. How low of a persistence do I need to maintain image detail?
We want to minimize motion blur during eye tracking, as mark explained. To not lose any resolution, we need to reduce blur below 1 pixels. Maximum smooth pursuit for humans is around 100 degree per second (Edit: Wikipedia claimed 30°, bullshit)
Source: publication/19100553_The_upper_limit_of_human_smooth_pursuit_velocity
So with blurbusters' law and given horizontal resolution, field and of view and maximum pursuit velocity we get:
Image
persistence = (H-FoV)/(100°*H-Res) seconds (red = limited by visual acuity)
Therefore we need sub-ms persistence. Also see this thread: http://forums.blurbusters.com/viewtopic.php?f=4&t=148

For example, THX recommends 26-36° for TVs. Monitors are usually viewed wider. Oculus DK2 has 100°/960 pixels per eye.

2. How low can I set persistence and still avoid flicker?
Roughly speaking, to avoid visible flicker blackout duration should be shorter the inverse of the critical flicker fusion rate.
For refresh rates higher above CFF this is always true, and for lower refresh rates we need to increase persistence.
See the other thread in Area 51 for how to calculate CFF: http://forums.blurbusters.com/viewtopic.php?f=7&t=333
Image
persistence = 1/Framerate - 1/CFF

For example, Oculus chose 3ms@72Hz and 2ms@75Hz for their DK2 headset, and want <1ms@90Hz for CV1.
In the graph above, this would be roughly equal to the yellow line. Mind that in virtual reality CFF is higher than for monitors.

3. How can I avoid eye-strain and headaches?
Interestingly, discomfort is not necessarily result of visible flicker perception. Cells in our brain fire at a rate of 40-70 Hz. Low frequency, low persistence flicker in this range inhibits stimulus recognition. So ultra short persistence at refresh rates below 70 Hz should be avoided. Reduced effect are measurable up to 147 Hz.
Source: http://web.mit.edu/parmstr/Public/NRCan/nrcc38944.pdf
Last edited by ScepticMatt on 30 Mar 2014, 07:12, edited 1 time in total.

User avatar
Chief Blur Buster
Site Admin
Posts: 11647
Joined: 05 Dec 2013, 15:44
Location: Toronto / Hamilton, Ontario, Canada
Contact:

Re: Understanding Persistence: Strobed & non-strobed, CRT vs

Post by Chief Blur Buster » 29 Mar 2014, 21:56

Very good research, ScepticMatt!

I should create a whole new section of BlurBusters, centering around this stuff, and also credits you (or even you writing some articles I put in a new section on the main site, if you express an interest in doing do.)
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

Image
Forum Rules wrote:  1. Rule #1: Be Nice. This is published forum rule #1. Even To Newbies & People You Disagree With!
  2. Please report rule violations If you see a post that violates forum rules, then report the post.
  3. ALWAYS respect indie testers here. See how indies are bootstrapping Blur Busters research!

ScepticMatt
Posts: 36
Joined: 16 Feb 2014, 14:42

So what persistence should I choose (cont.)

Post by ScepticMatt » 30 Mar 2014, 06:11

I think a good way to do this is to post research here, and discuss and fix it first (i.e. mini-"peer review")
Then after a while, with the info collected, post an blog article later.

Edit
4. How low can I set persistence and still achieve the screen brightness I need?.
First, what screen brightness do you need?
For office work, screen brightness should match ambient light levels, ranging in around 80-120 cd/m^2.
For watching movies in a brightness controlled room, STME recommends 41-75 cd/m^2 (55 optimal).
Image
Sources: http://www.ecse.rpi.edu/~schubert/Light ... hapter.pdf
http://www.spectracine.com/Pdf/dark.pdf

Effective screen brightness is equal to backlight brightness multiplied with duty cycle, the percentage the screen is "on". Duty cycle is equal to persistence divided by framerate. So the shortest persistence can be achieved with maximum backlight brightness and highest refresh rate. Should higher persistence be necessary we increase persistence by the same proportion we decrease backlight brightness (i.e. double persistence, half backlight brightness and so on)

Armed with this knowledge, we can calculate the minimum persistence needed:
60 Hz:
Image
85 Hz:
Image
100 Hz:
Image
120 Hz:
Image
144 Hz:
Image
persistence = target-bright / (max-bright*framerate)
Therefore persistence is usually limited by backlight brightness and refresh rate.

For example, the ASUS VG248QE has a maximum brightness of 412 cd/m^2.
So to achieve 100cd/m^2 at 144Hz, we need 1.68 ms persistence, far higher than the 0.3 ms we want to avoid blur.
So we need to increase the refresh rate, increase backlight brightness or decrease target brightness by a factor of ~5.
Or more likely, a combination of the above.

Post Reply