Here's a real example of a 1000hz display... I think

Advanced display talk, display hackers, advanced game programmers, scientists, display researchers, display manufacturers, vision researchers. The masters on Blur Busters.
jlafarga
Posts: 24
Joined: 06 May 2014, 23:56

Here's a real example of a 1000hz display... I think

Post by jlafarga » 07 May 2014, 00:21

Hello everyone! this is my first post here so I thought what better way to start than to share with y'all this rare jewel I recently stumbuled upon
http://www.youtube.com/watch?v=vOvQCPLkPt4
this video is not new but it shows an ultra low latency touch system which I think when the guy says they've lowered the latency to 1ms the display is actually running at 1000fps, yes, one thousand, notice nowhere is said it is though, but here's what makes me think so: first and foremost in order to get the 1ms he's talking about you need 1 second / 1000 samples = 1 millisecond.
second: the display is a screen lit with a DLP projector (the only display technologies that can deliver 1000hz and higher are OLED and DLP, a DLP can theoretically go up to 8000hz for binary images, this is why they went with a DLP projector).
third: they built a custom FPGA controller for the projector (this you'll find in the accompanying paper, the paper focuses on how humans perceive touchscreen latency and nowhere it mentions how fast they run the display).

And finally here's the paper:
https://dl.dropboxusercontent.com/u/740 ... 1-jota.pdf

Saludos everyone!
Jlafarga

spacediver
Posts: 505
Joined: 18 Dec 2013, 23:51

Re: Here's a real example of a 1000hz display... I think

Post by spacediver » 07 May 2014, 01:02

jlafarga wrote: this video is not new but it shows an ultra low latency touch system which I think when the guy says they've lowered the latency to 1ms the display is actually running at 1000fps, yes, one thousand, notice nowhere is said it is though, but here's what makes me think so: first and foremost in order to get the 1ms he's talking about you need 1 second / 1000 samples = 1 millisecond.
Welcome jlafarga!

My brain's hurting trying to think about this, but by that logic, wouldn't any display that had a true 1ms response time have to be running at at least 1000hz?

I really loved that video - it shows how we can clearly distinguish 10ms from 1ms latency. Think about moving the mouse in an FPS and you can see how 1ms is far superior!

Trip
Posts: 155
Joined: 23 Apr 2014, 15:44

Re: Here's a real example of a 1000hz display... I think

Post by Trip » 07 May 2014, 02:17

spacediver wrote:My brain's hurting trying to think about this, but by that logic, wouldn't any display that had a true 1ms response time have to be running at at least 1000hz?
Even though 1ms gtg is probably not the real pixel transition time in a lot of modern displays. It is just the response time of an individual pixel it can transition from one color to another in 1/1000 second. It is not how often the pixel is refreshed because that is just the refresh time.
Those overclockable korean monitors are a nice example that response time of a pixel can be slower then the displays refresh time. The pixels are transitioning and while they are transistioning they are refreshed again. So when the refresh is too fast the pixels can never settle on their colors fast enough. Thats also the most likely reason they wont release an 120 hz ips display since most ips displays struggle to mantain an 1000/120 = 8.33ms gtg response time. A lot of people stop overclocking those korean monitors at around 96hz to cancel out this effect.
So no a display does not have to be running at a 1000hz to get such response times.

User avatar
Chief Blur Buster
Site Admin
Posts: 6509
Joined: 05 Dec 2013, 15:44

Re: Here's a real example of a 1000hz display... I think

Post by Chief Blur Buster » 07 May 2014, 12:18

spacediver wrote:My brain's hurting trying to think about this, but by that logic, wouldn't any display that had a true 1ms response time have to be running at at least 1000hz?
We need to be specific, as GtG ≠ persistence ≠ latency
Response via GtG (pixel transition time) is not the same thing as response via persistence (Motion Picture Response Time / MPRT).
And even with that, we need to be specific within each.

1ms GtG - no, GtG is independent from refresh. You can have GtG shorter or longer than the refresh cycle.
1ms persistence - no, unless you want flickerfree (a sample-and-hold display would need to be 1000Hz for 1ms persistence).
1ms latency - only if you want trailing-1ms latency at absolutely all times (e.g. always having a refresh on screen less than 1ms old).
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

       To support Blur Busters:
       • Official List of Best Gaming Monitors
       • List of G-SYNC Monitors
       • List of FreeSync Monitors
       • List of Ultrawide Monitors

jlafarga
Posts: 24
Joined: 06 May 2014, 23:56

Re: Here's a real example of a 1000hz display... I think

Post by jlafarga » 07 May 2014, 12:40

btw the latency mentioned in the video is the latency of the whole system not the display input lag nor response time, in fact DLPs have response times in the order of microseconds. Also I might add a couple of reasons as to why I think it goes up to 1000hz.

1 - Notice the slo mo footage of the 1ms latency, play it frame by frame and you'll see every frame the cursor moves which indicates that it must indeed be running at a high refresh rate, again they don't say how fast they captured the footage but my guess is that they recorded it at least at 240fps.
2 - In order to maintain the 1ms latency between moving a finger and the cursor following it you must capture that finger at 1000hz AND refresh the display at 1000hz.

Saludos!
JLafarga

User avatar
Chief Blur Buster
Site Admin
Posts: 6509
Joined: 05 Dec 2013, 15:44

Re: Here's a real example of a 1000hz display... I think

Post by Chief Blur Buster » 07 May 2014, 21:47

jlafarga wrote:btw the latency mentioned in the video is the latency of the whole system not the display input lag nor response time, in fact DLPs have response times in the order of microseconds. Also I might add a couple of reasons as to why I think it goes up to 1000hz.

1 - Notice the slo mo footage of the 1ms latency, play it frame by frame and you'll see every frame the cursor moves which indicates that it must indeed be running at a high refresh rate, again they don't say how fast they captured the footage but my guess is that they recorded it at least at 240fps.
jlafarga wrote:2 - In order to maintain the 1ms latency between moving a finger and the cursor following it you must capture that finger at 1000hz AND refresh the display at 1000hz.
Good stuff.

Now, there is a special trick that avoids the 1000hz read / 1000Hz output requirement: low persitsence
You can use input reads perfectly synchronized with output skipping (aka "black frame insertion"), to achieve ultralow 1ms latency at lower rates. e.g. 1ms persistence. You can achieve 1ms lag with lower-frequency polls such as 100Hz or 50Hz. But it won't be absolutely always trailing 1ms latency, since the time since the last visible sample can be more than 1ms.

This would work better with CRT displays or with zero-latency rolling-scan OLED displays. It would not work well with persistence-lowering strobe backlights found in new 120Hz gaming monitors such as LightBoost/ULMB/Turbo240/BENQ Blur Reduction (strobe backlights necessarily needs to wait for an LCD to finish refreshing in the dark, before flashing the backlight on a fully complete refresh cycle).

Self-Explanatory Thought Experiment: A 50Hz zero-latency input read synchronized with 50Hz zero-latency display flickers (low persistence display such as CRT or rolling-scan OLED), would be simply akin to running waving a hand while running a 50Hz xenon strobe light (dance bar; or scientfic xenon strobe light, etc). It doesn't matter what your physical hand is doing between the xenon strobe flashes; you don't get any added lag. Just a stroboscopic effect. Now, ultra-low latency at low read rates, is achieved in the same way -- making sure persistence is equal or shorter than latency itself. (e.g. 1ms persistence and 1ms latency). Motion blurring moves the perceived midpoint of the moving object (e.g. mouse pointer) backwards to the center of the motion blur, which is 1/2 persistence ago (e.g. 8ms of motion blur means the object midpoint is 4ms ago, which creates a sensation of ~4ms of input lag during 8ms persistence). But shorten the persistence, and you have less trailing blur, so the midpoint of the motion-blurred object (by eye-tracking motion - as explained http://www.testufo.com/eyetracking ...) is closer forward to the time of the input read, achieving 1ms latency at read cycles of 10ms.

Black frame insertion is demonstrated at http://www.testufo.com/blackframes and perfectly demonstrates the latency-reducing effects of lower persistence. Observe that the UFO with black frame insertion is "ahead" by half a persistence cycle -- meaning it has less perceived lag. Lowering persistence is easier with hardware-based methods (e.g. rolling-scan OLED) since you can have persistence shorter than a refresh cycle (instead of a sample-and-hold LCD which increases latency sensation via its naturally high persistence). Both UFO's are in _exactly_ the same position (under high speed camera), yet the human eye sees the bottom UFO of the two UFOs (the one with the black frame insertion) being perceived as being "ahead" (lower lag), because its trailing blur is shorter. Strobing reduces/eliminates the perceived latency caused by persistence during low-frequency input reads.



You see, as I invented these motion tests and have also pointed a high speed camera at them, I understand this stuff a lot better than many researchers do because like some people think naturally geometrically, other people think naturally mathematically -- my mind naturally thinks motion/persistence/blurring (and I correctly predict effects even before I invent/write my motion tests!). It would be cool for more scientists to write papers on these things (you are very welcome to cite my due diligence / work on this matter -- contact me if required).

There may be some perceived latency effects caused by the stroboscopic effect, but it would be exactly the same as doing a real-world xenon strobe backlight (chopping up the visuals but without adding any latency to real life). The lack of trailing persistence eliminates the latency effects caused by on-screen objects continuing to be visible for more than 1ms after the input read. So synchronized low-latency black frame insertion or equivalent (e.g. CRT or rolling scan OLED) with low-latency input reads, is a way of keeping low latency at a lower poll rate. Having more persistence creates an additional sensation of latency, by keeping things visible on the screen (from the last input read) for more than 1ms.

TL;DR: You can achieve 1ms perceived latency at lower input/output rates than 1000Hz, via low-latency low-frequency input reads (e.g. 100Hz input reads at 1ms latency individually), and low-peristence low-frequencey output (e.g. 100Hz output at 1ms persistence & 1ms latency; aka 1ms flash). Mathematically, you are essentially using low persistence (strobing) to achieve an on-screen equivalent of a real-world xenon strobe in a totally dark room; and input (real world physical movement) follows output (real world illumination) without lag despite low frequency. The on-screen cursor would follow the finger without any noticeable "trailbehind" caused by motion blur of high persistence.
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

       To support Blur Busters:
       • Official List of Best Gaming Monitors
       • List of G-SYNC Monitors
       • List of FreeSync Monitors
       • List of Ultrawide Monitors

jlafarga
Posts: 24
Joined: 06 May 2014, 23:56

Re: Here's a real example of a 1000hz display... I think

Post by jlafarga » 07 May 2014, 22:25

Thanks for the responses, I'm not 100% sure they ran the thing at 1000hz, but these are the refresh rates I suppose they used:

100ms -> 60hz ?? (I think this is where my previous reasoning was a bit off, for 100ms latency you'd need a framerate of 10hz which it doesn't seem to be the case, maybe they did do it at 60hz but added some extra latency to simulate the mobile phone or tablet latency they talk about)
10ms -> 100hz
1ms -> 1000hz

so what do you personally think they actually went with in terms of refresh rates?

User avatar
Chief Blur Buster
Site Admin
Posts: 6509
Joined: 05 Dec 2013, 15:44

Re: Here's a real example of a 1000hz display... I think

Post by Chief Blur Buster » 07 May 2014, 22:29

jlafarga wrote:so what do you personally think they actually went with in terms of refresh rates?
I am not sure, but most LCDs are 60Hz. For 1000Hz, they would have needed to use a vision-research 1000Hz DLP projector, with a high-frequency infrared sensor for touch. I know Microsoft Research has used specialized high-frequency monochrome DLP projectors. Some vendors such as Viewpixx sells high-refresh-rate DLP projectors (500Hz available in PROPixx at http://www.vpixx.com)

That said, if one is using common touchscreen LCDs -- those would be sample and hold -- and have a persistence equivalent to a full refresh cycle (60Hz LCD = guaranteed minimum 16.7ms persistence), and if so, then they haven't accounted for the latency-lowering effects of low persistence (especially if one uses zero-latency black frame insertion, in the form of rolling scan -- CRT or OLED). Blur reduction strobe backlights (essentially hardware-based black frame insertion via strobe backlight) aren't fully zero latency because they wait for LCD to finish refreshing before flashing the backlight, so that won't solve the latency problem. However, both CRTs and rolling-scan OLEDs can be real-time efreshing directly off the wire (video cable) and would qualify as a way to have perfect finger following touchscreen at low Hz (far less than 1000Hz). Oculus Development Kit2 has repurposed a smartphone OLED into a rolling-scan OLED for low-persistence virtual reality, so we already have a real-world example already.

Image

Demo of motion blur: http://www.testufo.com/eyetracking
Demo of reducing latency/reducing blur via black frame insertion: http://www.testufo.com/blackframes
Head of Blur Busters - BlurBusters.com | TestUFO.com | Follow @BlurBusters on Twitter

       To support Blur Busters:
       • Official List of Best Gaming Monitors
       • List of G-SYNC Monitors
       • List of FreeSync Monitors
       • List of Ultrawide Monitors

jlafarga
Posts: 24
Joined: 06 May 2014, 23:56

Re: Here's a real example of a 1000hz display... I think

Post by jlafarga » 08 May 2014, 23:17

so what you mean by low persistency is that if you ran the display at say 120hz and flashed the cursor for 1ms your brain would interpret it as if though the cursor was 1 millimeter behind your finger ??? (if your finger was moving at 1 meter per second)

jlafarga
Posts: 24
Joined: 06 May 2014, 23:56

Re: Here's a real example of a 1000hz display... I think

Post by jlafarga » 08 May 2014, 23:26

one more thing, based on the slo mo footage (which it says is slowed by a factor of 8) you can sorta determine how fast they recorded it. Assuming the framerate of the final video is 30fps , then 30 times 8 would be 240fps which is the framerate they captured the slo mo footage. Now, on every frame you see the cursor changing position, that means they ran the display (which is a DLP) at at least 240hz.
And for it to show the cursor moving on every frame the camera and the display would've had to be in perfect sync which I doubt, thats the other factor that makes me think it ran at 1000hz.

Post Reply