4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Breaking news in the gaming monitor industry! Press releases, new monitors, rumors. IPS, OLED, 144Hz, G-SYNC, Ultrawides, etc. Submit news you see online!
Sparky
Posts: 682
Joined: 15 Jan 2014, 02:29

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by Sparky » 07 Jan 2015, 10:55

jts888 wrote:Yes, the industry needed to be shaken up, but G-sync seems to be dubiously over-engineering approach to the problem.

tl;dr: after matching a panel's perceived strobe/dimming levels, controller-created artifacts are easily avoided, and users are left only with determining and configuring their personally preferred motion clarity levels at different brightness x Hz ranges.

I get that it's a big overall undertaking bringing new technologies to market, but the underlying core engineering problem is vastly overstated IMO, mostly due to Nvidia being themselves and gamer "journalism".
Overengineered yes, but there's nothing stopping Nvidia from moving to an ASIC as cost becomes more important and the requirements become more clear. In the meantime, they have a lot more flexibility, and could in principle push out firmware updates to add features that would require a new ASIC revision for their competitors.
For visual reference.
As panel brightness goes up, the grey hold levels rise and the strobe pulse area (usually width, since intensity will normally be maxxed) gets bigger.

The blank+strobe+trailing transition interval is the panel's minimum frame interval, say 1/144 s ~= 8.3 ms.
Perceived (average over narrow time) brightness for the strobe-in-dark-box must be made to match the continuous lower hold level brightness.

At high frame rates, the hold periods go away or are too short and dim to be noticed, and exhibited behavior is perceptibly indistinguishable from tradition fixed-Hz strobing displays.
At low frame rates, the strobe pulse can be made arbitrary shorter/fatter until it fills its time window, becoming indistinguishable from sample-and-hold displays if de-accentuating choppy motion is desired.
I'm aware. Here's a relevant link: http://www.blurbusters.com/faq/creating ... blerefresh

As for your suggested implementation, one thing to keep in mind is that you're probably already using a higher frequency PWM to set the brightness of the high persistence mode, and you're going to have a lot less PWM resolution to play with at low brightnesses(1% duty cycle is as easy to distinguish from 2% duty cycle as 50% is from 100%.) You could get around this a few ways: by using a very high clock as the base of your PWM, temporal dithering, or a second voltage domain. In any case, it's going to be more complicated than exposing backlight duty cycle on an existing scaler(which is already done, for brightness control).

jts888
Posts: 28
Joined: 05 Jan 2015, 21:36

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by jts888 » 07 Jan 2015, 12:46

Sparky wrote:Overengineered yes, but there's nothing stopping Nvidia from moving to an ASIC as cost becomes more important and the requirements become more clear. In the meantime, they have a lot more flexibility, and could in principle push out firmware updates to add features that would require a new ASIC revision for their competitors.

I'm aware. Here's a relevant link: http://www.blurbusters.com/faq/creating ... blerefresh
Interesting, thanks.
Simpler algorithm variations are also possible (e.g. keeping a square wave, and using only pulsewidth / pulseheight manipulation to achieve the blending effect, but without curve-softening). This is included as part of this general idea of blending from PWM-free at lower refresh rates, to strobing at higher refresh rates. The trigger framerates may be different from the examples (or may even be adjustable via a user flicker-threshold setting), but the concept is the same.
It looks like Mark and I reached the same fundamental conclusions, but we appear to differ on whether strobe sequences are purely additive over the baseline holding levels, and his diagrams indicate a desire to dynamically alter hold brightness under rising Hz input.

I contend that the simplest and most effective way to achieve the blending is by intentionally dropping backlight levels before and after the strobe with impulses resembling a Ricker kernel or similar, since hiding transition times is desirable anyway, it doesn't require altering the baseline hold brightness, and is invariant on net brightness regardless of frame rate or stutter.

The approaches shown in both article sketches would require buffering on the order of 1 frame time or longer to avoid brightness flickering whenever the frame rate unexpectedly changes.
Sparky wrote: As for your suggested implementation, one thing to keep in mind is that you're probably already using a higher frequency PWM to set the brightness of the high persistence mode, and you're going to have a lot less PWM resolution to play with at low brightnesses(1% duty cycle is as easy to distinguish from 2% duty cycle as 50% is from 100%.) You could get around this a few ways: by using a very high clock as the base of your PWM, temporal dithering, or a second voltage domain. In any case, it's going to be more complicated than exposing backlight duty cycle on an existing scaler(which is already done, for brightness control).
Sorry, I thought it was clear that the second voltage domain was what I was proposing. It's definitely more complicated than what panel vendors likely expose on current integrated backlights, but it's technically a trivial thing to add to new models if someone thinks to ask.

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

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by Chief Blur Buster » 07 Jan 2015, 13:10

jts888 wrote:It looks like Mark and I reached the same fundamental conclusions, but we appear to differ on whether strobe sequences are purely additive over the baseline holding levels, and his diagrams indicate a desire to dynamically alter hold brightness under rising Hz input.
(For new visitors without previous reading, see variable-rate strobing chapter via Electronics Hacking: Creating A Strobe Backlight)

<AREA51 type="display engineering">
I think both your idea and my idea would work -- it's both different means of accomplishing the same effect.

They aren't purely additive (due to non-linear response of vision). But it definitely behaves additively over high-frequency timescales (say, 10KHz) -- they most definitely behave additively. There's actually a scientific law associated with this: Talbot-Plateau Law covering average brightness over flicker in an additive fashion, and this also stays true with non-regular flicker, whether full cycle (0-100%) or partial cycle (brightness curves), too. Roughly speaking, it's number of photons hitting the eyeballs in a specific time period, at sub-flicker-detection thresholds (and that any averages over multiple timescales (e.g. 1/100sec, 1/200sec, 1/300sec, etc) averages equal each other with average variances below detectability thresholds, and all the timescales are far below flicker detectability thresholds). My diagrams are compliant with this law, and I think a properly implemented version of what you described would also follow this law too. So whatever works, works.

A super extreme case of this is blackbody radiation from incandescent sources (ultra, ultra, ultra high frequency random emission of photons). A lesser extreme case is persistence adjustments of monitors (strobe duty / strobe phase adjustments). A middle case would be a microcontroller intentionally algorithmically "randomly" flickering an LED. Random 10,000Hz flicker can be made to look consistent brightness, as long as the outliers are precisely pulse-shaped (otherwise it shows as faint flicker caused by fluctuating average). Human detectability of brightness changes can be less than 1% when you're paying attention to it (1% difference in an 8-bit palette is 2.5 shades apart, comparing say, RGB 255 versus RGB 253, which can be detectable on a properly adjusted monitor), but even that threshold varies depending on the frequency of flicker and human (some will be more annoyed by random flicker than others) so we'd need to target a rough baseline.

So, any random-strobing will probably need to be slightly pulse-shaped on the leading edges and trailing edges of strobes, if doing PWM-based strobing (increasing/decreasing strobelengths away from persistence target at microsecond timescales) to ensure trailing averages are consistent. Even 10 microsecond error can lead to noticeable flicker; I've witnessed it: 1.0ms persistence versus 1.01ms persistence is a 1% brightness change! Some beta monitors with less accurate PWM, flickered randomly during regular strobe mode, because persistence was randomly changing by 10 microseconds! When 10us variance outliers occurs in 1.0ms strobing, about twice a second, you see about two major random flickers per second of 1% brightness change. Especially in a full-screen Windows Notepad window, and especially in peripheral vision. Ouch. Let this be a lesson to monitor manufacturers in precision needed to prevent human-visible flicker (IRQ interrupts during microcontroller-code-driven PWM is enough to cause human visible modulations in brightness!!)...

Over low-frequency flicker, things get more complicated. A random 70Hz flicker is far more noticeable than a non-random 70Hz flicker. But a random ~10,000Hz-ish flicker would be hard to tell apart from a regular synchronous 10,000Hz flicker, for stationary non-eye-tracking--situations, assuming that any random segment (over a short timescale that's sub-flicker-fusion-threshold such as 1/200sec; but all thresholds should be tested so that 1/200sec averages equal 1/500sec averages, if possible) is exactly equal to any other random segment of the same random flicker. That's the engineering challenge; making sure it's pretty consistent over reasonable flicker-fusion-threshold timescales. That may mean shimmying with leading edges and trailing edges a little bit to keep things consistent over varying time periods. It's the same number of photons hitting the eyeballs over human perceptual timescales.

However... side effects such as wagonwheel effects and phantom-array effects of variable flicker, would actually cause different artifacts, and even beat-frequency artifacts as the flicker Hz varied past thresholds (e.g. beat-frequencying with RPM speed of on-screen wheels) and non-regular PWM artifacts with variable spacing (pursuit camera capture of PWM artifact) which is witnessed during non-eye-tracking situations.

A big problem is that variable refresh rates often go as low as 30Hz, which definitely fall very far within flicker detectability thresholds, and people's sensitivity to flicker will vary (and even people's thresholds of detecting randomness of flicker may also even vary independently too, even if it definitely follows the Law beyond a high Hz). This is where the big engineering challenges probably is, accomplishing as low-frequency-as-possible variable rate strobing. So a power user adjustment of where the blending threshold occurs (e.g. at a lower or at a higher Hz). And even a second adjustment, the amount of blending (how many Hz it blends over -- e.g. flickerfree sample-and-hold at 53Hz through fully flicerking at 87Hz, versus flickerfree sample-and-hold at 75Hz through fully flickering at 80Hz). Regardless of what blending algorithm is used (yours or mine), since it definitely needs to avoid single-dominant-strobe at 30Hz, while we definitely want single-dominant-strobe at 120Hz, if we're doing motion blur reduction strobing during variable-refresh-rate modes...

Pure OLED, in theory, should be much easier to do variable-rate-strobing. Rolling-scan strobes are done on fixed-refresh OLEDs (pixel-row-off passes chase-scanning behind pixel-row-refresh passes, high speed video). The rolling scan can be dynamically modulated during variable refresh rates -- since it is theoreteically simply dynamic adjustment of the rolling scan windows. Of course, OLED pixel response is a limiting factor especially in AMOLEDs -- due to transitor latency -- individual OLED pixel response can be a limiting factor as it can take several hundred milliseconds for the active matrix transitor behind an OLED pixel, to fully activate the OLED pixel (see graph), passive matrix OLEDs are much faster without this transitor switching latency as LEDs themselves can switch in less than a microsecond. Also phosphor-backed OLEDs (e.g. white OLEDs driving colored phosphors), found in certain models of OLED HDTVs, will have phosphor latency (typically <1ms, CRT-style). So, for good low-persistence AMOLED, you want OLEDs with primary colors, and accelerated active matrix transistor switching, and then it could be fast enough for on-the-fly rolling-scan-window adjustment (longer distance between ON/OFF passes) or by doing faster/slower scanning of OLED pixel rows -- to accomplish dynamic-rate strobing on OLEDs.

The closest thing to variable-speed flicker is plasma subfields on an ultralow-persistence plasma (e.g. certain Panasonic NeoPlasma panels, like found on the Panasonic VT50). During some scenes, the plasma flickers at a different frequency (due to multiple subfield impulses versus fewer bright subfield impulses, per refresh cycle) than when during bright scenes, by the way it compresses the bright impulses to reduce persistence. You can actually visibly see the flicker increase/decrease dynamically on the Panasonic VT50 television, but it's not always too annoying. The large black frame duty cycle between the surge of flickers, however, is the main factor in the flicker detectability, I believe. Either way, plasmas is the closest real-world modern case study of variable flicker behavior (at least from a dynamic duty cycle perspective). This can be used as an example case study of flicker tolerances, especially during variable-refresh-rate operation.
</AREA51>
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!

flood
Posts: 929
Joined: 21 Dec 2013, 01:25

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by flood » 07 Jan 2015, 16:58

im fairly confident that human vision is quite linear for the intensities in which we're interested

by linear i mean that the perceived luminance as a function of time can be modelled accurately by the convulation of the actual luminance and the impulse response

jts888
Posts: 28
Joined: 05 Jan 2015, 21:36

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by jts888 » 07 Jan 2015, 17:37

Wow, what a post. I'm feeling a bit happy my uninformed guesswork is pretty compatible with your well-informed analysis.
Chief Blur Buster wrote: Roughly speaking, it's number of photons hitting the eyeballs in a specific time period, at sub-flicker-detection thresholds (and that any averages over multiple timescales (e.g. 1/100sec, 1/200sec, 1/300sec, etc) averages equal each other with average variances below detectability thresholds, and all the timescales are far below flicker detectability thresholds). My diagrams are compliant with this law, and I think a properly implemented version of what you described would also follow this law too. So whatever works, works.
My inital guesses about vision response were that photoreceptor stimulation, being based on photons transferring energy to individual pigment molecules and relying on mediated recovery, is probably pseudo-linear until reaching saturation, that recovery is pseudo-exponential, and that longer-term brightness/darkness accommodation takes substantially longer than frame-interval time scales.

I had no doubt that the effects aren't mathematically clean models, but it seemed to me that dynamic logic could pretty easily be table-driven once the data is known like LCD DAC LUTs, etc., provide a sufficiently flexible LED backlight driver exists. I thought that my approach was the most easily realizable one possible that could meet the constraints I was aware of, but your info above has given me more to think about obviously.
Chief Blur Buster wrote: Random 10,000Hz flicker can be made to look consistent brightness, as long as the outliers are precisely pulse-shaped (otherwise it shows as faint flicker caused by fluctuating average).
...
So, any random-strobing will probably need to be slightly pulse-shaped on the leading edges and trailing edges of strobes, if doing PWM-based strobing (increasing/decreasing strobelengths away from persistence target at microsecond timescales) to ensure trailing averages are consistent.
...
Even 10 microsecond error can lead to noticeable flicker; I've witnessed it: 1.0ms persistence versus 1.01ms persistence is a 1% brightness change! Some beta monitors with less accurate PWM, flickered randomly during regular strobe mode, because persistence was randomly changing by 10 microseconds! When 10us variance outliers occurs in 1.0ms strobing, about twice a second, you see about two major random flickers per second of 1% brightness change.
...
That's the engineering challenge; making sure it's pretty consistent over reasonable flicker-fusion-threshold timescales. That may mean shimmying with leading edges and trailing edges a little bit to keep things consistent over varying time periods. It's the same number of photons hitting the eyeballs over human perceptual timescales.
I get that pulse width accuracy matters since it effectively is the pseudo-instantaneous brightness, but I'm confused what you mean by "pulse-shaped" strobes exactly. Are proper sinusoid/sinc/whatever waveforms actually needed to compensate for effects other than motion blur profiles? I had assumed the duration and width of a rectangular pulse could easily be sufficiently controlled to provide desired visual responses even in the face of residual non-linearity.
Chief Blur Buster wrote: This is where the big engineering challenges probably is, accomplishing as low-frequency-as-possible variable rate strobing. So a power user adjustment of where the blending threshold occurs (e.g. at a lower or at a higher Hz). And even a second adjustment, the amount of blending (how many Hz it blends over -- e.g. flickerfree sample-and-hold at 53Hz through fully flicerking at 87Hz, versus flickerfree sample-and-hold at 75Hz through fully flickering at 80Hz). Regardless of what blending algorithm is used (yours or mine), since it definitely needs to avoid single-dominant-strobe at 30Hz, while we definitely want single-dominant-strobe at 120Hz, if we're doing motion blur reduction strobing during variable-refresh-rate modes...
In this we definitely agree, and it's largely why I don't consider G-sync's current dearth of user configurabilty a positive sign for the platform's future.

Final asides:
Chief Blur Buster wrote: Pure OLED, in theory, should be much easier to do variable-rate-strobing.
I've given up on the realistic possibility of OLED strobing given that the element wear is a super-linear function of instantaneous power output and that even new continuous output displays exhibit pretty bad dimming and color shift in as quick as a year of moderate use. I'm afraid that strobed modulated backlight systems will be the best flat panel displays we get for a long time, but I'd love to try a super high-Hz scanning laser array VR goggles or similar.

Thanks again for your thoughts!

Sparky
Posts: 682
Joined: 15 Jan 2014, 02:29

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by Sparky » 07 Jan 2015, 22:25

jts888 wrote: My inital guesses about vision response were that photoreceptor stimulation, being based on photons transferring energy to individual pigment molecules and relying on mediated recovery, is probably pseudo-linear until reaching saturation, that recovery is pseudo-exponential, and that longer-term brightness/darkness accommodation takes substantially longer than frame-interval time scales.

I had no doubt that the effects aren't mathematically clean models, but it seemed to me that dynamic logic could pretty easily be table-driven once the data is known like LCD DAC LUTs, etc., provide a sufficiently flexible LED backlight driver exists. I thought that my approach was the most easily realizable one possible that could meet the constraints I was aware of, but your info above has given me more to think about obviously.
I think temporal dithering would be cheaper in production. Say you use 10bit PWM and your desired pulse width is 325.7 clocks per 1024. Instead of just rounding to 326, you carry the -.3 to the next pulse, so you get 10bit resolution every 1024 clocks, but 11 bit resolution over 2048 clocks, 12bit over 4096 and so on. This way you eliminate the need for an extra voltage domain, and perhaps can eliminate the need for a new ASIC, depending on how the backlight controller is implemented. This does assume that a small brightness change can be spread over a longer time than a large change in brightness, without being noticed. This may require an increase in PWM clock, but probably not as big an increase as would be required to just switch to a higher resolution PWM with the same visual result.

jts888
Posts: 28
Joined: 05 Jan 2015, 21:36

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by jts888 » 07 Jan 2015, 23:09

Sparky wrote:I think temporal dithering would be cheaper in production. Say you use 10bit PWM and your desired pulse width is 325.7 clocks per 1024. Instead of just rounding to 326, you carry the -.3 to the next pulse, so you get 10bit resolution every 1024 clocks, but 11 bit resolution over 2048 clocks, 12bit over 4096 and so on. This way you eliminate the need for an extra voltage domain, and perhaps can eliminate the need for a new ASIC, depending on how the backlight controller is implemented. This does assume that a small brightness change can be spread over a longer time than a large change in brightness, without being noticed. This may require an increase in PWM clock, but probably not as big an increase as would be required to just switch to a higher resolution PWM with the same visual result.
While this could very well be effective, you may not be aware of the ease of making a DAC from PWM output.
A 1st or 2nd order passive RC low-pass filter, made with a pair or two of resistors and capacitors, would yield more than sufficient accurate analog modulation in the 10-1000 Hz frequency range we're looking at for the backlight signal.

Sparky
Posts: 682
Joined: 15 Jan 2014, 02:29

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by Sparky » 07 Jan 2015, 23:50

jts888 wrote:
Sparky wrote:I think temporal dithering would be cheaper in production. Say you use 10bit PWM and your desired pulse width is 325.7 clocks per 1024. Instead of just rounding to 326, you carry the -.3 to the next pulse, so you get 10bit resolution every 1024 clocks, but 11 bit resolution over 2048 clocks, 12bit over 4096 and so on. This way you eliminate the need for an extra voltage domain, and perhaps can eliminate the need for a new ASIC, depending on how the backlight controller is implemented. This does assume that a small brightness change can be spread over a longer time than a large change in brightness, without being noticed. This may require an increase in PWM clock, but probably not as big an increase as would be required to just switch to a higher resolution PWM with the same visual result.
While this could very well be effective, you may not be aware of the ease of making a DAC from PWM output.
A 1st or 2nd order passive RC low-pass filter, made with a pair or two of resistors and capacitors, would yield more than sufficient accurate analog modulation in the 10-1000 Hz frequency range we're looking at for the backlight signal.
visually, I don't think there's much difference between high frequency PWM and a DAC. you can filter out the high frequency stuff, but it's the rounding to x bits that gets you big steps in perceived brightness at low duty cycles.

User avatar
glenster
Posts: 25
Joined: 20 Apr 2014, 08:13
Location: Pittsburgh, PA USA
Contact:

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by glenster » 10 Jan 2015, 05:05


User avatar
glenster
Posts: 25
Joined: 20 Apr 2014, 08:13
Location: Pittsburgh, PA USA
Contact:

Re: 4K IPS FreeSync monitor by ASUS [ROG PG27AQ]

Post by glenster » 12 Jan 2015, 04:22


Post Reply