Light Illusion is renowned for the level of support it provides to all its customers, and strives to provide an unmatched level of assistance.

These forums are provided to enable open discussions on all aspects of the Film and TV industry.

 | Forums | Register | Reply | Search | Statistics |
Display Calibration Light Illusion Forums / Display Calibration /

Display Stabilisation, & Volumetric Accuracy

Page  Page 1 of 2:  1  2  Next »  
Author Steve
#1 | Posted: 29 Mar 2018 10:47 | Edited by: Steve 
With the most recent release of LightSpace we have added a 'Stabilisation' capability, for use during display measurement/profiling/calibration.

This addition has come about due to some in-depth evaluations we have been performing on different display technologies, where the displays would just not profile and calibrate as expected.

Some time back we added 'Drift' to help compensated for both display and probe drift over the profiling duration, and that makes some sense, as display profiling is a stressful process for any display/probe combination, especially for the cheaper probes.

But, in performing recent display technology evaluations it became apparent that some displays suffer 'drift', or more correctly 'stability' issues that are far more aggressive than the expected 'drift' issues.

The issues appear to be heat related thermal instability.

To overcome this during measurement/profiling/calibration we have added the ability to insert a 'stabilisation' frame after every probe read.
This is nominally a black frame, but can be user defined for colour, as well as duration.
The patch is not 'read', it is just displayed for the set period to prevent thermal instability.
The stability patch is set after every real profiling patch to even-out the stabilising effect, preventing cyclical issues caused by a slow heat build-up, followed by a cooling down period, as would happen with the insertion of longer duration stability patches after xxx seconds of profiling.

This 'Stabilisation' process seems to work well for measuring/profiling/calibration.

However, when in real use, such displays will not have regular 'cooling' periods, and will immediately suffer thermal instability, especially if the display is being used in a colour critical environment, such as for grading, as the same image/scene will be held on the display for extended periods, causing thermal instability as the display screen heats up.

The display will therefore NOT be calibrated as expected when in real use.

This means displays that require the use of Stabilisation Patches for calibration should be treated with caution if attempted to be used in colour critical environments, such as for grading work.



Author Steve
#2 | Posted: 29 Mar 2018 19:34 | Edited by: Steve 
To add to this thread the following are graphs of two different displays that show a direct comparison of issue with image stabilisation and non-linearity - the output of the display is not linear and predictable when the input signal is changed in a linear and predictable way.

Both displays were profiled with NO Stabilisation or Drift, just using Anisometric Patch Sequences to match the random image display under normal operation.
(The graphs are generated with a system we use internally to assess display quality.)

The issues are obvious to spot.
Both displays were profiled to their native gamut and gamma, so are a direct comparison of the display's underlying capabilities.
The CIE graphs are 3D views of the Yxy data, with error vectors.
(The vectors show where each point 'should' be, while the coloured points show that actual measured colour location. The points are colour coded Green, Orange, Red, for dE accuracy.)


In the above 3D CIE graphs the volumetric errors with the WRGB OLED are obvious, with the 'error vectors' showing the inaccuracies measured during the profiling sequence, with the 'vectors' changing direction throughout the profile, with multiple 'input' points also pointing towards the same measured colour point. All very irregular and non-linear.
The LCD display shows few such errors.

LCD Normalised Cube WRGB OLED Normalise Cube

The above are 'Normalised Cube' displays of the same data, making it easier to see the issues.
You can see the 'scallop' taken out of the side of the WRGB OLED for example, showing serious inaccuracy compared to the display's native colour space.

And if we add in the Error Vectors, the errors become yet more obvious.

LCD Cube Vectors WRGB OLED Cube Vectors

What you need to understand from the above is that a good display would profile to itself near perfectly.

For displays with inherent underlying non-linear inaccuracy, as shown above when a display is profiled to itself (and not directly associated with thermal instability, which requires the use of 'Stabilisation' patches to overcome), a very large profiling sequence will be required, in combination with a very large 3D calibration LUT to overcome the non-linear issues.

WRGB OLED technology will inherently suffer such issues, due to the inclusion of the 'white' pixel, as this will distort the standard RGB colour channel relationship.

Food for thought!


Author jfinnie
#3 | Posted: 29 Mar 2018 20:20 
Really interesting way of visualising the display behaviour.
If (not sure from the description) you're only comparing to the measured gamut boundaries and gamma, and extrapolating the rest of the expected values, how do you know how much of this shape is due to drift / stabilisation issues and how much due to just the way the display tracks across the cube?
It would be really interesting to see the same cubes profiled in different orders, or with the stabilisation turned on / off.

Any chance of getting some more interesting visualisations like these into LS?

Author Steve
#4 | Posted: 29 Mar 2018 20:38 | Edited by: Steve 
This was just part of our initial display assessment.
(It was these results that defined the need for Stabilisation...)
So far the results with Stabilisation active prove all the above.
Just not had time to documentate the results - but so far they follow exactly as you would expect.

But, while the addition of the 'Stabilisation' patches assist with the thermal issues, the underlying calibration issues remain, as shown by profiling the display to itself.
Even with the 'Stabilisation' active, the display still profiles very poorly to itself, meaning a very large profile and 3D LUT would be required for any chance of accurate calibration, ignoring the thermal instability issues.

As for making the tools available beyond in-house, we would have a lot more work to do.
They are not what we would call 'user friendly, as is.
But, they are rather useful


Author Steve
#5 | Posted: 2 Apr 2018 10:21 | Edited by: Steve 
One of the other calibration software suppliers, who we do work with on occasion, has also posted information on viewing angles:

viewing angle dE Luma

See: n-user-settings-no-price-talk-4.html#post55957080


Author Steve
#6 | Posted: 2 Apr 2018 19:59 | Edited by: Steve 
And a Drift graph, to show the higher-frequency drift, combined with a low-frequency drift/shift that WRGB OLEDs suffer.
(This profile was with a non-contact CR-100, so there is no probe drift.)


And if you look at the different RGB tracks you can also see the way the colour temp of the display changes over time.
It starts warm (red/yellow) and ends cold (blue/cyan).
(This was a display that had been left to warm/stabilise over an extended period before profiling.)


Edited to add: Using the standard Sequential Patch Set within LightSpace will cause the above Drift result, due to the thermal instability issues within the display. The previous Anisometric patch set also caused similar issues, but the patch set has since been updated to provide consistent power demands, so limiting thermal variations during profiling. See below for background information.

Author Steve
#7 | Posted: 4 Apr 2018 10:53 | Edited by: Steve 
To better answer your question, here is a direct comparison with Stabilisation on/off.
The left-hand graphs are 'Stabilisation Off', and the right 'Stabilisation On'.

Again, each profile has been 'mapped' to itself, using the peak gamut, white point, and gamma values from each profile.
A 'good' display would generate a near perfect result.

Graph Graph

If you look at the normal CIE graphs above you can see the 'With Stabilisation' (right hand graph) has a perceivably higher number of 'Orange' dE points, so in general terms the display when profiled with the black stabilise patches is better. But, there are still issues...
(The Drift graph below backs this up.)

Graph Graph

The 3D CIE graphs better show the difference, with the general level of error as shown with the dE vectors being visibly less in the 'Stabilised' graph.

Graph Graph

Graph Graph

The Cube graphs are normalised versions of the 3D CIE graphs, and assist in really seeing both the underlying issues, and the greater issues caused by the thermal instability of the display when Stabilisation is not used.
Note that the graphs without stabilisation show how the displays will respond in 'normal' use.
You can easily see the 'scallop' out of the cube on the left hand side on both with and without Stabilise graphs.
But on the 'no Stabilise' you can see a step change in the data along the bottom edge, and the more random alignment of the dE vectors, especially to the right of the step change.

Graph Graph

Both graphs verify the above, with the 'Stabilised' results showing the underlying instability of the display over time, with a Sequential patch set.


Edited to add: Adopting the new LightSpace Anisometric patch set, the regular saw-tooth instability can be reduced, but there remains a visible thermal instability. See below.

Author jfinnie
#8 | Posted: 4 Apr 2018 13:28 
Very interesting Steve. So is the sawtooth in the drift graph down to the particular patch sequence? It makes me wonder if there would be benefit to doing the patch sequence in an order which avoided the sawtooth, as at each big change of direction there will be a few patches that the drift algorithm would struggle to help (may even hinder) because it won't be clear which way they should be adjusted.
Of course the accurate profiling is only one part...The drift swing in real use looks pretty scary!

Author Steve
#9 | Posted: 4 Apr 2018 13:47 | Edited by: Steve 
No, it would appear to be down to the way the displays work, and has nothing to do with the patch sequence, size, or anything like that.
Easy to verify as changing the 'patch sequence' produces the same basic results.
And as the Anisometric patch sequence used is basically 'random', you would expect a random saw-tooth, if that was an influencing factor.


Edited to add: James was right, and the use of a Sequential patch set will cause the saw-tooth Drift result.

Author jfinnie
#10 | Posted: 8 Apr 2018 09:58 
Interesting, so I guess it is some kind of crude temperature compensation which has discrete adjustment steps once a threshold has been crossed. Be interesting to know if the sawtooth can be affected by environment (desk fan, aircon, etc).

But I guess there is no point in trying to keep the panel cool for the profile if the end result will not match how the display behaves in actual use.

Author Steve
#11 | Posted: 13 May 2018 10:12 | Edited by: Steve 
After some very interesting discussions with James (jfinnie), it has become very obvious he is right!

The way the Anisometric patch sequence works causes a sub-harmonic frequency thermal drift in the WOLED display, half the frequency of the Anisometric patch set.
(Sometimes being so close to the program we can miss the obvious!)

A Piccy

The above shows how using a different CSV patch set, defined to maintain a better overall patch level, can remove the saw-tooth thermal instability response.

However, it does nothing for the volumetric issues.

A Piccy A Piccy

We will be updating the Anisometric Patch set sequence based on this.

Thanks to James Finnie for spotting this association with the patch sequence, and also DeWayne Davis at A/V Fidelity for running a lot of different tests for us.

Note: the drift graph is 'flatter' just because during profiling with the new patch set the average power of the patches is maintained to a consistent level. That will NOT happen in real use, where the thermal instability will be a real issue.


Author Steve
#12 | Posted: 13 May 2018 10:15 
It should also be said that the initial ramp on the right-hand Drift profile, with the CSV patch set, is down to the displays slowly responding to the patch sequence average power value.

We will be adding a 'Pre-Roll' function to LightSpace to better manage this too.
(For now, run the calibration sequence for 5 to 10 mins, and then abort and re-start.)


Author jfinnie
#13 | Posted: 13 May 2018 21:46 
Glad it was a useful discussion. Being able to have sensible conversations with the people who can "make it happen" is one of the reasons I love this package. Lightspace just keeps getting better and better .

Author Steve
#14 | Posted: 16 May 2018 16:10 | Edited by: Steve 
As a continuation of WOLED assessment, the following shows a volumetric comparison of a WOLED and LCD display when used for HDR, with a peak Luma of around 700 to 800 nits, again with both displays profiled to themselves to provide relative data that is easy to compare.

As can very easily be seen the WOLED has major volumetric issues when used for HDR, as the colour gamut luma cannot match the grey scale luma, causing a clipping of RGB Separation.

The LCD display has relatively consistent volumetric accuracy throughout.






Again, looking at the above, a good display would profile to itself near perfectly.
And would have an RGB Separation graph where the RGB plots follow perfectly the target line (all overlapping).
As can be seen, the LCD is 'ok', while not perfect, but the WOLED has major issues that are easily apparent, due to the inclusion of the 'white' pixel, as this distorts the standard RGB colour channel relationship - excessively at HDR brightness levels.

In very simple terms, what this means is the WOLED can never be calibrated for HDR...

Author jfinnie
#15 | Posted: 17 May 2018 21:40 
Really interesting pics; though having trouble getting my head around it a little bit.

So what does the WOLED graph mean in terms of a graph that might be considered more "standard"?

The WOLED I believe would follow PQ EOTF up to about 70% stimulus level for white. Up to what stimulus level would the colours follow the PQ EOTF?

Page  Page 1 of 2:  1  2  Next » 
Display Calibration Light Illusion Forums / Display Calibration / Display Stabilisation, & Volumetric Accuracy Top
Your Reply Click this icon to move up to the quoted message

Only registered users are allowed to post here. Please, enter your username/password details upon posting a message, or register first.
Online now: Guests - 3
Members - 0
Most users ever online: 316 [5 Feb 2020 22:54]
Guests - 316 / Members - 0
© Light Illusion - All rights reserved - Privacy Notice