Correlation/EDR/CCSS Files - how accurate?

Understanding Correlation Files: Accuracy, Suitability, and SPD2Matrix

There is a fundamental point about Correlation Files that is easy to miss, and worth stating plainly here: there is never any guarantee that a Correlation File will improve the calibration result on any given display.

A Correlation File does not make an i1D3 accurate. It applies a relative correction to the probe's underlying factory calibration, not an absolute one, so whatever inaccuracy is baked into that individual probe remains. This is why five i1D3 units, measuring the same display with the same settings and the same Correlation File, can still disagree by more than 2.5 dE00 at white, as shown on the i1D3 User Guide page. The file cannot fix what it does not measure.

So where do Correlation Files actually fit?
If you have a spectroradiometer - even an i1Pro or a Myiro-1 - you do not need Correlation Files at all.
Probe Matching to a spectro is always the accurate route, because it measures the truth for your actual display, with your actual probe, at that moment in time.
Nothing else does.

Correlation Files exist for the common situation where you have only a colorimeter and no spectro.
In that case the correct Correlation File may well improve your result compared to out-of-the-box - particularly on home TVs and mass-market monitors - but there is no guarantee.
Everything hinges on one thing: how closely the display the file was generated on matches the display you are using it on.

This is where SPD2Matrix becomes genuinely useful - not only for generating a Reference file for Probe Matching, but as a way to judge whether a Correlation File suits your display at all, before you rely on it.

Judging suitability - the concept of 'close'
When SPD2Matrix processes a Correlation File it extracts the R/G/B/W x,y values the spectro originally recorded when the file was made - the original patch readings, exactly.
You can then compare those against your own display: measure your R/G/B/W with your colorimeter, with no correction and no probe match applied, and compare the x,y values.
(x,y are the values the correlation correction acts on; luminance is a separate matter and not the point of this check.)
The concept to grasp here is "close".

Because you have no spectro, you have no absolute truth to check against - so this comparison against the file's own spectro-origin values is the best signal available to you. You are asking one simple question: was this file made on a display like mine?

• If your measured x,y are close to the extracted values, the file was made on a display much like yours, and it is likely to help.
• If they are wide apart, the file does not represent your display - and no way of using it, as an i1D3 Correlation File or as an SPD2Matrix Reference for Probe Matching, will produce an accurate result.

'Close' is a judgement, not a pass/fail number, and that is precisely the point.
Without a reference you are weighing similarity, not measuring error - but similarity is exactly what determines whether the file is any use to you.

Why a mismatched file can appear 'less wrong' than a Probe Match
This part surprises people, so it is worth understanding.

Used as an i1D3 Correlation File, the data applies a gentle, relative nudge.
It never imposes an absolute answer, so even a mismatched file leaves your reading close to where your raw colorimeter put it - wrong, but often only a little more wrong.
Used as an SPD2Matrix Reference for ColourSpace's Probe Matching, the same data becomes an absolute correction. ColourSpace takes the extracted R/G/B/W as the Reference, measures your display's R/G/B/W with the colorimeter as the Active probe, and builds a matrix that forces the colorimeter onto the Reference values.
When the file matches your display this is the better method - it is an absolute correction, and to all intents and purposes it is the truth.
But when the file does not match your display, that same strength forces your reading onto the wrong truth, and the result is clearly wrong.
So neither approach escapes a bad match.
They simply fail differently: the Correlation File nudges - a little more wrong - while the Probe Match Reference forces - exactly right when matched, and clearly wrong when not.

Why a modest spectro still beats a fine Correlation File
This one often seems backwards, so it is worth explaining: how can a modest 10nm spectro like an i1Pro or a Myiro-1 be more accurate than a Correlation File that may have been generated on a high-end 2nm spectro?

The answer is that the file's resolution is not the point - what each one measures is.
A spectro, even at 10nm, measures your display, with your probe, right now, and produces a direct, absolute correction for that exact combination.
A Correlation File, however finely it was originally measured, was made on a different display and a different probe.
Used on your setup it applies only a relative nudge to your colorimeter's factory calibration - it never corrects your probe's own error, and individual i1D3 units can differ from one another by more than 2.5 dE00.

The two largest error sources in colorimeter calibration are that unit-to-unit probe variation, and the mismatch between the file's display and yours.

Only a direct spectro measurement addresses either.
So a 10nm spectro measuring the truth for your setup beats a 2nm measurement of someone else's display applied as a relative correction - the fine resolution of the file's origin spectro is thrown away the moment the file is used on a display it was not made on.
(The one place 10nm can cost you a little is on very narrowband sources - RGB lasers, some spiky lamps - where a coarse spectro can under-resolves sharp spectral peaks. Even then it is still measuring your actual display and probe, so the Probe Match remains valid and still beats a mismatched file.)

The bottom line
Correlation Files only genuinely work when the display the file was generated on is spectrally near-identical to the display it is being used on - the same emitter technology, with similar primaries and white point. A WOLED file covers WOLEDs; it will not save a lamp projector, or the wrong WOLED panel generation.

SPD2Matrix lets you test that match before you trust it, and - when the match is good - produces an absolute Reference correction that is as good as the truth.
What it cannot do is invent a match that is not there.
And ultimately there is no substitute for a spectro, even a modest i1Pro or Myiro-1, especially given how widely individual i1D3 units vary from one another. If accuracy genuinely matters, Probe Match to a spectro.
Everything else is a useful convenience - not a guarantee.

Steve

Due to questions that have been asked I have expanded the above post to hopefully make the information clearer.
Feel free to ask any questions.

Steve