Tuesday, November 08, 2022

The EIZO CG2700X UHD colour accurate monitor

I took delivery of my demo CG2700X Eizo monitor over the weekend and ran it through its paces; suffice to say I am pretty impressed.  

It’s a 27” IPS (so modern LCD) 3820x2160 res panel (so not full-4K) but good for TV, and notably smaller than the 31” FSIs and EIZOs so might fit on a QC desk better than the big guys.


So, first thing is to make a matching profile for the Klein K10A tri-stim probe; I've often written before how a photometer is only as good as the profile you make for it with a spectroradiometer - a CR250 in my case.


Interestingly the spectral power distribution is different from the engineering sample I had half a day with over the summer - Eizo tell me they changed the polariser for production units.
So, for calibration I profiled the display in it's native state (so no gamut imposed by the monitor, all colours as saturated as it will make them).


From that (which is 17^3 - around five thousand colours) I made a rec.709 LUT - 100Cd/m2


Rather splendidly, even though this is a brand new model, ColourSpace was able to talk to the patch generator inside.




It is a single-panel LCD so will not have the dark, inky-blacks you’re used to from a Sony HX-310 or Eizo Prominence (but you could buy a dozen of these displays for the price of one of those!).

In terms of wide-colour-gamut/HDR it has decent emulation for HLG and DolbyVision, topping out at about 550Cd/m2 (unlike the full 1,000Cd/m2 you expect from an Eizo Prominence or Sony HX-310). So could be used for edit/QC/ingest etc. for those standards, but probably not for final grade/mastering.


first column is 10-bit values, second column is light levels

Cages - still important for broadcast QC types; It has three sets of markers; they can be independently sized and positioned. It has presets for all the common aspect ratios and action/graphics safe areas, but you can set them as you want. You can set all line widths from one to six pixels and choose the colours independently.
I’m not sure if it could be any more flexible?! Aside from circular cages, that is!

















My demo unit is safely in a flight case and you guys were on the list for offering it to for assessment; you’re welcome to have it for a few days and I will arrive with it to demo some footage.

Friday, May 20, 2022

Why I wouldn't buy a Sony BVM-HX310 in 2022 (if it was my money)

I wrote something similar in 2018 - see Why I wouldn't buy a Sony BVM-X300 in 2018 (if it was my money) which was more of a comparison of OLEDs for HDR monitors vs. dual-layer IPS panels (and specifically the gen.1 Eizo Prominence; the CG3145).

Spin forward four years and I'm no longer working for a reseller but doing my own thing (probably a third of which is colour - calibrating and building LUTs) and I've recently become an Eizo reseller myself (full disclosure!). 

There are essentially three ways you can hit the requirement for HDR in a monitor - and I am surprised how for some people in the industry it is still an alien concept and yet for others it's 100% required. You have to be able to achieve 1,000 Cd/m2 peak white, a reliable D65 white point, ideally >80% of rec.2020 colour space and the SMPTE 2084 EOTF (the PQ curve), along with HLG and maybe even SLog3 and CLog2 (although those ones are really only occasionally needed on set). BUT, most importantly is good black performance as dynamic range is the difference between the lowest difference that can be faithfully represented and the peak light output. Dolby says this should be > 800,000:1 - you really need those dark, inky blacks.

  • OLED - these displays have no trouble with proper blacks (just turn off the sub-pixel elements and no light is output) although the reliability of OLEDs as they approach black is problematic. OLEDs also suffer burn-in - see the 2018 blog for more.
  • FALD LCDs - like the Apple XDR and others you can modulate the LED backlight elements in an effort to overcome LCD's failing of letting light leak past the TFT sub-pixels; you can never get good blacks from a single layer LCD (IPS technology or otherwise) so maybe shut-down the light from the backlight when you need black. The problem with this is that the LED backlight array is never as hi-res as the display panel and so at hard black-white transitions you get a halo effect; very noticeable on star-fields and the like. It's why even the very best Samsung TVs don't look as good as LG and Panasonic OLEDs. In the case of the Apple XDR the resolution is nearly 20Mpixels but it has only ~500 modulated backlight elements.
  • Dual-layer IPS-LCD - Panasonic manufacture a dual-IPS display assembly (4096x2160 resolution) with backlight that the Sony BVM-HX310 and the Eizo CG3146 Prominence both use. The benefit of having two IPS panels is that any light that leaks through the first will definately be caught by the second. In practise this givens those dark, inky blacks that you normally only get from OLED. Turn off the lights in the room, put up a full-frame black field and you really can't see where the bezel finishes and the screen starts, if you've never seen an LCD doing that well with black you'll be suprised.
So, you might think that the choice between Sony and Eizo would be just down to brand loyalty or price (the Sony does list for a few thousand more than the Eizo, but nobody ever got fired for buying a Sony!). However, the Eizo is not just a Sony with a different badge; for my money it improves thus;
  1. Colour workflow – when I have to calibrate an HX310 (or X300) the client will typically say “…we need to have it ready for 709, DCI-P3, DolbyVision (2084/PQ) and HLG” – with a Sony that is the best part of a morning’s work with lots of adjusting bias and gain values, making measurements and re-adjusting; more akin to a 1990s CRT. The Prominence is an entirely LUT-driven monitor and to set one up for those four standards takes ~15 minutes. Even standards that the monitor does not support out of the box (Clog2 or Slog3 as examples for camera monitoring) can be easily added with ColourSpace by uploading into one of the spare LUT slots. Closed-loop calibration with either ColourSpace or Eizo’s (free) ColourNavigator 7 software is automated and quick (like things should be in the 21st century!)
  2. Warrantee – the Prominence has a five-year on-site replacement guarantee. If your CG3146 fails four years and eleven months after purchase Eizo will ship a replacement and repair/replace yours free. This includes colour accuracy and faulty pixels (try and get Sony to stand by a two-year-old HX310…!) - I asked my man at Eizo how many times they've had to break out the stand-by monitor; "...we haven't yet had to" - no failed Eizo Prominence monitors in the UK in five years! You can't say that about Sony's.
  3. Uniformity – I have a customer who has had to send an HX310 back due to poor (i.e. outside EBU 3325 specification) luminance uniformity. The Prominence has Eizo’s DUE (digital uniformity equaliser) which is configured at the factory to give better than delta-0.5 L u’ v’ (as measured on CIE1976) across the whole panel. - here are the measurements taken off an out-of-the-box HX310 a few months ago;

(before you comment, I know these are CIE1931 values in x y Y) - for proper uniformity measurements you need to do CIE1976 L u' v' - I did convert these before calculating the euclidean distance between patches on the gamut. These give rise to a large-scale luma uniformity error >5% which exceeds SMPTE 3325's spec (which is here)

As an aside here is a grab from another bit of monitor uniformity work I did last year;

Anyway, I have a demo CG3146 which I can leave with you for a week (along with lots of hero HDR footage that you can playback off my BlackMagic Hyperdeck 4K - Arri's camera reel, BBC's HLG and Eizo's demo footage) - I also include Sony's SLog3 reel and a LUT loaded into the monitor! 






Friday, December 31, 2021

SDi physical layer measurement for 3G and 12G; a video presentation.

In the last couple of months I've had to do both training and fault-finding for SDi physical layer measurements. Below is a cut-down video of the half-day training; just so you can get feel for my training style.



I've also been working for a sports broadcaster tracing a problem they have with their incoming OB lines (all via a telco's private cloud J2K SDi route) - the problem was more fundamental than you'd think, but I got to the bottom of it.

I'm very pleased to have a Leader LV5490 with the physical layer measurement option; I can just rock-up at a clients and within minutes be able to give them proper eye measurements with jitter (both 10Hz wander and 10/100KHz-filtered readings). 



Thursday, December 30, 2021

ASUS ProArt PA32UCX monitor - LUTs etc.

My, my; it's been eighteen months since I paid any attention to this blog. Possibly the longest quiet period since I started writing it in 2003! Anyway - it's mostly down to work (I started Media Engineers at the start of 2020; a few weeks before the pandemic started).

Back in 2019 I was approached by ASUS to point a probe at their new PA32U (the first of their 32" monitors to carry the ProArt product name). It had a lot of issues and I wrote up my findings here. I also made a video showing my LightSpace profiling. 

In fairness ASUS issued a firmware update that took in my recommendations about DolbyVision and HLG. Also, in fairness they made their API available to the guys at LightIllusion which is just the thing - monitor manufacturers (almost) universally seem unable to make decent calibration software. 

SO, spin forward to the autumn on 2021 and an old pal sent me his new ProArt PA32UCX monitor and asked me to set it up for two SDR profiles and two HDR profiles;
  1. Rec.709 with a gamma of 2.2
  2. Rec.709 with a gamma of 2.4
  3. HLG - Rec.2020 colourspace with HLG 1.2
  4. DolbyVision - ST.2084 curve
At this point it's worth noting that I didn't have high hopes for the monitor for the following reasons;
  1. It's a single-layer IPS (LCD) display with an LED backlight,
  2. It's a FALD (so zone'd backlight) - like the Apple XDR it's an effort to make an LCD have a higher dynamic range. A typical 10-bit LCD panel (like an Eizo CG319X) can achieve about 1,000:1 but with a zone'd backlight that can be in excess of 1,000,000:1 but with the downside of halation around edges and transitions. Compared to dual IPS (think Eizo CG3146 or Sony HX-310) or OLED (typ. Sony X300) it's poor-man's HDR.
  3. They sell it on the strength of it having "quantum dots"(!) yet it can achieve about the same percentage of DCI-P3 colour space as a modern non-quantum LCD or OLED.  If QD actually exists then surely it should approach laser-projector like primary colours and so get close to Rec.2020 colour primaries? Non-intuitively you need monochromatic primaries to be able to get the largest colour-triangle and that's the promise of Quantum Dots - but since this panel does not have monochromatic primaries where are the quantum effects?!
So - my first thought was that rather than using the hunt & peck four-way controller I'd like to control the monitor from it's software. So, off to download ProArt Calibration 2.0 - first on my Windows 10 calibration laptop and then my other work laptop (15" MacBook Pro 2015 model) and finally my ancient workshop Windows 7 machine - none of them could communicate with the monitor.



I tried several different USB-C cables and eventually I had to resort to a powered hub to get Windows to recognise the device. However, at no point could I get the ASUS software to talk to the display (across three machines, three OSes). Thankfully ColourSpace was able to address the monitor and drive the internal patch generator (more about that later!) - but, first job was to use my CR250 (spectroradiometer) to make a matching profile for my K10A photometer. The reason for this is that the spectro is ultimately accurate but very slow and not good near black whereas the tri-stim K10A is great near black, fast (typ. sub 1 sec per read) but is bedeviled by Specral Power Distribution issues (the K10A is an RGB probe). BUT, with a profile made with the CR250 on the display concerned you can impose spectro accuracy on the photometer.
So, here's the matching profile - for my K10A on this PA32UCX


So, the next job is to profile the monitor so we can make the various calibration LUTs. For a 17-point 3D LUT you need to measure around 5,000 colours (17x17x17) and so even with the fast Klein probe you're looking at two hours. So - I set ColourSpace to drive the internal test signal generator on the monitor and went to make a coffee. When I looked in an hour later I realised something was very wrong; essentially the internal TSG does not seem to generate any blue?!


So, I had to break out my FSI BoxIO which I normally use for patch generation and start again. Two hours later I had a profile and could use ColourSpace to generate LUTs that corrected the monitor to the two rec.709 USER slots.  I saved out the various profiles as Builder Colour Space files (.bcs) and along with the LUTs I made you can grab there at https://tinyurl.com/yej2cs5j


Now to the HDR settings and I discovered that you have to toggle the HDR-flag in the HDMI stream to get the monitor to switch into HDR mode which is plainly stupid for something that isn't a TV! No matter; break out the AJA Hi5-4K+ and use that to switch the monitor into both HDR modes.


Repeat the profile in ColourSpace and then generate the DolbyVision and HLG LUTs and use CS to upload them to the monitor. The results are not bad; here's a photo of the Eizo's luma scale in DolbyVision mode (so display-referred) and it shows around 650 Cd/m2 (not the >1,000 as their website suggests) once calibrated. In RAW/uncalibrated mode it can hit more than one thousand.


Last thoughts;

  1. If you put it into HDR mode, then switch the i/p to an SDR signal, disconnect/re-connect (which you have to do) it will then let you recall one of the SDR USER settings (so 1 for 2.4 gamma, 2 for 2.2 gamma) BUT it never takes the backlight back down to SDR levels - so you get rec709 with 500Cd/m2 white. You have to manually wind the "Brightness" figure back from 100 to 10
  2. "Brightness" is mislabelled - it should be "Backlight" or somesuch
  3. Brightness is actually called "Black Level"
  4. Their software proved useless - without ColourSpace I would have been left high and dry. 
  5. All this fiddling about took days (whilst I was doing other things) - I would not want to have been faced with this at a client's site. I won't be taking bookings to calibrate these monitors.
It's like a computer monitor they've hit hard with a hammer to kinda behave like a broadcast display but they haven't listened to everything the broadcast guy told them - I would not buy this monitor - for Rec.709 I'd use an Eizo CG-series and for HDR I'd use an LG or Panasonic OLED TV.

Wednesday, August 05, 2020

JVC DLA-Z1 4k projector; terrible calibration software!

I was recently asked to calibrate two JVC projectors for rec.709 and DCI-P3. These projectors have a modest amount of colour adjustment in their remote interface, but none of the factory presets are particularly accurate. For calibration JVC have their own software which is terrible! Why people don't integrate with LightSpace (particularly since Steve and his team are very keen to help manufacturers) I'll never know.
Anyway, before detailing two days of frustration in a couple of grading rooms it's worth reminding ourselves about the difference between Spectroradiometers (AKA "spectro) and Colourimeters (AKA "tri-stim probes")
  • Spectroradiometers measure wide band light energy - everything from 380nm (or lower) - very deep blue through to 740 (or higher) - very deep red. They are slow to make a reading (many seconds) and do not cope well with low light levels. 
  • Colourimeters measure just three wavelengths (just like your eyes) - which we'll typically refer to as Red, Green and Blue (but really are X, Y, Z colour matching functions) and so are vulnerable to metameristic failure (a mismatch between the primary colours generated by the display device and the filters used in the colourimeter) BUT they are fast (my Klein K10A can make a read in less than a second) and they are accurate all the way down to near-black.
So, best practise is to use your spectro to make measurements of primary colours (and peak white to be sure) and use that to calibrate the tri-stim. After that you have the speed and black performance you need with the accuracy of the spectro imposed. I tend to do this every time I encounter a new display even though the K10A comes with a lot of factory profiles and the trick the Klein uses is that their filters are very close to average human vision and so any metameristic mismatch between the probe and the display is close to perception which is all important.


Now, onto the JVC software, the first thing they neglect to tell you is to not run the network setting in DHCP mode if you want to do calibration; the projector tries to renew the DHCP lease every hour and so it's likely you'll loose connection and have to restart the process...


Next you have only two choices of low-end probe - the "DTP special" - the Spyder, and the only slightly better Xrite i1Pro2 spectro (I happen to have one of those) but bear in mind all the things we said about spectro earlier.
The Xrite needs to calibrate itself to it's supplied white reference tile every time you use it BUT the JVC software has not implemented that functionality - so, you have to load up some other software (I use Sony's monitor colour balance software), but LightSpace, ColourSpace or several others would do - connect to the probe, trigger a calibration and then disconnect.


Now you have to position the probe to collect enough light to make measurements - most projector calibrating gets done from the operator's position and probes like the K10A have aiming lights to show you where they are pointing at the screen. BUT, after two days of experimentation I found I had to have to i1Pro2 as close to the screen as possible whilst avoiding it's own shadow. It's marginally improved by offsetting it horizontally (so the long edge of the probe is parallel to the screen) as the shadow is not as significant;



These projectors have a setting for the LD power - you can drive the lasers at three different power levels. At the highest the image is too bright for grading work; around 120Cd/m2 at peak white which would be fine for a 31" grading monitor, but not a projector. At the mid-LD setting you around 60Cd/m2 at peak white which although still bright is OK. Again, what the documentation doesn't tell you is that it take around half an hour for the power to ramp up or down between LD levels. 


Finally, there is one other setting that can really kill you ability to get decent reads with the Xrite i1Pro2 probe and that's limited/full-range video on the input settings - yes! They have sited the internal patch generator before the video range decoder! So, with all this in mind if you don't;
  1. Have the probe as close to the screen as possible,
  2. Have the projector in High LD power mode for at least half an hour,
  3. Have the video input set for full-range video,
Then you will not be able to read a decent way up the 2.4 gamma curve for rec.709 (and the red channel is particularly affected). Have a look at the light levels as they are read;


This results in some terrible response in the resulting profile with the red channel in a terrible state, incorrect low end response and clipping close to black. 


However, if you get those three things right (above) then you get decent reads close to black and a proper response for the range;



How much easier this would have all been if you could use a tri-stim probe like the Klein but by limiting the software to using a Spectro you are bedevilled by low-light issues. Having to do your calibration at High LD and then switch the projector back to mid-LD when your done is silly. 
I suppose the reason it's like this is that these projectors are aimed at high-end domestic/board-room/lecture-theatre applications and not film & TV. The fact the software defaults all SDR gammas to 2.2 seems to indicate this and not having LightSpace support (when LightIllusion have offered to do all the API donkey-work) is unforgivable in the professional display device.

Sunday, May 10, 2020

Kickstarter projects; three out of four ain't bad...

Along with KickStarter there are numerous crowd-funding sites and I've ploughed a bit of money into several. As well as electronics projects I have funded a few artists to record albums and have been very pleased. 
Here are four projects I backed, three of which came out really well and one which kinda got half-way there. The thing you have to remember is that backing things on KickStarter is not like buying something - you have to fully accept that some project just won't deliver.


  • PockEthernet is a tester for ethernet and IP networks. Serious network people use a Fluke DTX-1800 (I used to have access to one) - it's now discontinued, but like the replacement DSX-series all TDRs (Time Domain Reflectometers) are expensive (a few thousand pounds) but if you want to certify an install it is expected. At the other end of the spectrum you have the £50 DC testers that just make sure there is continuity on each of the eight legs and really just allow you to have some certainty in termination polarity etc. The PockEthernet is a half-way house with some TDR capability (not sure home accurate is it) but nice record keeping. Above a little DC-tester (like a ModTap or others) it can do some IP testing; POE, DHCP, VLAN tags etc. and so for me is ideal. 



You interact with it using a BlueTooth-attached app running on 'phone or tablet computer



The measurement reports can be over many circuits (so testing a whole patch-panel at once is do'able) and you can email/save as PDF from the app.


You can even brand the reports with your logo

  • BeeLine bike navigator - I often see folks with their smart 'phone in a waterproof wallet as a bike GPS. That's great, but when I'm cycling somewhere I'm not entirely familiar with I often like to find my way but certain in the knowledge that as I get closer I can make better navigation decisions. The BeeLine is a bluetooth attached smart compass that tells you what direction to go and how far your destination is. I've been using mine for maybe eighteen months and it works really well. It is stable and accurate with good battery life.

wiggly route; it was a Sunday afternoon!


  • Pebble Smart Watch - Although the Apple Watch is undoubtedly a miracle to technology I never felt it was for me; the biggest problem is the battery life; two days at best. It also seems to need a lot of curation. Friends who use them are constantly attending to them and I only really wanted a second screen for my 'phone with good notifications, health tracking and control of media players. The Pebble does just those things really well and nothing else. The battery on mine (Pebble Time Steel variant) lasts for more than a week and when they went bust at the end of 2016 I bought a second one just in case. They charge in about an hour.

The lockdown has been great for sleep but very bad for exercise...


I always return to the same watch-face "Graphite Too" as it has everything I use and is clear.

Thankfully after Pebble went bust and got bought by FitBit a group call Rebble acquired all the source-code etc and have been supporting the watches with new firmware and online services since.
  • Oscilloscope Watch - I know what you're thinking; what a daft idea! I've written a lot about this one in the past because I did get a very janky alpha-version (3D printed case, very early build of the software etc). Still, five years on and the project is still live on Kickstarter and so we live in hope!

Thursday, March 12, 2020

Modifying Blackmagic 6G routers for quiet(er) operation!

You can't deny the value in BMD SmartVideo Hubs - they are a fraction of the price of traditional broadcast video matrices. They have appalling return-loss on the BNC inputs and their control system is very simple (although in lots of cases that's a benefit). The temptation is to stick them in desks in edit, grading and audio suites, but they are noisy! The reasons are;

  1. Cheap, low air volume fans
  2. Tiny holes in the chassis through which to try and pull enough air
  3. No control of the fans even though the ones they supply have a tach output

'scope is showing the tach o/p of one of the fans, yes, I was routing video!

Even though the cheap/noisy fans BMD fit have a tach output it clearly isn't read by the hardware as the fans run at full tilt from power-on. This one had been on and routing video for a couple of hours (with the lid on) and it's like sitting next to a vacuum cleaner.
So, quick look at RS and filtering by size, volts and then listing by highest air volume & lowest noise I got these Papst fans - they also have a tach output (I had no plan to use that) and more importantly are induction-start motors (so they will run on much lower voltages; I had a feeling I could simply control them with a potentiometer with a similar impedance to the coils).

getting them ready to fit in the same JST 1.5mm pitch headers as the stock fans, 10K pots

fitted to replace the stock fans - I had to ream-out the screw holes in the fans for the screws to fit, double-sided tape for the pots.

The other issue is the tiny holes they have in those cases for airflow. With a bit of extruded aluminium and grill material you can get a good look.

Make sure you don't put another piece of equipment directly above it!

So, proof of the pudding and all that; I ran the stock unit for a couple of hours, pulled the lid off and took a photo with my thermal camera and then did exactly the same after the modifications. The results speak for themselves; the client has these in their audio suite and game me four more to modify.

before & after - running cooler and maybe 20dBs quieter

As an aside I found driving these fans at a constant 8v produced the best results.

Friday, November 29, 2019

Rigol Ultrascope software and Windows

Ever since abandoning the faithful Tektronix 2245 oscilloscope I've been a fan of Rigol digital 'scopes; compact and a load of functionality for modest money (FFT and 1Gig samples/sec in my little DS1052E).
Rigol have been less than stellar in keeping the Windows software current and so here are some cobbled-together instruction (from http://www.milkcarton.com amongst others - but his website is often down?).

  1. Download Ultrascope for your particular series (so DS1000E in my case)
  2. Download the Windows driver (had to find this on the Way Back Machine!), Extract these two files, then go find the device in the Device Manager. Update the driver and point it to the directory where you extract the driver files.
  3. Next, download the NI-VISA Run-Time Engine (v5.0.3 as of this writing). Beware, this file weighs in at 71 MB. Install the VISA runtime with the default options (you could probably get away with just installing the USB portion, but I didn’t try it).
  4. When the NI-VISA installer finally finishes, you might be prompted to reboot. I skipped this step :-). Run the Ultrascope software, and click on Tools –> Connect to Oscilloscope. I was prompted with a list of devices, with none of it making much sense, except the first option “USB0…”




Sunday, October 06, 2019

Experiments with white light (it's complicated!)

I've often run a day's training course for broadcast engineers who want to get up to speed with calibrating monitors and projectors; typically to rec.709 but increasingly to P3 as HDR and 4K/UHD are becoming a thing.  One of the principles I've always struggled to get over is Metamerism; that inability to see/measure colours correctly if your measurement device (camera, eye) is tristimulus and your source of illumination does not have a daylight-like spectrum (so LED lights, typically).

A few month's ago I got one of Chris Wesley's excellent home-brew spectroradiometer kits; from now on referred to as the ghetto-spectro.  Read Chris's excellent documentation about how you can make really quite accurate spectrum measurements with modest parts so long as you can accurately calibrate the thing - and this is where the spectrum of Mercury comes in useful. Mercury has two peaks in the visible spectrum at 546nm and 436nm and you can guarantee that a compact fluorescent bulb will have a decent amount of mercy in it.

the ghetto-spectro pointed at the mercury containing CFL bulb on my workshop bench

the measured output showing the various peaks of different elements

the image from the diffraction grating in the iPhos

So, watch Chris's video which tells you how to calibrate to the two Hg-peaks, and pay special attention between 540 and 550nm as Terbium lurks very closely to the 546nm peak (green) of Mercury.
Terbium is at 543nm, very close to Mercury at 546nm

OK, now I have a calibrated spectro I can turn my attention to experiments with white light and perception. I build a box with two isolated sections, painted inside with a very reflective white primer paint. In the left-hand cavity is one of those RGB-mixer bulbs based on LED technology (and controllable from an app; very 2019!) and in the right-hand section is a broad-spectrum white light. 




Thus equipped I can now mix the RGB values in the left-hand side to produce a white light that matches the right-hand side from my perception. As you can see; the camera in my iPhone does not agree! BUT, I promise you, to my eyes the two white are a really good match. I have spent may years "racking" studio cameras (matching their colourimetry for live TV shows so that the lighting director doesn't shout at you!) and eye-matching displays (typically a good domestic TV to a grade-1 broadcast monitor) - I have a better eye for colour than most.

So, at this point I should show the spectro output for the two light sources;


the right-hand broadband white light; reasonably continuous spectrum



the RGB-mixer bulb; three clear peaks


So; I took photos using three different cameras; an iPhone 8 using the native Apple photo app, a low-end Android tablet using the Google photo app and a 2015-vintage Fuji Finepix 5600 bridge camera. All three rendered the RGB-white differently (remember , that to my eyes it's the same white as the broadband white bulb) and they also minimized the differences in the colours of the juggling thuds I used as colour references in the two box sections.

From the iPhone 8




from the Android tablet




from the Fuji Finepix





Finally I should make a note of how my perception of the colours varied;

I need to think about this a bit more to relate the spectra of the two bulbs to the likely sensitivities of the cameras; but, it does show that observer metameristic failure is a things!