KEF Blade/Reference Meta whitepaper may paint 2014 non-Meta Reference in worse light than it deserves

[capslock]

I had basically written off the idea of buying old Reference series speakers based on the Blade / The Reference Meta whitepaper that showed a graph with the old Reference having slightly over twice the midrange distortion of the new one. Until I reread Amir's review of the non-Meta 4c that was amazingly low in midrange distortion:

KEF Reference 4C Review (Center Speaker)

This is a review and detailed measurements of the KEF Reference 4C Center home theater speaker. It was kindly purchased new by a member and drop shipped to me. 4C costs US $7,500. Apology for using stock picture. Speaker weighs 100 pounds and I got interrupted mid-review with our home...

www.audiosciencereview.com

I carefully reread the 2014 and 2022 whitepapers. My conclusion: The new Meta driver seems to be about the only change. The case with its constrained layer damping and the LF drivers appear to be unchanged. That leaves the MF/HF driver. It sports:
- the Meta absorber on the HF
- Neo magnets on both MF and HF (as opposed to ferrite magnets in both instances on the older UniQ)
- The split gap geometry with a heavy copper ring in the space between the two top plates and tiny aluminum rings above and below the top plates on the new MF vs a more conventional undercut design with heavy aluminum rings above and below the gap
- The decoupling between basked and magnet on the Meta UniQ

Let's have a look at the measurements:

So we the 2022 whitepaper says there was between 0.1 and 0.2% THD between 500 and 2000 Hz at 90 dB/1 m on the old design and the new one runs at about 40% of that. However, the old whitepaper shows between 0.07 and 0.04% distortion which would mean the old driver is in fact easily in the same leage as the now one and may even pull a little bit ahead.

Is there a good explanation why the Reference 1 MF should be so much worse in 2022 than the Reference 5 MF was in 2014? As far as I know, the UniQs are identical across Reference 1, 3, 5, 2c, and 4c, so the 1 and the 5 and 6c should have near identical midrange performance.

Speaking of which, Amir's measurement at 86 dB/1m is more like 0.04 to 0.02%. Ok, that was at lower level. How about 96 db/1m?

Well, there are a few tiny peaks approaching 0.2%, but mostly, it stays below 0.1% and even below 0.5% across wide bands. I'd say this is highly compatible with the plot from the 2014 whitepaper.

Next, inductance

The 2014 whitepaper gives us inductance only down to 500 Hz and at + and - 3 mm cone displacement. It is probably a safe bet that the rest position inductance is very close to these values, so about 0.17 mH at 500 Hz. The 2022 paper has a funny vertical axis but clearly it places the 500 Hz inductance of the older driver at or above 0.3 mH. Again, how can this be true?

Last example for today, the new spider is supposed to have done away with a 650 Hz kink of about 2 dB that is strangely shown in the 2022 graph of the 2014 driver but not in the original whitepaper (and the older graph looks less smoothed if anything, just look at 4 - 7 kHz):

Unsurprisingly, there is no hint of that 650 Hz kink in Amir's measurement:

Amir did find small blips in the impedance plot but they are at 500 and 920 Hz and not particularly evident in the FR plot either:

So what is going on here? Is there a good technical explanation or was this marketing getting ahead of engineering?

 

[capslock]

Next, flexible decoupling chassis. The new UniQ has flexible coupling between basket and magnet. This should get rid of a nasty peak in baffle vibration that the 2014 reference supposedly had at about 350 Hz. - 60 dB volocity may not sound like much, but depending on how much of the baffle resonates, it may be a much larger area than the MF driver has, so it might end up as a significant peak on the FR, except it doesn't in Amir's measurements:

 

[davidbosch]

Hi @capslock , I wrote the 2022 Blade and Reference Meta white paper when I was the lead acoustical engineer for that project. It was edited, formatted and some sections co-written by my colleague Ben Hagens, Senior Product Training Specialist. We produce the white papers entirely in R&D. Usually the lead project engineer writes it and the rest of us (we’re 6) proofread it. It is a very small but intense operation that happens in the span of about 3-4 weeks before the market launch of a speaker and therefore it’s a bit of a beast to go through. We don’t really have any input from our marketing colleagues, apart from some of the images, the nice renders, which we do get from them. All the rest of the images, graphs and the data are put together by us in R&D.

Thanks for raising these inconsistencies. Sometimes we do make mistakes when putting together the data or in showing it. When I did the R Meta white paper someone pointed out some mistakes and I had to republish it two times. In case we find more mistakes, I apologise and we will try our best to more carefully check every claim before we publish the next white paper.

I also want to thank you for reading the white papers so carefully. I am constantly trying to push them onto people (partly because they were such an effort to write) and I feel some might just skim through the juicy bits and scroll down directly to the spinoramas.

In the white papers we do try to report as clearly as we can on the improvements we have designed into our speakers and show any data that can exemplify a result we want to show. The idea of the white paper is for people (customers or just audiophiles) to see what we do in R&D and to see some evidence of the technology we develop and we hope they come across as more than marketing exercises.

Now let me try to answer all your points as best as I can:

• In Reference Meta, the changes are: Uni-Q driver both MF and HF, crossovers, LF acoustic damping, MF acoustic damping, MF bracing
  
  
• In Blade Meta, the changes are: Uni-Q driver both MF and HF, crossovers, Uni-Q steel can, added steel can CLD bracing, MF acoustic damping, LF acoustic damping
  
  
• The 2022 Uni-Q has new tweeter, including motor, voice coil and tangerine, tweeter gap damper, MF motor and voice coil, flexible decoupling chassis, surround, spider, thinner lead braids, and the addition of the metamaterial absorber for the tweeter, together with the waveguide.
  
  
• Both the 2022 and 2014 Uni-Qs have Neo magnets on MF and HF. The 2014 Uni-Q has a pot design for the MF motor while the 2022 Uni-Q has an outer ring design for the MF motor to fit the rear tweeter waveguide. The main goal with the new MF motor was to flatten and widen BL(x) and reduce Le(x) for a similar or higher BL(0).
  
  
• The Reference 1 Meta vs non-Meta THD% is when both systems are doing around 90dB around the midrange at 1m. The frequency response is not the same in both so a more correct and fair comparison would have been to match the SPL with EQ but given the results I thought this was not needed. I did go back through my archive and found the measurements of both speakers, I used the line standards so I’m pretty sure they’re both representative. They’re matched roughly across the upper midrange but the Ref 1 Meta has a bit of a tilt in its response so this meant 1dB more of voltage. Even so, the THD difference across the MF is what I put in the whitepaper. I’m sorry this doesn’t align with the Reference 5 in the 2014 white paper. If I have time, I could have a look into this. There is even the chance the 2014 Ref 1 line standard was broken and I was unlucky. Hopefully not.

• The inductance plots do raise an eyebrow I agree. It didn’t occur to me to have a closer look at the 2014 whitepaper when writing the 2022 one, but the 2022 results shown are using our latest FEA steel characterisation, especially permeability, as well as the algorithm to calculate this, which has been a big development, so these results should be much closer to reality. A lot has happened here in 10-11 years since the 2014 whitepaper in terms of our FEA capability so there’s a chance those results are not viable anymore. We did find the model of the 2014 Uni-Q in our archives but it’s a bit difficult to piece together what happened forensically. In any case, I had a look again at the FEA simulations we did of both motors in 2022 to get the results shown in the whitepaper and they’re exactly what you see there. I think the important thing to see is the relative values in both motors, where the same process has been used to obtain results.
  
  
• The spider resonance graph in the 2022 whitepaper is an FEA result, what happened here is that the stiffness of either the suspension or the surround material (or both) must have been lower in the simulation than in reality, then that’s why you get the resonance at a lower frequency of ~650 Hz instead of ~900 Hz, but the effect of the resonance on the SPL is definitely there and is comparable between the two models since the only variables changing are the geometries of the soft parts (and the motor excitation). In the measurement shown in the 2014 whitepaper this would be the kink at around 900 Hz. You can see this resonance in Amir’s data in the impedance as you pointed out and a bit in the Uni-Q nearfield. I’m afraid the 1m SPL curves are too ragged for this to show clearly. The 2022 MF has a much smaller suspension so its mass is lower and therefore its resonance is higher for a similar material, so it’s still there but at a higher frequency and doesn’t show in the SPL as clearly, which was the idea behind the improvement.
  
  
• I took samples this morning from both drivers from my cupboard and measured them on the 2pi baffle. I didn’t make the baffle smooth so HF might have suffered but you can see the resonance in the 2014 MF SPL.

• For the flexible decoupling chassis, the data that you see in the graph are actual laser velocity measurements I think on the side of a Reference 1 cabinet near the MF cavity with a prototype Uni-Q with the rigid chassis and then the flexible one and with and without damping. You’re quite right that given the enclosure surface area a small vibration might excite air efficiently. As for how clearly that shows in the SPL, we could have probably produced an SPL graph from a complete system FEA simulation but that’s a bit too much work for one graph and unfortunately even the cleanest anechoic measurements are ragged enough (or not anechoic down to that frequency in our case) to make small resonances look obvious, even if they’re well present. We thought laser velocity measurements on the cabinet were an acceptable way to benchmark this one.

I hope this clarifies your questions to some extent. I understand it can be frustrating when one reads technical texts where there are mistakes or inconsistencies. We always welcome feedback and this is useful to make sure the next white papers do their job well.

Lastly, if you get a good deal on 2014-2021 Reference speakers, by all means go for it, they’re excellent speakers and we’d never trash on them, quite the opposite.

 

[capslock]

Hi David, I never expected a reply from KEF, let alone from the lead engineer, and I am floored and humbled that you even went back to check data and remeasure drivers. I have thoughts and additional questions on most of your points, but I got home too late to write something that will still make sense in the morning. I'll probably be more like next week before I can get to that.

Let me add that I applaud KEF publishing these white papers. I have learned a great many things from them, and when I had a few questions on the R Meta whitepaper, one of your colleagues kindly answered them. I hope I didn't step on any toes with my first post. The 2022 white paper indeed felt like a bit of a hit job on the old Reference but I couldn't really believe this given your proven culture of openness and excellence in engineering and openness. I'm glad that it turned out not to be one!

 

[AM88]

Unbelievable!!! Only in ASR.
Having someone from a speaker company reply so thoroughly and quickly on such a geeky subject is really gold standard stuff. Also hats off to capslock for bothering to read the white paper and post questions. Inspires me to do the same.

 

[Akronos]

You guys are amazing. I loved reading this.

 

[phoenixdogfan]

Yeah, this is great stuff. Kudos to capslock for his probing analysis and to davidbosch for his prompt and thorough reply. Makes me wonder what the going price on a pair of OG Reference 1's are, because they do look mighty tasty and are bound to be selling at a deep discount.

 

[Vacceo]

Unbelievable!!! Only in ASR.
Having someone from a speaker company reply so thoroughly and quickly on such a geeky subject is really gold standard stuff. Also hats off to capslock for bothering to read the white paper and post questions. Inspires me to do the same.

Dr. Oclee Brown and one of the engineers behind the DSP implementation of the LS60 have provided detailed responses about different inquiries on these forums.

KEF is not just a competent and science-based manufacturer, their designers are some of the most open, responsive and transparent you will find. We are lucky to be able to ask them directly.

 

[DavidEdwinAston]

Should my Sierra LX's ever fail, I will replace them with KEF, purely on the basis of this thread!

 

[obsessive9119]

Should my Sierra LX's ever fail, I will replace them with KEF, purely on the basis of this thread!

Yeah, this thread makes me feel even better about recently buying a pair of LS50 Metas.

 

[tktran303]

One way to further improve the performance of the MF of the UniQ is to put in a full thick (enough) copper sleeve. I've discussed this in passing with Jack.

Or go active and current drive. Or MFC with adaptive filtering. There's already an algorithm that is capable of performing adaptation without chewing too much compute- it's Fast Sub-band Adaptive Filtering

Or at least try any/all of the above in the next prototype. ie. don't discriminate- just use every tool at your disposal.

Once you get to mass manufacturing then you can simplify and cut costs. But for the TOTL product, one must not be afraid to break conventions or discard practices based on legacy/historical ideas.

Let me know if you'd like me to put you in contact with Mr. FSAF...

 

[davidbosch]

I'm more concerned about the internet being full of pictures of these new woofers with thin aluminum drivers that are bend and cracked It's like the motor and voice coil have enough travel(X-max), but the rubber surround doesn't really, and that is where thin aluminum bend and in some cases break.
Can you tell me is that abused units?
Although people swear they didn't overload them.
And that the warranty rarely covers such damage?
In my opinion, the coil should be checked first at a service center, and if it's burned, that's a clear sign that the speaker is being abused and that off course can't be covered by warranty conditions.
I would really like someone who is familiar with the fact and is a member of the company, even better an engineer, to explain to me what the problem is with the thin aluminum or the reinforcements that were obviously supposed to strengthen the diaphragm but created additional fatigue when bending the already thin aluminum?

Hi @Marin there is already a different thread on this topic and @jackocleebrown VP of Technology at KEF replied with relevant information: https://www.audiosciencereview.com/...have-a-serious-design-flaw.61187/post-2242261