Siri's Killer Note
Copyright 2009 Troels Gravesen

Singing the note D (293.66 Hz) and why it may pose a problem to loudspeakers.


A few of the recordings used in speaker evaluation. To the left Siri's Svale Band and Mahler's 4th used specifically in this study.

For fine-tuning of crossovers, I obviously have a range of well-known CDs and vinyls to hear how my speakers perform and to correct tonal balance in the final crossover. One of the CDs are Siri's Svale Band (Norway) and on track 4, You Make Me Fell Like a Natural Woman, she goes solo and a few seconds into the song she sings: " - and when I knew I had to face another day -". The word "another" is a "clean" D, 293.66 Hz, and she sings it very loud and most speakers simply can't cope with it - when it also has to be played loud.
To see what was going on with this note I made a FFT plot of the sound and it turned out like seen below. Multiple harmonics reaching beyond usual point of crossover up to ~4700 Hz. Some of the harmonics are even higher in amplitude than the basic note!


The note "D" on the piano.

Left: Siri's note D when singing the word "another". Basic note is 293.66 Hz - with multiple harmonics. The CLIO actually says 295.9 Hz, but I guess she hits the note pretty well. To the right the signal taken directly from the line stage. It's difficult to hit exactly the same moment of recording, but basically we see the same pattern as what comes from the speaker.

This is what it sounds like: Click here to hear wav file (400 KB). Click here to go to page on creating a test track for speaker evaluation.

What I think Siri is doing is actually overstraining her voice, producing a multiple of overtones. I'm not talking artistic quality here. I very much enjoy what she's doing, but for comparison I did the same thing with Lisa Della Casa from Mahler's 4th Symphony, 4th movement. Lisa's vocal is tougher to follow on a FFT plot as we have the orchestra "blurring" the image, but here and there it was possible to "see" what she was doing. First of all she generally sings an octave higher than Siri and most noticeable, the notes are extremely clean compared to Siri's D note. I can play this part of Mahler's symphony loud without speakers getting into trouble. Obviously Lisa's "notes" have overtones creating the particular "sound" of her voice, but not nearly as extended and high in amplitude compared to Siri's killer note.

There are a number of observations in connection to this:
- A number of (modern) drivers with kapton voice coil former have trouble reproducing Siri's note compared to speakers with aluminium voice coil former.
- Small (mid-) drivers have trouble too due to insufficient membrane area creating nonlinear distortion, or - "basically due to insufficient linear volume displacement capability" as a well-known open-baffle constructor once put it when asked why a 4" driver didn't make it in an open baffle.
- A number of fullrange drivers do it fairly well, e.g. Supravox 215RTF, vintage SEAS 21 TV-G and 17TV-something but surely not the Lowthers and vintage Wharfedale Super 8. It has to added that fullrange drivers only do this well when either listened to off-axis or in a 2-way set-up with a point of crossover above the usual 3 kHz point.
- Beaming is another important issue here; that is when wavelength approaches the diameter of the radiating cone.
- I often do both series and parallel crossover in a new construction and I have never experienced one type doing better that the other with regard to this particular problem. Despite this, I feel certain that crossover topology (order of filter) plays a significant role too due to harmonics being spread all over the crossover region and the inability for higher-order filters to deliver correct impulse response.

Statements like these take a few comments because they are based and a limited number of observations as I don't have an infinite range of drivers and I haven't done a large number of experiments, but for what they are worth, here are my observations:

Kapton voice coil formers
With the quest for low compression and proper ventilation, most manufacturers these days use quite long voice coil formers making a significant distance from the actual voice coil windings to the cone. Having a long tube here may create resonances not experiences before where the cone would be only a few millimetres from the windings. The standard two-layer windings provide significant strength to the voice coil former, but having a 15-20 mm thin kapton (plastic) tube transferring the energy from the voice coil to the cone may pose problems. Together with all other moving parts of a driver, an energy build-up (resonances) may take place and be released once mass vs. frequency becomes too high and we see a rapid decline in acoustic output. It may look like this from an otherwise very high quality 5" middriver:

First of all it's a bit of a pain to make a smooth roll-off from this driver at the intended 2.5-3.5 kHz point of crossover and no matter which tweeter I add, it won't do Siri's "D" from 2nd and 3rd order filters, series or parallel, BUT, running it 1st order - with a notch-filter to reduce the 5 kHz peak - does it! Fortunately I had a tweeter with magnetic oil that allowed an - almost - first-order filter and although having to run the tweeter with inverted polarity, it goes very well compared to the higher order filters. Generally it takes darn good midrange drivers and tweeters to cope with 1st order filters.

I have done the ScanSpeak 18W/8545-00 and K00 drivers many, many times and the -00, having an aluminium voice coil former, has always had a better upper mid compared to the -K00, which is fitted with a kapton voice coil former. The -K00 may produce a speedier bass but at the expense of upper mid quality.
There are exceptions to this like the ScanSpeak 15W/8530-K00 also having a kapton voice coil former. I've never had serious problems with this driver and it seems like certain combinations of materials and overall construction of the driver may add to the problem. Maybe the sliced paper cone is more tolerant compared to the paper-carbon composite of the classic range of midbass drivers from SS.

Too small mid-drivers
If the membrane area isn't big enough we may have un-linear distortion. The driver simply can't play loud enough. No big deal, choose a bigger one as a starting point.

Fullrange drivers:
My experience here is limited, but a few drivers I have tested (some vintage) do Siri's D fairly well. Which points to the problem of phase and impulse correctness. The Supravox and vintage SEAS 21 TV-G and SEAS 17 TV-something (from an old B&O speaker) all do it well - until a certain level where the thin paper cones simply can't cope with the power input and the sound gets rough and congested from serious cone break-ups. No wonder. The lack of any crossover in the 2-5 kHz range with the inevitable problems of phase shift and impulse distortion, etc. tells its story. It also appears the whizzer cone fullrange drivers are not the way to go. The mechanical crossover in whizzer cone drivers doesn't seem better than a poorly constructed higher order passive crossover. Often we see sharp phase shift in the treble range from whizzer cone drivers and they do not please the ear.
As mentioned above, beaming plays a role for fullrange drivers and when listened to strictly on-axis they don't do too well either.

All in all an incomplete and complex set of observations.

Experiments on crossover slopes

Above a crossover simulation for a 3-way crossover. During optimisation minor changes were made to the bas-mid section, but this had little impact on overall sound. Overlaying the plots makes a messy picture, but I think it's useful to illustrate the experiment. To the right I have added thick lines to enhance the basic crossover topology for the mid-tweeter section:
# 1 is a 3rd order filter, # 2 is a 2nd order filter and # 3 is - an almost - 1st order filter.

Polarity of drivers:
# 1: With reference to bass driver (positive polarity) M and T have negative polarity.
# 2: With reference to bass driver (positive polarity) M connected with negative polarity and T connected with positive polarity.
# 3: With reference to bass driver (positive polarity) M and T connected with negative polarity.

To cut a long story short:
3rd and 2nd order filters can't cope with Siri's "D" although 2nd order is better than 3rd order.
The 1st order crossover does it significantly better than 3rd and 2nd order. Going through all my reference CDs, the 1st order crossover did all difficult tracks better than I've heard before.
1st order filters here turns out to be close-to-1st-order on the electrical side and close to 2nd LR on the acoustic side. True 1st order filters are indeed very difficult to achieve for almost any driver as they need to be fairly linear well beyond point of crossover.

More experiments:
a. An LCR circuit was applied to smooth tweeter impedance and this has no impact on performance, i.e. distortion measurements and sonically even when driven at excessively high levels (tweeter with magnetic oil used in experiment).
b. Adding an all-pass filter to # 3 to have mid and tweeter connected with same polarity, did not enhance sonic performance either. It did sound a little different maybe due to the signal having to pass two 47 uF standard polyprop caps. I was surprised not hearing any subjectively sonic degradation of treble performance from this - - but had to revise this impression later in final set-up (proper room and amplification).

Updates:
14-01-2010: Moving some 130 kgs (!) speakers and JungSon amp into our living room quickly required an update. The all-pass filter just didn't perform once the XX2 cartridge was in front. My first impression of the all-pass filter just didn't hold. Guess the standard polyprops after all can't deliver the level of transparency needed and at high levels treble got rough. Hmm... So, back to previous crossover version with inverted tweeter polarity and I added the impedance flattening circuit to the tweeter - and a little further attenuation (1 dB). Better. Much better! Which points to the fact that speakers should always be tested in different rooms, with different equipment, etc., before the final crossover is put on paper.

What are we going to make of all this?
Are we going to ditch all speakers not using 1st order filters? I mean, these pages - and most of what is found on the web - are full of speakers with higher-order filters. I don't think so. No speaker is perfect anyway and even the best are full of compromises.
There is an abundant amount of literature available on 1st order filters and for some people it has taken almost religious proportions. I find this website useful to get a clearer view of the pros and cons of filter topology:
http://sound.westhost.com/ptd.htm#intro. Not easy bedtime reading, but filter theory will never be easy to understand.

Most drivers can't cope with 1st order filters due to the huge overlap between drivers and in many cases we risk frying our tweeters when cranking up the volume. I have tried many times during crossovers modeling to see if 1st order filters are possible, and in most cases it can't be done due to phase and frequency response in combination with inadequate placement of drivers, i.e. the front panel would have to be stepped to acoustically align drivers (aligning assumed position of voice coils doesn't work), something that is not always needed, but often is when dealing with first order filters.
Some people think that a first order filter is a coil and a cap and that's all. It isn't, unless we settle for serious compromises. Good 1st order filters are often a serious amount of crossover components to liniarise impedance, to smooth roll-off characteristics and to equalise the usual anything but linear frequency response. Sometimes it pays off, sometimes it doesn't.

Vandersteen is a guy who for decades has worked on 1st order filters and unfortunately his speakers are not common on these shores - because I'd like to hear his constructions. What it has taken for Vandersteen to get this far is stepped cabinets, specially designed drivers, complex crossovers and not least below average speaker efficiency of 85 dB (3a), something that doesn't make more than average dynamic speakers. There's a good interview here, and the "problem" is that the guy is right, but it comes at a price. And let's make one thing clear: The use of complex crossover to realise e.g. 1st order filters does not count for low efficiency. Many people think that complex crossovers suck up energy and reduce efficiency. Not true. The reason for low efficiency is only when we require deep bass from small speakers. View a 1st order filter for a 3-way speaker from Thiel. Quite a lot of components to realise a 1st order filter! Lots of - presumably - all-pass sections, impedance linearisation, etc.
The reason for low efficiency of Vandersteen speakers is due to targeting good low-end extension from a fairly small footprint and to some extent the use minimum baffle. We get what comes from the driver - and no more. Very contrary to the wide baffle systems like the Sonus Faber Stradivari.

Dynaudio has always made a trademark of 1st order filters although they don't talk too much about it any more, maybe because some of their speakers really are not true 1st order constructions - or maybe because the discussion on 1st order filters really doesn't hold commercial value due to the complex issue, leaving most customers bewildered and confused.

Thiel is the last I want to mention of the manufacturers that are true believers in 1st order filters and - like Vandersteen - has done at lot of research in developing drivers that can cope with this.

Unfortunately Dunlavy's speakers are no longer around, another true believer in 1st order filters.

Siri's "D" is not a typical problem. Fortunately most music does not to such extent expose the inability of our speakers to perform this kind of reproduction. But it does tell that the majority of our speakers really do not sound right and that time and phase distortion are issues that should be addressed whenever possible. Based on modeling, a large number of drivers will require stepped front baffles to achieve 1st order roll-off, even sometimes for perfect 2nd LR roll-off.
We have learned to live with speakers where few of these problems have been addressed, and should someone tell you he has made a speaker that does it all right, take great care. This speaker doesn't exist.

This page
- may be revised from time to time when more experiments and observations may require changes to statements and conclusions. Don't think I think I've got it right and feel free to mail your comments.

Response from Peter Wojcik
First of all let me thank you for the advice on the CSM construction, I knew tweeters were very similiar but i didn't realize they were so identical. Thanx again.
Your article on Siri's killers note, I found it very interesting and intreaging. For some reason I always was a fan of low order x-overs not knowing about the problems higher orders would create.
Once you mentioned Vandersteen, red light in my head went off, because a few months ago I was doing a job out of town about 300 km away and as usual after work I would go and check out all of the local electronics stores.. One of the stores I visited had a pair of
Vandersteens Model 1C, so the plan became crystal clear that was gonna be my starting point of 1st order research.
Investigation: Armed with my copy of a cd I went off, and just as I suspected and you predicted the Vandesteen played the track perfectly with out even a hint of trouble. Lucky for me the same store also caarried my favourite speakers:
SonusFaber, Auditor M performed flawlessly, same as Guarneri Memento.
Next went
Minina Vintage it played well but not as good as the other two. And by the way Auditor M  & Guarneri are my all time favourite speakers. And they are based on two drivers that you know so well .S.S 15W and Audio Technology.
Next day i visited another dealer  and I listened to
Totems -Hawk, it performed great but Dynaudro X12 Excite didn't, which makes me wonder if it really is 1st order filter. Since then I have auditioned BW XT2, Gershman X-1R, Thiel CS1.6 and Vienna Acoustics Haydn Grand. They all performed flawlessly, which confirms your suspicion about 1st order filters. I have listened to other speakers from same brand with 2nd or 3rd order filters and they all had same problem in the same spot as you wrote in yout article.
The original Auditor from Sonus faber  used unmodified  version of 15W a Vifa XT and was 1st order filter as well, so i hope you don't get offended by my humble suggestion to try modelind 1st order cross over for your Ellam XT, which by the way I'm enjoying immensly - thank you.
If you come up with some new x-over for that project and are not 100% to your satisfaction I would love to see  it even if you decide not to post it on your website. I never realized that all the speakers I've enjoyed through the years and somehow made an impact on me, had one thing in common --- they were mostly 1st oder filtered.
I hope you come up with some new 1st filters for existing or new constructions. Great work on your part which I have enjoyed so much. Thank you. I am almost finished with the CSM project and as soon as can I will send you some pics of what I have done.

Response from Joe Rasmussen
It's interesting that at least one designer, Troels Gravesen, has heard audible effects from the type of former that the VC is wound on, preferring Aluminium over Kapton. Could it be that the former has a significant effect over the VC's heat distribution? He also notices the same effects on drivers he considers having an inadequate Sd, the cone area. If the conceptual theory is correct, then larger Sd will lessen the audible effects of non-linear VC heat distribution.
Read more here:
http://customanalogue.com/elsinore/elsinore_24.htm