Monday, August 3, 2015

Calibrating the Super Tweeters

Not in all the 5 years that I have owned the Elac 4pi super tweeters have I set their level by a measurement comparing their output level to that of the Acoustats.  I have set their level by ear (of course) and by overall system response measurements, but those don't give very much detail about the high frequencies where the super tweeter is operating.  I high pass the super tweeters at 20kHz and the Tact RCS spectrum analysis I have most often done has one point at 18kHz and one point at 20kHz and I wonder about how accurate they are since microphone angle has a huge effect on such frequencies and for stereo measurements I point the microphone straight forward.  Also the Tact microphone calibration itself has been called into question.

It is not easy to accurately measure levels at 20kHz and above.  But for the purposes of determining relative levels, it was simple enough to temporarily change the high pass to a lower frequency and test some signal that is easier to hear and/or measure.  I had been thinking about doing just that for several years, and that is what I finally did last Sunday.  I made a 30 minute aiff file with a steady 14080 Hz tone (A9).  I figured this was good to test tweeters: high and away from any low cutoff, and yet still not too high to be dangerous or inaudible to me.

I used Sox to make the file with the command:

sox -n a14k.aiff synth 30:00 sine 14080 sine 14080 gain -20 rate 44100

For some reason that wasn't acceptable to Sonos.  So I ran the file through Triumph, which had no problem re-writing it.  Then it was OK to Sonos.

At 73 dB playback level (6 inches from the Acoustats) it was just barely audible to me.  At about the same distance by eyeball, the Elacs measured about 3dB less, about 70dB, in both speakers, with the DCX crossover set to 10khz.  (The Elacs may also have their own internal high pass at 4kHz.)

It was difficult to do this near field measurement because it is highly dependent on distance, and I don't think right at the surface is valid (though…it might be as good or better than what I did).  I was thinking of attaching a ruler to the Galaxy SPL meter, but then I decided this measurement was good enough.  I was very pleasantly surprised, almost incredulous actually, that I had set the tweeters this close to the level of the Acoustats (and that is a very long story…they've been set all over the place on the DCX and further confounded with the Parasound amp's own attenuator, and a pair of Harrison Labs attenuators that I may have used different versions and sometimes had turned the wrong way).  By this primitive measurent, it seemed likely that I had not been setting the Elacs way too high, as I had been doing at one time and feared I was still doing.

It did occur to me that any near field measurement as I was doing was likely to be unimportant because the Acoustats have far more total radiating area.  By that standard, the Elacs might need to be set not just 3dB higher but 10-20dB higher.  I couldn't really figured that.  I tried measuring from my chair but the it was hard to get a good measurement without cranking the tweeter level much higher and I decided not to do that this time.

For awhile I tried setting the tweeters 4dB higher.  But, just to confound things further, I reset the crossover point down to 18kHz and I'm not quite sure if I had previously used BE24 but somewhere along the line I did that too, now that I was no longer worried that I was overdriving the hell out of the Elacs.

Well this indeed seemed to clarify the Sax in Jesse J's Lovesong, which had been seeming a touch dry on my system as compared with Luther's.  I have often found (or I should say, I though I found in sighted listening tests) that the Supertweeter actually removes harshness and glare, and that seemed to be the case here too, at first.  But then I did find it overbearing, and dialed the super tweeters back to 2 dB higher than the previous level (see below).

Well after this I couldn't stop playing with the adjustments during each thing played because it didn't sound or figure right one way or another.  One thing for sure, with the crossover reduced to 18kHz, the level raised 1-4dB, and the slope changed to BE24 or LR24, the output of the super tweeters has become clearly audible near the speakers themselves, which wasn't necessarily true before.

And it seemed like I was hearing huge differences with the slope changes.  BE24 did have a clarifying intensity, but also seemed to add some roughness (possibly boosting the 6kHz region, or causing comb filtering at more audible frequencies).  It was instructive to listen to the tweeters without the Acoustats playing, then the differences in the sounds of each option was far clearer.  The LR crossovers sounded by far the most refined, with LR48 the most refined of all.  I believe I had previously been using LR24, which reduces the super tweeter output to nearly nothing (but what there is is high high high).  But with LR48 at 18kHz, the effect sounded very top heavy.  The BE24 and LR24 seemed to have the least top heavy sound, a natural balance.  I ended up with the LR24 which sounded good by itself and also blended well with the Acoustats giving a very clear yet refined sound without coarseness.  The current adjustments as I am writing this:

LR24
19.2 kHz
-3.3dB
6dB attenuator

Note that this was -2.3dB by the old standard where I was boosting the DEQ for the panels by 1dB.  That seemed to work weirdly I determined.  With 1dB of boost, an input of -2.5dB caused peak clipping at the digital output.  I dialed back the boost to allow higher levels (and Monday I have been using +3dB on the Tact (96.9) when playing SACD from the Denon which peaks at -4dB).  I have to make sure I'm not loosing any digital gain because I need pretty much full digital level out of my DAC's for a good loudness.

(The difference between the output of the Onkyo which now powers the Aragon, and the DCX which powers the Parasound, is on the order of 12dB.  Oh, and the Acoustat 1+1 are about the most inefficient speaker ever, in the 70's!  Whereas the Elac speakers my tweeters were matched to were in the upper 80's.  So actually a combined attenuation of only 9.3dB seems to small, intuitively I would have expected 24dB or far more, and before these tests I wondered if I wasn't setting the level way too large and causing a huge ultrasonic peak.  Update: now I wonder if the Elac crossover was at 14k, and my measurements off by 6dB or more as a result of that--in which case I now have a slightly larger ultrasonic peak than before, but in which case it must not have been as large as I feared anyway.  I need to retest with higher frequencies.  And I should record amplifier drive levels and push the levels harder now that I know the Elacs can officially handle 400W continuous and 600W peaks, or at least a later one did.  So I should be able to reach above 80dB and make proper listening seat measurements.)

So when dialing back the DEQ boost to 0, I reduced all the DCX settings to compensate, putting the tweeters at -3.3 from -2.3.  When the day had started it had been set to -4.3.  So this was a 2dB rise as seemed conservative from measured levels, with twice that much sounding too much.

This is almost like having super tweeters for the first time.  You would have thought I'd do this kind of calibration the day they arrived.

The 19.2kHz LR24 seems intuitively consistent with the limitations of the Acoustats FWIW, just where a super tweeter should fill in.  20kHz was probably too high.  I suspect the actual rolloff of the Acoustats is close to 24dB/octave also, including 6dB/octave from the resistor in the crossover, the trannies, and the mass of the speakers.

Also contrary to earlier fears about one being bad from ribbon warping, both super tweeters measured the same in level and sounded the same.  (I've seen some ribbons even more warped than my left tweeter.)

Here are the specifications for the Elac 4PI Plus, which I think was a later model.  Mine is more impressive looking, and I think a very similar earlier model, except it did not extend to frequencies higher than 35kHz as later models did, but possibly extended lower, perhaps to 4kHz.  Well according to these specifications, not for my exact model but similar, the maximum power handling is 400W continuous and 600W peak.  They should be safe with the current Parasound HCA-1000A amplifier, but maybe weren't safe the the Acurus A250 amplifier I was using before which can produce 500W or more into 6 ohms rated impedance.  At least once, due to some technical error, I was clipping the A250 into these speakers at peak.

Other specs show sensitivity variable from 84-92dB at 2.83V (this model had sensitivity and crossover controls, I believe mine has fixed crossover but I'm unsure of the frequency).  The crossover is selectable for 10, 12, and 15 kHz.  Frequency range 10-53 kHz.  Impedance >10k ranges from 8-3.5 ohms.  Weight is 4kg (about the same as mine).

Here is an advertisement for what appears to be my exact model, the Elac CL 4Pi Plus.    I had been afraid it was only sold as part of a specific full range speaker, but here it appears to be sold specifically as an add-on module.  The weight is 5.5kg, frequency response 3.5k-35kHz, power handling 400/600W,sensitivity 88-91dB, minimum impedance 7.5 ohms at 25kHz.  It says Crossover Frequency 7,000Hz and I believe that refers to a built-in crossover which is not adjustable in this model.

WRT Bessel "filter" (it's actually a class of filters some say) something like that may form the basis of the crossover for the Spica T-50, which has excellent time response but ripple in the frequency response around the crossover according to some.  Because of ripple, it hasn't generally been recommended for crossovers.  The "bessel high pass" does not retain the perfect linear delay that the theoretical low pass does, it's merely the "maximally flat" case given the amount of phase shift which must occur.  Actually since LR24 has Q of 0.5, and BE24 is actually slightly higher, the LR crossover would have more gradual phase shift I would think, but I'm thinking it might be a more constant phase shift.

In my case, given that I am correcting the low pass of my panel speakers, I want a matching time response around crossover to that of my panels.  If the panels roll off like a pair of second order butterworths in cascade, and they probably do something not entirely unlike that because of cascading rolloffs in the transformer primary and secondary circuits, an LR24 or similar high pass on the tweeters would provide the correct inverse time response, and the flatter Bessel filter would be too-flat in the time domain for correct cancellation and cause some comb-like responses.  Perhaps that was the grundge I thought I was hearing.  But I would find it hard to believe I correctly sorted that out by ear, and I still suspect my selection of LR24 was more related to loudness than anything else, it was less loud that the BE24 but more loud that XX48.  Perhaps another crossover function would work better at a different loudness level.  I have not tested that.  But what I am doing seems intuitively correct or close to it, and sounds good.

Update:

[This was held up in the draft queue unexpectedly a long time.  Perhaps because all too soon I was fiddling with things again…  And then the re-cabling changed all the levels.  And so on.]

After a long time with LR24 on the super tweeters, I decided that was too hot.  So I rolled it up to BUT48, which is like the next step less rolled up than LR48--the historical setting until a couple months ago.

When I re-cabled to permit use of the Audio GD Dac 19 on both amplifiers, I chose to increase the gain by 2.3dB in the sub and super tweeter crossover rather than decrease the gain for the panels.  Which is all to the good as far as gain structure.  So what was a -3.3dB on the super tweeters is now -1dB, not because the relative level has changed.  Also the subs are now at -11.7dB.

Speaking of subs I felt that the SVS were contributing too much above 80 Hz.  So I made the crossover asymmetric as I often have.  So I've messed up the calibration I did a few months ago, and this probably isn't the right way to do things, but I think I like the sound better, maybe.  Anyway the SVS are crossed with LR48 while the Acoustats are crossed at LR24, both still at 80 Hz.





No comments:

Post a Comment