EMM-6 response, and system response, corrected

All my previous measurements of the EMM-6 microphone with various tweeters were wrong, because I had the Tact analog input sampling rate set to 44.1kHz instead of the 96kHz that I normally have it set to.  On Monday morning I started to think about the steep rolloff above 20kHz, and it occurred to me that could be the brickwall filter for a 44.1kHz sampling rate...and it was.  

For the longest time I had thought about something else, the bandwidth of the ATI 1502 power amplifier I use on the supertweeters.  It does not have quite the bandwidth of my Hafler 9300, Parasound, or Krell amplifiers.  But the power bandwidth is specified as 40khz, and I believe the small signal bandwidth is higher, 60kHz or so.  And that would be just the 3dB down point.  But even when Class AB power amplifiers roll off, they generally do not roll off like a brickwall filter.  Hmm, brickwall filter..

With this change, my system now peaks around 22kHz, and by the end of the calibration file for the microphone at 25kHz, the response is just a bit lower than the peak.  Then, where the calibration file ends, you can see a 4dB drop because there is no more calibration past that point, and that's where the serious rolloff begins.  The true response above 25kHz is much higher than shown, the microphone is seriously rolling off by that point without any applicable calibration.  (I intend to fix this by adding in an extrapolated microphone calibration up to 50kHz.)  Note that I also made this measurement with 96kHz sampling rate in ARTA and Focusrite, which is getting resampled by the Tact at 96kHz.  Most previous graphs were made with 48kHz 

The system now shows "useable" response up to 40kHz, which is barely lower than 12kHz.  If I had  calibration to 40kHz, it would be much higher than 12kHz.

FWIW the bass response is also "useable" to the limits of measurement shown here, about 13Hz, and probably down to about 8Hz or so.

("Useable" response means noticeable or something like that, which I would interpret as being roughly within the same  +/- 20dB window as the rest of system response.)

This measurement also shows the effect of yet another time alignment, performed on the evening of August 8, after I decided the time alignment done a week earlier and reported here a week ago, was wrong.  Some of the ideas in those articles were probably wrong also.  I plan to write an update on Time Alignment later today.

But you can notice here how the response from about 40Hz-300Hz is not perfect but stays within a clear range with uniformly spaced bumps, that are fairly small as these things go.  With something like octave averaging it  would be virtually flat.

This measurement was made with 10 averages.  Notice also the 0.1% distortion at 1kHz, rising gradually above that and below 200 Hz, but not above consistently above 1% until below 15Hz or so.

Testing the EMM-6 Microphone

After the incorrect polarity shown in the ARTA time record of my speakers (it appears that ARTA generates an out-of-polarity transient, and doesn't correct this by default in the Time Record display), the other curious issue raised in recent measurements is that my tweeters appear to peak just above 20kHz and then response plunges down after that.  But I had chosen the Dynaudio D21AF tweeters precisely because they are specified to 40kHz and virtually flat to that point.  What gives?

[Update: What I didn't discover until after writing this post: I had set the analog input sampling rate on the Tact to 44.1kHz, so obviously there was a brick wall filter at 20kHz, the effect of which I attributed falsely to the microphone.  Normally I keep all sampling rates set to 96kHz, but on some previous occasion I had reset this one, which I rarely use except for testing purposes, to 44.1, or a power interruption occurred and it went back to the default.  Moral: check everything, and I mean everything, even things you "never" change or forget about.  And expect that a lot of your serious "findings" are really the result of technical errors.]

I have verified that the Focusrite Scarlett has essentially flat response at least to 23kHz in the 48kHz sampling mode.  I have only just discovered that the Scarlett also supports higher sampling rates, up to 192kHz, so I would expect it to have a cutoff even higher than 23kHz.

But could it be that my D21AF tweeters have aged, or the response graph shown in the litereature was wrong?

I also measured the VIFA tweeters I used on the back side, which are supposed to have fairly flat response well above 30kHz.  Those also appeared to show plunging response above 20kHz.

I suppose those might not actually meet their specs either.

I was lured into the belief that my testing microphone has response to at least 25kHz, because the calibration file provided by an independent calibrator goes that high.  The microphone, btw, is a Dayton Audio EMM-6, calibrated by Cross Spectrum Labs.  (The original NTI microphone was not available.)

So I pulled down my Elac 4PI tweeters from storage.  These are supposed to have flat response to 35kHz, and they very nicely made, and it's possible I might start using them again.  The new steeper crossovers I am using might prevent stimulating the metallic resonance around 6kHz.  Rapping on the speakers, there is no indication of any metallic sound from the ribbon foil only a slight bell-like sound from the round top cover, which looks like it could be removed which might enable me to straighten the foil on one unit.

Once again, I put the EMM-6 within a few centimeters of the tweeters, right up to the foils.  And measurement still showed plunging response above 20kHz.

I'd say I've proven that the plunging response is caused by my measurement microphone, despite the calibration to 25kHz (and I am using the calibration file in ARTA, which only shows a few dB loss at 25kHz).  Online sources do show the EMM-6 generally rolling off fairly fast above 20kHz, but only about 10 or 15 dB by 30kHz.

In this review, the EMM-6 measures fairly closely to the Earthworks M30 up to about 30khz.  But why does the Earthworks, supposedly flat to 30kHz, so why does the Earthworks show 10dB loss at 30kHz?  These tests were done just like I am doing, stick the microphone as close as possible to a supertweeter.

This worries me in several ways.  What if I were to get an M50, as I hope to do soon, and find it measures no different?  I'm thinking what is going on here is that the electret element in the EMM-6 has deteriorated in such a way that the high frequency extension has fallen.

I have a few Bruel & Kjaer microphones with measured response up to 50kHz.  But I've always wondered how to use them, and it now appears they require being mounted to a Bruel & Kjaer preamplifier wand which supplies the unusual 200V polarizing voltage.  I do happen to have 2 funky surplus Bruel & Kjaer 2209 sound meters which have a built-in preamp and everything.  One of them actually works.  But for those, I had never figured out how to connect any cable to the output jacks which are most unusual.  They look like banana sockets but are smaller.  I once ordered a Bruel & Kjaer plug, a JP 0101, but it was too large.  It was very expensive--it cost me about as much as the surplus meters themselves--and I felt devastated.  At that point, years ago, I gave up.

Today I believe I have figured out that the plug required is the Bruel & Kjaer JP 006.  This is actually documented in the 2209 manual but somehow I overlooked it.  (It is not on the specifications page, but in the text describing the output and other jacks.)  I have now ordered one of these rare plugs from eBay.

I am not lusting over an Earthworks M50.  But given the cost, and the excess amount I've already spent on audio gear this year, I'm going to hold off until meeting some personal goals.

I put a fresh 9V battery in my Mitey Mike, and it's completely dead.  I bought it over 15 years ago but never even took it out of the box until last year, when it turned up as I was cleaning my climate controlled storage building.  To determine if problem was preamp or mike I angled the circuit board out and connected clips to the microphone and fed those to the Scarlett input, turned all the way up.  Still no signal on shouting or tapping or running sweep tests.  Finally, I noticed that just touching the microphone input terminals on the circuit board caused bit pulse.  It looks like the amplifier is working, but the mike is dead.  It could even be that I bought it this way and planned to replace the microphone, I recall seeing a loose tiny condenser about 10 years ago.  But like many projects, everything just sat until the related pieces are hard to bring together.

Then I tested the "IMP" microphone which I bought with LAUD3.  Online sources now describe this microphone as electret.  I simply hooked it up the the Scarlet, with level turned all the way up, and ran sweeps, shouted, and tapped.  No signal at all.  It might have required polarizing voltage, but the Fiji card is only equipped with 5V, so that's likely what it would have needed.  The 48V supplied by Scarlett might burn it out.  In fact, I recall this issue coming up before, it could be that I incorrectly attached the microphone to the Fiji with the 5V turned on, when it should not have been turned on, so actually I burned it out way back then.  Anyway, I dared not turn the 48V on now, for fear that would be way beyond it's safe operating region.

Likewise I'm somewhat afraid to hook up my Tact microphone to the Scarlett.  I recall the Tact used some lower voltage, like 15V.  48V might fry it.  I'm not going to risk this unobtanium microphone.  I always found the Tact measurements confounding, but still interesting, and all of my earlier time alignments were based on them.

So now I've got to focus on the personal goal I've set for myself before getting an M50.  That is, to read and perform the "Tutorial" for the HP audio spectrum analyzer that I bought this year.  The unit is powerful but not totally intuitive or self explanatory, and otherwise will just sit there for decades like a lot of my stuff.

I've basically "done" the first pass on crossover adjustments AND time alignment.  The fine details above 16kHz, for which I need a better microphone, are probably not that important.  I'm certainly getting response to 20kHz if not far higher, and the peak at 20kHz I'm seeing may also be bogus, since I see that on my Elac supertweeters just as much if not more, and without the small value coupling cap intended to raise response at 40kHz by 6dB on the cloth dome tweeters (which would be a tiny amount compared with the 20kHz peak I've recently been seeing, but may well be microphone error).

Focusrite Scarlett Input and Output Polarity are Correct

Tests over the past two nights showed the Focusrite Scarlett 2i4 has correct polarity on inputs and outputs.  The world of Focusrite Scarlett using musicians can rest easier now.

To perform these tests, I used a modified version of the SpeakerPop signal created for the SpeakerPop app.  My modified version has lower amplitude (safer) and no DC offset (safer).  I call this signal SpeakerPopSofter.

The SpeakerPopSofter signal looks like this in Audacity:

It looks like an inverse (downward sloping) sawtooth wave, but you need to examine the time scale.   The upward spikes only occur about once per second.  Most devices do not have flat response below 1 Hz, so instead of uniformly ramping down, as the digital signal does in Audacity, they will drop to zero at a rate determined by the low frequency response of the device.

Recorded by the ARTA Triggered Signal Recording feature, the results looked exactly like they should have.  I first recorded using a balanced cable from my Tact preamp into the Focusrite, and then with an unbalanced cable.

I played the SpeakerPopSofter signal through a Sonos Connect digital output, into my Tact 2.0 RCS Preamp and through its analog outputs (which I rarely use except for testing purposes...I use the AES digital outputs otherwise).  The Tact has both balanced and unbalanced analog outputs.  I tested those over 14 years ago to verify they are correct.  (The Tact also has a polarity control in the Menu.)

The balanced connection was direct from the Tact to the Focusrite through an XLR terminated cable made by Blue Jeans Cable.  For the unbalanced cable, I used an RCA to mono 1/4 plug adapter to plug into the combi inputs of the Focusrite.  I zoomed the entire 5 second recording to full scale, and in both cases the Speaker Pop signal appeared in positive polarity, with downward fall suggesting approximately 10 Hz multipole low frequency cutoff.

I strongly expected results would not work out this way.  Based on previous tests, and what seemed like reasonable assumptions, I expected to see the input polarity inverted.

But when possible one should always do the tests instead of relying on expectations.  Just before this polarity tests, I noticed I was having a problem with my bedroom video monitor playing through HDBaseT setup from my computer in the kitchen.  The video colors on my desktop were periodically shifting.  I'd seen this a few weeks ago but then the problem went away, I figured it might be caused by sunlight hitting the back of the monitor triggering some "autodim" feature, or it could be the 2 month old HDBaseT unit, which caused hum on the first system I used it so I swapped it to the bedroom system where the TV only connects to the audio via fiber optic, but perhaps the new HDBasteT unit, which I paid $350 for, had a fault beyond the lack of total ground isolation that was causing the problem.  I think it would still have warranty but especially now, it would be a hassle to ship it back for replacement, if they'd even agree it was faulty.  And then, the ultimate fear, my 4+ year old Samsung TV might have an issue!  I'd want to sort that out quickly as I might still have a few months on the extended warranty.  BUt mostly, I suspected the new HDBaseT unit, under the idea that what has changed most recently is probably the culprit.  Well, I swapped the new and possibly faulty HDBaseT unit with the old one that I had used in the bedroom for years, and, guess what, the problem went away.  Typical audiophiles would call it a day after such a test, and never use the now "proven" faulty unit again.  But one should never assume that in the process of swapping things other things might have changed too.  Maybe it was just the power cycling!  Or maybe it was the time of day, or the temperature of the units involved that shifted just enough during the tests, or the quality of the AC power.  Anyway, there are an infinite number of possible relevant tests one might do, depending on your lack of confiedence in them.   But the easiest would be just to swap back to "A" after trying "B" after "A."  Making the minimum reliable test sequence A/B/A, never just A/B.  So I did swap back, and guess what, the problem went away!  Now, sadly, it may come back, and if it does come back I might be inclined to call it done at that point.  But in this case, there's one more fact I knew.  As I was starting to replace the units, I started pulling out the HDMI cable, but then decided I didn't need to change that, I'd just use each unit with the HDMI cables they were currently with.  But at that point, I started plugging the HDMI connector back in, and it might have been a little crooked.  So THAT might have been the problem right there, and perhaps even the first time.  I've ordered a higher spec HDMI cable whose connector might be more secure.  So never jump to conclusions, always do A/B/A, and try to remember if not write down every last detail.

Because there is still a possibility my speakers are out of polarity, as the SpeakerPopSofter signal was playing, I ran the SpeakerPop app on my iPhone.  It indicated positive polarity everywhere, strangely even in back of the Acoustats.  So I turned off the subwoofers (which are behind the Acoustats).  Then I measured positive polarity in front of the Acoustats, and negative polarity behind the Acoustats, just as it should be.  I hadn't actually tested the polarity in about a year or so, and many changes have been done since then.  (However, I always test or use cable marking to ensure speaker cables are connected in correct polarity.  Most of my equipment has been checked for polarity at one time or another, even my Oppo UDP 205.  The only thing I've ever found having incorrect polarity is an iPod.  Not trusting my hearing in these regards, I perform only objective tests, using an oscilloscope if possible, but speaker outputs are remarkably hard to interpret with a microphone and oscilloscope because of phase shift and reflections, so on speakers I generally use the SpeakerPop app.)

As this proved to my satisfaction that the input polarity of the Focusrite is correct, I then believed that it must then be that the output polarity is incorrect.  However, that was disproven on the next day.

This time I connected an oscilloscope to the unbalanced outputs of the Focusrite.  I played the SpeakerPop signal using the Groove Music app (the default music player when you click on music files) on my Windows 10 laptop.  The Focusrite had already been selected as the audio output, as I verified by clicking on the volume button.

However, I immediately discovered that outputs 3 and 4 had nothing but noise on them.  Only outputs 1 and 2 had the speaker pop signal.  I used Input 1 on a Rigol oscilloscope to record the signal, and I verified that Input 1 was not "Inverted" in the input dialog.

At this point (haha) I believe it is the ARTA program, for some strange reason, that puts out signals in inverted polarity.  I had long discounted that possibility, assuming ARTA was the most technically sophisticated and therefore likely correct item in this story.  But apparently not.

However, I do not recall which outputs I previously used when testing the Acoustats!  Possibly I used the 3 and 4 outputs, which did not put out a signal using Groove Music.  I will need to check that using ARTA.

Update: I must have used output 1 or 2 (first left/right pair on the 2i4).  The second set of outputs does not put out any signal using ARTA.  I have also directly verified that outputs 1 and 2 put out the ARTA "Transient" signal in negative polarity, using my oscilloscope.

So the polarity inversion when using Triggered Signal Recording is a "feature" of ARTA having nothing to do with the Focusrite Scarlett 2i4.

Several Nagging Questions

While doing the time alignment, it was beginning to bother me very much that the transient response shown by ARTA was out of absolute polarity.

Numerous previous tests have shown my Acoustats, as well as my whole system, to be in correct absolute polarity.

I was also beginning to be troubled by the apparent rolloff above 21kHz.  Response just plummets to the 23kHz cutoff required by the 48kHz sampling rate.

So I thought I would investigate both of these problems with loopback tests, simply feeding the output of the Focusrite interface back into the input, bypassing the system and microphone.

This showed that while (curiously) the response is not perfectly flat, but there is no evidence of rolloff to the highest frequency ARTA looks at, in fact the response is still rising at 23.5kHz where the plot ends.  The bass is still rising to its cutoff around 12Hz also.  Distortion is negligible compared with the speakers.

So ARTA and the Focusrite interface are not showing any rolloff above 22kHz.  Does the computed step response show polarity inversion?  Not at all (but actually, I needed to check the transient time record, as that is the speaker measurement that shows inversion, the computed step response never does).  This is what the step should look like with close to DC response you will never see on speakers:

OK, so now I remembered to check the time record with a transient:

It's out of polarity, just like my speakers when I measure them this way.  It appears somehow the polarity inversion is caused by the measurement, though currently I cannot tell if it is caused by the Focusrite interface or the ARTA program.  I suspect the interface most, which may have its XLR inputs wired reverse from US standards.  Actually I'm not using the XLR inputs in this test, I'm using a mono phono plug to RCA adapter plugged into the combi plug.  I would think the convention of tip being hot would apply regardless of the XLR wiring, but maybe not.

Also notice the digital ringing, which makes this transient similar to the one I recorded on the speakers, except I had the Acoustats in reverse polarity so the transient looks like normal polarity.

I then tried recording the "Speaker Pop" signal, which I have slightly modified.  This signal is used by the Speaker Pop app, and consists of a series of steps followed by slow recoveries.  In the original version, the pop is full scale and the entire signal is positive voltage.  Not all amplifiers will appreciate this kind of signal, nor many speakers.  I've modified it by reducing the level to -20dB and eliminating the DC offset.  In Audacity, it looks like triangle waves, but it's so slow it's all pops and not waves as such.

Running the time record feature in ARTA, I simultaneously played the speaker pop signal on the Acoustats and the tweeters, which produced this:

It's almost unintelligible, except you can sort of see downward in-polarity ramps like the speaker pop signal.  But it could be mistaken.  Then I realized I should turn the woofers and tweeters off and just look at the Acoustat response.

First of all, it's a pretty clean transient, despite the crossovers being applied.  Once again, the linear phase crossovers.  But also, it has been recorded out-of-polarity through the Focusrite.  This suggests the inputs to the focusrite are where the polarity reversal occurs.  Note that the original signal here is not the ARTA transient, it's the actual modified speaker pop signal, but because of the highpass filter on the Acoustats just quits abruptly rather than ramping down to zero, so it looks like a transient.

To summarize:


1) In normal test the polarity is reversed (using XLR plug from microphone).
2) In loopback test, the polarity is reversed (using 1/4 plug through RCA adapter).
3) With speaker playing positive impulse, Focusrite/ARTA records negative.

Since the loopback reverses polarity just like the normal test, the audio system including speakers does not reverse (unless the XLR and 1/4 inch plug inputs are reversed from each other, which seems doubtful).

This means the polarity reversal almost certainly occurs in the Focusrite input or the ARTA time recorder.   To distinguish those two, I'd need to compare different program or interface.

Still, for a publishable result, I'd want to replicate the reversal using an electronic interface.  There's also the issue of using the 1/4 jack vs the XLR plug.  To be absolutely certain the speaker is not at fault, I'd want to be sure those two have identical polarity.

Now about that tweeter rolloff, I'd long assumed it was caused by the narrowing of the beam of the tweeters.  So to test that, I put the microphone in front of the Dynaudio D21AF tweeters, within 2cm, facing up (with 90 degree calibration file):

Tweeter is relatively far far hotter, no doubt because of this being made right in front of the tweeter.  However, the decline above 20kHz or so is still plainly visible.

Thinking the D21AF tweeters might have deteriorated from 35 years of age, I tried measuring the VIFA tweeters I use on back.

The result was virtually identical.

Note that the microphone calibration file goes all the way to 23kHz, and was made for this exact microphone by a specialist.

I need to test other tweeters with known response above 23kHz.

Tweeter alignment for linear phase crossovers

While the linear phase crossover didn't require much or any change to the subwoofer time alignment, there was a huge change required for the tweeters, as shown by these transient delay measurements:

Midrange transient delay time measurement

Tweeter transient delay time measurement

The tweeter response is delayed 55.3 ms less than the panel response, relative to the delays which had been dialed in during the first time alignment with ARTA in June.

Somehow, I had been thinking  the tweeter would be delayed more, not less.  But now it's obvious it should be delayed less, for one or both of these reasons: (1) the 96kHz plug in runs at a faster sampling rate, and (2) it has way fewer taps, less than 2000 available as it turned out.

I dialed in the 55.3 ms change suggested by these results, and then tried to maximize the cancellation at the crossover frequency.  

This was very difficult, possibly at least in part because I was still thinking about what I had always believed to be the acoustic crossover frequency, 17220 Hz (which is what I have dialed in for the tweeters).  True, I dial 20kHz into the panels, but they roll off before that, and the crossover has little effect on the panel output, they must effectively roll off at 17220 Hz I presumed, because that is thwat I dial in to the supertweeter response.

Little did I figure, or have I ever figured, that the tweeter response might affect the the 17220 Hz dialed in crossover.  I figured the tweeters didn't have much highpass or lowpass near that frequency.  But I should have thought more carefully.  For one thing, I'm using a cap having a 40kHz crossover on the actual tweeter inputs.  (I do this to provide "boost" all the way up to 40kHz, and also protection and hum elimination.)  That is little more than an octave away, so probably has some effect on the acoustic crossover of the tweeters.  As does the response of the tweeters themselves, which I had always assumed to be about 40kHz as in the advertising literature from 1980, but as we shall see...I should have thought of that also, it may be more like 22kHz because of deterioration of the damping oil in the tweeters.

Meanwhile, you will notice in all the attempted time alignment graphs I'm doing, there IS a big notch, but it's way below the 17kHz I was struggling to look at.  It's more like either 12kHz or 14kHz, depending on the graph.  Perhaps there is a separate phenomenon at 12kHz or something, and the real joint acoustic crossover is 14kHz.

Anyway, I was trying not to notice the 14kHz or 12kHz notches as I was making these measurements, with it only slowly sinking in that I should have been paying more attention to them.  Instead, I was trying to minimize the response at 17kHz, which is far into the area where the tweeter output is far away dominating the panel output, and therefore the time delay adjustment has little effect!!!

One thing I decided to do right off was figure out how to make screenshots on the PC itself rather than using my camera.  There are at least two very different ways to do this: using the ARTA "copy" button at the bottom of the page, or what I've found far more convenient, the Windows+PC Screen buttons.  What I didn't think about doing was keeping a separate logbook log of each photo I was taking.  This was a big improvement over the camera approach I used in earlier posts.

First I took a look at the response with the supertweeter turned off, so I could get see the panel acoustic crossover.  The Acoustats are out of phase so the subwoofer crossover cancellation is still shown.

But deciding where the acoustic crossover starts, the 6dB down point, depends a lot on what you define as the baseline.  If I take 1kHz as the basline, the 6dB down point might be only 8kHz.  But that clearly isn't the crossover, it's simply a slight general downward slope, the kind shown in most measurements of good sounding speakers.

Still, ignoring such absurdities, the question is really unanswerable just from this graph.  It could 12kHz, 14kHz, or the 17.22 kHz I had previously assumed.  Restoring normal polarity, the response from the subs down to the supertweeter crossover is remarkably smooth and uniform, then the supertweeter seems to take over around 12kHz.  Somehow that didn't yet sink in to my thinking.

Then, taking a look at the cancellation, with panels polarity reversed, it did not look good at all to my way of thinking.  There was no detectable cancellation at 17kHz if at all, suggesting to me things were way off.

This seemed to be a little better at 17kHz, and I noticed the notch below 14kHz was getting wider--which also looked like maybe a good thing, or maybe not.

But I was worried about going in the right direction based on this kind of scanty evidence.  I hunted around back and forth around the measured transient time with no avail at creating a notch at 17kHz. Finally I decided to look at the time record (note that Step Response was very confusing, time record was much clearer):

The thicker portion is where the tweeter output begins, and it is clearly about 2ms delayed.  Dialing that change in made it clearly better:

But then it occurred to me, the reason which this looks like a dual impulse is because I had the subwoofer turned off.  Turning the subwoofer on made it start looking like a real 2ms transient!

After a few more attempts to improve the tweeter overlap, I got this:

This was looking so much like a real impulse (as I'll show in a later post) that I was shocked.  In fact this would not at all been possible without the linear phase crossovers.  You would not see an outline of the 2ms transient as clearly as this.

So now I went back to the cancellation display (note that all the impulses above are with Acoustats reversed, so there would be some difference if that were corrected) and it looked like very little progress had been made at getting a 17kHz cancellation.  But there was a noticeable dip or something at 14kHz.  It might not be better than a previously measured response, but it was late and I quit here.  Perhaps some different engineering might be useful on this crossover, also considering the rise above 14kHz, which is not really correct, and some more of the fall above 8kHz might be corrected.  From the transiet picture above, it appears that this is already about as good as it's going to get without other changes.

Bass Alignment on system using Linear Phase Crossovers

While I was not overwhelmed by the improved fidelity of my system when I started using linear phase crossovers, I already had a good reason to suspect it would not yet have been optimized very well.  I had not done the new time alignment.

And it turned out, that the time alignment was indeed not good.

Without correct time alignment, there could be phase errors even worse than those caused by the nonlinear phase crossovers.  Since phase errors are essentially the only improvement made by using linear phase crossovers, this would make their benefits moot.

Since it takes some time (ha) measured in days or more to do a good time alignment, I hadn't even done the time alignment I needed to do when I started using the miniDSP's as my crossovers, or when I switched the woofer/panel crossover frequency to 125 Hz and slope to LR8.  So by the time I rolled out the linear phase crossovers last week, I was already behind the curve on this.  Here are the previous alignments I developed in June (using ARTA for the first time):

I started by measuring the leading edge of the midrange, tweeter, and bass outputs as I did in June, using the ARTA time record feature (just like a storage scope, no fancy FFT math) and a "transient" impulse of short width like 1 unit for the tweeter and longer like 100 for the subs and panels.  I later determined that the ARTA transient width specification is in samples, and 100 samples is 1/480 second or about 2ms.)  You may notice that the transient starts downwards.  I later determined this has nothing to do with my speakers, but is caused by an issue either in the Focusrite Scarlett interface or the ARTA program.  When ARTA computes a Step Response, the polarity is shown correctly, but the computed step response is useless for determining driver start time.  The bass response has a leading negative portion which needs to be ignored; it is caused by the low pass filter.  But because of the polarity inversion in the time record feature or interface, the leading negative portion of the bass response is shown as positive here.

Leading edge of panel output

Leading edge of tweeter output

Leading (in-phase) edge of sub output

I then proceeded to try to find the maximum cancellation point with the phase of one side (the Acoustats...since that is easiest) reversed.  I did not record the delay times for each as I was going (as I should have!).  However, somehow I ended up with a very impressive cancellation about where it should be.

Oops, no cancellation, forgot to reverse phase, note room notch at 92 Hz

Showing some cancellation just below 125

Better crossover cancellation near 125 Hz

Too far or wrong direction

Looking Good

Accepted!

While I was doing this, I was measuring "close" to 125 Hz.  It was only later that I realized it might have been better to be measuring at the very obvious notch just below 125 Hz.  That's what probably represents the acoustic crossover.  And I should also use that actual frequency as a phase compensation frequency instead of the nominal crossover value of 125 Hz.

Anyway, the last two are about the same, there may have been others in between and I furiously worked to try to repeat the earlier measurement, and while the second may be .02dB worse at 124 Hz, it's so close that the difference may well be random measurement error.

Notice how different the final cancellation result is from the first.  Clearly I've found the new optimum, or close to it.  I strangely seem to have ended up only 2ms of where I had been before the transient measurement, in spite of the 8ms change that the transient measurement suggested would be necessary.  I'm not sure why (lack of notes).  Here are the newly dialed in delay values:

Since both bass and midrange crossovers use the same miniDSP plugin and number of taps and centering method, switching to linear phase filters didn't affect the relative time alignment between the two at all, as I suspected.  (Tentative conclusion #1.)

I suspect it was the change in crossover frequency, from 100 Hz to 125 Hz, that makes the 2ms difference, and/or it could have been the LR8 slope.  (Tentative conclusion #2.)

Turning on the FIR and filling it with about 6000 taps, has made about an 80ms difference in the delay time, indentical to both bass and midrange.  (Firm conclusion #1.)

The center frequency of the steepest relative notch that appears as the correct time alignment delay is dialed in, represents the actual acoustic crossover frequency (which is otherwise fairly hard to determine).  (Firm conclusion #2.)

The best way to construct a linear phase crossover, is to use an ordinary IIR crossover then add a FIR phase compensator for the acoustic crossover frequency, as measured above.  (Hypothesis #1.)

Absolute Polarity Revisited

Ethan Winer has long argued against the audibility of absolute polarity, with the basic claim that the only reason why some people hear it under some circumstances, has to do with nonlinearity in loudspeakers.  In an early post on the subject, he provides some interesting test files.

More recently, however, Winer has updated this view with the fact that nonlinearity in the ear itself (and not the brain, btw) can make absolute polarity audible.  The IMD created within the ear varies with sound pressure level.  He has some newer test files that demonstrate this view, and show that it is not the brain doing this.

In my opinion, the best research ever done was published in the Journal of the Audio Engineering Society in the 1980's by Lipschitz and others.  The demonstrated that audio polarity was marginally audible.  It could be heard on test tones, or on recorded solo trombone on headphones and only on headphones.  (The trombone has one of the more asymmetric waveforms.)   On speakers, the differences between tests tones could be heard, but nothing else.  In no case was polarity clearly audible on more complex music.

Of course, high end audiophiles are interested in things marginally audible, though it should not be our primary focus.

But Ethan's later view suggests that individuals with greater hearing nonlinearity might be more affected by absolute polarity than others.  Nonlinearity probably correlates well with age and exposure related hearing loss.  So cranky old audiophiles who insist absolute polarity is one of the most important things in audio reproduction could have a point--for themselves!

However, all my attempts to prove the audibility of polarity changes on a cranky old audiophile who insists absolute polarity is the most important thing in audio ended in failure (despite the subject being 100% confident).  So in most cases, I still believe sighted testing bias is probably the best best explanation, as it likely is for most audio beliefs not founded in objective audio engineering and double blind validated audio science.

Sighted testing bias is a variant of expectation bias that applies to audio testing whenever people have independent means (which might not be concurrent) of establishing the test conditions.  Expectation bias is easy enough to understand--you get what you expect.  But often in audio, people didn't actually know what to expect until later, so they may argue that their expectations were not important.  Here the problem may actually be cherry picking results that fit an expected result.  Then people may argue they didn't have any such expectations beforehand, so how could they cherry pick results to match.  But if they only performed a limited number of sighted tests, the effect is similar to cherry picking, they simply stop testing when their final tests and expectations match.  Audio is extremely vulnerable to this as it is well established we may never  hear the same stimulus exactly the same way twice.  For that reason, audio science requires Double Blind Testing, where the tested person has no way of knowing the test conditions during the test, and repeats the test a predetermined number of times with accuracy reaching statistical significance.

All this testing is hard work, and I don't do it much myself.  But I don't claim to be re-inventing audio science either...I more or less accept generally accepted audio science results, only with the small disclaimer that small enough differences may be subject to false negative test results.  So there might be currently unexpected differences, but they would only likely be very small and usually unimportant.  Audiophiles are typically claiming the reverse, transcendent improvements in every recording from changes that have no basis in audio science or engineering.  Claims like this are best doubted until further validation.

My main area of "quackery" concernts the effect of ultrasonics past my 16kHz hearing limit.  I believe they are important.  But I often, myself, wonder if this is actually true.

Other than that, I am mostly merely trying to achieve ordinarily accepted audio engineering principles, such as wide bandwidth, low distortion, and low noise.

I always try to maintain correct absolute polarity.  It's probably not a big deal, but it's an objective detail which might as well be correct.  However I don't obsess over each piece of equipment or every recording--to find the optimal polarity for it.  In my experience, very few pieces of digital equipment have inverted polarity, and not many analog equipment either.

The only digital piece of equipment I'd ever known which reversed audio polarity was an iPod.  I've tested many disk players by Sony, Denon, Pioneer, and Integra, and not found one that inverted polarity, contrary to claims of a friend of mine.  (Who then surmises they have circuits to detect and correct polarity on test signals.)

Perhaps until just now.  The new equipment I am still investigating, it could be unusually important in the annals of absolute polarity...