Friday, September 20, 2024

"New" sub polarity is non-inverting

 I strongly believe now that the reversal in sub polarities I did during the Janus 1.0 system tuning was correcting a number of earlier mistakes, but the polarity of the subs is now non-inverted, which is what I intended them to be.

I most recently replaced the plate amplifier in the right sub.  That was a couple years ago, and I decided I'd do it right this time for sure.  So before hooking up the plate amplifier, I connected an extension wire to the sub driver so I could stand in front of it and watch the woofer go in or out as I connected a battery to it.  I had never done that test before.  Once the plate amplifier is connected, you can no longer do such an easy test.  If you try to use a polarity matching app as I have done, you quickly find it is not reliable and even probably wrong at subwoofer frequencies.  This is because room reflections become part of the mix before you've even had one "cycle" at low frequencies.

That right sub has a polarity menu item, and it is now set to "Normal" (it had been "Inverted" for some unknown previous amount of time, perhaps since I did the plate amplifier replacement 2 years ago).  Right now I don't remember why I set it to Inverted or perhaps it was a mistake.

The left sub, OTOH, has its polarity inverted because (so it seems) I installed the replacement plate amplifier incorrectly.  I knew this was the case the first time I replaced the plate amplifier, so I ordered a polarity reversing XLR cable to connect it with, rather than having to take the plate amplifier out all over again (which requires moving a lot of stuff around).  That was about 12 years ago.  I used the polarity reversing cable on the left for awhile, but at some point (while I was still using Behringer DCX 2496 as crossovers, and they have polarity control) I decided on a trick: I'd reverse the polarity on the left and then reverse the polarity to both subs, making them both correct.

In 2013 I started using Behringer DEQ 2496 as crossovers (because they have a digital output option, allowing me to use high end DACs) but they do not have any polarity control (though it seems now that older ones are polarity inverting), and then in 2021 I started using miniDSP's as crossovers.  Somehow in all these changes, I lost track of the fact that the left sub polarity needed to be inverted (as with the inverting cable) to be in non-inverting polarity.  I also replaced the left sub plate amp a second time, but I think I may have made the same mistake as the first time.

The problem I faced recently had nothing to do with absolute polarity (which, for subs, I now think isn't even an issue at all) but with the relative polarity between the panels and the subs.

With the 24db/octave filter, the top or bottom polarities can be reversed without having any effect on the frequency response.  Linkwitz pointed out that the "inverted" phase connection with the 24dB/octave filter has the lowest group delay, but I've always believed in having everything in non-inverted polarity simply because it's possible and more intuitively nicer.

(Although it's possible that at some point I decided to use the inverted sub polarity instead.  It should have had no effect on frequency response.)

I was thinking he 48dB/octave Phase Linear Linkwitz-Riley crossover I am using now has that same property, you can connect the drivers either way and get the same frequency response.  However, I now know that is not true, it makes some difference which way you do it and you get some cancellation at the crossover frequency doing it incorrectly.  So it actually seems that the "Phase Linear" FIR filters I am using don't have the same property in this regard as the normal Minimum Phase LR filters.  Perhaps it's because they don't have any phase shift at all, so the phases are either matching or not, there is no in-between.

Mind you, the crossover is so steep the "cancellation" (which isn't perfect either) is confined to a very narrow band around the crossover frequency, and can mostly be masked with EQ (as I had been doing without knowing it) mostly.

Sunday, September 15, 2024

Janus 1.1

The stereo was sounding great on Saturday night, and I was doing some rare serious listening.  But in the morning the weird looking left channel response was bugging me, compared to the right.  And it had been much better looking, so it seemed in the last post, just before I added the last 182 Hz boost.  Then the lower midrange somehow got all messed up.  I had to solve this riddle.  And by Sunday afternoon my knees were no longer hurting from earlier days of adjusting (even using my 3" foam kneeling pad).

First I dialed out that boost at 182 Hz, which was seeming to be the culprit.  But it wasn't.  With no boost at 182 Hz it looked much worse overall, however the region 500-1000 Hz was perhaps slightly better, mysteriously.  Very slightly.  Not worth it.  Only then I did a good re-measurement of the 182 Hz boosted version for comparison, and it was mysteriously looking quite a bit better than a few days ago just by chance (not EQ), and obviously better with the boost.

Left, 182 Hz Boost removed

Left boost restored, "Janus 1.0", maybe not so bad

I tried half as much boost.  That was neither here nor there.

So then I just tried fixing the issues in the midrange while leaving the boost in place.

There was a depression centered around 400 Hz (it was a deep depression in the measurement I made a few days ago, today it was a fairly mild depression).  And there was also a -3.5dB notch at 437 Hz.  So I eliminated that notch, the the depression looked much better.  I left another -2.5dB notch at 366 Hz untouched because 300 and 400 were about the same level with the 437 Hz cut removed.

Now that I'd given up the 437 Hz notch altogether, I had an available PEQ I could use for a notch at 833, which had been used in earlier smoother iterations, but (mistakenly it seems now) given up at one point as unnecessary.  I might have intended to remove the 437 and ended up removing the 833 instead.

I moved the 903 Hz boost up to 1013 where it had been, but that left 900 depressed and added to the peak following 1khz.  So I found a happy medium where 900 and 1000kHz are about the same, tuning the boost to 957 Hz.  It's a very narrow boost so it doesn't raise the 800 Hz (I didn't change that).  For the smoothest response above and below 1kHz I increased the boost from 4dB to 5dB which gave the flattest response around 1kHz.

With that, I needed to increase the depth of the notch at 654 Hz because that was bubbling up again.  It had been precisely that I before but for some reason I'd then rolled it back.

And that was it, I got a smoothness from 500-2000 Hz which was similar if not better than the pre-182-Hz-boost response I'd measured a few days ago, but without sacrificing the 180-300 region which that boost fixed.  Actually this is about the smoothest I recall ever seeing the Left Channel.  Still not quite as smooth as the Right but pretty close.  It's now good enough that I think I can relax on the EQ'ing for awhile.

Left Channel, "Janus 1.1" EQ

Janus 1.1 PEQ's (only left panel changed)

One notch was a restoration of a notch I had figured out by sweeping and was using earlier, the notch at 833 Hz.  I basically just added that back and it helped so I'm thinking removing it was a mistake.

The boost was moved to a frequency based on the RTA effect and not sweeping this time.  That was what I actually do a lot but not what I claim to be my method because it's leaving the ultimate hinge frequency to a wider than necessary analytical technique.  But I've actually done that in many cases like this before.  Sweeping the mids and highs is not easy at all because the distances from micro peaks to micro valleys (which you would see in a very slow sweep graph, it would be "thick" because of all the ups and downs) becomes a smaller and smaller portion of an octave, and the frequencies you can choose for the PEQ's get wider and wider and there's not much you can do to get really "in there."  It is possible it might be better if I moved the boost a tick or two higher or lower based on sweeping or, gasp, sound.  But it could not get any flatter RTA, since that was being optimized.  I recall when I was sweeping that there was a dip at 903 Hz and also around 960 or so.  When you get up to 1khz, it doesn't sound like it needs boosting at all.  So the 957 may actually be just right and it's certainly pretty close to what I would have chose by sweeping.











Friday, September 13, 2024

2024 'Janus One' EQ adjustment

The basics have now basically been done for the current EQ and level (and phase this time) adjustments.  And I've given it a name to leave a milestone here, "Janus One."

It's sounding wonderful, a huge upgrade, even though I thought it was incredibly good (and basically better than anything else I've heard in a while) before.

I've eliminated the "Boost" aka "Listening Position" EQ.  Now it's called Normal and used for everything except movies--which have guests and use peripheral seating where there is maximum room gain.  Normal sounds way over the top at those positions, however audiophiles might find it ok at a meeting.

I've been able to eliminate Boost because the new EQ is so well tuned.  It isn't "sloppy" like the old Listening Position EQ, which featured boosts on top of boosts in some cases.

It's much smoother overall for many reasons.  A big part of it is the deliberate choice to NEVER use Graphic EQ's (GEQ).  GEQ's tend to increase roughness because they aren't precisely dialed in over existing depressions, and therefore often amplify peaks--and peaks amplify with gusto.

Each way and channel uses only 10 or fewer Programmable EQ's (PEQ), mostly as notches, but with a few highly tuned boosts also where they appeared to be needed.

They were all tuned using an oscillator, frequency counter, and chairside Digital Equalizer (DEQ) with all the critical listening being done in the listening chair.  Measurements were done with 1/6 octave 16k bin 'Analyzer' app, for the first time held in the center head (or sometimes ear) position with a microphone stand and new phone mount.   These are big improvements over previous tunings, where I often sat on the floor in front of the front DEQ's while tuning in the frequencies and deciding on their need for attenuation, usually using the DEQ itself to figure out what the critical frequencies were.

In the process, I discovered that it worked substantially better in both channels to invert the polarity of both subs compared to their previous positions (which in one case had been "inverted" and is now moved back to Normal).  The sub markings are in question, as I perhaps installed both replacement plate amplifiers correctly, but they somewhat agree with the ambiguous Speaker Pop app which becomes very unreliable with subs, which is no surprise because I originally set them up using Speaker Pop.

Anyway, in recap, I think it's good to go over what the change has been.

When I started, the listening position response was very bass shy UNLESS I engaged the "Boost" EQ.  That added a bunch of GEQ boosts to the normal set of PEQ cut used in all settings.  With that selected, I got a pretty rough and tumble bass response in the Right Channel:

Right Channel, Boost EQ, July 13

When I finished Janus One, there was no longer any Boost EQ as I had designed the NORMAL eq to have sufficient bass at the listening position without using any GEQ boosts (but using some very small PEQ boosts).  When I finally got the sub phase correctly set, I measured this:

Right Channel, Normal EQ, September 12

Wow!  Very smooth overall (especially from 20-400 Hz) and closely fitting what I think is the correct room curve.  No need for a Boost curve and not too bad elsewhere in room or house (just a tad too much bass elsewhere which often works like a pleasant loudness curve for background listening).  In one intermediate adjustment I had the bass completely flat with the midrange, but I decided the bass was far too weak sounding, so I raised it by increasing the sub level.

The left channel had been looking much better than the right channel, I originally thought, but by the time I'd finished the right it was clearly no match anymore.  (Now I wonder why I ever thought this was good.)

Left Channel, Boost EQ, August 1

The finished Left Channel Normal EQ isn't quite as smooth as the right is still, but much nicer for the most part.

Left Channel, Normal EQ, September 12

The bass from 20 Hz to 300 is very smooth, but above that there's some roughness.  The only way to fix 400 Hz is with a boost near there, I thought the region around 1kHz looked smoother in some previous measurements.  I try to avoid boosts and they need to be done very carefully, so not done yet.  Careful listening tests might be warranted too.  This includes "PB1" the current latest Panel EQ, which only has one boost in the left channel (but looks like it might need 2 or 3, I think now.)

Looking at it now, I think this could use some more work asap, it's still no where near as perfected as the Right channel.  I was bogged down for a week with the phasing issues so I couldn't quite see that.

While doing this EQ I learned at least one lesson for sure.  Always try inverting the phase between subs and panels, even if you think you have the correct phase.  And take a good measurement to compare them.

And many others I've noted along the way.

One part of identifying this milestone is to write down ALL the details so it's not lost later.

Details:

XLR polarity inverter on Left sub channel
Right Sub set to "Normal"

Left Sub level: -7
Right Sub level: -10

Sub DAC: -6.5
Panel DAC: +9.5

Panel DEQ Gain Compensation: +3.0

Tact Levels:
R 0
L -2.0

Panel DEQ Delays
L: 10.0 ms (design default)
R: 10.0 ms

Sub DEQ Delays
L: 8.72 ms
R: 8.40 ms

(Honestly these delays don't look small enough compared to the panels, there should be more delay differenceI think, but these were dialed in years ago after some testing and thinking then.)

NO GEQ's whatsoever!  (Except in Movie mode presently)

Left DEQ's, sub on top

I see now that there are actually 3 boosts in the left panel, and the upper frequency ones aren't doing very well.  The chairside PB1 memory is identical.  This looks like a mistake perhaps, did PB1 get overwritten by PB3 by mistake???  Looks like a week or two of rework is needed on left to banish the remaining lumpiness, not just a new boost or two.  But I've decided my knees need to rest from kneeling so I'm not going to do it now.  That's going to be "Janus 1.1".  And also asap I should start measuring with my calibrated microphone and ARTA.  I believe my iPhone 8S and Analyzer app is accurate enough for most purposes, especially when supported in my new microphone stand mount, but super fine tuning around 1kHz and above might benefit from something even better.  The beauty of using the phone analyzer is that it's so fast I can just whip off measurement after measurement, only a fraction of which have been displayed here, which is invaluable for mass EQ tuning.  You've got to test every little change because actually it's all virtually unpredictable.  (Some supercomputer with a description of every object in the room in precise detail would be the minimum requirement.)  Cut and try is the only hope.  Once you've more or less zoned in on the correct notch and boost frequencies, you can take the time for more exacting but time consuming measurements and even finer tuning.

I notice now that there are cuts near the 400 Hz depression, which I thought I had eliminated...

The better tuned Right channel does have only one boost in the panels (and it's just the low one at 133 Hz).

Right DEQ's, sub on top

The Movie presets simply add a set of GEQ's to an earlier set of PEQ's with no boosts.  (The PEQ's are not very critical, but the sub version should not have any boosts which may go awry in the far field.  It would be better to re-do the PEQ's for the back seating position from scratch so they don't need GEQ's, but that's another big project for later, heck I'm not even done with the Normal yet I realize now.

Left Movie GEQ


Right Movie GEQ

The subs also have Dynamic EQ (dEQ), a form of limiting.  It rolls off the deepest bass (to prevent overexcursion) when the level is too high.  SVS probably already has such a thing built in to their plate amplifiers to keep their speakers from blowing up, but it's not as limiting and protective as mine.  I investigated this a few years back, and determined I got the best results with two identical dEQ's in sequence, that way it can compress the maximum amount fast enough when needed.  And I eliminated the standard compression DYN because it isn't needed on the subs and I don't think it sounds good.  3 dEQ's was too much so I only use 2.  The only thing that may be out of adjustment now is the Threshold level, which depends on sub DAC and sub amp levels.  It can be re-tuned by watching the indicators when playing a bass test track.  On most recordings it should not trigger at all.


Sub Dynamic EQ page 2

Sub DEQ page 3

**** Wait!

I thought I had remembered much better looking Left channel response.  And indeed I had a stunningly good measurement on August 29 in an earlier post.  This was made using the chairside DEQ (shown in photo) so it automatically had the polarity changing adapter I added later (though I added it to the sub not the panel, the effect on response should be the same).

Left Channel, August 28



Left Panel peq's, August 28

The only difference I see here is that I subsequently added at boost at 182 Hz, replacing a 0dB filter at 824 Hz.  It's hard to believe that would cause a depression at 400 Hz, or 1kHz depressed compared with 900 Hz.  Perhaps the Chairside was not showing the EQ that was actually operating because it needed rebooting.

Running through all the published spectra and adjustments shows nothing inexplicable.  I had a mound around 1khz when I had a wide filter at 1013.  There was consistently a depression at 900 Hz before I moved the boost there instead of 1013.  Fine tuning that nearly leveled the whole region.  All looked believable, it didn't seem there was any hangup.

Stuff in the room may have been moved around a bit, but not that much I would think.  It still looks like the point where everything goes south is when the 182 Hz boost was added, which doesn't make any sense, but seems worth trying.  Tiny time delays might have been involved (I found the front panel DEQ's had 0.1ms greater latency, don't recall if that was with or without optical connections though the optical connections made no more difference than that, if at all).

Anyway, that's a wrap for the Janus 1.0 summary.  We can start to imagine Janus 1.1 now.





Monday, September 9, 2024

Re- Doing the Polarity

Now it's looking like I can get much nicer system response in both channels...with the polarity of either subs or panels (but not both) inverted from the present settings.  When either was unintentionally inverted with the chairside DEQ, I got nicer looking curves, and notably with NO dip at or around the crossover region.  With the front panel DEQ, which was not inverting, I struggled to get such nice curves and came closest when I adjusted the delay to be around 4 ms more than previously determined settings (setting which make sense and they also approximate the sub backward distance, as I couldn't see it any better than that anyway).  That 4ms is just enough to cause a polarity inversion at the 125 Hz crossover frequency.

Both were checked before in a variety of ways, including a phone app combined with its downloaded test file.  (I never trust such things very much.)

But to really know what's going on, it would be best to measure polarity myself.  And now I have an oscilloscope in the living room just ready for such a task.  I just need to hook up a microphone and microphone preamp.

I'm sure the sub output, even without equalization, will look quite different from the original polarity test signal I have been using in the past few days (either of my self produced files: SoftPolarityTest or LoudPolarityTest2, which are a series of pulses spaced by about 5ms).  It will be long and flat and even cyclic because of the rolloff and crossover of the sub.  Turning off the EQ (but not the crossover, which is not a good idea on my system) may be helpful in making the correct polarity more visible.  I might also need to record a test with longer pulses, the current ones probably don't generate much low frequencies.  My original tests were made by analog recording one of my oscillators (not the present one, which might make it easier) because I don't know of any software that does this sort of thing, not Audacity for sure, and I couldn't find any such plug-ins either.


SoftPolarityTest output from panel DEQ (including crossover)

The 4x orders of the Linkwitz-Riley filters permit both drivers to be in phase with one another.  Thus it is true for the 4th order (24dB/octave) however for that crossover, less group delay is achieved with the out-of-phase connections (I never used them anyway with 4th order, preferring to have everything in-phase just in principle).  Now that I have 8th order Linkwitz Riley, I believe not only can I use the in-phase connection, but it achieves less group delay that way as well, because that characteristic alternates with higher orders.  (IIRC.)  However, I am not sure if this is affected by the phase corrected (FIR) filters I currently use.  It was my understanding that they would simply output everything in phase all the time.  So therefore phase would not be optional--in phase would be required.

If there is some flaw in this which means I must use out-of-phase connections, I want to know that too.  Perhaps I can change the crossover to fix that.  Because the current pair of speakers lacks problems in the 80-100 Hz region, I could use a 4th order crossover again if needed (though...I think it also causes complications with wall bounce).  I could also sacrifice the phase correction thing too (if I went to 4th order, for which phase effects are still inaudible).

So the best case would be if the current sub polarity is incorrect.  It was actually somewhat ambiguous with the phone app.  Then I could just fix that.  But if it turns out that the ways work better being out-of-phase with one another, well I will do that for now but think about changing the crossover as well.

And if I do make them out-of-phase for now, which should be the in-phase part?  Opinions vary.  I don't think I could hear a difference but I suppose I should try, if that becomes necessary.  My preconceived idea is that the sub should be out-of-phase and the panel in-phase, which is the opposite of opinions I have heard from the Polarity Pundit, my friend George Louis.

The kind of solo trumbone and other asymmetrical sounds where absolute polarity has proven in some cases to be detectable mostly contain a lot of high frequencies.  In the bass, it's all a jumble of whooshing reflected sound.  The leading edge of anything, even a low frequency sine, but with acoustic harmonic sounds even more so, represents a quick blast of high frequencies because the starting and stopping of anything, it's abruptness, represents high frequencies, whereas the extension, duration, of everything represents its lowest frequencies.  The only such thing as a "pure sine" would be one that started at the beginning of time and never ends.  We normally don't think in such terms, we truncate into bits, and say "see sine wave here" and imagine that duration in isolation from the world--a world unto itself--but in the real world a sound must start and stop, and those are not actually the same as the rest.

*****

I doubt absolute polarity is reliably identified by ear.  High quality experiments published in JAES indicated otherwise with some exceptions, such as solo trumbone on headphones.  Mixed music, or virtually anything played in a room, has too much complexity for polarity effects to be audible.

But nevertheless, I do take some effort to preserve it.  It was a shock therefore to discover this week that my long trusted (since 2006 !) digital processor was out of polarity, at least in the earliest version (which I may well have had several of at the time).

I had never tested it before, I just couldn't believe it would be wrong, especially through the digital connections.

Even this time, it took a long exploration of other ideas before I finally saw with my own eyes that it was a simple polarity difference, not so much a difference in latency.

So while I named 'Eisenson's Revenge' the kind of issue where you are at some moment having made some major mistake, but thinking about so many other things you haven't caught it yet, not so much that you are deaf or whatever, this isn't really that.  In this case I presumed my device to be correct for 18 years.  I even used it to do polarity experiments!  (However I failed to even hear a difference, so it wouldn't matter if I had the conditions reversed.)

George was constantly telling me I needed to test these things, everything.  Of course, test it his way, by listening, which is highly unreliable at best, if not completely random.  But he was right I should have tested this before now.  So this error needs to be named after him, when you just assume the polarity is correct and it isn't.

And further, by mixing DEQ's of different vintages over time (as units burned out and were replaced) further compounds the problem because some units would be out of relative phase with the other units, causing issues in systems with multiple ways.

Leading to the situation in the past two months, where I was adjusting with one way out of phase, so compensating for errors caused by the phase inversion, a complete waste of time.  Though it's only the points closest to the crossover frequency which seem most affected.

However, then the same is true about the sub inversion.  Which I'd have never discovered if the chairside DEQ weren't out of polarity.  In addition to the faulty DEQ, somehow the subs seem to be inverted, and have been for some time.

*****

Getting the microphone hooked up to scope hasn't worked yet.  I can't find the 9V AC (!) adapter for my M-Audio microphone preamp.  It's not clear my Focusrite is working right without a computer connected but just a USB power supply.  I've seen no microphone output on my scope yet.

Meanwhile, I have a few Oscilloscope apps on my phone, so I tried one, SignalScope X.

When I ran my "SoftPolarityTestRight" through the right channel (I don't run the LoudPolarityTest through speakers because it's dangerously loud but works ok for electronics) I was able to get an image which shows a very nice 1ms impulse looking almost square and perfect, trailed by bass and reflections


Right Channel playing Soft Polarity Test

But when I ran the same test signal through the right sub by itself...it's simply inaudible.  The test signal simply doesn't have any bass at all, it's all 1kHz and above.  The test signal does have DC since it only goes in one direction, but that DC gets eliminated in many ways, including the servo loop on my Aragon 8008 BB amplifier, and the bass transformer in the Acoustat.  For that matter, I don't think the Emotiva Stealth DAC passes DC either, and it's certainly filtered out by the crossover for the panels and the DEQ.

So I ran a different test signal, the one that goes with the SpeakerPop app.  It's not a series of spaced pulses, it's a series of very long duration (OTOH one second) sawtooth waves.  Running that, the subs produce a very weak sounding "buzz" sound.   Once again, there is very little bass at normal frequencies  20-200 Hz in such a signal.  And again, the near DC part is filtered out in many ways.  (In fact, to avoid stressing my amps and speakers which I did a few years ago with this signal, I created a "normalized" version with no DC offset and reduced amplitude, and that is what I actually played.)

Still, there was sufficient output to capture it unambiguously in SignalScope X.  But the result is not easy to understand.

Right Sub playing normalized Speaker Pop triangle wave

Is that in normal polarity or inverted polarity?  There's no easy answer.

At first I thought this showed the phase shift above and beyond the crossover frequency all the way up to 20khz, some of which might be leaking through the crossover and sub speakers.  But that doesn't make sense because such output should be too weak to measure, I'm barely able to measure anything at all.

But then later I realized this is showing the effect of room reflections on the bass.  Bass tones are so long that you can't even play one full cycle without the room reflections butting in.

I verified that the source of the wrapped sub tones is NOT the crossover or DEQ.  I viewed the analog output of the front sub DEQ (which follows the crossover so that's in there too) on the scope, it it was a low frequency sawtooth, just one positive impulse fading into the background on each cycle.

Speaker Pop electronic output from Sub DEQ

Studying the sub output again, I pondered whether I should consider the very first stroke as the polarity of the output, the first large stroke, the last large stroke, and so on.  I couldn't make up my mind about this.  But then I also noticed that the relative widths and sizes of the up and down strokes varied a lot with position in the room.  If you were taking the first large stroke, or something like that, the polarity would invert going from very close to the subwoofer, to the listening position, and then once again to the kitchen doorway in the back of the room.

Right sub Speaker Pop at different location

Right sub Speaker Pop at different location

Which is making all these stroke things seem like reflections.

Well, how can you tell what the polarity is with a jumble like that?  I should note that I got similar results with the Speaker Pop app itself (which uses this same test signal).  I got one polarity near the speaker, a different polarity at the listening position, etc.  So maybe the app is having the same trouble here as I am in interpreting the waveform because of all the included reflections.

Way back before subwoofers, I first learned to test the polarity of speakers by connecting a battery to the speaker terminals.  I think my friend Bro showed me that.  I was leery of the idea, thinking the battery might harm the speaker since the speaker had very low impedance at DC.  Bro assured me it was not a problem, the battery itself would be limited by its own internal resistance to not put out enough current to damage a speaker driver.

But I can't do that with subs now, it seems.  I can't just connect to "the speaker terminals."  I must go through the very fancy electronics that forms the "plate amplifier" of the subwoofer.  The latest generation on my right sub has only line inputs, balanced and unbalanced.  If there were speaker terminals, they would just feed in through the amplifier in the same way, but with more attenuation.  And most certainly the amplifier does not pass DC, etc.

Now the other way of figuring out the correct polarity was to see, experimentally, which gave you the loudest output at the crossover frequency.  I worry that might not be correct in some cases, but generally speaking it should be correct.

*****

I tried using the Speaker Pop app which was easy because I was already playing the (normalized) Speaker Pop tone 

The Right channel was very inconsistent, though it mostly seemed like negative polarity close to the subwoofer, positive polarity at the listening position, and negative polarity in the kitchen doorway.

The Left channel was a bit more consistent, almost always positive polarity near the subwoofer.

I tried playing both sub channels together and inverting the polarity of the right sub.  At first, inverting the polarity of the right sub made it sound "weird" and I thought that couldn't be right.  But when I measured the "weird" sound, it was about 2dB louder.  It was more like a thrack than a tap, and I was beginning to think this sawtooth like impulse *should* sound like a thrack.

I also noticed that indeed it seemed more centered with the Right channel set to negative polarity in the Tact.

I went to the SVS app for the right sub to change the polarity there...and as I was changing it, I could have been wrong, but just before I "changed" it I noticed that it seemed like it was already in the "inverted" polarity so nothing changed when I "changed" it.  I repeated the previous tests in both the Normal and Inverted to be sure, and confirmed that the previous measurements corresponded to the "inverted" mode in the SVS sub app.  So it appears that until now I had been running the Right sub in the "Inverted" mode of the sub electronics without knowing it.

This might have happened when I replaced the plate amplifier a couple years back.  There was a wire harness and the best thing was to test it both ways.  Back then I used the Speaker Pop app and getting the usual ambiguous result, I decided that "Inverted" was correct because that was how it measured at the listening position.  Well, now I see that having the right sub set to Inverted in the SVS app was exactly the wrong thing to do, and was at the bottom of the DEQ inconsistencies and the fact that the inverted DEQ actually worked better on the right channel than the correct polarity DEQ I normally use.  But not so much on the left.  (So now I believe that the Right sub had been out of polarity, but not the Left.  The Left channel was different but not worse when I used the chairside DEQ.  When I started this post and investigation I had believed that both subs might be out of polarity.)

All these years though, I must point out, that it actually sounded fine, even with the imperfect EQ and out of polarity sub.  Making one of these mistakes isn't like leaving the water running.  It's more like it's hardly noticeable in the overall scheme of things.  And this is more true for me than usual because I'm using such a steep crossover, the effect of panel and sub being in the incorrect polarity has relatively small effect, localized around the crossover frequency, where I could easily equalize it out.

But now, with both SVS subs in their "Normal" polarity as they should have been, and using a non-inverting DEQ, I believe the polarity issues have been corrected.  (Even though I still don't have a good test signal that makes this easy to see on an oscilloscope.  There's just a preponderance of evidence that points this way.)

Speaker Pop has been a very useful tool in these regards despite it's ambiguity in some cases.

*****

It occurs to me now that I *can* use the polarity inverting DEQ as a chairside DEQ so long as I invert the phase of the particular way it is operating.  So if I use the chairside DEQ on the panels, I simply invert the panels, or I could invert the subs and the Tact to achieve the same effect (ignoring absolute polarity, which I think I can do mostly).  It's an inconvenience, and it also means I can't do fast A-B testing where I had one adjustment in the chairside and a different adjustment in the front.  But it's still possible for month long adjustment sessions, as I am just now (hopefully and finally) completing soon.

*****

Oops!  I thought it was all done, I was now going to "finalize" the PEQ's for both channels not that I figured out the polarity issue which was dooming the right channel.

So for the "End of Eq'ing" post I was going to lead with the measured spectra with the front panel DEQ's for both subs and panels.  But something was clearly wrong with the right channel again!


Right channel, no panel EQ

I hadn't re-loaded the '2024' preset into the panel DEQ.  For doing some test, I had loaded the 'Inital Data' preset, and just left it that way.

So I re-loaded the 2024 preset, and the Right channel was looking pretty good, and you can see the benefits (and/or losses) from my EQ'ing, which I think is overall a plus.  Notably the most depressed point in the upper bass, 160 Hz, is right on the line with 1kHz.

Right Channel, latest 2024 presets

And I measured the left channel, it was as looking pretty good.

Left Channel, latest presets

But then I noticed something quite wrong.  Though I had changed the polarity of the right sub at the panel, the preset still had the 14ms delay I had found to work the best before I had done that.

Re-measuring the Left channel with that 10ms (default design) delay and it looked horrible, with a huge depression around the crossover frequency.

Left Channel, Design 10ms Delay

And that gaping depression could be solved with either 14ms delay (which I had just removed) or polarity inversion, just as the right channel.

But what about my that/thrack test which made me believe only the right sub must be inverted and the other one not?

Well, instead of repeating that possibly erroneous test, I played the Phasing Test from Stereophile CD2 with the panels disconnected.

That also was very ambiguous.   It did possibly seem that with both subs inverted, the out-of-phase was more out-of-phase, and the in-phase more nicely centered.  But when I measured loudness, when I played the out-of-phase with only one sub inverted (which now seems wrong) it measured about 1-2dB softer on peak, and the in-phase about 1-2dB softer on peak, exactly as it should have, compared to having both subs inverted.

I am thinking we can't go by the phasing test here because Atkinson's bass has lots of weird harmonics and phase shifting going on.  If I switch the phase during one bass tone, the results are nearly unpredictable.  I also think some out-of-phase cases are louder than they should be because they excite certain twisted tangential and oblique room modes more.  I should mention in all these phase loudness tests, there was a significant amount of room rattling going on.

I'm thinking the numbers here which are unequivocal are that the right sub must also be inverted from it's former condition to avoid a depression in the crossover region which happens otherwise.  And the effect on what little imaging I can hear from the stereo subs is (because it's highly ambiguous) is either acceptable or slightly superior.  I suspect on low continuous bass tones around the crossover frequency the improvement would also be as noticeable as with the pink noise.

When I re-measured the Right Channel with the spurious extra delay removed (so back to the 10ms design delay which every other speaker way uses as a baseline)  it looked even better, what I thought I had achieved with the chairside DEQ before discovering all the polarity issues:


Right Channel, 2024 preset, 10ms design delay

That passes with flying colors!

In the older generation SVS left sub plate amplifier (prior to bluetooth app) there is no "polarity" control, only a "phase" control.  I'd rather not use such things especially not knowing how they work.  Is 180 degrees the same as inverting polarity?  It might be adding some internal delay instead, or performing some other kind of complex operation.  Anyway, I set it to 180 degrees, with the delay back to 10ms, and it wasn't looking as nice as with the 14ms delay.  In the right channel, no delay was as good as the real polarity change.

Left Channel, Design Delay, SVS Phase 180

I recall using the SVS "Phase" control before and determining that it was useless.  This is only barely as good as the 14ms delay version shown above and only slightly better than the 10ms design delay with no inversion.  Similar results could have been achieved by adding about 2ms more delay, which is perhaps what it actually does.

What needs to be done is to invert the polarity through the balanced cabling.  I think I had and WAS using (and perhaps still?) a polarity inverter in one of the channels.  It was mounted midway between two balanced cables so not as to hang on any xlr connection.  It will take some work to see if there is such a beast already connected now (and remove it) or needs to be added (in which case, I will have to find the old one or buy/make a new one).

Oh boy.

*****

Fortunately I easily found my 2 XLR polarity inverter adapters.  None appear to be used on the existing left subwoofer connection; though it is marked with an "S" label, I think that means "Subwoofer" and not "Switched."  So it will actually be quite easy to add the polarity inverter tonight right behind the sub DAC where I can easily switch it out when needed.

So the plan is to switch that in, change the "Phase" control back to 0, and re-test.  Hopefully then this phase (and post) will be completed.

*****

As expected, using the polarity inverting adapter is much better than using the "Phase" setting.  I'm getting near perfect bass, main problem now is in midrange depression where there's no filter (looks like it needs a boost).

Left channel, XLR polarity inverting adapter

So that's that, supposedly inverted polarity seems better in the left sub and perhaps right.  In opposition to the SVS setting (fwiw, since I did amp replacements and may have erred) and the notion that I have a "Phase Linear" and also "Linkwitz-Riley" crossover.  Perhaps it's due to phase shift on both sides or one in particular.  Perhaps the Speaker Pop app really isn't worth a damn on subs for determining polarity in the first place...and I'm not sure anything is.  And since that was what I used to set up the subs themselves, their nomenclature might now be wrong too.

(Note: I "changed" the polarity on the Right sub, but I changed it to Normal instead of Inverted.  So, nominally the right sub has correct polarity...if I installed the plate amplifier correctly...which is highly uncertain...but perhaps that should be the default assumption here...I've long used phase reversal on the left sub because I think I installed the first replacement amp wrong, but then tried to get the others right?  So, anyway, whether one or both subs are in "inverted" polarity now is a big enough question mark so as not to labor under the belief I am using inverted sub polarity...I simply do not know, but some indications suggest that, like the Speaker Pop app, and my own attempts to measure it have so far been futile and I grasp that a suitable measurement might not be easy.)

This phasing between panels and subs is clearly right this way, however.  The center is more solid, bass is more solid, etc, and yet without being "too loud" either, remaining effortless like never before because less cancellation.

Given the uncertainties here, I'm not sure how much to be disappointed that my "linear phase" crossover may not actually be "linear phase" at all.  If you have to invert the polarity of one side (and it's not actually clear I'm doing that) you cannot possibly have truly linear phase.  There must be at least 180 degrees of phase shift somehow.

If that's all there is, it's hardly worth worrying about.  That's how much phase shift you'd get from the simplest possible crossover, 6dB/octave, which also requires an inverted side.

Linkwitz investigation suggested that phase shift was not audible until you got beyond 4th order crossovers, which would be giving you 720 degrees of phase shift.

But it does raise questions.  If my crossover isn't actually linear phase, just how much different from linear phase is it?  A standard minimum phase 8th order crossover would have 1440 degrees of phase shift.  I still believe I must be nowhere near that with my "linear phase" version, but I can't prove it, and maybe it's worth more exploration.

One tack would be to develop tools to see what the polarity of the subs actually is.

Another tack would be to try different crossovers.  I could try rolling back to a standard minimum phase 4th order Linkwitz-Riley crossover.  It probably wouldn't be very different, and in theory would give me both in-phase and out-of-phase options.  Since, in theory, the drivers are in phase quadrature either connection supposedly gives the same frequency response, but the in-phase connection actually has more group delay than the out-of-phase connection, which is pretty mind boggling.  Even then, spurious phase shift from other sources including the drivers themselves might well mean, if examined as carefully as I have done here, that the out-of-phase connection actually has better frequency response too.

And a third tack would be to explore the phase shift itself.  To do that, I'd need to use much more sophisticated tools, like the Arta software I have.

I'm willing to put that all on the back burner now.  I have pretty good (extremely good in my opinion) sound already.  Improving the phase linearity of my system, which is now apparently not perfect, is not really a high priority except for my ego.  I believed I had solved phase problems, and in fact I have not.  Even if I did, without making other things worse, it probably wouldn't be a big deal.  Frequency response is #1.

On that score, there still are a few more wrinkles that call out for fixing, perhaps the biggest being the depression around 400 Hz in the left channel.

And macro adjustments such as raising or lowering the bass plateau aren't necessarily completed either.  Now that I've raised the crossover range output by "inverting" the subs, it's possible the overall bass level can be lowered again.  Only extended listening can guide that.

So while this audio project is not entirely completed, it will never be, and I think it's worth putting a marker her with all the remaining issues and ambiguities by giving this new phase alignment a name, enabling me to move on mentally at least.   I've decided to call it "Janus One."  I haven't used a crossover requiring me to make even a nominal phase inversion in over 16 years, and I didn't intend to do that when I adopted the present crossover.  But I never truly explored all those earlier crossovers in the past they way I have this time.  Perhaps they would have worked better with a phase inversion too.  I simply never bothered to test it until now, which is one more reason this is better than ever before.  I've never had measured frequency response as smooth and well configured as at present.









  




















Thursday, September 5, 2024

The (last?) wrinkle

[Note: the "previous" post was actually finally completed days after this one.  The fine tuning done here still looks pretty good, but as I determined in the previous post, it will all have to be rechecked after I re-check the sub polarity and the sub delays, which will be the subject of future posts.  And the "delay bug" I describe here wasn't really a delay bug after all, it was a polarity inversion bug.]

Having worked around the strange left channel delay bug again, this time without messing the right channel, I'm back to trying to solve the weird lower midrange of the right channel.

I decided to keep the 133 Hz boost, because the resulting bass curve is the smoothest I've ever seen to 200 Hz, and with about the right room curve too (in fact, it looks as good as anything I could imagine).  But above that, and up to 2kHz, it looked like crap.

So I pulled out the notches above the 133 Hz boost until the 2.7 kHz cut (and more about that later perhaps...) and swept for the biggest peak.  Here was what the Analyzer spectrum looked like:

Right with lower midrange PEQ's removed

By the picture the 600 Hz peak looked biggest, which swept to be 599, so I set the closest notch at 597 Hz with 1/6 octave and -6dB.  That swept OK.


Then I added more notches at 833, 563 and 734 following similar process.  All along I looked at RTA's to get an idea of where the next peaks might be.

These are all different numbers, perhaps in part because I'm not starting with a boost near 1kHz.

Now the RTA was still looking pretty rocky, even rockier than before maybe, with the previous peaks all below 1kHz and the depressions even further down.  So I backed off each of the PEQ's, reducing the attenuation, until the peak was just at the 1kHz level.  Then the floor below the peaks was brought up too.  Sometimes I narrowed or widened the peaks (but I already chose narrow PEQ's when they were near depressed areas).  In the end, the 833 Hz cut no longer seemed necessary at all and was just depressing the two bars lower than 1kHz (and sometimes even 1kHz itself), which are still slightly depressed even without it.  I think it was largely subsumed by the 734 Hz notch for the 733 swept peak, which persistently showed up on the Analyzer as 710, seemingly one of the most persistent peaks.

The result uses only the 133 Hz boost (for the left panel anyway) compared to the previous 2 boosts and the 3 boosts used in the right channel.  Yet it looks pretty nice on the RTA and sweeps pretty nicely too.  (I don't even much notice the remaining peaks and depressions except at very very slow sweeps.)  There's still some lower midrange depression underneath the peaks but it's reduced as much as it can be without boosts.  The deepest valley tells me it needs a mere +4.5dB boost.  I swept that area and it seemed perfectly fine.  Perhaps there is some microscopic null but I didn't find it.  In sweeping everything is going up and down all the time, and it seemed no different there.

Right with new midrange filters

Just to show the improvement, here's what it looked like before:

Right with previous filters (measured day before)

Similar underlying shape but the tolerance has been cut at least in half by the new filters.  The earlier filters achieved a kind of artificial smoothness around 1kHz (but not AT 1kHz, which stuck out like a thumb) with a 3/4 octave boost at 1kHz.  I think it's better not to resort to such things.  But without it, that set of filters made 1kHz sucked out.  The new set of filters never sucked out 1kHz in the first place.  I think that HAS to be better.  I now wonder if I could do the same thing in the left channel, which has a 900 Hz boost.  But I don't think so, that was very carefully analyzed last month.  There was a deep depression there which carried 1kHz with it.  But I found no such depression in the right channel, only a peak which mysteriously went away when the other filters were applied.  Here are the new filters:


Right Panel PEQ's with 1 boost (PB1)

Wednesday, September 4, 2024

Balancing...no back to DEQ difference problem

 After copying over the new boosts from the chairside DEQ into the front DEQ, I measured (Unweighted) the sub and panel outputs separately for both channels.

Left/Right balancing by SPL seemed to occur when I attenuated the left channel by about 1.2dB, instead of the 2.3dB I have been using.

However, when I switched to my ultimate "Balance" test, I was right back to the 2.3dB.  That test is the Speaker Phase test from the Stereophile Test CD 2.  When balance is exactly correct, the out-of-phase (both in John Atkinson's voice and bass guitar) sounds equally loud in both channels, just not centered.  Somehow this test is extremely critical, even the smallest imbalance points it one way or the other.  You can't tell much from the actual in-phase center image which says roughly centered over a 2dB range.

I'm not sure what explains this difference, or why there should even be a 1.2dB difference in measured levels.

I tried setting the right sub level 1-2dB higher because the right channel actually seemed less bassy.  But after some testing that way, I moved it back to the earlier -10dB default.  It's not even clear if the right channel is less bassy, it could be more bassy at some points.

I re-measured both channels (subject to the 1dB bass reduction I did yesterday by changing the sub DAC to -6.5dB from -5.5dB...a few days ago it had been at -7.5dB).

Left Channel (end of Sep 3) cursor at 1kHz

Left Channel, cursor at 35 Hz

The different cursor positions make it easier to see how high the bass plateau is.  Overall this is the smoothest response I've ever achieved, and fitting within the desired room curve very well (only a little bit of up and down all the way).  Though 4khz is a bit high perhaps, it doesn't need as much attenuation as 3kHz to sound good (Gundry-Linkwitz dip).

The right channel has a similar bass rise, but the plateau keeps rising all the way down to 20 Hz.  The bass looks very smooth, but the midrange looks very bumpy (which I don't understand because I just flattened it in July and August).  The upper midrange is curiously smoother.

Right Channel, end of September 3

At this point, I believe the right channel midrange 250Hz-2.5kHz needs to be fixed before going any further.  10dB peaks and valleys are not acceptable (though I've lived with them many many times before if not most of the time).

The EQ is fairly simple in the greater midrange from 200-2000 Hz.

Right Channel front panel DEQ

I'm going to start by turning off all the PEQ's in the 200-2000 Hz range and sweep for new ones.

The current PEQ's in this range are notches at 251, 647, and 853 Hz, and a boost at 1013 Hz.

It's probably going to need a boost below 500 Hz.  The left channel now has a boost at 272.

Left Channel front panel DEQ

Both channels have similar cuts at 251 Hz and 654 (or 647) Hz.  To get the flattest response in the left channel I lowered the boost to 903 Hz and made it very narrow so as not to exaggerate nearby peaks, whereas I used a wide boost in the right channel at 1013 Hz which exaggerates the peak at 1.2kHz (?), but otherwise it's actually pretty smooth right around 1013 Hz, but has awful peaks in the 600-800 Hz region as well as an awful depression around 400 Hz (curiously very close to where I have a deep cut in the left channel).

I hooked up the chairside EQ to the panel DAC and loaded the pb2 preset (the latest panel work).

Checking the values in the right PEQ quickly, they looked identical.

But when I measured the response with the chairside DAC, it looked much flatter than the right channel measurements I made previously.

Right with chairside DEQ

Now that's more what I remember.  There is still stuff around 500-1000 that needs fixing, but the rest of the curve is fine.  I measured this with my new "2 dukes" standard where I position the phone two fists (plus a small but comfortable gap) between the back of the chair and the phone.  I think this is slightly more accurate (considering the main microphone is at the other end of the phone still) than my old standard which was 2.5-3 dukes (measured with a remote control).

Also I am measuring consistently at center now, whatever demerits this has I think it's basically ok, in my own testing it made very little difference.

I measured it with a completely new adjustment (move the stand arm out of the way and bending the flexible mount, then moving them back) and with 2.5 dukes spaces to get another measurement to be sure this wasn't a microphone positioning thing:

Right remeasured at 2.5 dukes

Basically the same thing, actually the 2 dukes looked better.  I then reverted to the front panel DEQ, and that shows the error, a suckout around the crossover point.

Right with front panel DEQ

I know what this is.  It's caused by the front panel delay being in Main, which it needs to be in order to fix the left channel.  So for the front panel DEQ, the left channel wants Main delay, and the right channel wants Aux delay.  Somehow this doesn't apply to the chairside DEQ, perhaps because it's a different vintage DEQ, or perhaps because I'm using optical I/O.

This calls for one thing.  RTFM for the Behringer DEQ 2496.

I've found no way to adjust left and right channel delays separately (despite having obviously separately windows) nor any reasonable explanation for what the Main and Aux selections do.

*****

Well, according the the manual the Main selection applies to the "Main" (ie main analog audio) outputs.  The Aux applies to the Aux outputs (there are also aux analog audio outputs).

What applies to the digital outputs?  It doesn't actually say in this section.  But it appears from another I/O menu that the Aux and digital outputs are tied together and you select what Aux/Digital outputs are connected to in menu item 3.  So, one would think that whatever applies to Aux applies to digital.

When I started this exercise, in fact right up until around mid September, I had the front panel DEQ set to Aux.  Mistakenly (?) I had the chairside DAC set to Main, and I did the tuning there.  When I copied the tuning to the front DEQ, it no longer worked, but I found that selecting the Main delay worked.

Now, with the delay line selected for Main, the delay processing is not available for Aux.  The manual says you get one or the other but not both.  You select the channel delays, and which output they are used for.

So it would seem the chairside DEQ is giving the digital outputs no delay at all.  Which basically can't work, because I've designed my system so the subs have less delay (around  8ms) and the panels always have 10ms delay to allow me to set the sub delay to anything.  What matters is that the subs have around 2ms less delay because they around 2 feet further back.  (Actually, this was fine tuned with impulse analysis using ARTA.)

So it can't work with anything other than 10msec delay, and yet the both channels work fine that way on the chairside DEQ with the delay set to Main (which should mean the delay is then "not available" for the Aux/Digital outputs).

Well it seems I must have understood this operation of the delay selector, since prior to mid september and going back who knows I had both the panel and sub DEQ's set to Aux.  But I neglected to do that for the chairside DEQ (and still) all the tuning I do there assumes no delay.

So now the rub seems to be that when I was fixing the right channel on the chairside DEQ, I was not fixing it correctly, I was throwing in a no-delay condition as I had been doing on the chairside to get the same response.

What it appears I need to do is reset everything to Aux, and re-tune the Left channel to work with the 10ms delay instead of working around the lack of it.

(Even though the right channel is problematical, the nature of the problem requires me to fix left channel.  Life is like that sometimes.)

What I did for the right channel to get the smoothest bass response ever up to around 300 Hz was to boost the 133 Hz region.  That worked with the correct delay.  It makes sense because the Acoustat is at a lower level than the sub (because room curve).  What I did for the left channel was NOT to boost 133 Hz (which was elevated already) but instead to boost 180 Hz.  Well, that worked, but it appears now it was really a work-around.

With everything on Aux, it should be the left channel that looks crappy on everything and the right channel looks fine on everything (well, except for that weird stuff 500-1000 Hz on the right channel).

(A weird angle here is that the Aux signal can also be selected to follow the main output, which is in fact the default on the setting on my units.  So in that case, does the delay apply to both?  It appears not, because in all cases changing the delay from Main to Aux makes some difference or another, and under that interpretation it could not make any difference.)

I set both DEQ's to AUX.  The right channel is smooth to 300 Hz in both chairside and front panel DEQ's.  No depression around 125 Hz as would happen with an improper delay.  But the left channel is not.  It shows a serious depression around 125 Hz, only with the front panel DAC.

I'm back to the "error" I thought I fixed a few days ago when I put the front panel DAC into Main Delay.  That wasn't fixing it at all, it should be in Aux to work correctly.  There is still an utterly inexplicable difference between the two DEQ's in the Left Channel.  All the PEQ's are exactly the same and there are not GEQ's and I just tried another reboot of the front panel DEQ and it didn't help.

I *could* try to fix it by varying the left channel delay in the front panel DAC.  Apparently I can chose which channel to change the delay setting for (or both) by holding the A button (more intuitive than it sounds because the A button is the one immediately in front of the Left Right selector button, which defaults to being just one button for both).

Since setting the front panel DEQ to Main delay fixed the problem in the left channel before, and since that should be setting the digital delay to zero, I tried to just change the delay to zero in the left channel, which should have the same effect there as setting the delay to main in just that one channel.

And it worked.  It now shows the incredibly smooth bass and conforming curve to what I believe is the best room curve using the front panel DEQ still in the Aux Delay setting.

I have no idea why this hack is necessary.  The use of optical connectors has no effect on the right channel, which measures the same with the correct 10ms delay dialed in as with the chairside DAC.  A delay time bug which only affects one digital channel, where it erroneously takes on the Main instead of Aux delay switch polarity?  That would be very weird.  This problem can't have anything to do with the actual polarity (etc) of the subs etc because that is not what is being varied.  Only thing being varied is the DEQ and the type of connections to it.  (I double checked the volume settings on the DAC and they are identical for AES/EBU and OPT.)  And it only affects one channel.  Weird.

I need to be vigilent if some future change or power cycle or something changes this back to the expected behavior, then my hack today would introduce the problem it's fixing now.

Another rule probably is to never do anything permanent without measuring it, to be sure you aren't locking in some kind of error (yours or the units).

Over time (and over this two month so far episode) I've seen lots of things change inexplicably.  I previously didn't measure carefully enough to know it wasn't measurement error, but now I use a stand for my iphone.

I can't imagine keeping up with all these variables without measurement.  Few serious speaker designers design without measuring, they measure while designing then listen and then decide if it needs to change, etc., just as I'm doing (now at least).

*** September 5

Sadly, this wasn't the end of this.  When I was testing the updated panel DEQ balance with the Stereophile Test CD 2, using the phasing test (which is finer than the channel identification) I noticed that in-phase sounded out-of-phase and out-of-phase sounded like outer space.

Funny I hadn't noticed that all night long in background music listening yesterday.  (Eisenson's Revenge.)

I rolled back both panel channels to the exact same designed delay (10ms) and my system was back to normal.  Except that that the pesky crossover depression in the left channel is back.

It looks like it's just going to have to be fixed with another EQ adjustment.  Trying to adjust it with delay changes was a hack that has run it's course.

While I was here, though, I tried to get the balance more right.  I've been applying a -2.3 db attenuation to the left channel (in the Tact, the only place I can easily do this) to make the channels sound balanced, but A-weighted measurement show only -1.7 or so should be needed to correct for the level imbalance, so I've always wondered, why do I have to give it more attenuation than -1.7dB?

So I tried setting the level to the correct -1.7 dB, and applying various delays to the left channel also.  Less than 0.02ms made no easily detectable difference.  (That would be like moving the speaker a couple inches.)  I could complete "fix" the balance error inconsistency a delay between 0.2 and 0.3 ms.  But then I wasn't sure if it sounded right.

I delayed the left sub by a corresponding amount.  I still wasn't sure it sounded right.

After some fussing with different levels and different delays, I gave up and went back to the default delays and the -1.3dB adjustment.  At least it's not messing with the sound much.

But adding such tiny delays would have no effect on the bass crossover, since the cycle time of 125 Hz is 8 ms, it's going to be way out of phase by that point just like I noticed when I sat down to listen to the phasing test.

But it might be worth trying some of the smaller delays, up to the perhaps beneficial 0.3ms, on the panels to see if the flat-ish response can be restored by such small changes anyway, because they might be correct anyway.

Another possibility...use optical connections on the front panel DEQ.  As to why that would fix one channel and not break the other I have no idea.  What else could explain the differences between the two DEQ's?

*****

Thursday night I was thinking of removing all the old PEQ's from the front DEQ, and then entering them in the exact same order as in the chairside unit and see if that fixes the problem.

(I still ponder the math and don't quite believe that a difference in order would affect the results.  But it's still the best explanation for the difference between DEQ's.)

But instead I wondered if I could fix the problem by lowering the left channel boost from 182 Hz to 133 Hz or something like that as used in the right channel.

Since I already had the PEQ's dialed in, and they're going to be reset anyway, why not just try messing with them a bit.

So I swept the left channel 100-300 Hz with the 182 Hz boost turned off, and indeed 133 Hz was the depression.  So I put a boost there.

The result was very disappointed.  It was still depressed at 125 Hz, the crossover frequency despite the very close boost.

I moved the boost right to 125 Hz and that didn't create a nice looking spectrum either.

Since the boost was already at maximum, I then tried removing the huge notches in the sub at 129 Hz and 144 Hz.  Then I tried changing their depths to get the system response flatter (which meant keeping a bit of the narrow cut at 129 Hz).

Fiddling with those 3 things, the sub notches and the 133-ish panel boost (which I ultimately moved to 162 Hz), I was able to get more reasonable looking response than before.  This would be OK if I absolutely needed to use the front panel DEQ and didn't care anymore that it didn't match the chairside DEQ.

Left after readjusting sub and panel filters

Left Sub and Panel filters

But I don't think it is as nice as what I've dialed into the chairside DEQ, if that would only work in the front.  And I do care that the chairside and front DEQ's should be the same, otherwise I can't do very good experiments with the chairside DEQ.

Remaining theories about why the two DEQ's don't match are now down to:

a) the order of PEQ's does matter and is very different.  Notably I put the boost on top of the front DEQ, and it's on the bottom fo the chairside DEQ.

b) Using the optical connections introduces a significant delay* in both channels (otherwise they'd sound out-of-phase) that has a terrible effect on the left channel, but virtually no effect on the right.  (It's weird that it would have no effect on one channel, but sometimes things are like that.)

(Which has to be on the order of 10ms or so...artificially adding 10ms of relative anti-delay fixes it completely, as I previously discovered.  But the 10ms of delay needs to be added to both channels.)

c) Using the optical OUTPUT of the miniDSP is different than the AES output possibly because of some setting in the miniDSP

d) One of the DEQ's has some sort of digital bug

Thinking about (b) a bit more...  Remember I "fixed" the problematic left channel by going to Main delay, which removes all the delays from the Aux (and presumably digital) delay.  BUT, when I did that, it DID make the Right channel worse, and in exactly the same way--a depression at the crossover frequency.  So there is evidence that making the DEQ delay different would have that kind of effect, not fixing both channels at the same time as now happens with the chairside DEQ.  But the change is in the different direction.  The chairside DEQ does have the 10msec designed delay dialed in.  If there were a delay for using optical connections, it would be on top of that, not subtracted from it.  So to test this theory, what I need to do is dial in MORE delay, like another 10msec but possibly not exactly that.  Perhaps unlike removing delay, adding more delay wouldn't mess up the right channel, but would fix the left channel.

A flurry of tests on Friday morning was very interesting.

Starting with the front DEQ, I tried adding more delay to the 10ms design delay in the range 2-10ms to the left channel.  10ms eliminated the crossover depression, and 6 and 4ms did almost but not quite as well.  At 2ms the crossover depression was back.

In the right channel, adding 4ms delay introduced a significant crossover depression, but 3ms looked OK.  3ms looked OK in the left channel too, so we have a "winner," I could make chairside and front panel DEQ's match pretty well in both channels with a 3ms delay added to each in the front panel DEQ.

So this is consistent with the idea that the optical I/O is adding about 3ms latency (because of the optical conversion boards, not the speed of light).

However, then I hooked up the chairside DEQ with all AES connections.  It still looked fine at the 10ms design delay in both channels.  Adding more delay as I had with the front DEQ messed it up.

Conclusion: it doesn't seem to be related to the optical I/O, which makes no difference for the chairside DEQ.

So what about the PEQ order?  I carefully copied all the chairside PEQ's into the front panel DEQ.  The result was the same as before this copying: a depression at 125 Hz.

Conclusion: it doesn't seem to be related to the PEQ ordering either.

Final tentative conclusion: there is a vintage difference in the two DEQ's.  The chairside DEQ has more or less latency in the +/- 4ms range.  About +3ms seems to fit the existing data pretty well, although it could be less latency by a comparable amount instead of more.

Anyway, fine tuning the latency in the front panel DEQ is basically the only thing I can do to bring the two DEQ's into alignment, and it now looks like I may be able to do it reasonably well (for +/- 2dB differences at most, in both channels--which is the tricky part).

As I've already determined, making the PEQ's different in the panel DEQ isn't enough to help, I'd also need to modify PEQ's in the sub DEQ, which would make all future testing with the chairside DEQ impossible or very difficult.

Q: Wasn't the delay originally tuned for the front panel DEQ?

Yes, when I tuned the delays it was with the front panel DEQ (though I don't remember exactly if it was the current unit).  But since then a lot of EQ and other things have changed.  Large amounts of EQ can apparently change delays in the 3ms range.

As of right now, the (greater???) delay in the chairside DEQ works better, there is little or no crossover depression despite many other factors.  In principle I could try to match the chairside with the front, but since the chairside DEQ is working better, I now think I want to match the front to the chairside, which somehow landed on a somewhat better delay.

It was actually very very hard to set the bass/panel time alignment.  It was very ambiguous, probably because the majority of the bass isn't being seen from the sub driver but from reflections from the front wall which is 2 feet behind it.  I may have previously matched to the driver, but matching to the wall behind may be more important for the frequency response.

Left panel, front DEQ, 10ms (default) delay


Left, front DEQ, 12ms delay

Left, front DEQ, 14ms delay

Left, front DEQ, 16ms delay



Left, front DEQ, 20ms delay

14ms is clearly the best here, and very similar (or better!) than my chairside fine tuning.  I'm shocked it makes so much difference.  But now I think like this: 8ms is one full cycle of 125 Hz, 4ms one half cycle, and 2 is 1/4 cycle.  To go from best (full add) to worst (full cancellation) it takes 4ms.  So my default position was (pretty close to if not) the worst, and 14ms is (pretty close to) the best.

One problem is that there will be another "best" point at 6ms, 4ms less than my default 10ms value.  Random noise measurements alone are probably not good enough to determine whether 6ms or 14ms is the true best choice.  Listening might nail it, or some kind of impulse testing.

There are two things I need to figure out.  1) What is the true best optimal panel delay (relative to the current 8.x ms sub delays, it all could be renormalized around a 10ms panel delay all over again afterwards).  2) What is the latency difference between the chairside and front panel DEQ's.

As long as I'm doing "optimal crossover addition" (which may depend a lot on PEQ's) as the test for optimal delay, should I optimize both channels?  I need to keep panel delays identical, but the sub delays possibly need adjustment relative to each other wrt the new EQ's anyway.

Of the above curves, 14ms looks closest to what I was getting with the chairside DEQ.  But 14ms sucked out the crossover frequency in the right channel, so that can't be it.

Right, DEQ ?, 10ms delay


Right, front DEQ, 14ms delay

Right, front DEQ, 16ms delay

13ms looked pretty good in both channels.

Right channel, front DEQ, 13ms delay


Left, front DEQ, 13ms delay

At 13ms, the left looks closer to the chairside than the right, so perhaps slighly smaller delay would bring closer agreement (since the right gets worse after leaving 10ms, backing up to that would make it better).

It would be reasonable to try 12.5 ms and maybe 12.75.  We know 12ms is not good in the left at all so we shouldn't go there.

Once that is figured out, I can easily determine the other listening test case by simply subtracting 8ms.  At this point, I don't even know if what the chairside unit is doing is correct.  It's just a coincidence that the difference in delay turned out to work better in my pink noise testing, but it could be that it's actually being delayed either too much or to little by some multiple of 8ms and that could still be the case.

Testing the chairside vs the panel DEQ's at plus and minus conditions, it should be possible to tell which one matches the chairside better AND which one sounds better (which might not be the same), and once I've developed that skill, I might try 8ms further...

8ms delay is enough to seriously upset balance, among other things (so I could try split DEQ testing too).

I've been pondering whether I could do some kind of technical testing, say with impulses and digital recording from the electronics...but it still sounds easier/better to work acoustically if I can now.  I don't need microsecond matching here, 0.2ms is good for practical use of the chairside DAC though 0.02ms would be wonderful, that's the finest adjustment available on the DEQ.  (0.2ms is already down to about 3 inches, 0.2ms is 0.3 inches of sound travel say if you were mechanically adjusting driver distances.  At low frequencies it's hard to adjust things that finely with typical impulses, which look very rounded coming out of subs, using digital recording or analysis.  I think I ultimately relied heavily on the measured distances, which I could easily measure to high accuracy but they're fraught with issues such as the planar speaker being so big there's not just one "spot" where they add.  Plus things may have moved since then without me updating the adjustments.)

It occurs to me that the optimal adding at the crossover frequency should be true regardless of the bass and panel relative levels, and especially because I use a Linkwitz-Riley crossover which has highest adding in the listening axis, and further because it is a phase-corrected Linkwitz-Riley crossover.  Basically I could do a pretty good job tuning the relative sub/panel delay just by optimizing the crossover frequency response.

There are some issues though.  The first is that there is more than direct sound involved.  I do need to optimize the delay for the direct response, but not necessarily the reverberant response.  The only way to really test for delays is with impulses (which is what I did last time, around 2022 IIRC).

For now, I'll just assume since I'm measuring at one spot, that's what counts, I'll ultimately try to optimize bass by adjusting the sub delays.  But I also have to get the DEQ compensation delay.

I can't read the date code on my front panel DEQ, but it's not a paper label it's printed.  I believe that was a later vintage when they just stamped them out in big batches.  The date code on my chairside must be one of the earliest at 0203.  So it stands to reason the front panel DEQ might have less latency, and require some to be added in to match the chairside.  Meanwhile, the sub DEQ is an earlier vintage than the panel one, a factor I'd never thought about before.  It may have been swapped in to replace an even earlier unit at some point, but it looks like the panel DAC was the last one swapped in.  The sub time alignment may have been done prior to that swap, or not, I don't recall.

It would be less likely for the newer unit to have more latency, and therefore require delay to be subtracted.  (Though, right now I don't know if the front panel DEQ is actually earlier.)

****

I tried gridding (delays of 13.0, 13.1, 13.2, etc) and searching several times on Saturday.  It did not work out.  Usually I gave up or got confused before too long.  I took dozens of spectra, some looked like they were going in the right direction, but I never found anything overall better than the 3.06 delay difference (10 with chairside DEQ vs 13.06 with front DEQ).  Those only seemed a little different, but still significantly different.  I was going to write it all up but got confused over which spectrum was which.

Chairside DEQ at 10ms delay



Front DEQ at 13.06ms delay

The front DEQ has a little more upturn just before the 125 Hz crossover (or maybe it's that 125 is slightly more depressed), and a bit more downturn afterwards.  Varying the front delay it was easy to get the relative pre-125Hz upturn bigger, but not smaller.  When it was made smaller, it tended to go with the lagging part turning down even more.

I was stuck in something like gambler's fallacy as I was doing this.  It seemed the big break was just around the corner, but never was.

So today I decided to go the entirely other approach.  I decided to hook the two DEQ's to an oscilloscope (which I can do from their analog outputs) and run a polarity test signal (all positive pulses).

Of course I also needed to set both "delays" to Main since I was using the "Main" (analog) output.

It was a very sensitive test to small timing differences.  I found a near perfect match closest to 15.16 on the front panel DEQ, and at 15.14 it was significantly different.

Delay match at +5.16ms at front DEQ


Delay mismatch at +5.14ms at front DEQ

This kind of delay match however can't distinguish between leading delay and following delay because it's a set of evenly spaced pulses.  You can't tell whether which pulses to match!  (Note that the wiggling of the impulse has to do with all the PEQ's, when I tested w/o those PEQ's it was a very basic looking pulse.)  So we don't know if any pulse matches the forwards one or the backwards one in the other series.  (If I had been smarter, I'd have made the test <pulse> <space> <pulse> <2 x space>.)

But since the crossover frequency is 125 Hz, that repeats every 8ms, which means that 15.16-8 or 7.16 should be another matching point (which might be better).  Which one could be determined by a listening test or better measurement test.

But when I ran the pulses, I noticed something else that that I neglected to make clear in these photos.  The two DEQ's are out of polarity with one another!!!  To make these photos I had to invert the polarity of one of the signals at the scope.  It is the chairside unit which appears to be out-of-polarity (pulses going downwards without inversion) while the front unit appears to be correct.

Chairside DEQ, all AES connections, default preset, Emotiva DAC, scope inversion off

I figured this had to be a measurement error of some kind, and tried to isolate it.

I tried changing both DEQ's to the default clear preset (no PEQ's).  The pulses got more square but the inversion remained.

I tried connecting the chairside DEQ via AES like the front instead of optical.  The inversion remained.

I tried swapping the XLR to RCA adapters being used.  No change.

I tried switching the 20dB attenuator on the back panel of the DEQ.  The inversion remained.

This is extremely concerning in many ways.  There is no control in the Behringer 2496 DEQ that controls polarity, either for one or both channels.  I don't understand how this could happen except for:

1) a bug in some vintages of the Behringer DEQ (notably the chairside unit appears to be a very early vintage).

2) Somebody altered the analog stage of the Behringer DEQ swapping the XLR pins (or maybe one of the units was sold in EU or UK, where such difference was standard).  I bought the chairside unit used so I don't know anything about its history.

This all could be (and probably is I think) limited to the analog outputs, which I never use anyway.  What really counts is what happens at the digital outputs.  And the only way I can check that is by connecting a DAC to the chairside unit.  (The main panel DAC is currently muted for safety, I don't want to blow up my speakers with a test signal.)

So I hooked up a spare DAC, and sadly it also made no difference.  The inversion of the signal occurs in the digital domain in the chairside unit regardless of AES or Optical input or the preset chosen.

(Note the AES and optical inputs originate from miniDSP which might have inversion associated with either output.  But apparently not, they are the same.)

The broader disturbing thing is that it appears I cannot trust any random DEQ to not do polarity inversion.  I'd always assumed they were all correct, but some (and I don't know how many) are not.

A less disturbing thing is that apparently I cannot use the current Chairside DEQ except for adjustments where polarity inversion is not an issue.  It is a very serious issue for the sub/panel crossover, however the differences may be limited to a narrow region around the crossover frequency.

I also cannot use Chairside DEQ for serious listening, only for temporary adjustment purposes, and at lest a half octave away from the crossover frequency.

Now it's perfectly clear that no adjustment of the delays will get the two DEQ's to match.  In fact, where it shows the greatest match on the DEQ's, it might actually be the worst match, because the delay is partially compensating for the polarity inversion.

But the results I get on the front panel DEQ are the results I get.  That DAC has the correct polarity.  And the current results are already pretty good in both channels when the delay is set in the range 13-15ms.

****

One trick I could do to use the chairside DEQ is to swap it with whatever crossover way of the system is NOT being adjusted during that adjustment, and compensate that way by reversing the polarity again at it's output.  Oh boy.  If I were to permanently swap the sub DEQ with the chairside, assuming sub DEQ is correct, and then compensate at the sub controls, that would make the chairside match the panel DEQ but no longer match the sub DEQ.

A benefit of having (at least?) one DEQ that's polarity inverted in the digital domain is that I could use it for on-the-fly inversion by simply selecting which DEQ is being used.  However, that's useless for me because I have 2 ways, panels and subs, which BOTH need to be simultaneously polarity inverted.  I can't just polarity invert one of them, unless I had two such out-of-polarity DEQ's not already in use.

****

So making the DEQ's match is a hopeless endeavor.  This spare DEQ can only be used in certain ways, just not for "chairside" purposes without large reservations (which I am now staring at, the final adjustments around the crossover frequency including the delay itself cannot be done on the chairside because it has mismatching polarity).

To avoid all the kneeling necessary to try each little change to the DEQ, I would have to move that DEQ to the chairside position, which is dooable but also problematic.

Possible improvements:

Raise front DEQ temporarily on stand (interferes with stereo image)
Relocate front DEQ to chairside (must be moved back for parties)
Use midi (still may have to use controls a lot)
Get new(er) DEQ with correct polarity.
Figure things out with less trial and error (haha)

****

Just working with the front DEQ (and kneeling on my 3inch foam pad) I can wrap the adjustment project up for now with just a few more adjustments.

The primary one of these is setting the sub/panel delay correctly.

My current tests can't even for sure tell me if I'm in the correct 8ms ballpark.  I need a polarity test that allows me to be sure of things like that, using asymmetrical delays.  One must then match up no only the pulse directions, but the pattern of delays between them.

Now a single impulse would be just great, if it could happen at the moment you want and be stored using storage oscilloscope mode.  But that's hard to do.  It's easier just to have a test as I just described, which can be freely running just like the freely running oscilloscope.

All you need is pulses that are widely enough spaced that the width between pulses (or shorter series of pulses) is greater than the unknown latency.  So if I want to decode an unknown latency with 8ms repeats, I need merely make a larger gap.

20 ms might work great, or even 200ms

It turns out my existing polarity test has about a 5ms gap between pulses.  Therefore there should be matches every 5ms.  The matching on this test has nothing to do with my crossover frequency (as I was just thinking).

It's good enough if the unknown latency is < 5ms, but it isn't.

I didn't look for one...but there might be a match right around 10ms...which might mean there isn't any delay difference at all!  I need a bigger gap to be sure.  It might only be the sad polarity difference that was causing all the trouble, and I was trying to offset it by just enough delay, which would be exactly 4ms, which is why I seemed to be ending up close to 14ms in both channels.  THAT mis-matching did have to do with the crossover frequency.

The matching with my pulse test is related to the pulse gap in that test, not the crossover frequency!

People say the latency of DEQ 2496 is about 1ms.  It's likely then that even early units did not vary by 4ms, and that varying latency wasn't the issue for me at all, only the polarity difference.

I need to re-check my results above and see if there isn't a match right at 0ms delay difference rather than almost 5 (which happens to match my test too well).

And then check latency/delay of sub unit.

But I've also wanted to know for a long time about the maximum output level of the Emotive Stealth dac.  So I checked that first, using a test signal with a sine wave (200 Hz) generated at maximum level (which I re-checked in Audacity, it showed clipping on every peak when I set the clipping threshold to 1, but no clipping at all when I set the clipping threshold to 2, meaning at every peak it just barely touches the maximum digital level, this test had been generated in either Audacity or Sox).

Playing this signal, I set the Emotiva DAC to +12dB gain.  That would be about 8V output.  And it showed no clipping whatsoever.

Emotiva at +12dB output level

I wondered how sensitive this test was.  Might there be clipping and I just wasn't seeing it?  So I artificially created digital clipping using digital gain in my Tact preamp.  I was able to detect the clipping at a level of +0.2dB.  I took this picture at +0.5dB:

+0.5dB digital clipping

Note that this second test is not showing any kind of clipping in the actual DAC.   There is no way to know with these tests how much additional analog headroom the Emotiva DAC has above 8V output.  It may have some additional headroom to handle inter-sample-overs (ISOs) produced by its own digital filter.  I'd need to run a test which produces ISOs to be sure.  J-Test does that pretty well.  Or 90's recordings by Steely Dan.  The maximum theoretically possible ISO would be at +6dB, but many digital filters might not ever go that high.

This doesn't matter for me because I don't really need the 8V output anyway.  My Aragon amp reaches rated power at 1.7V and reaches the end of its headroom well before 3.4V.  But I may run the DAC at +12dB of gain anyway, and I just wanted to be sure it can output to at least the nominal 8V level, in case there are ISOs up to that level.

Then I proceeded to re-compare the latencies between the chairside and front panel DEQ's, but this time looking for alignment of the two outputs near 0ms instead of near 4ms.  In fact I found such an alignment with the panel DAC set only 0.1ms lower.  For some reason, the older chairside DAC actually has less latency rather than more as I had been assuming.  Perhaps in fixing the output polarity problem in later DEQ's, Behringer added one more stage to the output line amplifier which increased its latency by 0.1ms.

Panel DEQ at alignment with chairside DEQ set to 10ms

This 0.1ms is effectively inconsequential at the subwoofer crossover frequency of 125 Hz.  0.1ms corresponds to a frequency of 10,000 Hz and might have tiny effects down to 1khz at most.  Still I generally at least try to set delay times closer than that.  It would be the same about as about 1.5 inches of speaker distance.

So I was wrong about different delays causing the differences around the sub crossover frequency.  The only thing that is doing that is that the chairside DEQ is out of polarity while the front panel DEQ (much newer) has correct polarity.

What about the front sub DEQ, which is an intermediate vintage.  The date code on the sub DEQ is 1411 vs the 0203 for the chairside DEQ.  The panel DEQ appears to be the ultimate vintage in which they didn't bother with date codes anymore.  It could be that the chairside DEQ is the 2nd year of production whereas the sub DEQ is from the 14th year of production.  The DEQ was still being sold for something like 20 years after its introduction (and now there isn't any kind of replacement).

The front sub DEQ was exactly like the front panel DEQ.  It required the exact same 0.1ms difference in delay times to match the chairside DEQ.  And it had the correct polarity too.

I'm very happy that my two front DEQ's match perfectly despite having somewhat different vintages.  And of course I'm very disappointed that my chairside DEQ has the wrong polarity, and mildly disappointed that it has 0.1ms less latency.

Sadly because of the differences, I cannot use the chairside DEQ for the final adjustments around the crossover frequency.

Speaking of which, the first thing is to determine the correct sub polarity.  Since I was getting better looking results with the chairside DEQ which inverts polarity, and I struggled and failed to produce as good looking results with the panel DEQ which doesn't invert polarity, it suggest the subs have been in the wrong polarity all along.

Only after I've figured out the correct polarity, and then the correct delay, should I get into the final tuning of the notches and boosts I've added around the crossover frequency, and only then can I get to the balancing questions.

***** End

Addendum:  It seems I can still buy a DEQ 2496 brand new (!!!) for $279 from Sweetwater.  They project expected delivery in January 2025.  Many ebay sellers are selling parts-only units for more than that, and quite a few of the pre-owned units have date codes with leading 0 and are therefore might still have the polarity inversion bug.  One ebay seller is selling a "new in box" unit for over $800 collectible price.

I have about a half dozen broken ones myself (currently I have 4 in use not counting the chairside one, and I'd be using 5 except the fifth one broke a few years ago...it was doing the sub EQ in the kitchen which is now approximated by my HTR.  I started to repair one last year but got hung up on the lead free solder, which I wasn't experienced enough with yet.  I could fix that one soon.  But I have no idea how many of my broken ones have the polarity inversion bug, which seems to have been fixed in the newer ones.