Thursday, March 28, 2013

More speaker fine tuning

Some of the response curves showed a bit of peaking above 6khz, that had me concerned that I was too much on-axis with the acoustats.  One needs to be slightly off axis for the best sound, this can be tuned best by ear, but measurement sometimes helps too.  I have felt quite often in the past few months that my close-up position was too much on-axis, having a slightly peaky sound.

So I moved both speakers slightly straighter, by first angling out the super tweeters about a half inch, then moving the speakers to match.

After doing this I also measured the speaker distances to the nose-position microphone.  I was very surprised to find how short the distances are now, about 40 inches to the center of the panels.  But the left side was clearly shorter, I first measured 36 inches but 38 seemed more accurate, whereas the right side was at 40 inches.

So I moved the left speaker back as much as I could, which wasn't much, hardly an inch.  When doing the full system response curves, there was still a slight gap between the impulse in the right and left channels, the left channel still seemed to measure about 0.03 ms faster as if it were about 6mm closer.  BTW, that is a fraction of an inch, about 1/4 inch.

I decided to fudge this one with the Behringer DEQ.  I dialed in short delay of 0.03 in the left panel and  added 0.03 to the existing delay for the left tweeter also, figuring it might be more delayed now.  It seemed that after I set the "unlink" option globally, I could set short delays separately for each channel. I had not figured that out before.

This yielded a measurement in which the impulses for the two channels exactly lined up.  But in listening, it seemed slightly skewed to the right.  So actually I reversed the DEQ setting, dialing in short delay for the right panels only of 0.03 ms, and undoing the delays I had previously added on the left.  That sounded correct.  I can't explain why, perhaps I don't have the microphone positioned correctly or my head is slightly asymmetrical.  In principle I could have moved the panels.  I chose not to do measurements after making this final tweak by ear, since the measurements seemed to steer me wrong on this one.

Another change I made was to reduce the bass levels on both sides.  The bass has been sounding a bit boomy, and the measurements showed  a rise in bass below 80 Hz and the left channel rising above the right in the bass below about 40 Hz, so I took about 1.5 dB off the right sub and 3.0 dB off the left sub.  That seemed to make both the measurements and the sound better, but I subsequently noticed that the Tact measurements are inconsistent, even averaging 40 trials, sometimes the right channel exceeds the level of bass in the left channel all the way down to 20 Hz, and other times left channel has more below 50 Hz.  So some additional relative subwoofer adjusting may be required, about a dB or two, but the current adjustment is an improvement in reducing the boominess if not level matching.

Saturday, March 23, 2013

Tact good for tweeter alignment

While I have concluded that the low frequency resolution of the Tact 2.0 measurement program (or more precisely, the display of the measurement program, but it might be the program itself, because of the type of pulse it uses and the number of bins) doesn't work well for subwoofer time alignment.

But it works very well for supertweeter time alignment.  A tone burst (which may be partly digital artifacts) appears in the supertweeter channel.  And using two channels is fine, I was wrong about the delay being auto-adjusted to make the leading edge of the tweeter signal line up with the leading edge of the Acoustat signal.

After the first measurement, it appeared that the supertweeter was lagging by about 0.16 ms.  So I adjusted that exactly (for some reason, I used the "short time delay" menu in the Behringer) in the right channel and got this picture, where white is the tweeter and yellow-orange is the Acoustats:

Yes, for some reason the leading edge of the acoustat signal appears to go down, out-of-polarity.  But I believe the main part of the pulse is what follows, and it goes up.  I still don't understand this.  But looking at the above picture you see that I have lined up the leading downward pulse from the panels with the leading edge of the squiggly burst from the super tweeter, which might go up (depending on which pixel you look at, some of the leading edge of the tweeter signal looks like digital artifact pre-ringing which can be ignored.  So I ignored tiny pixels, but chose the first decent looking line as the leading edge, which I admit is a judgement call.)  There is a bit of ambiguity here as to where the tweeter signal really starts, but we are within a few 0.01 ms here.  When I started, the tweeter burst was half way further down the screen, and that was a mere 0.16 ms difference.  For absolute perfection here, a better measuring device and/or listening may be required.

After doing the above measurement, I realized that the two supertweeters were not correspondingly positioned for the two panels.  I would have to adjust the other one to match this one.  But then it also occurred to me it might be better to push out both supertweeters all the way, so that the front edge of the stands for supertweeters line up with the stands of the Acoustats they are next to.  There is nothing magic about getting the two stands to line up, but it is a more reproducible positioning than most others (except having them line up on the back side) which is helpful for practical reasons.  I often move one or both supertweeters out of the living room for  parties.  If I have them calibrated for any particular position, it helps to make that position an easily reproducible one.

And since I am adjusting the delay anyway, I don't have to time align the positions of the two speakers for the reason people not having digital systems must.  I can choose to optimize the relative position of the panels and supertweeters for other reasons than actual time alignment.  In addition to the reproduciblity issue described in previous paragraph, there are also issues related to dispersion and diffraction.

Basically you don't want the supertweeter firing from behind other speakers, inside a hole as it were, because it's like talking through cupped hands.  If the tweeters are actually slightly forward of the Acoustats, that is helpful in reducing edge diffraction related to the sound projected by the supertweeters.  On the other hand, it could increase edge diffraction related to sound eminated from the Acoustats.  But that doesn't matter as much here for several reasons, the most important being that as a figure 8 speaker the acoustats don't signficantly project sound to the hard right or left, that's a null.  For another, the supertweeters are omnidirectional, which is exactly the opposite, they will product loads of diffraction and other undesirable addition effects when there are nearby boundaries.  Plus, one takes advantage of their omnidirectionality if they are slightly forward of the other speakers, getting more that 180 degrees of free radiating angle directed toward the user rather than in the other direction.

So I moved both the supertweeters out like that, and calibrated both channels like the above for correct time alignment achieved by digital delay.

I listened a bit to radio, KRTU because KPAC was playing opera, and the new setup is wonderful.  Somehow it is both more transparent, more spacious, and more relaxed.  I can also move my had a bit either way without the image seriously distorting, instead the image shifts gradually as I move my head.  This is all to the good, and I think having the supertweeters moved out and time aligned digitally is a big improvement, and a successful day's work.

Using the Tact to adjust crossover time delays

At long last I'm firing up the living room Tact 2.0 program (which runs on a Windows PC) to adjust the time delay that time aligns my Acoustat panels and SVS subwoofers.  I am looking at frequency response too.

Unfortunately for me the Tact program is oriented around room correction (not really measurement) so much that it's unclear to me whether the impulse shown after measuring my system is a full cycle (0, +1, -1, 0) because the tact is simply showing what it recorded, an image of its own test signal, or whether it shows a full cycle because of group delay in my Acoustat speaker, and if the Acoustat had no group delay the Tact would be showing a standard impulse that goes from 0 to +1 and then back to 0.  When I use the Liberty Audio Suite, it plays a signal that looks nothing like a simple pulse, but LAS then mathematically transforms it into a perfect step pulse graph if indeed the unit being tested has perfect pulse response (as most solid state electronics does, or comes close to, while speakers are usually far from it). I think that's called deconvolution, but it might be called convolution (I get the two confused).

Worse, I had a specific idea for adjusting the time delay between the Acoustat panels and the subs.  I reversed the amplifier connections for the acoustats.  Then I mute every speaker driver except the Left channel Acoustat (which is actually playing on the right, because I reversed them) and the right channel sub.  Then when I run Tact in full stereo mode, it plays both drivers for the right channel but as if they were left and right full range speakers.  Then, in principle, I could verify and adjust time delay, by comparing the onset of left and right channels in the impulse picture the Tact produces.

However, because the Tact is so oriented around measurement, it uses some trigger to put both channels into alignment even if they are not.  Or at least that is what it seems.  Whenever I do this measurement, the bass looks like it starts about 6.0 ms (millisecond) later than the panels.  Which is very strange because the speakers are only about 3 feet apart, and I was already applying about 3.4 ms of delay to the panels to compensate.  So the expected error, or difference, should have been less than 1 ms.  But as this measurment appears, there is also a box showing the relative time difference between the channels (which the Tact is compensating for).  And it shows a really big number, like 20 ms.  So once again I don't know what the impulse picture means.  Is it showing the raw measurement before the 20 msec correction it is ultimately applying in the chosen correction number?  Or is it showing the two channels adjusted, as it thinks they should be?  If it is showing them after correction, I can't use this measurement.

Because the Tact seemed to be showing about a 6ms gap with the subs starting later, I increased the time delay on the panels, but for some reason I don't now remember, or possibly just lapse of memory, I increased the delay adjustement (in the Behringer 2496 DEQ) for the Acoustats from 2.47ms to 5.53ms.  I would have made the second 5.47 but the adjustment is course and that exact number is not available.  Anyway, that seemed to have little effect on the graph.

So then I tried bigger adjustments, 9ms, 12ms, 10ms, and in every case it wasn't clear whether it made the bass alignment with the treble impulse better or worse.  Those numbers made no sense at all, the required delay should be between 1 and 4 ms.  But in every case it still looked like 8-12 ms of additional adjustment was needed.

One problem is that the bass may start more slowly, for various reasons, especially the crossover, but also it's limited high frequency response.  It should, I believe, begin moving the instant current is applied, but at first slowly, then building up to a full wave.  The resolution of the display is low enough, and there is also noise, so it is ambiguous where the subwoofer output really begins.

After messing around with the primary "Measurments and Correction" page for hours, the only one that actually allows you to do measurments (and you must have a non-bypass correction number selected, it won't let you do measurments in bypass mode) I finally went over to the dual-domain page, where you can load the previously made measurement, and there it seemed I could get a pretty clear image of the treble impulse, and see that it started around 11.8 ms in the right channel.  (This was with all crossover settings restored to original settings.)  Then I played the bass.  Well, it could have been correct, you could see some rising, maybe, in the bass response at 11.8 ms, where the red line is in the graph below.  But it could also be as fast as 2ms, or as slow as 15 ms, it's hard to be sure, because there is noise and the initial start might be slow, very slow.  True, at about 15 ms it really starts going, it's clearly going at that point, though not as strong as later, there's a cycle of reduced output before the full response builds, it's typical for speakers, especially bass speakers being crossed over, to respond that way.  I have to think that on a higher resolution plot, the very initial slow part, corresponding to the limited high frequency response of the subwoofer, there would be a more clearly visible starting point, and it would be very close to the red line at 11.8 ms, because I was already compensating for the delay with digitial delay in the crossover which should have been accurate to a few inches, which would correspond to about 0.1ms or so, because sound travels 1foot in about 1 ms.  On the other hand, frustratingly, I can't be sure, because it may well be that even with the crossover and all filters in the SVS subwoofer turned off (as usual, except the sub 15 Hz filter I am required to run with only one port filled) there is time delay in the electronics, that's different between the sub amp and the acoustat amp.  I can't explain that much difference in delay happening from analog elctronic processing.  I can explain the 11.8 ms delay as about 6 ms from distance to the microphone, and about 6 ms fixed in the digital processing from the output of the Tact through the Behringer DEQ.  But that should not vary between the treble and the bass.

Anyway, as the graph below shows, the Tact measurement system is useless for setting the delay on the bass, because the starting point of the bass is ambiguous with such limited resolution.  And the graph below was made with crossover turned off, and the bass does have HF response to about 300 Hz, so I think a good measurement system would clearly show the starting point better than this.  Part of the problem here...the Tact has a simple impulse that has limited low frequency information, hence, limited low frequency resolution.  A Maximum Length Sequence system, like liberty audio suite, uses chirps, which have a better spectral distribution than clicks, allowing greater bass resolution.