Wednesday, January 19, 2011

Crossover Redesign, Day One

The 3 day MLK weekend featured cold rainy weather, so I had the perfect alibi to spend 2 days measuring, re-thinking, and re-designing the crossover of my main Living Room System.*  This effort is still continuing (now in something like Day Three or more as I am writing this) and I think it has already been extremely beneficial to the sound and will soon make possible a Wonderful New System once I rerun Room Correction.

(*Nominally I tend to use Linkwitz-Riley crossovers whenever possible; they have many advantages starting from the fact that the drivers are always in phase with each other.  Currently I use the LR48 slope built into my Behringer DCX 2496 crossover; in addition to being a Linkwitz Riley crossover it is very steep and the drivers are kept in the same polarity.  But quite often I adjust crossovers to perform additional frequency countouring by using different crossover frequencies for high and lowpass, etc.  This is needed to correct speaker and/or room response, and that is the subject of this post.)

The state of things in late 2010 had clearly gone downhill when I made the measurements (see back a few posts) that showed startling lack of bass response below 100Hz.  What happened?  I had not only previously measured this as actually boosted below 100Hz but was using a Tact RCS (Room Correction System) 2.0 system to realize my "room curve" which is essentially flat with bass boost below 30Hz.  What happened was that over the course of the year, small changes tended to roll back the bass levels.  Why were these small changes being made?  Because despite reasonable measurements and RCS, the bass response had time smearing properties because of the previous way I had "designed" the crossover between subwoofer(s) and electrostatic panels.

That was waaay back in 2008, I had played pure bass tones from my Kurzweil synthesizer.  One of the concerns that I had then was lack of bass in the lowest regions to be reproduced by the panels, in the 100-200Hz range.  I used Ivie IE30, Behringer DEQ 2496, and other devices to assess the frequency response and possible ways of setting the crossover.  But that was so long ago, I can't really remember much about it except that it proved to me it was OK to use a single subwoofer crossed as high as 86 Hz.

Then later, when I bought the second subwoofer in January 2010, I decided to use higher frequency crossovers to avoid damaging the panels like I had in March 2009 (Acoustat panels cannot arc but you can burn up the transformers with enough sustained high volume levels).  To allow me to play as loud as possible with no worries, I pushed the crossover frequency all the way up to 121 Hz.  (Previously, I had experimented with crossovers as low as 40Hz.)

By this time, I was already using Tact RCS 2.0 room correction, but I felt I could get the best results out of correction by providing plenty of "material" (sound output) for the RCS algorithms to work with.  I would make up for the lack of uncorrected response in the region 100-200Hz by raising the subwoofer lowpass all the way to about 165Hz.  So the uncorrected response was now rather "thick" in the bass, but I had no fear about this believing that RCS would fix it up.

Well, RCS did seem to fix it up (though I no longer have any measurements to prove that, wasn't blogging then).  But the RCS corrections were apparently very fragile.  If you weren't sitting in the correct spot (which moved to mid-room) the bass could get quite boomy.  So over time, I turned down the bass, leading to the current situation.

All new work is starting from NO RCS CORRECTION to make it more robust this time.  RCS will be added to a good working system to make it even better, rather than being added to a system which merely has enough "material".

Pink noise measurements like the ones in the previous post strongly suggested that the high bass lowpass was not good.  When the lowpass was brought down to 121 Hz used by the highpass, response got a lot better.

For several days, I reveled in the idea that maybe I should stick with idealized crossovers, just set the controls so high pass and low pass are the same.  But it has passed now, because I found something even better.

Even with low and highpass set to 121 Hz, there was still considerable peaking around 100Hz as shown just below.

So I got to thinking, what if I lowered the lowpass even more.  Now this is really going in the opposite direction from what I did last time, but I thought now it might be worth testing.

Sure enough, I can lower the subwoofer lowpass as low as 72Hz and still get seemingly nice response up to 121 Hz where the panels start crossing out.  In fact, it is still slightly peaky if I do that, but the peak is lowered to about 80Hz and much smaller.

Uncorrected, this is far better sounding than the straight 121Hz crossover, and has much less peaking in the 100Hz region.

Now some people might complain that I am not crossing over "correctly" because high and low pass frequencies are not set the same.  But the proof is in the pudding, I get far flatter response separating the low and high pass.  Can I get away with this, and should I?

Yes and yes.  It turns out that, very conveniently, speakers and rooms have errors mostly of the type called "minimum phase".  If you add a minimum phase error to a minimum phase correction that perfectly cancels it for flat frequency response, it turns out you have also perfectly corrected the phase response as well.  (Of course, nothing is ever "perfect" but it can be close.)

Now what is happening in this case?  I believe the room has a strong resonant node in the 80-110Hz region caused by ceiling bounce (the ceiling ranges from 8'2" to 9'8").  This has half wave nodes in the 45Hz region, and full wave nodes in the 90Hz region.

A much smaller factor may be that the subwoofers have resonance in the 80-110Hz region.  I have not tried to measure such a resonance, but I do sense that the SVS PB-13 is not as perfectly flat to 200Hz as the website suggested.

I currently think the room resonance is the larger factor, and it is very convenient that I can cancel it out by pushing down the lowpass cutoff of the subwoofer even below the highpass cutoff of my panels.  I have used this trick previously in other systems, but never documented it as well as I am doing here.

Later in the day, the strange notch frequency around 4-6K began to bother me.  But first I asked myself "is this real, or an artifact of the iPhone RTA measuring device."  I tried running a spectrum of my bedroom system, based on Revel M20 speakers, which I had previously measured with an Ivie IE30 as being nearly perfectly flat above 1Khz.

Well, that spectrum showed exactly the same notchout at 4-6K, followed by a boost from 8K-13K, followed by a notch, followed by a boost up to 20K.  In short, it looked almost exactly the same.

I then checked a spectrum I had run of a friends system (a very committed audiophile) in December.  Same features in the response above 4K.

Three systems with the same response above 4K?  Not likely, especially as I had previously measured one of those systems as being flat, not what was being shown.  I conclude that the iPhone RTA measurement above 4K can't be trusted, or it means more-or-less flat if it shows the pattern shown above.

I think the bass response  of RTA is fairly trustworthy, however, and it looked different on each of the three systems tested.

But it is good to beware, even if you like to do objective audio measurements as I do, that any particularly measurement can be, and usually is, flawed in more than one way.  That's also why I have many different measurement instruments that do roughly the same thing, because they each reveal different things, and not one of them is perfect.  If I had lots more money, I could afford more perfect measuring instruments, they are extremely expensive.

A real time analyzer with lots of bands is one of the most handy tools for setting up an audio system.  However, it can also be misleading.  It could be misleading even if it had flat response.  Frequency coutours of a speaker or room are obscured by the fixed frequency bandwidths.  One can get much closer to the actual sound by using a sine wave generator.  Plus, with a sine wave generator, I can make spectrum level and continuity judgements by ear, and also use one of many Class A microphones I have to measure sound pressure level.

On day two, I brought out the B&K Function Generator to see what I had done and to make it even better...

No comments:

Post a Comment