Monday, November 23, 2015

System tuning requires iterative process

For several years now, I had not used my GenRad Oscillator because I thought it was broken.  I finally got around to removing the MDA 0.25A fuses, and found they measured about 5 ohms.  Then I looked up the cold resistance for such fuses, and it's rated as 9 ohms.  So by the book these fuses could still be good.  I put them back into the oscillator, and, It Worked!  Sometimes things are just like that.

This is the kind of oscillator best used for speaker system tuning because it has a big full turning logarithmic frequency control which sweeps the decade 20Hz-200Hz, and going a bit lower than 20Hz actually.  So it's easy to do useful frequency sweeps at any speed, and the resolution is immense.  I have other oscillators, such as a B&K with a linear control, and it's nearly useless for speaker tuning.

Now I had really intended to do measurements, and possibly parametric EQ calculations, by Room Eq Wizard.  I did my first measurements with REW in January this year.  And I haven't done any since, because there is some complication in setting up the computer, interface, microphone, and stand.  And it could be argued that is far better than my very old fashioned approach of sweeping an oscillator by hand and adjusting EQ's to fit by ear.

But just because I got the oscillator working, which was a little miracle, I went ahead and checked out the oscillator.  I wasn't planning to spend a whole day, just a few minutes.  But I ended up only finishing the tunings (they're only finished when you can write them down and get them recorded and measured without being forced, simply forced by one's own ego, to change them again) about 20 hours later, with 8 hours sleep and breakfast in the middle.  Now in the process, I did listen to a lot of music also, which then sometimes motivated me to make more changes.

So like I said, many times, including after the first hour or so, I tried to write down the adjustments, and move on.  But just going over them, I couldn't leave them alone.  I couldn't believe they were done correctly, I must have been mistaking the Q as I often do, not fully remembering the correspondences to bandwidths.

So that's how it went, just changing them, sometimes just because I thought they must be wrong.  Often making the changes, at first, or in the middle, I didn't bother to verify.  But then towards the end it became clear that verifying is the key step.  And not just at any one frequency, but sweeping up and down at various speeds.  And that's because the correction process is far more complex than you might think.  It's far more complex than most automatic systems can actually handle.

That's because the speaker and room system has a lot of possible states.  And those states, the position of compressions and rarefactions in the air at any one time, and the positions of everything in the room, and so on, are very very complex, even if ultimately "linear" it is far more complex than any single set of measurements can reveal, and more complex than a single deconvolution can address.  (And the simplification in most FFT's of what is going on in the nearly absurd.  You absolutely cannot do a tuning based on 1/6 octave FFT, though that can be an additional useful test and verification.  But imagine a problem with 10,000 variables, and you're modeling it with 30.)

It's very much like this, you have a particular peak, so you make an exactly equivalent notch to notch it out.  But it doesn't notch it out.  You're left with at least half as much as you started with, most likely, and possibly slightly shifted in frequency and q, or split into two separate frequencies and Q's.

So the only way a correction can be done is iteratively, not just in the mathematics but in the measurements.  You make one measurement, do one set of fixes, make another measurement, make another set of fixes to the fixes, and so on.

Since automatic systems generally just do one measurement and pray for the best, they can't possibly be as good as obsessive kid with his oscillator, sweeping up and down until all the bumps are gone, at least if he had infinite time.

Unfortunately, I was so unsystematic in everything, I can't assuredly say that my work here is truly great, though it's a vast improvement over the way things were before.  I really should do R.E.W measurements along with the oscillator sweeps, and so on.  But for now I am going to let it go, because I've already invested too much time in this for now, and it's actually pretty good too.  FWIW, here's the smoothed 1/6 octave measurements from Analyzer app on my iPhone:

That might look rough, for one you can only blame me for about 120 Hz and below, and it was far rougher as well as more complex at first.  The flat shelf from about 40Hz to 100Hz would be several huge peaks, falling off at the bottom for no good reason, and hugely bulging at 100Hz to sound like mid fi.

In the process of doing this, I discovered that my disc players, not often used, had channels reversed.  I think you can't blame the audiophile with so many different things that can go wrong, and constantly experimenting, for getting lots of things wrong.

Oscillator system tunings, November 2015

LR24 at 80Hz using two 2nd order Q.707 filters in combination
Right Only, 120 Hz, 1/2 octave, -4dB

32Hz +1.0dB Q2.8
40Hz -4dB Q6.3
45Hz -6.4dB Q4.5
57Hz -3dB Q4.5
69Hz -3.5dB Q7.9
96Hz -8.4dB Q6.3

32Hz +1.5dB Q3.2
40Hz -7dB Q7.9
46Hz -9dB Q4.0
51Hz -4dB Q7.0
71Hz -8dB Q7.1
LR48 (oops!?  I thought I changed that to LR24 to be symmetrical with panels)
-10.6dB (both)

Super Tweeters
Set down 6dB after monumental bulge on iPhone analyzer, still about 4dB bulge, believed correct for wider dispersion pattern in random noise.

No comments:

Post a Comment