posted 18 April 2003 02:04            

sonicmaximizer.kym

I went to the US Patent office to see if I could find anything on this after first having learned that it was invented in the early '80s by an engineer named Robert Crooks, who worked for Lockheed Aerospace before joining the Barcus-Berry Co. He did several rounds of inventions, culminating in the final patent #4,638,258 dated Jan 20, 1987.

Well reading patents is no fun, but I spotted a schematic diagram showing op-amps and R-C networks. I did the analysis on how this circuit works, and found that it does essentially what my Sound block does here, except for the fact that his filters are Bessel filters, while our Kyma filters are Butterworth. It makes a slight difference in the phase response.

What does this sound do?

1. Strip out the low frequencies and high frequencies with 2-pole lowpass and highpass Butterworth filters, cutoff frequencies of 600 Hz and 4 KHz.

2. Boost them with some gain, one for the highs and another for the lows.

3. Recombine them with the negated input signal. (???)

Point #3 seems to be the crux of how it really does its magic. The default gains in the VCS of this Sound are set to +6 dB. Some compensating attenuation is used in the final mixer to prevent DSP overload, most of which comes from the bass region.

When you do the analysis for +6 dB boost, what you find are two things:

1. The signal is boosted only slightly across the mid and high frequencies, about 5 dB, with peaks at 600 Hz and 4 KHz. So this makes it act like a mid-range emphasis and a kind of high presence filter. I think that has a lot to do with why people claim they love the sound -- that presence filter emphasizes the air in vocals and the snap in drum strikes,

2. There is a phase shift -- a lot of it, ranging from +180 degrees at the bass region and ramping down to -180 degrees at the highest frequencies. It is almost a straight line ramp in log-frequency. That supposedly corrects for speaker induced phase misalignment. But truthfully, I don't use speakers very often but it sounds terrific in headphones too.

If you use less than 6 dB boost in the high and low bands you find that it induces a lot of phase cancellation with the original signal (not surprisingly). So 6 dB seems like the minimum necessary gain in both bands.

Now I have a hard time believing that headphone speakers would behave the same way as a 3-way studio monitor, and I don't believe such a simple solution really works to "clean up" anything in particular. For one thing, I know from having measured my friends' homebrew speaker systems that they vary all over the place.

But here is a perfect example of something very simple and crude working magic on our minds. Just like in visual processing, sonic processing doesn't need to be much more than a hint of correctness in order to astonish our senses.

It works! and it works very well.

- DM

[ Woops! I see I had some old numbers in the wrong place and the analysis on the lowpass was incorrect. The cutoff should be about 63 Hz, not 600 Hz, if you want to match Robert Crooks version. I uploaded the corrected Sound.

But frankly I think it sounded better with a 600 Hz cutoff... At any rate the amplitude response is now peaked at the low and at the high end, and the phase response is still large at each end, monotonoically declining from low frequencies to high, but it isn't nearly so straight line ]

posted 18 April 2003 11:15            

I originally modeled it as a simple audio mixer with 3 inputs as you can see. But on closer inspection I see that may be incorrect. I now see what he is trying to accomplish, and op-amps are the way to go. He basically inserts the expected load impedance for the input leg to the final op-amp and then uses simple resistive feedback. The effect should be to cancel out the frequency dependency in the load impedance. Very clever!

So how do we accomplish this in the digital domain on Kyma??? And what the heck is that Sound that I just invented?

posted 18 April 2003 13:37            

I find that in my haste, I often stumble on really interesting things, but it takes me a while to figure out what's going on...

Now that it is really a high pass filter, the sound is quite neutral to my ears... hmmm... So I went to the BBE Web site and tried out their online demos for what it sounds like.... about the same as mine.

Is it me? or is there really very little difference in headphones? Anyone out there have a BBE SonicMaximizer?

posted 18 April 2003 17:38            

So now, it looks like BBE is on to something here. By inserting a filter into the input node of an op-amp and using a resistor for feedback, they create the reciprocal transfer function of that input filter. When the output of the op-amp feeds the speakers which have that same input transfer function, the result should be more flat.

Creating the reciprocal transfer function is simple for an op-amp. Anyone know how to do it in the digital domain on Kyma? (short of finding the poles and zeros and forming an inverse filter?)

- DM

[BTW... this is DECONVOLUTION!! I never looked at op-amps that way before, but they can be used to deconvolve a signal with a known transfer function. So the digital answer must lie in the domain of digital deconvolution... a terribly difficult undertaking.

What I want to find is a way of performing this on the fly with an arbitrary input network, just like an op-amp does, but I want to do it on Kyma. ]

posted 18 April 2003 20:05            

bbe.kym

I tried creating inverse filters, but I knew that would be trouble -- it was! Deconvolution always magnifies the worst parts of the spectrum, and inverse filters apply enormous gain where the signal is weakest. Bad news!

This version is a pretty close approximation to the inverse filter representing my HD600 headphones. It does apply an inverse phase shift, and it boosts those frequencies where the response of the headphone is down. It also happens to boost bass, but this is a byproduct of trying to get a large enough negative phase shift on the bass frequencies.

I tuned for phase response first, amplitude response secondly. It just so happens that the amplitude response is more or less correct too.

This version uses the Kyma shelving filters. The default low band processing has a cutoff of 63 Hz and about 12 dB boost in a low shelving filter. The high band cuts off at 4 KHz with a boost of 6 dB.

... and yes! it does make a noticeable difference.

- DM

[But now this begs the question... Recording engineers use real speakers and studio monitors to mix, so if I cancel out the headphone EQ, does that make things overly bright and too bassy? How can one ever know if a recording was mixed using BBE technology in the final mixing?]

posted 19 April 2003 17:32            

A simple PseudoDifferentiator Sound with the Scale of the Delay Sound set to around 0.7. Simple!

This produces an essentially flat phase response in the headphones from around 40 Hz to well above 10 KHz. Interestingly, the amplitude response is likewise flat to plus or minus a few dB over that entire range too.

Why does this work? Well here's how I figured it out...

What's a headphone to the circuit? A big coil! What neutralizes a coil's AC reactance? Series Capacitive reactance! What is the transfer function of a series capacitor? It's F all the way. Just the opposite of a series coil. But that's waaay to much attenuation at DC and low frequencies. But we'll get to that... What does a capacitor act like? A differentiator!

Now for that gain problem. Putting a zero right at the DC frequency makes waaay to much attenuation at the low bass end. So we can lighten up on that a little by modifying the differentiator to move its zero away from the unit circle. In other words, make that input scale factor to the delay block something less than one. I find 0.7 works well. When you do this, you need to apply a little post gain, about +6 dB, after the mixer for the input signal and its scaled delayed value, so you get back up to the levels you had before.

This works wonders and it is very simple.

posted 27 April 2003 02:29            

It is difficult to know without consulting a patent attorney just where the boundaries are on the ideas advanced by the original patents. I often feel the patents are overly broad, and that granting exclusive license to mere ideas is something that really gets my goat! I am personally offended by mathematics being owned by anyone. I will grant them the right to physical expressions of those ideas. If only I were Emperor!! Heh!