Circuit analysis, anyone?

[SSLtech]

Can anyone tell me in terms simple enough for my aching noggin to grasp, how this circuit works?

The bit that I'm struggling with is the phase-juggling stuff on the left... it's repeated a lot, but for an example, R6-R11, U1A and C1 and C2

Thanks,

Keith.

 

[gyraf]

All-pass filters for delay? Maybe to compensate for mic elements not being at the same point in space? :razz: just guessing...

 

[SSLtech]

Supposed to be phase shifting of some sort... it's a UHJ converter.

Oh, that brings to mind another question that I had to ask... I don;t see how that whole "time coherence through delaying capsules can possibly work... I'll split it off into another question I think.

Keith

 

[PRR]

this page halfway down "UHJ Converter"

The phase-shift networks are approximate; I dunno why he's matching caps to 1% when a 2 or 3 stage shifter can't cover the audio band much better than +/-20 degrees (I may be wrong). That level of precision was entirely fine for some early quadrasonic-on-2track work. I don't kwno what UHJ Converter is supposed to do or how accurate it needs to be.

Ah... it appears that UHJ is similar to antique quad techniques: UHJ link

The UHJ (standing for "Universal HJ") system of encoding directional sounds into two or more transmission channels, developed in conjunction with the British Broadcasting Corporation, conveys the Ambisonic soundfield to the final listener, and was designed to achieve outstanding mono and stereo compatibility. No mono listener should lose any important sonic information, no matter where the producer decides to place it in space. In addition, a stereo listener tuned into a UH J broadcast not only hears a full-width stereo presentation, but also enjoys an enhanced sense of depth and focus not possible with conventional stereo transmission systems. And, of course, the listener equipped with psycho-acoustically optimized Ambisonic surround sound decoders hears the precise pattern of surround sound intended by the producer.

UHJ Two- and Multi-Channel Systems
In its simplest form, UHJ uses the same two recording or transmission channels as conventional stereo, utilizing both amplitude and phase to convey a full 360-degree, horizontal sound stage.

 

[bcarso]

I get a kick out of the 5% values being padded to get exact values. It is true that you can now get 1% tolerance 5% values, but the schematic at least says nothing about this.

Also the caps are usually no better than 5%, although with only two values you might be able to get 2% parts a bit more easily/cheaply based on the quantities, if you were building a bunch of these.

KEF used a lot of Friend filters and used 1000pF just about everywhere. They got a deal on 1% parts.

 

[SSLtech]

[quote author="PRR"]The phase-shift networks are approximate; I dunno why he's matching caps to 1% when a 2 or 3 stage shifter can't cover the audio band much better than +/-20 degrees (I may be wrong).[/quote]

I thought I read that this thing was supposed to be linear across the audio band to within a degree or two... the mistake may be on my part however; I'll look it up further.

Keith

 

[PRR]

> how thic circuit works?

Ignore the R-C crapola.

If you look at the top-left opamp, it has the standard "Inverter" feedback on one input. The other input has an R-C mess to source and ground. Replace that with a switch that can either connect the "+" input to the source or to ground.

With "+" input strapped to ground, it obviously works as in inverter.

With "+" input strapped to source, it works as a follower. That is not obvious, but if you work it out you will see that it is so. Even if the "strap" is a resistor (much smaller than the opamp input resistance, which in this case is true).

That R-C crapola connects the "+" input to source or ground at different frequencies. Combined with follower/inverter action, you get big phase shifts. If you do the math right, the amplitude does not change (much).

Here is my stimulation:

Because the simulator thinks all cycles of a sine are identical, it "jumps" as the phase goes through +/-180 degrees. If you kept track of each cycle, the true phase-plot is a sorta-straight line covering many hundreds of degrees.

BTW: 5SPICE don't suck, though it breaks some old habits.

 

[SSLtech]

[quote author="PRR"]That R-C crapola connects the "+" input to source or ground at different frequencies. Combined with follower/inverter action, you get big phase shifts. If you do the math right, the amplitude does not change (much). [/quote]
Brilliant. That's exactly the analysis/explanation that I wanted.
[quote author="Ethan"]PRR - Will make seemingly complicated things as simple as possible (but not any simpler) [/quote]
Precisely! -

Thanks!

Keith

 

[luckylocky]

At first sight it seems to be a so called "Phase Rotator".
The allpass filter adds symmetry to the signal. This is used in bradcast to maximise signal levels that go through lots of gain control, compression and limiting.
The problem ist that especially male voices have lots of asymmetry (the rms level of the positive halfwaves and the rms level of the negative halfaves do NOT equal). This causes loss of headroom.

See http://www.w3am.com/8poleapf.html

Cheers Patr!ck

 

[PRR]

> used in broadcast to maximise signal levels

That's a darn interesting fact, all-pass faze-shift destroying male-voice asymmetry. But it isn't what SSLtech's box aims to do.

As I read it, it is a variant of the old quad-on-stereo matrixes like SQ.

First: take a straight (not panpot) stereo recording. You have two mikes and two channels.

Play in two speakers, you have a left-right illusion of space.

Sum the two channels, you get a mono signal that emphasizes the center of the band.

Take the difference of the two channels, you cancel the center of the band and bring out the sides and the room reverberation.

You can wire four speakers on a stereo recording and get a very interesting 4 channel (left, right, front, rear) presentation.

But "quad" is usually not Lt-Rt-Ft-Rr. We like FL-FR-LR-RR.

To over-simplify, we can shift some channels 90 degrees when we matrix from quad to stereo, and then un-do the shifts to matrix from stereo to quad. It works, pretty-much. The phantom channels tend to have only a few dB separation, not at all "discrete". But this is enough to fool the ear into hearing things in four directions. And it does not need new disk formats. And it plays in 2-channel or 1-channel speakers perfectly acceptably (no missing instruments).

To get an effective 90 degree shift, we seem to have to run hundreds of degrees of shift over the whole audio band, but starting at different frequencies in each channel. So one channel may be 600 deg at 20KHz, the other may be 690 deg at 20KHz, difference being 90 deg and nearly constant over the whole audio band.

Now that I think on it: this is a sine thing so few-degree errors do not cause large decoding errors.

Essay on various Quad formats, though without explanation.

This page has more, including this morsel (under SQ): "the level of Right Front signal in the Left Total equation is zero and vice versa. This means that front speaker crosstalk is at a minimum and the front stereo image is fully preserved. ... As we have said, Lf and Rf emerge intact. However the signals sent to the rear speakers contain substantial amounts of all the other components, giving seperation between front and back and diagonally of only 3dB.... This is a severe limitation of the SQ matrix system and means that some blurring of images is inevitable with the basic decoder circuitry."

This page lists several good essays. "Understanding the Scheiber Sphere" may be the most general. "Ambisonics Decoded" summarizes the difference between kludges like SQ and general acoustic location theory.

Encoding SQ at home is good general explanation of phased matrixes.

The Prologic (tm) Compromise is good history.

Ambisonics: The Surround Alternative is a good survey of the last 30 years (plus a new idea the author is pushing).

 

[Svart]

OW! my brain hurts! :shock:

 

[SSLtech]

Dawdling on the assymmetric point for a minute:

My assumption has been that with assymmetric waveforms, the real reason for selecting the polarity of the waveform was to keep equivalent transmitter average RF power higher, without raising the peak output...

By way of a poor example, if you have a square wave (let's pretend that one could reach the transmitter!) but with an unequal mark:space ratio. If the wave has a 3:1 mark:space, then it spends 3 times as much time at full RF output powet t does as at zero RF. The average RF output is therefore 75% of the peak permitted power, noise is lower at the receiver since the average signal is higher, so the signal-to-noise is lower... (hugely simplified) Inverting the polairyt of this example would result in a 25% average RF output: a clear case of one polarity being better than the other.

If the phase rotator -by breaking up the unified timing of the fundamental components of the signal- restores a 'balance' to the previously highly asymmetrical waveform, this 'accidental advantage' is lost... but that's not the same as 'headroom problems'. -Also, if the preceived level is higher at the receiver, the signal-to-noise might also be improved..

I don't know for sure, I'm just speculating based on assumptions that I'd held for a good long time. -If I'm off-base I'd appreciate the corrections.! :grin:

Keith

 

[bcarso]

Svart, I could direct you to a bunch of patents if you really want to get OD'ed!

 

[PRR]

> By way of a poor example

It is indeed a poor example and a fallacious argument. But I can't debate it without a chalkboard.

Shifting phase to give a less spiky wave will, generally, improve S/N at clipping.

Male voice (and some reed sounds) have asymmetry too.

In FM, this does not matter: FM is generally symmetrical. (SSB also?)

AM has a special property: you can't modulate the negative side more than 100% (by definition and because it clips) but you can modulate the positive side as much as you want. 200% positive modulation is perfectly possible (if the modulator is oversized!).

A related situation that older geeks may understand: If you cut a mono phonodisc with vertical modulation, say 0.001" deep, you can modulate down to 0.002", 0.003", 0.004" if you have the cutting strength; but you can only modulate up 0.001" to 0.000" before the cutter comes out of the wax master. So you can cut +0.001" but down 0.002+". If you want to maximize recorded level, you find asymmetric peaks and flop them so the peaks cut down, not up.

If your male-voice asymmetry peaks land on the positive side, you have 6+dB better S/N at clipping. Around the 1950s, AM broadcasters used this fact; one of the CBS boxes will detect significant asymmetry and flop the phase to put the peaks on the positive side of the modulation. Gross abuse of this plan (especially with under-boosted modulators) causes splatter on adjacent channels, so the FCC ruled that only 125% positive modulation is allowed, and reminded us of the general rules of cleanliness and non-interference.

I had never seen phase rotation used to bury peaks. Seems to me there is a finite chance it could increase peak height; but I suspect that on average it will bury more peaks than it brings up. And this is not about asymmetry, just peakedness, which is characteristic of most speech/music sounds.

 

[Svart]

Bcarso,

The brain hurting is only a sign of it growing, much like muscles... :green:

now if I couple that with lots of beer to kill the weaker brain cells then I can selectively breed better, stronger braincells!

:guinness:

In other words, send the links my way. :thumb: