Graphic EQ Polarity Question

[Bill Mountain]

In a graphic eq such as this:

http://sound.westhost.com/project28.htm

The 10k pots for the eq controls straddle the inverting and non-inverting inputs of the opamp.  What does this do for the polarity of the signal?  It seems as if the polarity will change depending on whether the frequency is being boosted or cut.  I'm working on a design where I split the signal and one side will have a graphic eq on it and I'm trying to figure out what the polarity will be so I can mix it with the other half of the signal without any phase issues.

Any help???

[PRR]

> It seems as if the polarity will change depending on whether the frequency is being boosted or cut.

No. That plan is always non-inverting.

Throw out the pots. Connect one filter unit to one or the other bus, extreme action. When connected at the + input, the filter loads-down the 2K7 series resistor, dip. When connected at the - input, it forms a divider with the 2K7 NFB resistor, boost.

There is also a phase-shift inherent in any simple analog filter, but in this case only about +/-90 deg at the steepest part of the bell-curve.

Put the pots back in. Same thing, but less, reducing to no-action in center.

It "follows", near enough.

[Bill Mountain]

Thanks for the explanation!  I'm glad you mentioned the 90 degree shift in filters.  My buddy is an engineer and he was telling me about this but that I shouldn't worry because I wouldn't be able to hear it.  But if I have my signal go through 2 filters then wouldn't I be at 180 degrees?

[PRR]

Polarity is largely inaudible unless you mix a twisted signal with the same signal not-twisted.

[Bill Mountain]

Well that's my fear.  If each half of the signal goes through separate filtering there seems to be a chance of phase issues when I combine them later.  I will have to experiment.  I just prefer to get as much figured out before I turn on the iron as possible.

Thanks for your help!

[R.G.]

Polarity is a slippery issue any time filtering and mixing get into the picture.

Paul is right (as usual) about the phase shift in filters and the audibility of phase. Each reactive element (capacitor or inductor) has the ability to cause a 6db/octave rolloff filtering action. The math describing this involves a got-to-infinity point in one way of graphing the response in complex algebra, so this is called a single-pole response. For a one-reactive-element filter, the phase shift approaches but never quite gets to (i.e. is asymptotic to) a 90 degree shift as frequency changes through the rolloff point of the filter.

Since phase varies smoothly before, during and after the rolloff frequency of the filter, "polarity" in the sense of inverted or non-inverted doesn't have much real meaning. It's not one or the other, it's a smooth slide between zero degrees and nearly 90 or 270 degrees or between 180 degrees and nearly 270 or 90. When you add a second pole (i.e. reactive element in the filtering) you can get phase changes asymptotic to 0 to 180 degrees, with no real place where it's true or inverted. Add more stages, you get more smooth phase variation.

However, when you have lots of phase shifting going on, if you add back some non-messed-with signal, you can get cancellation at the frequencies where the smoothly varying phase shifts add up wrong, and reinforcement at the frequencies where the signals add up right (that is, in phase). Now you have loudness variations, and that is audible, where the phase changes by themselves were not. This is how phasers and flangers work.

It is an issue  with speaker crossovers, where there are possibly sharp filters trying to cut the amplitude of a signal sharply at a given crossover frequency, and still have the two split signals add to a constant amplitude. It's easy to wind up with a cancellation or peak in the middle of the crossover where the speaker outputs get added back together in the air in front of the speakers. It can be very complex to get filters that provide both sharp cutoffs (and the many poles that implies, and the many degrees of phase shift THAT implies) and addition back to constant amplitude.

Another place that phase gets into it is time delays. Flangers use short time delays to deliberately cause peaks and notches in the amplitude response. When the delays get much longer than the audio period, "phase" starts being almost meaningless. You still get reinforcement/cancellation, but it's not clear what that means when the output signal is delayed by many cycles from the dry signal. The dry signal is different in general by the time the delayed signal gets back to mix with it, so things get really complicated.

But I'm rambling. I've thought a lot about what phase means on long delays, and that's what made me come up with the description. Human ears can't hear absolute phase (true/inverted) very well if at all. They can hear mix-to-cancel, but with single pole filters you generally don't get there, and maybe not even with two-pole filters because these are 180 degree asymptotic.  Three and more pole filters, yep, you can hear cancellations if they're there.

[ashcat_lt]

You'd probably start to hear the cancellation in the mix as the relative phase shift appoaches 180, but by that point the response is way down, so cancellation would be minimal.  This is where that "3:1 Rule" - more precisely named the "at least 9db rule" - comes in.  Basically, if one signal is at least 9db greater than the other (at the freqs of interest) then the phase interference will be inaudible.

Also, what else are you doing to these signals?  Phase interference depends on signal correlation.  RG touched on it.  The two signals really need to be approximately identical except for phase (or short delay) to interfere in a noticeable way.  If, for example, you're distorting the bejeezuz out of one of the signals before mixing them, the two signals won't resemble each other enough to interfere noticeably.

[Bill Mountain]

RG and Ashcat thanks for the info.  I'm working on a tube/ss bass preamp that splits the signal after the fist 1/2 of a 12AX7.  Signal 1 goes to the second 1/2 of the tube and then to a passive FMV tone stack.  Signal 2 goes to a graphic eq then the 2 sides will go to a simple mixer circuit.  This will hopefully allow me to get either a classic sound with a pleasing mid scoop and little bit of grit (by overdriving V1B) or a modern clean tone with active eq.  I hope to be able to blend these signals.  I picture getting a smooth vintage tone then blending in a bit of the modern signal to bring back the mids just a hair.  I already know that my 2 signals will be out of phase so I'll have to flip one of them but I was trying to figure out what any eq's, hpf's, or tonestacks would do to the polarity of the signals.

Thanks for your help and please let me know what you think of my idea.  I just found out that Marshall does something similar with one of their bass amps so it seems it's not as original as I had thought!

[PRR]

> I just prefer to get as much figured out before I turn on the iron as possible.

WHY??

OK, it is good to figure-out how to light the heaters, pull plate current, and "pass audio". Basic stuff.

The 12AX7 and BMT tonestack is an obvious module which you should have and can build.

The graphic EQ is an obviously useful thing to build.

Mixer is semi-trivial.

But when it comes to combining modules for "sound", pre-thinking only takes you so far, and over-thinking may lead you astray.

Talking about sound is like dancing about architecture. Thinking is little better.

Fer example: my engineering analysis says that adding a broad mid-boost to the heavy mid-scoop Fender tonestack does little more than just turning-up the Fender tonestack MID knob. So "why bother?"

OTOH, my artist/designer intuition whispers that it may not be that simple.

Reconciling my two minds, I wonder about the width of Elliot's passbands.

My pragmatic side says to build the two modules, tube-preamp and graphic EQ, plus frills like mixer and knobs and jacks, with ample space on a big board so that various variations can be tried conveniently and compared quickly. If/when a plan comes together, then transfer it into a stage-worthy box.

JS Bach could compose short ditties before putting ink in his quill, far from his instrument. Many other fine composers have to sit at their instrument and dink-dink-dink a composition by trial and re-try.

I _did_ miss your original point that you were mixing signals of different EQ.

FWIW: the Fender BMT tonestack is ALL phase-shift, since for most popular settings it slopes most of the audio band.

The phase question did have me wondering. And I have an idiot assistant to do math.

Here is a simulation of what I think your plan is.

The 12AX7 stages are wideband and low-THD, near enough for phase/response questions, so I simmed them as gain-of-inverting-50 blocks. As you say, there is an extra inversion, I used an inverter in place of Elliot's buffer to come out right. I simmed just enough of the EQ to give a midrange boost. The stock EQ caps gave an awful narrow bump, I broadened the bell-curve. R15 fixes a simulator stupidity, with 1/4db "error".



Three outputs: Mixed, toobe amp, EQ output. Top plot phase, bottom db response.



The classic Fender tonestack with typical setting (red) gives -40deg to +30deg phase over the guitar band. It IS interestingly smooth.

The Elliot EQ (blue) gives minimum-phase action: a simple analog EQ network "must" give phase-shift proportional to EQ slope, and these slopes lead to +45/-45deg an octave apart.

The Mixed output (green) really just splits the difference, moderates the extremes of both paths. As-is, it is a mellow Fender TS with a 2-3 note popout on the upper strings (and low-string harmonics in this range). I suspect it wants ear-tweaking of bump frequency, bump width, and mix ratios to find something musically useful.

But no horrible phase collision/cancellations.

This was an ideal analysis. In real life, your first tube output may be as high as 30V peak, while the chip EQ won't eat more than 10V-13V peaks cleanly. So you may have to pad-down the EQ to gain of 0.3, while the tonestack-tube path has gain near 10, so there may be more gain-trim needed to pass audio cleanly in proper proportion.

[Bill Mountain]

Holy hell that was thorough!

I was going to set the V1A up so that it didn't clip the graphic eq and I was going to set up V1B (maybe only 12dB boost) so that the passive tone stack at noon would have a similar level to the GEQ channel.  I forgot to mention that I was planning on only have bass and treble controls on the tooobe side and and only a 2 or three bands on the GEQ side.  I've even though of simple bass and treble shelving controls with simulated inductors on the GEQ side and leaving the mids flat (well I guess on the bass control only).  I never boost mids because my fat fingers produce plenty on their own.

It's early and I should be performing my domestic duties so I'll have to come reread your post later to soak everything in.

Thanks again.  This was most helpful!