Phase rotator subcircuit

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

atavacron

Well-known member
Joined
Jan 28, 2009
Messages
344
I'm trying to figure out the simplest way to set up a real phase rotator, for integration into a larger mixer that runs on ±15V. Let's assume that 0° to 180° (or -180° to 0°) is all that is necessary, and that a conventional flip can happen outside of the subcircuit. The source is a direct coupled op amp, the output needs to be the same phase.

I haven't simulated this, but I am told that one could make a wide-ish bandwidth 90° version of the signal by cascading amp sections, to crossfade or mix in with the original signal. I think that means at least twelve sections, maybe up to sixteen if you wanted it within .5°. That doesn't seem realistic, for noise floor reasons, even if you used a SOIC quad such as the OPA1604. I've been working with DIP LM4562s, which would translate to both an unrealistic amount of PCB real estate and power supply draw.

I understand that a second-order allpass filter with a dual reverse log control would work to focus the phase crossover at a particular point in the spectrum, but that doesn't quite hit the mark. Radial's Phazer seems to be a more flexible version of this, where they're blending an inverted LPF in with the original signal. It would be tough to cover 20Hz-20kHz otherwise, even with a 100K pot controlling the RC section of an APF.

Obviously the Little Labs IBP is the elephant in the room. I am not seeking to compete, or make a DIY clone, but it is a reference point.

Is there a slick way to simply turn a sine into a cosine, with low noise and something like 25mA drive?
 
I may not understand your question properly but there are some simple one pot and one op amp circuits that deliver continuously variable between same polarity and opposite polarity.

JR
 
If you want it to continuously and smoothly sweep, the usual JFET circuit like in the MXR phase 90 pedal is actually not a bad way to do it. It uses 4 amps but I think it could be made quiet. If you run on +-15V (and keep the signal level high-ish so that you're using it), lower the impedance from say 10K to 2.2K or so and use a good sturdy amp that can drive the lower Z like RC4580 or TLV4172, I don't see why noise performance would not be quite good.
 
Thanks for your interest, John. The only one- and two-amp solutions I’ve seen involve varying the corner of an allpass, which (as you and others have noted on a handful of threads here and on GS over the past decade) doesn’t vary phase across the audio band, but rather sweeps the effect above or below the corner.

I am looking to vary the entire 0 degree audio band against a 90 degree copy of itself. Please correct me if this is the wrong approach.
 
squarewave said:
the usual JFET circuit like in the MXR phase 90 pedal is actually not a bad way to do it

That’s one backup plan, yes, sans LFO. I’m a little unclear on what modern part to use for the JFETs, and how to create the totally nutso taper required to access the entire audio band [EDIT: ...without switching cap value]. Range could be more than 100K that way, which is good. With a dual pot it’s either that or 1M, which I wouldn’t want as a noise source.

The source in my application is an LM4562 that wants to see at least a 600R load (681R shows up throughout the mixer), so I would definitely be taking filter resistances as low as they can go.
 
atavacron said:
That’s one backup plan, yes, sans LFO. I’m a little unclear on what modern part to use for the JFETs, and how to create the totally nutso taper required to access the entire audio band.
Sounds like maybe you want a logarithmic converter. They actually make ICs for precision and temperature compensated log conversion like LOG104. They're strangely expensive but you could make one out of some op amps and transistors for cheap that would work well enough. You're not performing scientific measurements and you might need anti-log anyway in which case you might have to roll your own.
 
squarewave said:
They actually make ICs for precision and temperature compensated log conversion like LOG104.

That’s great to know. Is there a preferred version of the Phase 90 circuit (or similar) that I should be looking at?
 
I'm not familiar with versions of anything like that. But I think they're pretty much all the same. They're just conventional all-pass circuits. Performance will depend more on things like the driver circuit (like using the mixing in half the signal trick to linearize the JFET) and matching transistors.

There are a number of permutations that I can think are mostly theoretical but they are:

1) Use vactrols instead of JFETs. Vactrols already have a log behavior. You would just put all of the LEDs in series and then use the usual op amp / transistor / resistor current sink circuit to make a good voltage to current converter. Then all vactrols get the same current. As you apply a small amount of current, resistance drops precipitously. At higher currents resistance is low but doesn't change as much. So with a capacitor series element and vactrol shunt element, the log behavior should be right I think.

2) Use a VCF IC like V3320 or SSI2144 or whatever. Noise performance probably isn't as good as some of the other permutations but it's good enough for most things and frequency control is probably fairly precise which means the sweeping sound might be better.

3) Use THAT 2180A VCAs to make all-pass filter sections. I can't think of how this is done off-the-top-of-my-head but I'm pretty sure it can be done. Noise performance of these VCAs is actually pretty good but after 4 stages I'm not sure it can compete with a simple JFET or vactrol circuit.

Anyway just brainstorming ...
 
squarewave said:
Anyway just brainstorming ...

That’s exactly what is needed! I’ll take a look at all these options, cheers.

Ironically the first thing I drew when I started working on this project was a 2180A-based crossfader. It’s moved light years away since, but at least I’m familiar. The A version is not too shabby actually, there are things you do to accommodate it that basically allow it to not be much worse than a typical 5532 section in noise performance. And of course it’s already log taper.
 
squarewave said:
1) Use vactrols instead of JFETs. Vactrols already have a log behavior. You would just put all of the LEDs in series and then use the usual op amp / transistor / resistor current sink circuit to make a good voltage to current converter. Then all vactrols get the same current. As you apply a small amount of current, resistance drops precipitously. At higher currents resistance is low but doesn't change as much. So with a capacitor series element and vactrol shunt element, the log behavior should be right I think.

Before I start messing with fader math, did I get the topology right here for two first order filters and a dual vactrol? I was trying to figure out how to do it with two MFB second-order filters (or even one), but couldn't figure out what one value could be varied in that case. Frankly I'm a little confused about whether the notch is advantageous....not really trying to create a phasey effect, just going for correction here.

[EDIT: Don't bother.]
 

Attachments

  • Screen Shot 2020-10-19 at 7.42.42 PM.png
    Screen Shot 2020-10-19 at 7.42.42 PM.png
    59.6 KB · Views: 34
atavacron said:
did I get the topology right here for two first order filters and a dual vactrol?

As far as I can tell, the only two-pole that can be varied with a dual vactrol, and then turned into an all-pass, is the Sallen & Key low pass. What do people think of this? Attached. Any reason to invert the dry instead of the wet at the diff amp?
 

Attachments

  • Screen Shot 2020-10-20 at 3.07.36 AM.png
    Screen Shot 2020-10-20 at 3.07.36 AM.png
    124.4 KB · Views: 24
atavacron said:
Frankly I'm a little confused about whether the notch is advantageous....not really trying to create a phasey effect, just going for correction here.
What are correcting for exactly? Are you trying to tweak phase cancellation in speakers or something? So you're applying this two one channel and not the other?

Note that if you swap R and C on the non-inverting input of the standard all-pass circuit you switch between leading phase shift and lagging phase shift. So you could make an all-pass for each channel, one with leading and one with lagging and with one dual pot you might get a wider differential effect.

Another thing you can do is combine the leading / lagging bits into one circuit. Meaning do R and C in series into R and C parallel shunt elements to get 360 degrees from one amp. See TAOE by Horowitz and Hill page 416.

If I were doing this, I would probably spend a week messing around in LTSpice before even thinking about parts, layouts and PCBs.

PS: There's also another circuit where you put a pot across the inputs of an all-pass type of circuit which is the one I think maybe JR was describing but I can't seem to google it at the moment. Depending on what you're trying to do exactly that might be of interest.
 
atavacron said:
I'm trying to figure out the simplest way to set up a real phase rotator, for integration into a larger mixer that runs on ±15V. Let's assume that 0° to 180° (or -180° to 0°) is all that is necessary, and that a conventional flip can happen outside of the subcircuit. The source is a direct coupled op amp, the output needs to be the same phase.
If you explained what you want to achieve, we could probably orient you in the right direction.

I haven't simulated this, but I am told that one could make a wide-ish bandwidth 90° version of the signal by cascading amp sections, to crossfade or mix in with the original signal.
When cross-fading, that's a phaser effect, which creates notches. By varying the corner frequency of the APF's, you end up with the well known swishing effect. I don't think that's what you want.

I think that means at least twelve sections, maybe up to sixteen if you wanted it within .5°.
It is indeed possible to achieve 90° phase shift over the audio spectrum using a rather large number of APF's. Why do you want this phase-shift to be accurate? If you want to use it for source alignment, you rather need a constant delay, i.e. a continuously varying phase-shift.
BTW, delays can be simulated with APF's.


I understand that a second-order allpass filter with a dual reverse log control would work to focus the phase crossover at a particular point in the spectrum, but that doesn't quite hit the mark.
That's also what the IPB does.

Radial's Phazer seems to be a more flexible version of this, where they're blending an inverted LPF in with the original signal.
Actually, that's what ALL APF's do, so Radial's claim is not so extraordinary.

It would be tough to cover 20Hz-20kHz otherwise, even with a 100K pot controlling the RC section of an APF.
Would it? Radial do it with a range selector.

Obviously the Little Labs IBP is the elephant in the room.
Really? It's the same APF everybody uses. Much of the difference is in the mktg blurb.

Is there a slick way to simply turn a sine into a cosine, with low noise and something like 25mA drive?
You need something that introduces continuously varying delay, so the choice is either cascading APF's in analog or cascading APF's in digital.
 
Like Abbey said, it can be done with APF, you can get a 90° phase shift across the audio band with some ripple (something like +/-2° depending on how many sections you use). I made my own code in which I used numerical optimization to calculate the time constants of each APF to achieve the 90° phase shift across the band, last time I checked I had to use something like 12 APF in total

The only problem I see with this method is that the group delay at low frequencies is significant with the amount of sections I used, something like 15ms at 20Hz, so if you have several mics, for instance in a drum kit, that amount of delay can make a difference to the sound, it does drop to a much lower value at around 100Hz, how will this affect the sound? I haven't done enough testing... If you use less sections the group delay will be lower but you'll have more phase ripple, pick your poison...

Do you need constant phase shift or constant group delay across the audio band? You haven't told use what are you trying to do. If you want to correct time allignment between signals you may want to go digital, its so much easier, programming a delay in a DSP is childs play....
 
That's cool that people want to know what the thing does! I hadn't assumed anyone would.

It's a two-channel mixer in the 500 series format, for standard racks, 51x racks, or especially racks with aux I/O. The phase adjust subcircuit would be on an insert, available on the frontpanel. You could patch any digital room correction in instead of the (definitely analog) subcircuit if desired. The subcircuit would exist solely to phase rotate one channel against the other channel, and of course there's a bypass and a 180° switch. FYI, the subcircuit had to become its own unit, and it's stereo -- you can see the "Channel 2" tie coming from the voltage control pot on the current version, attached. I know I keep changing the schematic and it makes it hard to comment, sorry! At least there's less to constructively criticize now?  :D

I know that a huge APF array is unrealistic due to noise, and that a two-pole swept APF with capacitor switching will work for most any tracking scenario. Top and bottom snare, inside/outside kick, two guitar cab mics, bass DI and cab, etc. But would it work on program material -- i.e. parallel compression in a mastering scenario? Would that be better handled by digitally correcting the original signal to the group delay of the smasher that's being blended in, such as with a nice PA time alignment rack unit? I kind of feel like if the reason your parallel compressor plus eq chain sounds smeary is because of its phase, that correcting the entire phase might still be the way to go, and that a discrete amount of time doesn’t really get at the problem.

I'll have to respond to some of the more specific questions/suggestions a little later on. There are good ones and the detail is appreciated.

By the way, the patent for Industrial Research Products' Transversal Equalizer is totally amazeballs.
 

Attachments

  • Screen Shot 2020-10-20 at 7.07.28 PM.png
    Screen Shot 2020-10-20 at 7.07.28 PM.png
    195.1 KB · Views: 30
user 37518 said:
last time I checked I had to use something like 12 APF in total

I really need to get LTspice working on my ancient laptop. What happens with 16 APF sections? Were yours made from two pole LPFs summed with the original signal, or something else? Would spec’ing the filter between 40Hz and 20kHz be a better move? You gotta roll off around 40 for vinyl anyways.

Four quad SOICs on a custom SIP plugin PCB is about the most real estate I can imagine. Maybe that would be a handy little thing for people to use in other designs. One wouldn’t want a bunch of caps in the signal path, so I suppose that puts HP- and BP-based APF sections out of consideration.
 
atavacron said:
I really need to get LTspice working on my ancient laptop. What happens with 16 APF sections? Were yours made from two pole LPFs summed with the original signal, or something else? Would spec’ing the filter between 40Hz and 20kHz be a better move? You gotta roll off around 40 for vinyl anyways.

Four quad SOICs on a custom SIP plugin PCB is about the most real estate I can imagine. Maybe that would be a handy little thing for people to use in other designs. One wouldn’t want a bunch of caps in the signal path, so I suppose that puts HP- and BP-based APF sections out of consideration.

With 16 APF sections you would get less phase ripple in the passband but more group delay at low frequencies, I used standard 1st order All Pass Filters https://en.wikipedia.org/wiki/All-pass_filter#/media/File:Schem_All-Pass_Filter_Producing_Lag.png.
 
atavacron said:
The subcircuit would exist solely to phase rotate one channel against the other channel, and of course there's a bypass and a 180° switch.
I still fail to understand what this would do that a simple APF couldn't.

Would that be better handled by digitally correcting the original signal to the group delay of the smasher
What group delay? Do you mean correcting latency of a digital thing?

I kind of feel like if the reason your parallel compressor plus eq chain sounds smeary
Does it?

is because of its phase,
This would need to be substantiated.

that correcting the entire phase might still be the way to go,
"entire phase" Huh??

By the way, the patent for Industrial Research Products' Transversal Equalizer is totally amazeballs.
I don't see anything in this schemo that shouts "breakthrough". Combining two 1st-order APF"s? Wow!
Maybe there are some extraordinary claims in the patent text...
 
Isn't the whole point of this thread to come up with something resembling a miniature (adjustable) delay, so that e.g. drumkit mics can be set straight after-the-fact?

for the record, I never got the ibp working like I feel it should. Nothing anywhere near moving mics, be it drums or gtr cabinets.

On a side note - yes, the Transversal seems very interesting, but no one has come up with a schematic so far. Pic related - i admit that it's aesthetics sparks some interest in me..

/Jakob E.
 

Attachments

  • Transversal Equalizer.jpg
    Transversal Equalizer.jpg
    204.6 KB · Views: 30
Back
Top