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.
abbey road d enfer said:
I still fail to understand what this would do that a simple APF couldn't.

D, the schematic I came to in post 14 is my current understanding of how to do a swept two-pole without inversion (other than the one that you choose manually) with the least number of amps, with a cap switching option and lowest possible noise, based on the easiest (for me) method of sweeping two channels with one pot.

The discussion of how to do something else, like an analog group delay correction, is a different scope.
 
gyraf said:
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.

Yes. I believe the post 14 schematic does that, at least with as much elegance as I can come up with, but not in any way that hasn’t already been done by the IBP or Phazer.

Your experience with the IBP is not unlike my experience with the Phazer. Which is why I was originally asking (admittedly without great clarity) whether there was a better way, such as creating a complete unity gain 90 degree copy of the spectrum to fade against the original signal.

On the IRP topic, I’ll find that patent again. It’s all regular  first order LPF-style inverting APF filters in a row. The 0 degree ones all pick off via resistors to a bus that feeds an inverting amp. The 180 degree ones do the same, to a second inverting amp. Both then combine to a diff amp.
 
Quote from: atavacron on October 20, 2020, 10:04:01 PM

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

abbey road d enfer said:
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...

Abbey, the schemo below his quote is not the IRP EQ. 

We have one of the IRP EQs here and despite how cool *I* think it is, nobody uses it.
 
atavacron said:
D, the schematic I came to in post 14 is my current understanding of how to do a swept two-pole without inversion (other than the one that you choose manually) with the least number of amps, with a cap switching option and lowest possible noise, based on the easiest (for me) method of sweeping two channels with one pot.
It's a basic 2nd-order APF. Cascading two 1st-order APF would give the same performance.
The only finesses in this schematic is that the corner frequency is DC-controlled, but that could be done also with two cascaded stages. DC-control using LDR couplers is not new, and the pitfalls are known, particularly matching the two elements.
There is no evidence that a 2nd-order APF produces better results than a 1st-order for source alignment.
In addition, I don't understand the use of a stereo version.
 
abbey road d enfer said:
There is no evidence that a 2nd-order APF produces better results than a 1st-order for source alignment.
In addition, I don't understand the use of a stereo version.

Don't know if it produces better results or not but a 2nd order APF has a wider group delay bandwidth than two 1st order APF daisy chained in which the combined group delay is the same as that of the 2nd order APF.
 
57sputnik said:
Take a look at the Studer 90 degree filter:

https://groupdiy.com/index.php?topic=62866.msg796469#msg796469

Thats exactly the circuit I have been discussing, it uses 8 1st order APF, in my opinion that is not enough if you want something like +/-2° phase ripple from 20Hz - 20KHz, which is probably why the Studer filter is spec'd at +/- 3° phase ripple from 30 to 16KHz
 
user 37518 said:
Thats exactly the circuit I have been discussing,

I have a dumb question (for anyone really). What happens if the sections are summed in parallel? Asking because noise floor. Studer schematic attached.
 

Attachments

  • 5CB02F48-1B93-406A-A1AC-F7C1E6E9D2D2.jpeg
    5CB02F48-1B93-406A-A1AC-F7C1E6E9D2D2.jpeg
    124.6 KB · Views: 11
Dumb question number 2...just stop me wherever it’s wrong.

2a) Does the bandpass-based MFB allpass have its fl and fh 90 degrees from 0, or 45 from 0 like a regular bandpass?

2b) If you vary the resistor to ground, are you varying both the fl and fh between 45 and 90, or between 90 and 180?

2c) If the filter were comprised of two like sections in series, would the resistors to ground vary the fl and fh 180 degrees? You can see where I’m going here.

2d) Does that defeat the phase adjustment purpose entirely by turning the sections back into regular ol’ bandpasses, or does the fact that the second section is inverted mean that the dry signal stays at unity, thus maintaining the allpass unity gain?

This whole question is based on a Jim Williams comment on a GS thread wherein he reverse engineers the IBP in one sentence, saying essentially to put two MFB allpasses in series and use a dual 100K log pot in place of the two resistors to ground. It took a really long time for me to figure out WTF he was talking about.

That doesn’t improve on the IBP or Phazer, I’m just trying to understand if it’s valid. And maybe it would open the door to putting a bunch of fixed MFB allpass sections in parallel instead of in series (which I am gathering doesn’t work with LPF-derived first order allpasses).

Like...a 16:4:1 sum. With JFETs to ground. Which would be a better starting point than the IRP/Studer/etc series approach for a circuit that could rotate the phase of many bands at once. If that were a more desirable way to phase correct mics — and maybe other things — than sweeping the fc of a LPF-derived second order allpass....y’know?
 

Attachments

  • 49592224-4184-4F6B-ABBA-2A42E530A74B.png
    49592224-4184-4F6B-ABBA-2A42E530A74B.png
    301.8 KB · Views: 6
user 37518 said:
Don't know if it produces better results or not but a 2nd order APF has a wider group delay bandwidth than two 1st order APF daisy chained in which the combined group delay is the same as that of the 2nd order APF.
Is that right? In the particular case of Q=1, the response of a 2nd-order APF is exactly identical to that of two cascaded 1st-order APF's.
I'm not sure departing from Q=1 results in better performance.
I'll keep simulating.
 
atavacron said:
Which would be a better starting point than the IRP/Studer/etc series approach for a circuit that could rotate the phase of many bands at once. If that were a more desirable way to phase correct mics — and maybe other things — than sweeping the fc of a LPF-derived second order allpass....y’know?
That's the crux of all this thread. Your intent, as far as I can understand it, is to compensate micing distance, which is a pure delay matter. Trying to compensate it with a constant phase-shift is just inadequate. Constant phase-shift means decreasing delay vs. frequency. Do you think the mic gets closer to the source when frequency increases? I don't think so.
That's why source alignment is easily done in DAW's with a simple delay.
Delay implies phase-shift increasing with frequency.
The typical IBP type gizmos apply variable delay, but they make it slightly less variable by using several cascaded APF's. They manage to get increasing phase-shift over a specific BW by scattering the characteristic frequency of those APF's.
 
atavacron said:
I have a dumb question (for anyone really). What happens if the sections are summed in parallel? Asking because noise floor. Studer schematic attached.
When you sum the outputs, you get 45° shift, with about 2dB amplitude ripple.
Now, you have to understand that this circuit produces two images of the input signal that are not constantly phase-shifted, but they are within 90° of each other.
This is used by Studer to produce sine-cosine outputs, that are entered in multipliers for applying the well-known equation cos(Awt)²+sin(Awt)²=A, in order to extract the rms amplitude of the signal. A basic flaw in this scheme is that the value of amplitude A appears with a frequency-variable delay.
 
abbey road d enfer said:
That's the crux of all this thread. Your intent, as far as I can understand it, is to compensate micing distance, which is a pure delay matter. Trying to compensate it with a constant phase-shift is just inadequate. Constant phase-shift means decreasing delay vs. frequency. Do you think the mic gets closer to the source when frequency increases? I don't think so.
That's why source alignment is easily done in DAW's with a simple delay.
Delay implies phase-shift increasing with frequency.
The typical IBP type gizmos apply variable delay, but they make it slightly less variable by using several cascaded APF's. They manage to get increasing phase-shift over a specific BW by scattering the characteristic frequency of those APF's.

I agree, what the OP seems to want to achieve is constant group delay which is achieved with linear phase shift VS frequency as opposed to constant phase VS frequency. A series of APF will also achieve a constant group delay VS frequency over a specified frequency range, however, the bandwidth of such delay will be in function of the delay time and number of APF sections, the wider the delay required and the more delay required the more APF sections needed. In my experience it is extremely difficult to obtain considerable delays over the 20Hz to 20KHz range using APF, which is why DSP is the desired tool for the task.
 
abbey road d enfer said:
Is that right? In the particular case of Q=1, the response of a 2nd-order APF is exactly identical to that of two cascaded 1st-order APF's.
I'm not sure departing from Q=1 results in better performance.
I'll keep simulating.

The following graph I just made in MATLAB shows a 10usec group delay obtained with one 2nd order APF (in blue) and in orange it displays the group delay of two 5usec 1st order APF in cascade, as you can see, there is a bandwidth difference. The 2nd order APF transfer function is that of a MFB Bandpass + Summing amp, the 1st order APF is the typical APF. At 20KHz that difference is around 10% and considering that both use 2 opamps it costs nothing to use the better 2nd order section.
 

Attachments

  • APF comparison.jpg
    APF comparison.jpg
    43 KB · Views: 8
user 37518 said:
considering that both use 2 opamps it costs nothing to use the better 2nd order section.

That’s encouraging. I’m sorry to ask another uninformed question, but what is the difference between VE summing the bandpass in with the original signal at a second amp (which in my understanding you just modeled), vs bringing the original signal into the non-inverting input of the bandpass amp (my attached schematic in reply #29)?
 
To clarify:

D has demonstrated at this point that correcting mic placement with many fixed frequency series APF is a non-starter, and user 37518 has demonstrated that a significant delay is not worth achieving in the analog realm.

So the thread comes down to two related goals, which are not the same circuit:

1) the best way to do what the typical phase widgets are doing

2) the best way to compensate for the admittedly-minor-but-occasionally-problematic group delay and/or phase shift of a stereo parallel compressor (which I still don’t know how to quantify, and I would think would vary with frequency)
 
atavacron said:
So the thread comes down to two related goals, which are not the same circuit:

1) the best way to do what the typical phase widgets are doing
What these things do is provide a suitable phase shift that is not equal to a delay. They just rotate the phase so the signal appears to be in phase, but in fact with several cycles of difference. If you're familiar with modulo-pi, you'll get what I mean. Because of that, this phase-compensation will be valid fover a limited frequency range. Increasing the order may improve the result, but I'm not so sure, since a phase-shift that put two signals in coincidence at one frequency may not do the same at other frequencies, aprticularly if they are not perfect harmonics..
These units work on the principle that choosing teh right shift will increase the bass response, which seems to be what everyone wants to achieve.

2) the best way to compensate for the group delay and/or phase shift of a stereo parallel compressor (which I still don’t know how to quantify, and I would think would vary with frequency)
This would be a concern if the phase-shift was more than 90°. for that, the compressor would have either a very poor frequency response, or be non-minimum phase. A digital compressor would be NMP. Some broadcast copressors are deliberatel NMP, those that include a form of delay in the signal path for achieving "zero attack time".
Most compressors used in music production are MP and have an excellent prequecy response, so teh output is in the same quadrant as the input, which gurantees optimum summing.
 
Alright, well, I integrated a two pole bandpass-based MFB allpass into this 500 series mixer design - made it minimal enough to not require a separate unit -  and am figuring out component values. One amp for filtering, one diff amp for summing. Same as doing the summing within the filtering amp, as far as I can tell.

Does indeed require a dual rev log pot, turns out 10k is a good call and you can stick with a BJT (LM4562 in this case). 100K would be fine with no capacitor switching, but then you’re into a FET (OPA2132 is the best bet there in a DIP).

As user 37518 modeled, somewhere else on the web it is noted that while the phase response for this particular allpass is two pole (180), the skirts are still the same as associated with single pole, i.e. -6dB/octave or -20dB/decade.

In my application I’ve added a fader after the bandpass stage, and am summing the original signal and the wiper output with a differential stage, which requires a buffer between the fader and the inverting input of the diff amp (if you’re keeping resistor values low). I believe this to be lower noise than using a VE sum, though it requires the same number of stages to get back to non-inverted. Bandpass > fader > buffer > diff amp vs bandpass > fader > VE sum > inverting amp. So you could call this a three-amp variable depth allpass.

The other option in the MFB arena is to just use the filter amp for summing, but then you’re stuck with unity gain for both the filter and the original signal (generally speaking). You could set up a crossfader between that result and the original signal, but you would still need a VE sum and inverting amp after, so that’s actually the highest noise option.

If anyone is interested in the future, I’m happy to send this portion of the schematic.

Thanks all!
 
atavacron said:
In my application I’m summing the original signal and the inverted bandpass response
After so many posts, I (think I) understand what you want to achieve. Basically you want to apply some phase correction to one of the signals coming into a summer, one of these signals is an insert send, the other is an insert return.
Trying to insert it in the return signal is a no-go. If the inserted unit is lagging, there is no circuit that can make it lead. Delay is possible, anticipation is...anticipation.
Of course, you understand the phase-correction circuit must be inserted in the direct signal path and track as closely as possible the phase response of the inserted unit?

This is because you assume the return signal does some funny thing to the phase. Have you assessed exactly what it does?

As I wrote earlier, an analog compressor that has no look-ahead delay is minimum-phase, so summing a MP signal with a NMP one is bound to produce notches. The APF is a NMP circuit.

Now if the compressor has a look-ahead circuit, that is what you must duplicate in your direct signal path.
 
Back
Top