Phasers: the more stages the better ?

[parser]

Hello,

I'm planning to build a phaser and I wonder if it makes sense if adding phasing stages **strongly** increases the overall quality of the device. What do you think about ?
I found mainly pedal projects with 4, 6 and 8 and - for synths - 12 stages. The second question is "why not building 24, 36 or even more stages phasers?" Maybe after 12 stages you don't feel the difference ?
Can you suggest me the best DIY project on internet ?
(I found this 8 stages phaser on MOS: http://musicfromouterspace.com/analogsynth_new/PHASESHIFTER2007/PHASESHIFTER2007.php)

Thankyou
 

[Mark Hammer]

Mike Irwin demonstrated a 24-stage phaser for me, using white noise as the signal, and it was both real AND spectacular.

HOWEVER, the value/virtue of having more stages depends on the bandwidth and complexity of the signal.  Much like a tree in the woods that makes no sound when it falls if there is no one there to hear it, sweeping notches in parts of the spectrum where there is no signal, does little that is audible.  So, in that spirit, 12 stages is great for synths or mixes, but of less use for a single guitar.

Consider flangers.  The most dramatic sounds are produced when the unit can sweep down to extremely short delays, much shorter than 1msec, or even a half msec.  The effect is for the sound to appear to get "infected" with notches as the circuit sweeps from way up high to progressively longer delays.  But there is still comb filtering potentially going on, just well above where there is guitar content, such that, as the delay increases (i.e., sweep starts to "descend"), it sounds like notches are progressively being introduced.

Another thing to consider is that the more stages one adds, the greater the risk of cumulative hiss and possibly oscillation.  We assume that each allpass stage is unity gain.  As such, the signal level at the end of N stages is the same as that entering the first stage.  That is what allows us to use feedback to emphasize the notches and peaks.  That assumption can be misplaced, even if one uses 1% resistors.  Remember that gain is multiplicative, so that 4 stages that each have a gain of 1.1x ends up with a cumulative gain of 1.46x.  If there is a tiny bit of gain in one or more stages, then feeding back some of the last stage's output back to earlier stages will result in amplification of what has already been amplified, leading to oscillation and annoying howl (1.46x multiplied by itself a few times quickly gets you into howling territory).  As well, any tiny bit of residual noise in the stages also gets amplified.

There are cures for this.  One is something you'll often see in units with 8+ stages.  That is a feedback capacitor in parallel with the feedback resistor in the last stage.  So, if one placed a 1500pf cap in parallel with the 10k feedback resistor in the last stage of a P90 with added stages, that would roll off starting around 10.5khz, decreasing the amplification of hiss if regeneration is used.

A second "cure" is to trim off the low end of the feedback signal.  Keep in mind that, for guitar, lower frequencies are generally greater amplitude than higher frequencies, such that recirculating the entire spectrum through lots of stages with a teensy bit of gain in one or more stages can quickly result in an amplitude that will severely tax the limits of any FETs used.  Those low frequencies will also be the ones that risk oscillation.

Personally, I find 6 stages to be the "sweet spot" for guitar.  That's not a hard and fast rule, just something I find sounds good and introduces the fewest problems.

All of that said, if one wants to make a "general-purpose" phase-shifter that can be applied to things other than electric guitar or bass, there is nothing wrong with adding more than 6 stages, so long as one attends to the various sources of risk noted here, in order to maintain a clean, non-oscillating, non-distorting signal.

[Elektrojänis]

This made me wonder if the notches of a phaser could be somehow made to get closer together to fit more of them to guitar frequencies?

Or are they too wide anyway and would just blend in to one big notch or something? I don't really think that would happen since the phase shift goes always through in phase between two out of phase spots... But is there something that makes it impossible to fit more notches in smaller frequency range?

[Mark Hammer]

The answer is yes.  Two-pole allpass stages result in notches that are more closely-spaced together.

Of course I can't talk about notch-depth/spacing without mentioning in passing the relationship between phasers and Uni-vibes.  Where phasers produce more defined notches, whose movement captures our attention, vibes produce wide-and-shallow dips that we tend to perceive as a sort of "animation" in sound, rather than something to pay attention to, largely because they don't strike us as being at any particular location in the spectrum.  Where feedback of phase-shifter stages produces an emphasis of notches and peaks, feedback does nothing of use for Uni-vibe circuits.  That's why they never have feedback/resonance/regeneration controls (unless they are designed to offer both phaser and vibe functions).  Finally, human perception is such that we have a hard time detecting that which exhibits little change.  As a consequence, the LFO rate on a phaser will generally always be capable of achieving MUCH slower sweep speeds than on a vibe.  Easy to notice that a well-defined notch and peak has moved over even a little bit at a slow rate of change, but MUCH harder to detect the nudge in a broad shallow dip that still covers much of the same part of the spectrum as it did a few moments ago.

[soggybag]

I can't answer the question: is more better. I will off that I built the Madbean Maestro phaser clone. This has 5 or 6 stages, there's a switch. You can hear difference between the two settings. It sounds good but I'm not sure it sounds better than a 4 stage phase shifter.

[Processaurus]

I bought some MXR phase 100 pcbs way back, some old factory leftovers, and tried a bunch of experiments. Cascading the 2x 6 stages into one 12 stage phaser sounded AMAZING on guitar once the feedback path ran like a typical phaser, from stage #12 back to stage #1. Deep and pronounced. That one is an optical circuit though, so a more transparent stage, less cumulative distortion (than FET) or noise (than OTA). The Mu-Tron Biphase is two 6 stage optical phasers that can be gained up to one 12 stage phaser, it's amazing on guitar.

More than 12, I think I'm with Mark, it seems like diminishing returns, I've tried some digital phasers with lots of stages and it wasn't mind blowing. The 4 stage is the classic sound. Working on an elaborate phaser it would be worthwhile to be able to switch down to less stages for the classic sound.

[Fancy Lime]

It also depends on the musical purpose, imo. A very slow two stage phaser like the MXR 45 or DOD 201 can sound great as an (almost) always on thing on bass or funky rhythm guitar. It just adds a little flavor to the sound without standing out as an "effect" as such. It's unimpressive on it's own but makes the band sound subtly more complex and "three dimensional". So maybe, just maybe, the answer is "more different phasers are better". Some with many, some with few stages.

Andy

[ElectricDruid]

Jürgen Haible's "Tau Pipe" phaser recreation is about the only analog phaser I know with many more stages - it has 20!!

It's an interesting design too, more similar to the Moog ladder filter than anything else - an allpass ladder, with both differential drive and differential signal outputs. Full details are here:

http://jhaible.com/legacy/tau/jh_tau.html

This is the actual phase stages and the signal path:

http://jhaible.com/legacy/tau/jh_tau_sch_page2_allpass.pdf
http://jhaible.com/legacy/tau/jh_tau_sch_page1_signal.pdf

That's not all of it. Jürgen was never scared of a lot of parts in his circuits!

Whether such a thing is worthwhile for a guitar is questionable for the reasons that Mark H gave. Maybe with a gnarly fuzz on it? I think it is a question of diminishing returns, and avoiding noise as the number of stages goes up gets more and more difficult.

[Rob Strand]

The question has be answered quite a few times on the forum already.

There's a trend of stronger phasing when you go from 2 to 6 stages.   When you get to 8 things start to change and when you get to 10 or more stages it starts to sound more like Flanger than a Phaser.

The key point is "better" is subjective.

[Mark Hammer]

I bought some MXR phase 100 pcbs way back, some old factory leftovers, and tried a bunch of experiments. Cascading the 2x 6 stages into one 12 stage phaser sounded AMAZING on guitar once the feedback path ran like a typical phaser, from stage #12 back to stage #1. Deep and pronounced. That one is an optical circuit though, so a more transparent stage, less cumulative distortion (than FET) or noise (than OTA). The Mu-Tron Biphase is two 6 stage optical phasers that can be gained up to one 12 stage phaser, it's amazing on guitar.

More than 12, I think I'm with Mark, it seems like diminishing returns, I've tried some digital phasers with lots of stages and it wasn't mind blowing. The 4 stage is the classic sound. Working on an elaborate phaser it would be worthwhile to be able to switch down to less stages for the classic sound.

Not a criticism, but a clarification.  The Phase 100 is mostly optical.  There are 6 swept stages that use optoisolators, but 4 additional fixed (unswept) allpass stages.

I have an RPH-10, which is essentially a glorified PH-2 with more controls.  Personally I don't find much of use with he 12-stage setting, when it comes to guitar.

[ElectricDruid]

Not a criticism, but a clarification.  The Phase 100 is mostly optical.  There are 6 swept stages that use optoisolators, but 4 additional fixed (unswept) allpass stages.

I have an RPH-10, which is essentially a glorified PH-2 with more controls.  Personally I don't find much of use with he 12-stage setting, when it comes to guitar.

I notice that the RPH-10 uses a NE571 compander to keep the noise down, as well as two Roland IR3109 filter chips for 8 swept allpass stages.

I tried building a phaser with the AS3320 filter chip, but found it a bit noisy - although the same was true with the chip arranged as a highpass filter too. Both allpass and highpass do nothing to cut out the hiss, so you notice it much more. As a lowpass filter it's no problem since when you close the filter, you remove all the hiss, and when you open it, it's masked by the signal. With highpass, the *opposite* is true, and with allpass the situation is somewhere in between!
Still, four swept stages on one chip is nonetheless very tempting, so perhaps I should wrap a compander around it and try again.

[Mark Hammer]

I wonder how the 3320 compares with the SSM2164. The latter is not specifically designed as a filter chip, but it IS four uncommitted OTAs.

[ElectricDruid]

Anecdotally, it seems like the 2164 is lower noise. Whether that's really true or not, I don't know. The 3320 includes buffers as well as OTAs, whereas the 2164 is just the OTA/VCAs, so you need external op-amps or something to make an allpass stage. The typical design uses *two* op-amps per allpass stage (one to make a lowpass filter, and one to turn the lowpass into an allpass). There's also a resonance VCA on the 3320, which you don't get on the 2164. Not that CV-controlled resonance is a big feature on a pedal phaser. Maybe for a Eurorack phaser module.

In the synth world, both of them benefit from higher signal levels and +/-15V headroom. Those things are going to help your noise figures a lot.

<Edit>One version of the 2164 allpass stage is shown in the (extremely good) SSI2164 datasheet, fig 14 on page 12:

https://www.soundsemiconductor.com/downloads/ssi2164datasheet.pdf

Their datasheets have gathered up all the synth wisdom of the last 30 years and put it together in one place. They're fantastic.

<Edit2>Just for completeness, here's another version by Osama Hoshuyama:

https://userdisk.webry.biglobe.ne.jp/000/024/65/1/Vcph0505.GIF

Notice this one doesn't use the extra cap and highpass-style input, and doesn't need an inverted signal. Instead, the lowpass output is mixed with the input signal to create the allpass, more like the allpass stage we're familiar with. Just for Mark H, he also shows a 2-pole version. Neil Johnson did a version of the same thing:

https://www.njohnson.co.uk/index.php?menu=2&submenu=0&subsubmenu=3
https://www.njohnson.co.uk/pdf/cesyg/vcapf-sch.pdf

[Digital Larry]

Somewhat OT, but when I implemented the phase shifter stage in SpinCAD, as usual I just used some existing code I found and made it adjustable.  I think the stock algo was a 6-stage, while I was able to expand that to 10 stages within the memory limitations of the FV-1.  And honestly I would not conclude that more stages is "better", just "different". 

Some other observations that may or may not apply whether analog or digital.
- apparently to prevent clipping in the all-pass itself, the signal level going in is heavily attenuated and then boosted up again afterwards.
- my biggest problem with the FV-1 phase shifter was trying to get a wide sweep which is more or less uniform, which boils down to trying to shape the LFO in a particular way.  I was never really completely successful.  What happens is that at one extreme of the sweep you get a very sudden "beyooww" sort of pitch bend.  With a chorus or flanger, generally using a triangle wave will give you a constant "pitch shift" either up or down assuming that there is a linear relationship between the LFO voltage and the delay memory clock frequency.  That's probably not a given either.

So this begs the next question, esp. for analog phasers/univibes/etc. 
- You have the circuit topology which dictates a phase shift/voltage characteristic.  My guess is that this differs between vactrol, VCA, OTA, FET implementations of the phase shift element.
- You have an LFO which in most analog cases is a sine wave or triangle.

So, what sorts of LFO circuitry have been implemented in analog phasers to optimize this for max sweep?

[Mark Hammer]

Anecdotally, it seems like the 2164 is lower noise. Whether that's really true or not, I don't know. The 3320 includes buffers as well as OTAs, whereas the 2164 is just the OTA/VCAs, so you need external op-amps or something to make an allpass stage. The typical design uses *two* op-amps per allpass stage (one to make a lowpass filter, and one to turn the lowpass into an allpass). There's also a resonance VCA on the 3320, which you don't get on the 2164. Not that CV-controlled resonance is a big feature on a pedal phaser. Maybe for a Eurorack phaser module.

In the synth world, both of them benefit from higher signal levels and +/-15V headroom. Those things are going to help your noise figures a lot.

<Edit>One version of the 2164 allpass stage is shown in the (extremely good) SSI2164 datasheet, fig 14 on page 12:

https://www.soundsemiconductor.com/downloads/ssi2164datasheet.pdf

Their datasheets have gathered up all the synth wisdom of the last 30 years and put it together in one place. They're fantastic.

<Edit2>Just for completeness, here's another version by Osama Hoshuyama:

https://userdisk.webry.biglobe.ne.jp/000/024/65/1/Vcph0505.GIF

Notice this one doesn't use the extra cap and highpass-style input, and doesn't need an inverted signal. Instead, the lowpass output is mixed with the input signal to create the allpass, more like the allpass stage we're familiar with. Just for Mark H, he also shows a 2-pole version. Neil Johnson did a version of the same thing:

https://www.njohnson.co.uk/index.php?menu=2&submenu=0&subsubmenu=3
https://www.njohnson.co.uk/pdf/cesyg/vcapf-sch.pdf

That IS a good datasheet!

[ElectricDruid]

So this begs the next question, esp. for analog phasers/univibes/etc. 
- You have the circuit topology which dictates a phase shift/voltage characteristic.  My guess is that this differs between vactrol, VCA, OTA, FET implementations of the phase shift element.
- You have an LFO which in most analog cases is a sine wave or triangle.

So, what sorts of LFO circuitry have been implemented in analog phasers to optimize this for max sweep?

That Jurgen Haible circuit I posted uses a full-blown exponential convertor ahead of the control input. If you tuned it and turned up the resonance, you could probably play a tune on the phaser alone, if you provided CV from an analog keyboard.

Slightly more common that that is some kind of "hyper triangular" waveform. For that you typically start with a triangle wave, round the peaks using one of the common methods (OTA distortion or a differential pair are good/cheap/popular), and then full-wave-rectify the output so you "fold the tops down" and finish up with a series of "bottom half of sine waves". Or is it fold the bottoms up to get "top halfs of sine waves"? Maybe that depends on which control method you're using. Either way, the basic idea is that you have an abrupt change at one end of the range, and a much softer smoother change at the other end. That can be used to slow down the abrupt BEEYOOW into something mellow and simultaneous speed up the bit at the other end where a sine wave seems to hang around doing nothing for a little while.

In my world, digital LFOs make this sort of stuff easy, so you can tweak the LFO waveform to suit the thing you're playing with. I put both the "bottoms of sines" and "top of sines" waveforms on my TAPLFO (Sweep and Lumps respectively). The StompLFO only has the Sweep wave, so if you need it the other way up, you have to flip it yourself.

[Rob Strand]

And honestly I would not conclude that more stages is "better", just "different". 

Same here.

2, 4, 6, stages have the "traditional" phaser sound.   The 2 stage is weaker sounding than the 6 stage.   Once you hit 10 stages and over the character of the sound changes.

Fixed stages move the notch positions, so does changing the caps (stage f0's for DSP) and don't contribute to the pitch shift.

[Mark Hammer]

So, what sorts of LFO circuitry have been implemented in analog phasers to optimize this for max sweep?

The earliest-issue Small Stones (and possibly some later issues) used a clever trick to provide LFO waveforms that varied with the purpose/nature of the sweep.

The "Color" switch simultaneously alters the resonance/feedback in the circuit, the speed range, and the LFO waveform.  When the junction of the 100k/180k pair is tied to V+, the speed range is faster, and the resulting waveform is triangular.  When V+ is tied to the junction of the 270k/100k pair, the waveform becomes more "hypertriangular" and the speed range is slower overall.

A clever trick is the use of the 10k/10uf pair at the LFO output.  The 10k needs to be there to keep a lid on current going to the OTAs.  In conjunction with the 10uf cap, however, it introduces a single-pole rolloff starting around 1.6hz.  The effect is to "soften" faster sweeps in both Color modes, doing so more the faster the sweep.  Whoever designed the Small Stone was smart smart smart.

[amptramp]

Just out of curiosity, I have seen phaser stages where all parts in each stage are identical (most of them) and some where there is a difference in capacitance (a few of them).  Does this make much difference and is it something to be pursued or something to be avoided?  I imagine they sound different but it is not clear which way would be "better".

[Fender3D]

Same here.

2, 4, 6, stages have the "traditional" phaser sound.   The 2 stage is weaker sounding than the 6 stage.   Once you hit 10 stages and over the character of the sound changes.

Fixed stages move the notch positions, so does changing the caps (stage f0's for DSP) and don't contribute to the pitch shift.

same here too...
Furthermore, 2 and 4 stages sound good even BEFORE distortion or overdriven amp, while 6 stages or more work better AFTER distortion