Analog style all pass with selectable frequency.

[KingofBeers]

This is great.

https://www.meldaproduction.com/MFreeformPhase

cheers.

[Girolad]

dunno if its a "analog" all pass but here

https://kilohearts.com/products/disperser

Already addressed in OP, friend.

To clarify, I've found the right plugin. Equilibrium seems to do the trick. I've yet to find the all-pass filter in Aries Verb. However I would like to understand those graphs and that real world electrical circuit a bit better. Should I start another thread?

I don't think the -180 to 0 all pass is possible in the analog domain, only the digital domain, right?

Sure it is.. you just invert the polarity to get another 180 degrees phase-shift (in which ever direction makes your plots align, since multiples of 360 are arbitrary as far as phase goes).

mystran, hmm, maybe I wasn't clear enough. My question is more about causal phase, with an analog first order filter, Its not possible to make the output precede the input. Only in the digital domain is that possible; that is, to make phase start at -180 (provided 0 represents a causal signal event). Disregard phase relativity between two signals and focus on one signal via one (emulated) electrical RC filter network.


With regards to "just invert the polarity" shouldn't the phase plot be mirrored about the x axis if this is done? as opposed to a vertical shift up/down by a constant 180 (as seen aeries verb). Whats the difference? Clearly phase isn't my specialty, I just would like to understand all-pass filters better.

Untitled.png

(attachment: ariesverb)

[Ichad.c]

I don't think the -180 to 0 all pass is possible in the analog domain, only the digital domain, right?

Sure it is.. you just invert the polarity to get another 180 degrees phase-shift (in which ever direction makes your plots align, since multiples of 360 are arbitrary as far as phase goes).

mystran, hmm, maybe I wasn't clear enough. My question is more about causal phase, with an analog first order filter, Its not possible to make the output precede the input. Only in the digital domain is that possible; that is, to make phase start at -180 (provided 0 represents a causal signal event). Disregard phase relativity between two signals and focus on one signal via one (emulated) electrical RC filter network.

Mystran is 100% correct and that's exactly how analog phase works as well, barring any non-linearity(distortion) and frequency warping of course. The thing that might be confusing you is the analog stuff! A lot of analog gain/filter stages have 2 versions: Like an inverting and non-inverting gain stage, the only difference is the output that is inverted in the one version - they sound exactly the same!

The reason to have 2 versions in analog design is for versatility/cost, say you need negative feedback somewhere in an analog product -> you would use the inverting version! If you used the non-inverting version, you would need another stage to invert the signal for negative feedback, which adds to increased parts and labour cost, and potentially add more noise.

Back to allpasses: In analog you get two versions: Leading(non-inverting) and Lagging(inverting) allpass.

Lagging Allpass = Inverted Leading Allpass.

P.S. That LC circuit you linked looks a lot like the classic Blumlein shuffler (my understanding of LC filters are sketchy at best). You would find the same effect in nearly most portable MP3 players since the early 2000s. Also have it in Media Player Classic in headphones mode, it's pretty common - it's as old as stereo

some fun reading: http://www.audiosignal.co.uk/Resources/ ... ing_A4.pdf

[Girolad]

Hi,

Thanks for your helpful reply.
Regarding gain stages (op-amps?) there can be inv and non-inv on one part to minimize cost, sure. With regards to the acticle you posted, I'm familiar with Gerzons work - I'm a great fan. Ironically I own the Realistic mixer he describes.

FYI, The LC circuit illustrated in my previous post isn't a headphones crossfeed (Meier, Bauer) at all, but actually a re-worked version of the classic shuffler - It was a short-lived solution / correction circuit devised in the 50s, intended for installation in all consoles capable of mono panning, which were relatively new at the time. Short-lived because its effect was deemed imperceptible (though mathematically perfect, not worthwhile, surprise, surprise). Michael describes it in his localization metatheory [1], it being specifically for 'fixed head' (sweet spot, 60/60) listening and/or mastering (the proper term is 'trans-aural' listening, where both speakers interact with both ears). There, he associates this shuffler under 'CDV Localization' theory, connoted after Clark, Dutton & Vanderlyn (the original inventors). It's really a glorified head-shadow EQ, the phase shift that the all-passes supposedly negate are approx 20deg and hardly worth fretting over, (unless your a square:P). The idea behind it is to increase crosstalk in the mid/treble range (because that subtends a greater angle than bass if the two types are panned by the same amount, this due to headshadow). This error can be also and more easily rectified with monitor speakers mounted horizontally with the tweeters on inside. Its useless for headphones and in Live/PA situations.


I should add that although the process in that article (with the realistic mixer) is termed 'shuffling', michael made a point in a subsequent paper [2], that this process shouldn't be confused with the CDV nor classic blumlein shuffling; as it involves a feedback element and therefor fundamentally different.

[1] General metatheory of Auditory Localization, 92' (most of the maths is beyond my comprehension, Michael being an Elite Oxford graduate and all. Interesting read nevertheless).
[2] Gerzon Applications of Blumlein Shuffling to Stereo, 94' (Essential read).

As for your explanation of the leading all pass, it now makes sense to me, however am I right in thinking that by 'inverted' you don't necessarily mean 'polarity reversal' ( i.e. ø )? It is my understanding that the 360 range on the left defines 'phase/time shift' and has nothing to do with polarity reversal.

[Ichad.c]

Hi,

Thanks for your helpful reply.
Regarding gain stages (op-amps?) there can be inv and non-inv on one part ....

Not in one part, it how you arrange things and not just op-amps -> tubes, OTAs, VCAs, FETs, transistors - basically everything. If done quite a bit a research and simulation of phaser pedals which almost exclusively use inverting allpass stages.

As for your explanation of the leading all pass, it now makes sense to me, however am I right in thinking that by 'inverted' you don't necessarily mean 'polarity reversal' ( i.e. ø )? It is my understanding that the 360 range on the left defines 'phase/time shift' and has nothing to do with polarity reversal.

Ah, I think I understand your problem now, you're assuming 'phase' is always connected to 'time-shift' - that is incorrect. Phase is an observation, basically like the number 1, a number is just a number, phase is just phase, there is no assumption on how it was created and there is other ways to create it that is unconnected to time. Time-Shift is a function, it performs an action, it can induce a change in phase in response to frequency or it can manipulate it-> heard of linear-phase filters? That is why in math you have different symbols for time and phase because they a two separate things.

Another example of phase-shift (non-time version) you'll find in a lot of modern synths, there are controls to change the phase of the oscillators. It is done by simple addition, not by manipulating time. Ironically an oscillator in DSP-speak is called a Phase Accumulator

Polarity reversal = Invert = linear -180 degrees phase shift -> these are all synonyms.

Lagging Allpass = Inverted Leading Allpass.
Leading Allpass = Inverted Lagging Allpass.

As far as basic 1st order analog IIR Allpass filters go, there are only two kinds. If you have one of them you can get the other by simple 'polarity reversal' ( i.e. ø ). Basically once you take out noise and distortion -> analog allpass = digital allpass. They are bound by the same system of physics.

[Girolad]

That is why in math you have different symbols for time and phase because they a two separate things.

Often wondered about that. So phase is used to compare relativity for example of two measurements in 'virtual space' to aid calculations, kind of similar to how dB is used then...... I think.

As for the all pass, provided the amp stage is non-inverting then this a 'typical' all pass (seen in the IBP Phase Adjust - already addressed in OP), the other inverted type could be equally typical, I'm not sure which is more common, for instance, regarding the illustration, I assumed that the lrc filters were non inverting, and the preceding / subsequent valves (not shown) to be either or instead.

What I'd like to debunk how the all pass interact in the illustrated circuit, in relation to the phase shift imparted by the filter.

In one unrelated example given in previous reference [2], in order to negate a phase shift, Micheal suggests the placement of a "complimentary all pass" at a frequency slightly below the 'intruding' low pass filter center freq. In terms of two signals parallel to each other, if say, L and R, thats easy enough, both are shifted equally. The illustrated example is trickier, its a mid/side array, comprising TWO all pass, not to mention the L-R,R-L side component as well.

I care not for implementing this circuit, its fairly unnecessary considering two monitors placed horizontally does more or less the same, to an extent (negate head shadow). I just found the circuit intriguing and wanted to share it.

Thanks for the input.

[mystran]

That is why in math you have different symbols for time and phase because they a two separate things.

So phase is used to compare relativity for example of two measurements in 'virtual space' to aid calculations, kind of similar to how dB is used then...... I think.

Phase is kinda like an analog clock, where you can certainly measure the "absolute" phase of some signal (or rather the phases of it's partials), but it wraps around every time it completes a full circle (eg. you can tell it's 3 o'clock, but can't tell if it's 3am or 3pm or if it's Monday or Wednesday). Inverting the signal amounts to turning the clock hand so it points to the opposite direction and again it's impossible to know (just by looking at the hands of the clock) whether you went forwards or backwards.

edit: In reality, it's even a bit more complicated, because in a real-valued signals there's always two clocks for every frequency, one of them going forward and one of them going backwards and if we put the 0-phase at 12 o'clock then we can only really see vertical position of the hand (eg. the cosine), because the horizontal positions (eg. the sines) of the forward and backward going hands sum into zero for a real-valued signal (in comparison to a complex signal where we would have both cosine and sine components).

[Girolad]

Good analogy thanks.

I was introduced to phase with the example of two wine waves in parallel, animated sliding in either direction relative to each other, waves can be made to appear relatively 'inverted' this way, as if one had been polarity reversed. Ive learned in this thread this isn't lagging phase because the wave has shifted in time (I hope i'm right here) wher as lagging phase is just a PolRev in place (negating any phase distortion from an op amp for example).

I read a bit on phase before OP btw, in attempt to differentiate group delay (which is time) from phase. From a stackexchange thread, apparently there is technically no 'start' of a sine wave, its infinite in space and time.

What I do know is that phase shifts can be used to destruct stereo signals acoustically. I'm after an elementary understanding of phase enough to understand stereo (beyond just this thread, don't worry), without doing too much maths which as you know, can escalate quickly.

The main reason for this thread was to figure out how the hell the two all pass interact in the shuffler circuit I posted previously (but sidestepped into phase).

[Girolad]

What about this second plot below though. Notice the graph is negative. These are supposed to be all pass delays...

4632Fig08.gif

The differential phase shift (top) makes sense. Acoustically, it creates a 3db 'dip' in intensity. Similar concept is described in patent 6683962, where a quadrature shift applied at bass frequencies to even out room modes . The graph above is the same principle but addresses acoustic cross-talk emitted from mobile devices instead.

The lines in the bottom plot descend and does not match the all pass already discussed. Any ideas as to why that is?

[mystran]

What about this second plot below, why is the line descending from left to right? Notice the graph is negative. These are supposed to be all pass delays...

Single pole all-pass always gives 180 degrees total phase-shift. By changing the polarity you can choose whether you want that 180 degrees at DC or infinity (and for digital filters "infinity" can be found at half the sampling rate), but it'll gradually ramp from one to the other and all you can really do is change the frequency at which this happens.

Phase-shift is usually plotted negative since any stable causal filter can only ever delay signals anyway (since.. obviously we can't look into the future). For measurements though, you might get anything, since you need an additional unwrapping (ie. undo the wrap-around, which involves a bit of guesswork unless you know the details of the system.. plus you might wanna factor out inversion 180 degrees which is not helpful if you want to look at the delays) to actually get the full picture.

To get the actual (phase-)delay at a given frequency (and you should probably calculate this from a "non-inverted" filter where phase-shift at DC is zero and treat the 180 degrees from inverting separately since it's not really "delay" as such), multiply the wave-length (which is the reciprocal of frequency) by the (unwrapped, but not really relevant for first order filters) phase-shift. It is worth noting that for very low frequencies (where the wave-length is very long), very little phase shift can give you plenty of delay, where as in the limit at infinity you will never have any delay whatsoever (except in case of digital filters where the "infinity" is found at half sampling rate, you get 1 sample of delay there per pole).

It turns out that for sufficiently low frequencies the delay is approximately constant. The lower you put the "cutoff" frequency, the more the low-frequencies are delayed, but this also lowers the frequency where the delay starts getting smaller (ie. can no longer be considered approximately constant). It's actually pretty logical, since we only have finite amount of phase-shift, so we can't really delay higher frequencies very much, 'cos we would need a lot of phase-shift due to their wave-lengths being short... where as for sufficiently low-frequencies we can get as much delay as we want, because we only need a little phase-shift if the wave-length is long enough.

In some cases you might also be interested in the "group delay" which is essentially the delay as applied to the envelope of the signal at a given frequency. Depending on application, this can be more meaningful and can be found by taking the derivative of the phase-delay with respect to frequency... meaning the steeper the angle at which the phase-shift changes around some frequency, the more the envelope will be delayed there.

There's probably something more I forgot to mention, but like.. maybe some of this answers your current questions, hopefully without raising too many new ones.

[Girolad]

I think I made the mistake of asking questions secondary to my initial question (now answered), about phase on a digital orientated forum i.e. KVR. best focus on linear cont-time filters I think, its all thats of concern here.

I understand phase shift at infinity, being small(er), is a limit in discrete-time audio though it depends on sample rate, for instance if the bandwidth of the discrete-time signal was 256*fs (2.8Mhz), this fast digital signal has more room available 'up there' theoretically and an all-pass can otherwise behave more like it would in continuous-time electronics.

Did you misread my question though? For clarity... see the phase curves below, all all-pass filters. I was told the difference between plots 1 and 2 is polarity in-place. Now plot 3 I'm told is also due to polarity, but how can this be? The curve descends from DC and this makes my head explode.

Untitled.png

Ignore the funky DC in plot 1, I've seen this for the low pass filter before in a filter sim program a while ago (maybe spice).

I took the first two 'measurements' with VST Analyzer (my innumerable thanks to christian budde for this free creation) and operated the loaded plugin therein at 192k for more continuous-time-like results at 20k.

[Ichad.c]

Now plot 3 I'm told is also due to polarity, but how can this be? The curve descends from DC and this makes my head explode.

Don't know where you got plot 3, so there is 2 possible explanations(speculating):

Unlikely explanation: It's what I like to call a reciprocal allpass, can also mathematically be called an 'inverse' allpass, but I don't want to stoke more confusion. These only exist on paper(but you can plot it with a symbolic math program),they are not casual/stable. In other words - you'll blow up your speakers

Likely explanation: Mathematics isn't as standardized as you might think. To borrow Mystran's clock analogy, sometimes AM and PM get mixed up. plot 3 is most likely plot 1(phase is a circle). For a real example: One of the most famous FM synths of all time is the Yamaha DX7, yet it isn't an FM synth, it uses Phase Modulation. The inventor is John M. Chownin, an acclaimed Stanford professor... So always take everything you read with a pinch of salt.

[mystran]

Did you misread my question though? For clarity... see the phase curves below, all all-pass filters. I was told the difference between plots 1 and 2 is polarity in-place. Now plot 3 I'm told is also due to polarity, but how can this be? The curve descends from DC and this makes my head explode.

Untitled.png

I think VST Analyzer just plots towards positive and your other plot is towards negative. You see both conventions around, but it makes no practical difference since stable causal phase can only ever go one direction anyway.

[Girolad]

thanks

[Girolad]

Hi,
Sorry for the delay.

Don't know where you got plot 3, so there is 2 possible explanations (speculating):

Unlikely explanation: It's what I like to call a reciprocal allpass, can also mathematically be called an 'inverse' allpass, but I don't want to stoke more confusion.

I got the plot from this page
https://www.maximintegrated.com/en/app- ... vp/id/4632
Can all-pass can be arranged to have an identical output to that of an integrator in the difference of a SD network? If that seems like a weird question, I'll elaborate on what lead me to ask it:

So far, I've identified two practical circuits that alter phase and in turn stereo acoustic response, as well as a third circuit which I will explain.
The first two circuits involve an arrangement of all-pass networks in parallel. The last circuit does not but the net acoustic result is as if it were.
1) Aforementioned lexicon circuit, (see attachment).
2) The maxim circuit (see above link).
3) Blumlein shuffler (not to be confused with the linear CDV shuffler circuit illustrated earlier). Termed here b-shuffler for short.

Because I lack math skill I prefer to rely on worded descriptions of these circuits in order to possibly better understand (and maybe explain to others, stereo-phase & acoustic sum).

The lexicon circuit (below) has filters cascaded after the all-pass. The filters purpose is to counter an acoustic null introduced by the subsequent quadrature shift between channels at the relevant frequencies, when summed acoustically in a room. At what frequencies this works isn't important for now.

Lexicon.png

I can only assume that the maxim circuit also exhibits a null seeing as it uses all-pass in parallel and mentions 'quadrature' (see link). It differs from lexicon in that it has a crossfeed element:

Maxim.gif

Recall the 1930s B-shuffler, it too applies a 90deg / perpendicular / quadrature phase shift, but this time in a difference signal, derived via SumDiff network. (I assume) the B-shuffler must then also have a null somewhere, even though a null isn't implied in the limited documentation that exists for it. The only hint, as taken from the EMI archives:

One method of obtaining a binaural illusion is to convert the low frequency phase differences of the pressure microphone outputs into amplitude differences. [...] phase differences would be automatically converted into amplitude differences thus reducing differences in output intensity of the two speakers.

Where intensity = sum of two acoustic signals at listening position.

Per original design, (below) the b-shuffler is a Sum/Diff network, therein a low passed portion of the difference signal is boosted relative to sum (rather the sum is attenuated by ~18dB). The LPF is an integrator with fc= 16-50Hz.

b-shuffler.jpg

The significance here is that 'LF difference amplitude boost' conflicts with blumleins statement, shown in italic in the quote above. As such its quite plausible to assume that an acoustic null is present in this effect as well.

Also, Maxim terms the circuit as trans-aural correction. Blumleins shuffler has never been termed this but 'binaural' stereo, because the descriptions match, (trans-aural = both speakers via both ears acoustically) then they are the same. The odd thing about the maxim is the complexity of the circuit, with crossfeed lines.

Any contributions?