Airwindows Ultrasonic Lite (and Medium): Mac/Windows/Linux AU/VST

[rafa1981]

At 44/48 using this plugin makes no sense IMO. It is just a 2-pole or 2x 2-poles set near nyquist with Q=0.707 (Butterworth). It won't have any rejection and introduce phase of issues. The same can be achieved with a regular EQ.

At higher sample rates the main characteristics of a filter of this kind should be to leave the passband untouched and to have very high rejection. Basically to be exactly like a downsample filter but having a bit more of freedom to place the cutoff higher.

44k to 88kHz is just one octave, so a single or two bipoles are not going to achieve a lot of rejection. 12 or 24dB. Even the original Ultrasonic is 60dB/octave. Going to 90-120dB rejection wouldn't hurt.

IMO a filter based on this concept should do nothing at < 48k, be FIR based by default (mixing, mastering) with linear-phase 90dB+ rejection and to offer IIR for when latency is a problem. Bonus if the amount of rejection/CPU could be configurable.

[jinxtigr]

Tokyo Dawn's got your back

Mine is different, because we disagree quite profoundly on a number of things

We agree on 'at 44.1/48k this plugin makes no sense' though. MAYBE at 48k. But mostly nah. If you're using this it should be because you're operating at a high sample rate.

[rafa1981]

Well, some things admit little discussion. An ultrasonic filter has to attenuate the non audible frequencies to an inaudible level (-90/-120dB)(if the were bounced back as aliasing) as fast as possible and to respect the passband as much as it goes.

A Butterworth is going to:

• Affect the phase of the whole passband.

• Reach the attenuation considered as non-audible (-90/-110dB) much later/gradually than a FIR (for the same CPU budget).

• The better the attenuation profile (number of poles) the worse the phase shift on the passband.

• Not introduce latency.

• Only have post-ringing.

A FIR is going to:

• Keep the phase intact.

• Reach the attenuation considered as non-audible (-90/-110dB) faster (for the same CPU budget).

• Introduce latency.

• Have pre-ringing and post-ringing.

And as this is an ultrasonic filter, the ringing won't be heard as the cutoff is over the audible range. Pre-ringing is mostly a problem of steep FIR filters at low frequencies -> sharp linear phase EQs.

[Everglide]

I'm not convinced by Dark, I think it sounds strange and so do some clients, a couple have mentioned "What did you do? Sounds all weird in the upper mids and highs etc." when the only difference was swapping NJAD for Dark for a week, too coloured for me.

Dark produces very low harmonics

[jinxtigr]

Well, that's why my choices around 'updating' something like Monitoring are always in the form of releasing a V2 with Dark rather than swapping out the existing plugin. I do have the additional two menu items, the Pad24 and Pad16 that work seamlessly with Tube and Tube2's clipping at exactly 0dB (or anything else known to clip at exactly 0dB), but if it was an issue, I could update Monitoring to have those as well: people's daily use of Monitoring would be on setting 0.0 most of the time, so retaining saved patches is not nearly as much a concern. I didn't do this as I didn't think it was that big of a change, but I didn't know whether there's people out there using some setting in Monitoring as a mix effect or template.

As for ultrasonic filtering, I really do not agree that an ultrasonic filter has to brickwall, or that it has to have pre-ring and not act like an analog filter would. Analog circuitry is LOADED with simple ultrasonic filtering and my intention is to make available the digital version of this, to use as it exists on large format analog consoles, and part of the concept is that you don't impose all the filtering at once, but distribute it across the entire mix. So we're gong to disagree about the idea that ultrasonic filtering (which correctly put, addresses aliasing, which is always undesirable) must itself approximate perfect defined behavior in every stage. I don't think it does. If people are looking for truly impressive aliasing suppression and advancing the state of digital processing, they can further escalate the sample rate and be doing things at 192k, at which point the 'very gentle' UltrasonicLite has enough room to be pretty spectacularly effective used in a distributed way.

I'm not suggesting that (though I've got Music Thing Modular chord organ modules running at a sample rate of over 768k: brute force IS an effective way to attack this problem, you do NOT have to assume you can make brickwall filters sound acceptable)

But I AM suggesting that you'll continue to get an alternate sound some may find desirable, through using Airwindows Console (every element of which contains ultrasonic filtering, in Console7 and PurestConsole2) and UltrasonicLite/Med/Full between stages that don't already have this filtering, at 96k (assuming a powerful enough DAW to do what you need to do at 2X sample rate). It turns out this is enough to make more minimalist, naive processing sound good. The end result will be different from doing everything at 44.1k and relying heavily on oversampling and brickwall filters.

You could very likely take the result and play it back through a weird audiophile DAC that also doesn't do brickwall filtering or reconstruction: those exist, and people are interested in them even at 44.1k but they'd be considerably more relevant at 96k playing back work that was done in the way I describe.

[cantaloupe]

Nice post Chris.

I'm bumping my previous request: Can you please explain which of your plugins have ultrasonic filtering built in (outside of Console7), and does having it built in mean that we don´t need to use any Ultrasonic plugins before those?
Thanks again.

A tip for Reaper users: There is a stock plugin called "Spectrograph" that quickly reveals ultrasonic frequencies. I found it very useful to put that on each bus to monitor how strongly i need to filter.

[rafa1981]

As for ultrasonic filtering, I really do not agree that an ultrasonic filter has to brickwall, or that it has to have pre-ring and not act like an analog filter would.

If it has to prevent aliasing, a digital-only phenomenon, it has to remove all the undesirable frequencies as best as possible and to respect the passband. Why is desirable for a digital filter to combat a digital phenomenon (aliasing) to act "analog"?

Analog circuitry is LOADED with simple ultrasonic filtering and my intention is to make available the digital version of this, to use as it exists on large format analog consoles, ...

I thought that this was an ultrasonic filter: a digital filter to prevent a digital-only phenomenon: aliasing. I see nothing "analog" about it.

Analog circuitry (medium) has infinite bandwidth in theory. The medium and circuits are not perfect and have a lowpass effect but it stretches to the MHz range and never bounces back as aliasing. I would say that a digital IIR Butterworth filter is very far from an emulation of an analog medium: the analog medium doesn't phase-shift the passband.

and part of the concept is that you don't impose all the filtering at once, but distribute it across the entire mix. So we're gong to disagree about the idea that ultrasonic filtering (which correctly put, addresses aliasing, which is always undesirable) must itself approximate perfect defined behavior in every stage. I don't think it does.

I thought that the idea of a digital ultrasonic filter was to be placed just after problematic effects generating supersonic frequencies. Why distributing it over a mix? Distributing it on the mix is not going to reduce the phase shift and reach more aliasing suppression.

If people are looking for truly impressive aliasing suppression and advancing the state of digital processing, they can further escalate the sample rate and be doing things at 192k, at which point the 'very gentle' UltrasonicLite has enough room to be pretty spectacularly effective used in a distributed way.

This doesn't change the fact that it will affect the passband. So I see no "advance of the state of digital processing".

m not suggesting that (though I've got Music Thing Modular chord organ modules running at a sample rate of over 768k: brute force IS an effective way to attack this problem, you do NOT have to assume you can make brickwall filters sound acceptable)

Without entering down the rabbit hole of if brickwall filters can sound acceptable, a supersonic lowpass filter does not operate on the audible band. If it doesn't affect the audio band it won't be heard.

But I AM suggesting that you'll continue to get an alternate sound some may find desirable, through using Airwindows Console (every element of which contains ultrasonic filtering, in Console7 and PurestConsole2) and UltrasonicLite/Med/Full between stages that don't already have this filtering, at 96k (assuming a powerful enough DAW to do what you need to do at 2X sample rate). It turns out this is enough to make more minimalist, naive processing sound good. The end result will be different from doing everything at 44.1k and relying heavily on oversampling and brickwall filters.

I don't see why making assumptions about what the user has behind on its processing chain and embedding ultrasonic filters in all effects, specially when such ultrasonic filters add phase shift. The track behind "Console" can be perfectly clean, so the extra processing is not minimalistic, quite the contrary: it will burn CPU "just in case". I see that e.g. on Console7 the ultrasonic filter is even enabled when working at 44/48KHz, so it will get the phase shift without the aliasing rejection. You agreed that at 44/48 the ultrasonic filter is useless, yet it is still present on those.

You could very likely take the result and play it back through a weird audiophile DAC that also doesn't do brickwall filtering or reconstruction: those exist, and people are interested in them even at 44.1k but they'd be considerably more relevant at 96k playing back work that was done in the way I describe.

I leave the audiophile stuff to the audiophiles, as I consider that world a business based on selling snake-oil to folks with money sensitive to audio buzzwording.

[jinxtigr]

Nice post Chris.

I'm bumping my previous request: Can you please explain which of your plugins have ultrasonic filtering built in (outside of Console7), and does having it built in mean that we don´t need to use any Ultrasonic plugins before those?
Thanks again. .

I have now updated the website to have tagging, with an Ultrasonic Filtering tag to denote all of the plugins that have that as part of themselves. So now you can look it up without having to dig through GitHub to do it

[jinxtigr]

Analog circuitry (medium) has infinite bandwidth in theory. The medium and circuits are not perfect and have a lowpass effect but it stretches to the MHz range and never bounces back as aliasing. I would say that a digital IIR Butterworth filter is very far from an emulation of an analog medium: the analog medium doesn't phase-shift the passband.

Having taken apart a lot of mixing boards and things… all those hiss reduction caps and stablize-the-opamps caps are not, not, not linear phase.

I would like to say 'trust me', but you really don't have to. I'm not offended if you're frustrated. I am just not going to budge on this. I'm after something that I do not get in naive DAW mixes, and I'll do whatever I have to, in order to get there.

BTW, KVRistans, I'm about to post the new Biquad… with zipper noise suppression through interpolating coefficients, like the synth guys do

See, I change my mind sometimes. Just not about analog circuitry being linear phase

[rafa1981]

Not frustrated, please not be offended. Just hopefully having a constructive technical discussion.

The capacitors on op amps, etc I doubt they are designed to affect the audio path and even if they do they will be a gentle 1 pole RC filter with I guess a very high cutoff. I can buy a 3-10 degree shift over the passband on analog from a single pole with cutoff on the high kHz/low MHz range, but not a 10 pole Butterworth near nyquist. It's been more than 20 years since I studied basic electronics. Maybe a concrete schematic for some console would help illustrating what you mean.

But anyways, this last message was a change of subject, this is the thread of an ultrasonic digital filter.

None of the technical questions on my last message have been replied. I hope it helps you improving, otherwise it is a time waste.