Wampler Pedalhead

[yeky83]

Seems to me it's a class-D power amp with Two-Notes style power amp modeling & IR-player built in?
Except its special sauce is that it measures speaker impedance (Dayton DATS ish inside), and uses that speaker impedance curve for the power amp modeling. Cool.

But what's machine learning about it?

 

[Boudoir Guitar]

But what's machine learning about it?

IS YOUR SPEAKER CABINET NOT A MACHINE?!?!?

 

[politoleo]

Seems to me it's a class-D power amp with Two-Notes style power amp modeling & IR-player built in?
Except its special sauce is that it measures speaker impedance (Dayton DATS ish inside), and uses that speaker impedance curve for the power amp modeling. Cool.

But what's machine learning about it?

Each amp (any, but tube in this case) behave differently and distort differently and sag differently, plus their presence/depth behave differently as well from each other.

That's where the machine learning part come into play

 

[Boudoir Guitar]

I'm sure glad we got the design guy in here to help clear things up

 

[yeky83]

Each amp (any, but tube in this case) behave differently and distort differently and sag differently, plus their presence/depth behave differently as well from each other.

That's where the machine learning part come into play

So the power amp modeling is machine learning derived, OK.

 

[JiveTurkey]

 

[Chocol8]

Ugh, sounds like these are not going to be shipping until later in the year. I guess that's more time to save up for the inevitable sticker shock come pricing announcement.

 

[Richard_G]

I'm Leo from Synergy, part of the DSP team, was invited here by a moderator to help answering some questions!!

Welcome, Leo! I’m glad you found your way over here.

(Getting info from you about Synergy stuff, the Tone King Imperial preamp pedal, etc., is one of the few reasons I still visit TOP («The other place»).)

 

[banned]

Each amp (any, but tube in this case) behave differently and distort differently and sag differently, plus their presence/depth behave differently as well from each other.

That's where the machine learning part come into play

Hi Leo, does the ML part have to do with fitting the speaker impedance curve? Or is it more related to the 6 power amp models? Or is it something else? Anything more that you can share? I still don't think I fully grasp what this product does.

 

[Frodebro]

Hi Leo, does the ML part have to do with fitting the speaker impedance curve? Or is it more related to the 6 power amp models? Or is it something else? Anything more that you can share? I still don't think I fully grasp what this product does.

It’s a stereo power amp that is compact enough to fit on a pedalboard. It has power amp modeling baked in, and the machine learning bit refers to plotting the speaker impedance curve of whatever cab you’re plugged into.

 

[dronerstone]

@politoleo depending on the overall implementation, it might be a non-issue, but how is the latency?

 

[politoleo]

@politoleo depending on the overall implementation, it might be a non-issue, but how is the latency?

0.75ms right now, but likely 0.4ms in production,

 

[yeky83]

Hi Leo, does the ML part have to do with fitting the speaker impedance curve? Or is it more related to the 6 power amp models? Or is it something else? Anything more that you can share? I still don't think I fully grasp what this product does.

It’s a stereo power amp that is compact enough to fit on a pedalboard. It has power amp modeling baked in, and the machine learning bit refers to plotting the speaker impedance curve of whatever cab you’re plugged into.

No, it's not referring to plotting the speaker impedance curve. See post #63, they used ML to model the power amp (how it distorts, sags, how the presence/depth behaves, etc.).

I like the premise of the product but don't like the labeling... unfortunately we live in a buzzword marketplace. When a product labeled "machine learning" has a "learn" button, one would assume the button does machine learning... but no.

 

[Frodebro]

No, it's not referring to plotting the speaker impedance curve. See post #63, they used ML to model the power amp (how it distorts, sags, how the presence/depth behaves, etc.).

I like the premise of the product but don't like the labeling... unfortunately we live in a buzzword marketplace. When a product labeled "machine learning" has a "learn" button, one would assume the button does machine learning... but no.

It plots that stuff against the speaker impedance curve of whatever cabinet you're plugged into. That's the only variable, otherwise you wouldn't have to utilize the learning function every time you changed cabinets.

 

[banned]

No, it's not referring to plotting the speaker impedance curve. See post #63, they used ML to model the power amp (how it distorts, sags, how the presence/depth behaves, etc.).

I like the premise of the product but don't like the labeling... unfortunately we live in a buzzword marketplace. When a product labeled "machine learning" has a "learn" button, one would assume the button does machine learning... but no.

It is sending a frequency sweep signal as part of the 'learn' routine - isn't it? Wouldn't that be to obtain the SIC? How could they calibrate the power amp just based on that?

 

[yeky83]

It is sending a frequency sweep signal as part of the 'learn' routine - isn't it? Wouldn't that be to obtain the SIC? How could they calibrate the power amp based on that?

A power amp model needs to presume a load (speaker), aka SIC.

Normally power amp models are based on some SIC that is unlikely to match your own particular speaker.
Some products like Fractal allow you to tune the SIC if you have the know-how.

This product measures your speaker's particular SIC using the swept sine wave learn routine (doesn't take ML to do this), then incorporates it into its apparently ML-derived power amp modeling.

 

[politoleo]

The full sweep is for full frequency response.

It does all measurements in real time using the input signal as generator however since the human player it's not a controlled environment and it's unknown what notes/frequency are going to be generated, the sweep serves as a starting point and to set soft/hard limit.


Please come at NAMM and play it, we are setting up some A/B comparison with a switcher, I agree there's nothing worse than buzzword!

 

[politoleo]

It plots that stuff against the speaker impedance curve of whatever cabinet you're plugged into. That's the only variable, otherwise you wouldn't have to utilize the learning function every time you changed cabinets.

Current and voltage feedback are the other variables other then input.

Speaker impedance changes over time it's not a static thing because of BEMF. By just measuring it once and applying you would make a great eq approximation but that wouldn't cover everything else!

 

[Frodebro]

Current and voltage feedback are the other variables other then input.

Speaker impedance changes over time it's not a static thing because of BEMF. By just measuring it once and applying you would make a great eq approximation but that wouldn't cover everything else!

I meant that the cab was the only variable, as far as what the Wampler needs to 'learn' and what is already baked into the firmware.

 

[politoleo]

I meant that the cab was the only variable, as far as what the Wampler needs to 'learn' and what is already baked into the firmware.

Oh sorry I got that wrong! Yes you're right