The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Discuss your workplace, instruments, amps, and any other gear.

Moderators: MattKingUSA, khz

gimmeapill
Established Member
Posts: 564
Joined: Thu Mar 12, 2015 8:41 am
Has thanked: 44 times
Been thanked: 8 times

The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Post by gimmeapill »

Creating a dedicated topic for this type of device:

Image

Review here:
http://www.soundonsound.com/reviews/kem ... -amplifier

Already discussed below:
viewtopic.php?f=24&t=16731&start=15

viewtopic.php?f=48&t=16539&p=77059&hilit=kemper#p77059

Feel free to post your thoughts on the theory of operation, hints on reverse engineering, how to build a poor man's version with a Raspberry Pi and a can of beans, etc...
Last edited by gimmeapill on Fri Feb 24, 2017 9:59 am, edited 2 times in total.
User avatar
sadko4u
Established Member
Posts: 986
Joined: Mon Sep 28, 2015 9:03 pm
Has thanked: 2 times
Been thanked: 359 times

Re: The Kemper Profiler thread

Post by sadko4u »

I think the first we should start to, it to understand the process of profiling: what does it do to profile an amplifier.
We also working with CrocoDuck on his theory of profiling. Not so quick but we're doing small steps.
LSP (Linux Studio Plugins) Developer and Maintainer.
CrocoDuck
Established Member
Posts: 1133
Joined: Sat May 05, 2012 6:12 pm
Been thanked: 17 times

Re: The Kemper Profiler thread

Post by CrocoDuck »

sadko4u wrote: We also working with CrocoDuck on his theory of profiling. Not so quick but we're doing small steps.
Yeah I am such a noob I am struggling with writing a band limited signal generator... :oops:

Anyway, there is a swept sine method for profiling nonlinear systems that I know it can work. I used it in my master dissertation to identify the response of audio amplifiers and it worked well. Small step by small step we should be able to get to a working plugin I think.
gimmeapill
Established Member
Posts: 564
Joined: Thu Mar 12, 2015 8:41 am
Has thanked: 44 times
Been thanked: 8 times

Re: The Kemper Profiler thread

Post by gimmeapill »

I never tried one or understand the low level technology, so let's start speculating wildly in layman terms ;-)

From what I got from the review, a profile is not a full amp model but a snapshot of one fixed amp+cab+mic setting - or whatever signal chain with an input and an output that can be recorded.
If you have 2 settings you like for your amp (like clean, and overdrive), you will then need 2 profiles.

The innovation here is that this thing can also capture the non linear part and approximate it well enough that even pro players like it.
To do that profiling, the Kemper sends a series of tests tones to that particular configuration (assuming at various levels and frequencies - to mimic what a guitar would do) and records the results.

Then what?
Could it be that it just generates one IR per test tone, then do its black magic to *merge* them all into one profile?
Assuming that this is the case: at play time it would then switch between IRs transparently by using the reference input tone as a trigger to move from one IR to the other?

That would mean a few individual requirements already:
- Having a file format that describes a bunch of IRs, and keeping track of what input tone generated which IR (=a database?) - that would be the profile.
- Being able to switch/morph from one IR to the other in real time (the convolution engine?)
- Comparing the live input signal to the database of combos test signals / IRs in real time.

If that's the case, a good way to start could be maybe to get a recording of those test tones - that could tell us what order of magnitude we are talking about there. 12? 120? 1200?
Then why not trying to *cough* *cough* open of those profile files?

Just random thoughts...
gimmeapill
Established Member
Posts: 564
Joined: Thu Mar 12, 2015 8:41 am
Has thanked: 44 times
Been thanked: 8 times

Re: The Kemper Profiler thread

Post by gimmeapill »

@Beck: This is not the exact same use case. Line6 devices don't allow you to model your own hardware signal chain, and their emulations are not as advanced. Pro players seem to like it because they can model their premium gear at home or in the studio and take only the Kemper when they go on tour. The glowing review from SOS will take you through the features better than I could:
http://www.soundonsound.com/reviews/kem ... -amplifier

But yeah, the price point is not targeted at hobbyists like us. As an emulation amp it would probably be more fair to compare it to a THR100: http://www.musicradar.com/reviews/guita ... 0hd-628051


Now, back to trying to understand what sadko4u & crocoduck are cooking. I get that this "swept-sine method" means "nonlinear convolution process". I don't grep the maths, but the description make sense.
What I don't get is how this thing can let you apply power-supply sag, tube bias and picking definition on a profile if it's made of IRs?
ssj71
Established Member
Posts: 1294
Joined: Tue Sep 25, 2012 6:36 pm
Has thanked: 1 time

Re: The Kemper Profiler thread

Post by ssj71 »

here are some of the profiling signal: https://youtu.be/RBylJkMXaqk?t=192
you can find other demos of profiling to hear more (that one was cut short).

I actually looked at one of the profile files. Its mostly binary blobs so you'd have to do quite the reverse engineering job to figure out what the different fields in it are.

EDIT: They have a clean and distorted set of profiling signals. Here is a more complete set of the clean test signals https://youtu.be/BWMG5DhFckU?t=158 (previous video was the distorted ones).
_ssj71

music: https://soundcloud.com/ssj71
My plugins are Infamous! http://ssj71.github.io/infamousPlugins
I just want to get back to making music!
CrocoDuck
Established Member
Posts: 1133
Joined: Sat May 05, 2012 6:12 pm
Been thanked: 17 times

Re: The Kemper Profiler thread

Post by CrocoDuck »

There's a lot of literature in nonlinear systems identification. See this google for example. Quite a lot of useful info pops up.

I will have a look at the Kemper maybe, I have the feeling it mixes traditional nonlinear system identification techniques with optimization routines. Anyway, I don't think we should necessarily aim to emulate the Kemper, but rather come with our own identification (profiling) algorithm and optimize that.

@beck: this would be able to create profile of every time invariant nonlinear system you want. Imagine you have your pedalboard with your favorite pedals set up. Now, you are going on holiday and you cannot bring the pedalboard with you. No problem! You profile the pedalboard and you get a plugin simulation of it running in your host. Wouldn't be that cool?
CrocoDuck
Established Member
Posts: 1133
Joined: Sat May 05, 2012 6:12 pm
Been thanked: 17 times

Re: The Kemper Profiler thread

Post by CrocoDuck »

From the original post:
gimmeapill wrote:Feel free to post your thoughts on the theory of operation, hints on reverse engineering, how to build a poor man's version with a Raspberry Pi and a can of beans, etc...
I think this thread is more about "how we can write some software and use some hardware to achieve the same?". It is not about discussing how cool or not cool is the Kemper (if not as just a mere reference). Which means that this:
beck wrote:Is it that interesting for us homerecorders? And overall, for that price?
is kinda out of topic. Again, we want to create a tool with similar functionality, not discuss about the Kemper Profiler itself.

Am I right on this? The thread title feels misleading tho. Maybe it was better something like "The Nonliner Audio Equipment Profiler Thread"... Dunno....
CrocoDuck
Established Member
Posts: 1133
Joined: Sat May 05, 2012 6:12 pm
Been thanked: 17 times

Re: The Kemper Profiler thread

Post by CrocoDuck »

gimmeapill wrote:Now, back to trying to understand what sadko4u & crocoduck are cooking. I get that this "swept-sine method" means "nonlinear convolution process". I don't grep the maths, but the description make sense.
What I don't get is how this thing can let you apply power-supply sag, tube bias and picking definition on a profile if it's made of IRs?
Sorry, I forgot to answer this.

You mean the Kemper? Sorry, I still have to document myself properly on what the Kemper does really. Had not a lot of time lately...

Anyway, seems like it is studied for tube amps specifically. Mi guess is that it has generic physical models of tube amps inside for which it tries to calculate parameters to obtain best agreement of synthesized VS measured response through optimization algorithms, plus probably some black box algorithm as well, maybe to get a starting guess for the parameters. I guess we will never know. The test signal say nothing obvious to me... The broad band sounds like MLS in a way and there are MLS based techniques for black box nonlinear system identification... the sort of pulse signal might be a sort of hysteresis cycle tester... But I might be utterly wrong.

The swept sine method is a convenient way to calculate parameters of a generalized nonlinear model instead, which applies to nonlinear systems in general. It would work also for a loudspeaker, for example, or a compressor unit. Appropriate parameters will produce a model with input/output relation approximating the one of the real system. The IRs actually don't convey the whole nonlinear picture of the system, but only partial information. They are transformed mathematically into linear filters in a network DSP which contains also static nonlinearities. The basic concept is as follows:

The static nonlinearities generate the distortion products, the linear filters correct their magnitude and phase so that they resemble the ones produced by the real system. The linear filters are built on the entirety of the higher order responses of the system, but they are not the same.

My dissertation supervisor and school office gave me green light to publish my thesis on the web, I will post the link here soon. Hopefully it will make things clearer... or much more more confused. Perhaps the latter...
gimmeapill
Established Member
Posts: 564
Joined: Thu Mar 12, 2015 8:41 am
Has thanked: 44 times
Been thanked: 8 times

Re: The Kemper Profiler thread

Post by gimmeapill »

Thanks Croco for the explanation of your work on non-linear audio systems analysis, and to all who contributed.
I'm looking forward to your publication (even though I don't expect to understand much more than the intro and conclusion).
At least it is good to hear that some of us are working to crack that nut ;-)
gimmeapill wrote:
Feel free to post your thoughts on the theory of operation, hints on reverse engineering, how to build a poor man's version with a Raspberry Pi and a can of beans, etc...

I think this thread is more about "how we can write some software and use some hardware to achieve the same?". It is not about discussing how cool or not cool is the Kemper (if not as just a mere reference). Which means that this:
Yeah, I didn't mean it as an advertisement, sorry for not being clear. I just found the technology interesting and became curious after seeing the name popping up left and right. But if any Kemper owner reads this, please do not hesitate to share your impressions.
Again, we want to create a tool with similar functionality, not discuss about the Kemper Profiler itself.

Am I right on this? The thread title feels misleading tho. Maybe it was better something like "The Nonliner Audio Equipment Profiler Thread"... Dunno....
You're absolutely correct, but I'm afraid "The Nonlinear Audio Equipment Profiler Thread" would not directly appeal to many end users. Let me edit a bit the title then.

Cheers,

LX
CrocoDuck
Established Member
Posts: 1133
Joined: Sat May 05, 2012 6:12 pm
Been thanked: 17 times

Re: The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Post by CrocoDuck »

gimmeapill wrote:I'm looking forward to your publication
There we go. By the way, I will be soon in contact with my Bachelors dissertation supervisor to check whether I can publish my Bachelors dissertation too (I think there shouldn't be problems with that as well). That's about a physical model for electric guitars. It is in Italian tho...
gimmeapill
Established Member
Posts: 564
Joined: Thu Mar 12, 2015 8:41 am
Has thanked: 44 times
Been thanked: 8 times

Re: The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Post by gimmeapill »

OK, I gave it a shot and as expected the maths don't mean much to me - sorry ;-(
At least, with the help of the wikipedia article https://en.wikipedia.org/wiki/Nonlinear ... tification, I see where you're going.
That's spot on - nice work!
glowrak guy
Established Member
Posts: 2315
Joined: Sat Jun 21, 2014 8:37 pm
Been thanked: 251 times

Re: The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Post by glowrak guy »

How about using visulaizations technology? Audacity gives us a flat linear view of a tone,
so how about cataloguing a useful range of tones, from a useful range of speakers,
but present them as rays emitting from a speaker, over the duration of the signal.

Once you can see and study visually the result of the signal, perhaps you can
work backwards to visualize the point where the signal impacts the cone,
and backtrace the stages of that signal through the amp circuit (bias/sag inclusive)
even to the pickup's output, and who knows, even to the string gauge and the pluck strength and medium.
'Personalized Players Profiling' (tm)

So many variables. Create an array of EQ and filtering controls to monitor and edit
the amp signal. At some point, the relationship between these controls and the visual array
from the speaker cone, should be able replicate closely, the 2D waveform, but having filled in
the dimensial blanks, so to speak, at which point you might have the basis of a working 'profile'.
Actually converting the linear flat 2d plane to a round, moving 3D array (or 4D counting the elapsed time of a test pluck?)
is where I remind myself that I'm just a dishwasher, stirring the mathematicians pot while they're out back
having a smoke... ...(which I may be blowing)...

Beyond that, Ken McLaren, of ACME Bargig/Headcase fame/infamy might be a good friend to consult.
He loves this stuff, and has proven to be quite good at it. I think he has a facebook related to Headcase,
if you're interested. Also, a topic at www.guitarampmodeling.com might generate some interest.
Cheers
tramp
Established Member
Posts: 2335
Joined: Mon Jul 01, 2013 8:13 am
Has thanked: 9 times
Been thanked: 454 times

Re: The Kemper Profiler thread

Post by tramp »

CrocoDuck wrote:The basic concept is as follows:

The static nonlinearities generate the distortion products, the linear filters correct their magnitude and phase so that they resemble the ones produced by the real system. The linear filters are built on the entirety of the higher order responses of the system, but they are not the same.
This is somewhat similar to what we do in specmatch.
The main difference is that specmatch is focused to re-assemble a specific sound of a band (of course, of a guitar part from it) instead profile a amp.
Guitarix modules provide the distortion, were the user needs to select the distortion modules and settings himself, and specmatch creates a IR file witch correct magnitude and phase to fit the sound perfect. It's easy to play with the settings, re-create a IR-file and do a A<>B test.
Unfortunately we've written specmatch in python, and it is a bit rotten now.
On the road again.
User avatar
funkmuscle
Established Member
Posts: 2801
Joined: Mon Jun 02, 2008 2:30 pm
Has thanked: 129 times
Been thanked: 31 times

Re: The Kemper Profiler and other Nonlinear Audio Equipment analysis and emulation tools

Post by funkmuscle »

glowrak guy wrote:How about using visulaizations technology? Audacity gives us a flat linear view of a tone,
so how about cataloguing a useful range of tones, from a useful range of speakers,
but present them as rays emitting from a speaker, over the duration of the signal.

Once you can see and study visually the result of the signal, perhaps you can
work backwards to visualize the point where the signal impacts the cone,
and backtrace the stages of that signal through the amp circuit (bias/sag inclusive)
even to the pickup's output, and who knows, even to the string gauge and the pluck strength and medium.
'Personalized Players Profiling' (tm)

So many variables. Create an array of EQ and filtering controls to monitor and edit
the amp signal. At some point, the relationship between these controls and the visual array
from the speaker cone, should be able replicate closely, the 2D waveform, but having filled in
the dimensial blanks, so to speak, at which point you might have the basis of a working 'profile'.
Actually converting the linear flat 2d plane to a round, moving 3D array (or 4D counting the elapsed time of a test pluck?)
is where I remind myself that I'm just a dishwasher, stirring the mathematicians pot while they're out back
having a smoke... ...(which I may be blowing)...

Beyond that, Ken McLaren, of ACME Bargig/Headcase fame/infamy might be a good friend to consult.
He loves this stuff, and has proven to be quite good at it. I think he has a facebook related to Headcase,
if you're interested. Also, a topic at http://www.guitarampmodeling.com might generate some interest.
Cheers
When tramp was starting Guitarix and I was assisting by testing things, I reached out to Ken for help as he lives about half hour from me. His health was an issue and he's no longer doing anything.
He couldn't offer help. It's been about two years now since I last spoke to him on fb
Post Reply