My Vision of the Future Guitar Amp

[RudeBoy]

Has anyone tried making, or thought of making a modeling amp that was a circuit modeler instead of a soundwave modeler? That point I'm getting to, is that a tube amp with digitally controlled circuits might stand a much better chance of imitating all the other amps out there.

A Laplace transformer instead of a Fourier transformer? I doubt you could tell the difference. OTOH modeling a circuit in the time domain has a certain attractiveness to it. It would be like a real-time Spice sim'.

(Assuming you could get the models), a closed-form solution for the whole amp in the time domain would probably do it. Whether a fast DSP could crunch it... (?).

[Steve Conner]

Well, strictly speaking, DSP can only do Z-transforms. If you want to implement digital versions of things defined in terms of Fourier or Laplace transforms, you have to convert them to Z-transforms. There are plenty of math-heavy explanations out there that explain how to do this, the classic example being when you want to design a digital filter based on an analog prototype.

The conversion is not perfect, and this is one reason why we have aliasing in digital systems. You don't really unlock the power of digital filtering until you start making filters that aren't based on anything in the analog domain, which is great news for cellphone base station designers, but bad news for amp modellers.

A Spice-like simulator works by turning the schematic into a set of simultaneous differential equations, then solving them with a numerical integration engine. Some versions of Spice will actually try to take audio input and output.

Problems with using this for amp modelling:

1. All Spice simulators that I know of like to vary the time step of the integration. It gets smaller at busy periods when there's a lot happening, and larger when the circuit is quiescent. Essentially the timestep is made just small enough that the difference between continuous and discrete time (ie, the difference between Laplace and Z transforms) can be ignored at all times. This doesn't sit well with the need to process audio at a constant sampling rate, and meet real-time guarantees.

2. Spice models are notoriously hard to get to mimic the actual component. I'm willing to bet that part of the "mojo" of tubes comes from parts of their characteristic where they deviate from the Child-Langmuir power laws and so on. How would you model this in a simulation? Linear interpolation (what Spice calls "piecewise-linear") may not be good enough, and higher-order interpolation can be fiendishly expensive in CPU power.

3. Bob Pease hates Spice.

Having said all this, I bet that with a good 32-bit floating-point DSP, you could develop a vacuum tube model based on polynomials, lookup tables, spline interpolation or whatever, and a coupling capacitor/tonestack model based on IIR filters, and hook instances of them together in a signal flow graph like Lego bricks. This is probably how Revalver works.

But good luck trying to model an output stage with NFB around it!

[Enzo]

I have never used Spice, so cannot speak from experience. I do enjoy reading Bob Pease, and am aware of his views on the stuff. I read the electronic engineering magazines trying to keep at least semi-aware of what goes on in the industry. SO I bump into SPice stories from time to time.

I recall one discussion wherein some circuit that ought to work just wouldn;t come together in Spice. They solved the problem by adding a resistor to the Spice simulation. It was grounded at both ends, but it was what it took to make the simulation come together. I recall another, no details, but something trivial that shouldn't matter made a difference. it was something along the lines of reversing the ends of a resistor.

To me, in some way I cannot quite articulate, that represents the problem of trying to simulate reality.

[tubeswell]

I have never used Spice, so cannot speak from experience.

Me neither - I haven't found a version for Mac so I haven't tried it.

But even if there was a Mac version, I probably wouldn't get around to experimenting with it because I'm less of a in-the-head type and more of a hands-on learner - which is riskier, but that's part of the excitement ain't it?

[Phostenix]

I had read an interview with Cliff Chase where he mentioned what software tools (generically) that he used to program the TigerShark DSP, but I can't find that now.


I read a TV trade magazine article today about LECs - Light Emitting Capacitors - 3' x 6' colored screens that emit their own light (for chroma-key backgrounds). That got me wondering if capacitive touch screens will soon have their own built-in backlight. Hmmm.
Making Green 'Cool', by James E. O'Neal

[Steve Conner]

Well, you have a choice of C or assembler through Analog Devices' VisualDSP development environment, or... OK, that's probably your only choice, and a license costs about $10k.

Unless they have something that ties in with Matlab now, but a license for Matlab isn't exactly cheap either.

[raiken]

Well, you have a choice of C or assembler through Analog Devices' VisualDSP development environment, or... OK, that's probably your only choice, and a license costs about $10k.

Unless they have something that ties in with Matlab now, but a license for Matlab isn't exactly cheap either.

Freescale offers a free Eclipse-based IDE for their Symphony audio DSPs so you can program them in C or assembler. I wish more manufacturers would follow suit. I really like the Eclipse/GNU tools, I use them all the time for ARM7 and ARM9 microcontrollers.

Randall Aiken

[Steve Conner]

That free Eclipse thing is what ships with the Line6 Tonecore DDK. Well, it doesn't ship: the instructions tell you to download it from Freescale, and it won't run unless you install Java and a bunch of other junk.

After all, that has a "Symphony" DSP inside, or a 56k as they used to be known before the marketing guys rebranded them.

[raiken]

That free Eclipse thing is what ships with the Line6 Tonecore DDK. Well, it doesn't ship: the instructions tell you to download it from Freescale, and it won't run unless you install Java and a bunch of other junk.

After all, that has a "Symphony" DSP inside, or a 56k as they used to be known before the marketing guys rebranded them.

Really? I didn't know they used a Freescale part, I figured it would be a Blackfin or Sharc. The Symphony DSP parts are nice.

RA

[Steve Conner]

As far as I know, Line6 have always used the Motorola 56k, oops, Freescale Symphony DSPs, in everything they've ever made. Certainly every Line6 product I've opened up has had one in it.

Someone just posted a picture here of the guts of a Digitech Whammy 4, and I spotted a 56k in there too.

I think the 56k is more cost-effective because it's a 24-bit fixed-point processor. The SHARC is 32-bit floating point, which is arguably better, but it makes the chip more complicated and expensive.

The Blackfin series are (IIRC) 16-bit fixed point, which is only really useful for comms or video.

[raiken]

The Blackfin series are (IIRC) 16-bit fixed point, which is only really useful for comms or video.

The Blackfins are 16/32 bit processors, and they blaze at around 600MHz for processors like the BF537. They have 32-bit registers, 40-bit ALUs, and 40-bit accumulators, so they can be used for audio. I know they offer audio starter kits for things like MP3 and other compression formats. I suspect you can get a lot done with 600MHz instruction cycles for 48kHz or 96kHz sampled audio. Now, 24-bit "quality" audio might be another story, since you'd probably want a bigger accumulator and a nice 24x24 multiplier like the one in the Freescale parts.

RA

[Old Tele man]

back in 2003...

Phostenix -- have you seen this paper? http://cegt201.bradley.edu/projects/...20Proposal.pdf

[Phostenix]

Thanks. I'll read it more closely when I've got more time, but it highlights what I think is the limitation of the "modelled" sound approach. He's done 12 samples of 12 volume settings. Makes sense for an old Champ, but what if the amp has Bass, Mid, Treble, Presence, Gain & Volume controls? How many different (interactive) settings must be modelled to truly represent that amp? How do you interpolate the sounds between the settings you've sampled? In my (naive & inexperienced) way of thinking, simulating the circuits & components involved would have a better chance ultimately of achieving a good digital clone of the original.

[BackwardsBoB]

I think the challenge will be to come up with the right ergonomics to make a complete amp/effects system that is truly 'guitarist-friendly', and practical/useable in an on-stage situation.

Yeah. That's the ticket. Go after the fat-cat musicians' money. We over-the-hill mid-life-crisis bedroom headbangers are tired of supporting the whole industry!

[Patrick M]

I just thought of an analogy I never have before. Playing a tube amp is like playing a violin, while playing a modeling amp is like playing a keyboard.

And I have yet to hear a keyboard that can REALLY fool me regarding grand piano sounds, but that doesn't mean they can't do some pretty neat things.

Very well said. I love my Tele and my tube amp - simple and sounds great