How good amp sim emulate real circuits? Amplitube '65 Twin reverb vs TH3 '65 Darkface vs LTSpice

[bmanic]

I'd say it's irrelevant but you can nevertheless still fill a bug report to IKM and say their model is 20dB off the target. That's a reasonable complaint.

Thanks for revealing your agenda - that's settled then.

Indeed. It's rather amazing how narrow minded and tribal some people can be. Here is an original poster who created an absolutely awesome, superbly detailed post where most of the hyperbole is held in check..

.. then we get people like stratum here pissing all over the efforts with complete ignorance. Amazing really. Reminds us why there are struggles in this world when there are people like this who can't appreciate the effort the OP put into the first post, no matter what ones own opinion is.

@stratum: Dude.. seriously.

@alteregoxxx: Nice and informative detective work there. Please know that I'm sure there are tons of people in the silent majority that thoroughly appreciate your post. I found it extremely interesting and thorough. Wasn't planning on commenting without some tests of my own but stratum here really triggered me.

[stratum]

Indeed. It's rather amazing how narrow minded and tribal some people can be. Here is an original poster who created an absolutely awesome, superbly detailed post where most of the hyperbole is held in check..

.. then we get people like stratum here pissing all over the efforts with complete ignorance. Amazing really. Reminds us why there are struggles in this world when there are people like this who can't appreciate the effort the OP put into the first post, no matter what ones own opinion is.

@stratum: Dude.. seriously.

OK guys, I have already said I had left the thread. Have fun and continue. Indeed that's a very interesting thread. I know what's wrong with it, but that doesn't mean I have the right the spoil the picnic. Acknowledged.

[DaveClark]

Hi alteregoxxx,

Thanks for your efforts to compare sims to SPICE and posting your results.

Perhaps I missed them in your description, but it seems to me that there are some important details missing. For example, what is the load of your LTspice circuit? Did you attempt to model an electromechanical speaker with some purely electrical circuit, or did you just terminate it with a transformer into a resistor? Or something else?

Similarly, did you simply turn off the cabinet modelling for Amplitude 4 and other sims? Or are you running the waveform into some other sim or circuit model? The Amplitude 4 results look very suspicious for not having been properly terminated. Have you asked IK Multimedia about this? You may think that you should just be able to disconnect the cabinet and that the model should still behave appropriately, but "appropriately" could mean different things to different people.

In your discussion about memory (memoryless) and time variance (invariance) you may have confused some readers. As you probably know, any circuit that has a capacitor in it has memory. You probably also know that the 65 Twin Reverb has feedback from the transformer secondary back to the 12AT7 tubes before the pentodes and that this is also memory in an even more obvious form. What people call "sag" is also a form of memory. So when you posted that there is no "memory" (in quotes), you seem to be writing about something other than what most engineers, scientists, and other technical folks refer to as memory when discussing linearity, invertibility, memory, time invariance, etc. From what you wrote, some readers may not realize that time invariance does not imply memoryless, that these are two different attributes.

You wrote that for the first plot, the TH3 and the LTspice simulations were not notably different. In my experience in working with simulations of modifications of tonestacks, I disagree, somewhat. The frequency of the smallest magnitude is different between the two. The amount of dip below the peaks is different. The relative peaks of bass and treble are different - almost the same for the TH3 but about 3dB difference for LTspice. The lower frequency peak of the TH3 is significantly narrower than the corresponding LTspice peak. As you surely know, small differences on log plots can be large enough to be clearly heard, and I would expect these to results to sound different to me. Yes, the Amplitude 4 curve looks like it's way off, but that may be due to the termination as I mentioned above, or something like that. (It's too far off to be a trustworthy result.) The TH3 result may also be off; that is, the comparison may turn out to be worse if terminated in a similar fashion as the LTspice simulation.

Because your are an EE, I am curious as to why you decided to run white noise into the circuits. Most EEs would probably run sinusoidal sweeps because it's right there in all SPICE versions. Was it just easier to compare the amp sims with white noise? You could have, of course, created a sinusoidal sweep waveform and used that for the amp sims. Just curious.

Finally, you did not show the phase response. Did you look at that? Some people think that phase response is not important, but not most EEs. I've taken sounds and fed them through EQs I've created which modify only phase. The result can be a very different sound indeed.

Thanks again for your efforts. I look forward to reading more about your work.

Regards,
Dave Clark

[alteregoxxx]

Hi alteregoxxx,

Thanks for your efforts to compare sims to SPICE and posting your results.

Perhaps I missed them in your description, but it seems to me that there are some important details missing. For example, what is the load of your LTspice circuit? Did you attempt to model an electromechanical speaker with some purely electrical circuit, or did you just terminate it with a transformer into a resistor? Or something else?

See my last answer, at the bottom of this reply.

Similarly, did you simply turn off the cabinet modelling for Amplitude 4 and other sims? Or are you running the waveform into some other sim or circuit model?

Simply turned off the cab, no other sim or circuit model after the preamp/poweramp.

The Amplitude 4 results look very suspicious for not having been properly terminated.

I firmly believe you can not have unproper termination in the vst amps sims world: if this could happen, you could not correctly use external impulse responses loaded in third parts plugins, as normally happens.
How do you think Amplitube or TH3 or whatever vst amp sim could ever interact "electrically" with a third part impulse loader?
I'll try to explain better; I can think of only 2 possible paths:

1)The developer of the amp sim did plan the interaction with the speaker load to happen when use their own native internal cab sims--> In this case, you should have really bad sound when turning off the native cab sim and using third party impulse loader vst and, obviously, electrical arcs, bad smells and, finally, fire because of the missing load on the transformer secondary.

2)The developer of the amp sim did plan that interaction to happen when use their internal cab sims but, moreover, the finicky developers coded their sims in a way that, when the user turn off the native cab sim, the code automatically "load" the power amp with a "general purpose" electrical load simulating a "generic" speaker as load-->Hence, no problem of unloaded power amp here so....

Moreover, as I wise person, before posting screenshots here, I also measured the frequency response of the Amplitube model with its own cab sim on, so, following your reasoning, with its power amp "loaded". And measured TH3 using this same Amplitube cab sim as "load" for TH3. Guess what? Amplitube was still 20dB off....

Have you asked IK Multimedia about this?

I did not, sorry.

You may think that you should just be able to disconnect the cabinet and that the model should still behave appropriately, but "appropriately" could mean different things to different people.

See my previous answer about load termination and amp sims.

In your discussion about memory (memoryless) and time variance (invariance) you may have confused some readers. As you probably know, any circuit that has a capacitor in it has memory.

Capacitors store DC values. Did you read In that discussion that I mentioned "bias point shifting" ?

You probably also know that the 65 Twin Reverb has feedback from the transformer secondary back to the 12AT7 tubes before the pentodes and that this is also memory in an even more obvious form.

Nope, that feedback is only...feedback. it's not memory because it happens instantaneously, the moment next to the one it happened it has no memory of what it did before. But we are defying in philosophy.

What people call "sag" is also a form of memory. So when you posted that there is no "memory" (in quotes), you seem to be writing about something other than what most engineers, scientists, and other technical folks refer to as memory when discussing linearity, invertibility, memory, time invariance, etc.

Again, "Sag" is a cool word to name guess what?....bias shifting.

From what you wrote, some readers may not realize that time invariance does not imply memoryless, that these are two different attributes.

Again, read carefully the part about bias shifting: I splitted the concept in two parts. I said you can have a time invariant electronic circuit, but also that if you shift its bias point, the same signal that before gave you a result, after the bias shift will give you a different one. Now, you can call it memory if you like to call it that way. Truth is that it is simply bias shifting. Put the circuit in the same exact bias point it was before it was altered by an external signal, power supply fluctuation or whatever, give it a signal and it will alway respond with the same output. That's all.

You wrote that for the first plot, the TH3 and the LTspice simulations were not notably different. In my experience in working with simulations of modifications of tonestacks, I disagree, somewhat. The frequency of the smallest magnitude is different between the two. The amount of dip below the peaks is different. The relative peaks of bass and treble are different - almost the same for the TH3 but about 3dB difference for LTspice. The lower frequency peak of the TH3 is significantly narrower than the corresponding LTspice peak.

Well, we're talking of a circuit that is more than 50 years old, subject to thousands of minor "mods". My LTSpice schematic is of aTwin Reverb '65 reissue. Where in the "original" Twin Reverb there were old 240Kohm pots now there are 250Kohm in the '65 reissue, some caps value are not standard anymore, so they too could be slightly different in the '65 Reissue...That's alone can justify some "little" differences. I can swear that measuring two identical built circuits never gave me same identical results!

As you surely know, small differences on log plots can be large enough to be clearly heard, and I would expect these to results to sound different to me.

You know...It was not me estabilishing we can consider 3dB differences small enough
Jokes a part, For sure they are not identical. Nor I expected them to be, as I explained previously.
Comparing that minor differences to a frequency response that is 20dB off....well...Don't you think one is a "very minor" difference, the other is a gigantic one ? Perspective is the key of life!

Yes, the Amplitude 4 curve looks like it's way off, but that may be due to the termination as I mentioned above, or something like that. (It's too far off to be a trustworthy result.) The TH3 result may also be off; that is, the comparison may turn out to be worse if terminated in a similar fashion as the LTspice simulation.

Uhmmm...I've no caught the logic....Th3 may also be off, but it agrees (Please, forgive my 3dB philosophy!) with LTSpice, Amplitube is massively off...Two results agree, more or less, one is compeltely off...I repeat, I do not believe termination has a relationship with all this.

Because your are an EE, I am curious as to why you decided to run white noise into the circuits. Most EEs would probably run sinusoidal sweeps because it's right there in all SPICE versions. Was it just easier to compare the amp sims with white noise? You could have, of course, created a sinusoidal sweep waveform and used that for the amp sims. Just curious.

Simply to compare, right! (No slope to set on Spctrum analyzer, no reasoning on time integration to choose etc etc etc...)

Finally, you did not show the phase response. Did you look at that? Some people think that phase response is not important, but not most EEs. I've taken sounds and fed them through EQs I've created which modify only phase. The result can be a very different sound indeed.

Please no! Please, do not call for phase response to come in!!
Audibility of phase response is a can of worm. I just don't want to start philosophals debates. Nevertheless, I'll give you my point of view: In some extreme cases, read large enough time shifts between spectrum components of a signal, it can be audible. But those cases cross the frontier with time delay audibility. All the other cases, I think phase delay is not audible. But that's just my opinion and I don't want to convince anyone that I'm right nor he/she is wrong.

Thanks again for your efforts. I look forward to reading more about your work.

You're welcome! I'll post some more screenshots, maybe in a day or two, adding in the party Peavey Revalver 4. I'll take the time to reply also to your first question with some examples.

[DaveClark]

Hi alteregoxxx,

Thanks for your detailed responses to points I wrote about.

1) You said that improper termination would cause problems with third-party cabinet IRs.

I agree that this is a likely result, but my question actually has two parts: What did you use for termination of the amp sims? What did you use for termination of the LTspice simulation? If these are different, then your results are not fully valid. It is very important to ensure that you have terminated the amp sims in the same manner as the LTspice simulation, or somethng very close to this. Unless I missed it, you didn't say what you did for LTspice. I have used a transformer model, no transformer model, resistive loads, etc. They all sound very different, much like what you are showing.

Much more likely, you have simply gotten the bass knob set to completely different values between Amplitube and LTspice without realizing it. From your plots, it appears to me that increasing the bass knob might put the red line right up into the orange one.

Something such as this is very wrong, probably something simple. That is why I don't really think that 20 dB off versus 3 dB off is a valid comparison of error. Not completely impossible, but very unlikely. Do you really believe that guitarists cannot hear that the bass is 20 dB too low? Even Amplitube DSP programmers should be able to hear that!

If I have time, I'll look at the Amplitube model you are working with and see if I can determine anything. I also have ngspice (modded to allow binary signals) and LTspice.

2) You said that in the VST world, you cannot have the termination incorrectly set.

Well you certainly can have this! I didn't bother to set the terminations to anything particular in my own guitar amp sims (LG2Amp, etc.) when the cabs are turned off. If you turn the cabs off, interfacing them to any external cabinet modelling program is your problem, not mine. I allow cabs to be turned off simply to remove the cab effect so that they can be used for acoustic guitar, piano, voice, etc. I did it to remove a filter, not with the intention to allow other filters to be used.

If users heard differences between amp sims with the same cab, they would attribute those differences to the amp sims' modelling, not to different terminations of the circuit, even though the latter may be correct. That is similar to what you did.

3) Unless I misunderstood you: You still insisted that there is no memory. You also said that feedback is instantaneous.

I am very surprised by this because these claims are simply not correct, assuming that you are using the technical definition “memoryless.” This is not philosophy at all; I am speaking about fact. A simple RC lowpass filter is not memoryless; look at the basic equation for voltage of a capacitor which involves an integral over time. In the digital world, most DSP implementations of circuits contain n-1, n-2, etc. and some contain y[n-1], y[n-2], etc. These are also not memoryless. It is the existence of memory that allows filters to work. Bias-shifting is not instantaneous. It can't be.

Capacitors cannot instantaneously change voltage, and inductors cannot instantaneously change current, hence reactive circuits have memory. Guitar amps have reactive circuits (and reactive parasitics) so have memory.

4) You say that phase discussions are usually philosophical or usually something like a philosophical discussion.

These discussions are often about two different things, so people talk past each other, as we say in the English language. It is true that absolute phase of an infinite single-frequency signal, as far as we know, cannot be determined by listening. (I suspect that one could almost prove this from logic alone.) However, when a temporary signal has more than one frequency, shifting the relative phases can audibly change the sound. Music is composed of many temporary sounds so can be audibly changed by shifting the phases of the constituent frequencies relative to one another.

I hope I understood what you wrote correctly and did not misinterpret what you wrote.

Thanks once again for your discussion. I'll let you know if I get some time to look at this myself, but I was planning to rewrite my plugins for CPU efficiency and clean them up a bit. (Lots of trial code is in all of them.)

Regards,
Dave Clark

[alteregoxxx]

Hi alteregoxxx,

Thanks for your detailed responses to points I wrote about.

1) You said that improper termination would cause problems with third-party cabinet IRs.

I agree that this is a likely result, but my question actually has two parts: What did you use for termination of the amp sims? What did you use for termination of the LTspice simulation? If these are different, then your results are not fully valid. It is very important to ensure that you have terminated the amp sims in the same manner as the LTspice simulation, or somethng very close to this. Unless I missed it, you didn't say what you did for LTspice. I have used a transformer model, no transformer model, resistive loads, etc. They all sound very different, much like what you are showing.

For LTSpice : Transformer model loaded with 4 Ohm resistor. And this is the only way I could do it, because I didn't want to see the speaker freq response coming in....

Amp sims have been measured simply putting their cab sims off. Please, could you explain me how many ways you know for " properly terminating an amp sim" ? You have no way of forcing an amp sim coded by someone else to "electrically interact" with the world after the blackbox the amp sim it's closed in...The only thing you can do is to turn its cab sim on or off...

And, let me contraddict you about the "They all sound very different": I can assure you that the frequency response of the preamp alone, in LTSpice, is 99% identical to the frequency response you can measure at the push pull transformer output.
Look for yourself, only 2dB off at 40Hz...:

Much more likely, you have simply gotten the bass knob set to completely different values between Amplitube and LTspice without realizing it. From your plots, it appears to me that increasing the bass knob might put the red line right up into the orange one.

Something such as this is very wrong, probably something simple. That is why I don't really think that 20 dB off versus 3 dB off is a valid comparison of error. Not completely impossible, but very unlikely. Do you really believe that guitarists cannot hear that the bass is 20 dB too low? Even Amplitube DSP programmers should be able to hear that!

Nope. Look at the screenshots...you can see the knob for TH3 and Amplitube...Same positions...Same for LTSpice...

"Do you really believe that guitarists cannot hear that the bass is 20 dB too low?"

Me for sure can. In fact all this started beacause I tought (and still think) Amplitube Twin Reverb 65 model sounds thinny and not like a real Twin. At least it didn't sound at all like the solid state model I realized and was testing on a breadboard with oscilloscope, spectrum analyzer, ears and a guitar...

Don't know...Maybe Amplitube DSP programmer are bass shy!

If I have time, I'll look at the Amplitube model you are working with and see if I can determine anything. I also have ngspice (modded to allow binary signals) and LTspice.

Please, do it.

2) You said that in the VST world, you cannot have the termination incorrectly set.

Well you certainly can have this! I didn't bother to set the terminations to anything particular in my own guitar amp sims (LG2Amp, etc.) when the cabs are turned off. If you turn the cabs off, interfacing them to any external cabinet modelling program is your problem, not mine. I allow cabs to be turned off simply to remove the cab effect so that they can be used for acoustic guitar, piano, voice, etc. I did it to remove a filter, not with the intention to allow other filters to be used.

If users heard differences between amp sims with the same cab, they would attribute those differences to the amp sims' modelling, not to different terminations of the circuit, even though the latter may be correct. That is similar to what you did.

So, in your models, if the user turn off the cab sim, you leave your modeled transformer unloaded? And the output of your modeled cirtcuits doen't oscillate?? Uhmm...now THAT is really strange!

3) Unless I misunderstood you: You still insisted that there is no memory. You also said that feedback is instantaneous.

I am very surprised by this because these claims are simply not correct, assuming that you are using the technical definition “memoryless.” This is not philosophy at all; I am speaking about fact. A simple RC lowpass filter is not memoryless; look at the basic equation for voltage of a capacitor which involves an integral over time. In the digital world, most DSP implementations of circuits contain n-1, n-2, etc. and some contain y[n-1], y[n-2], etc. These are also not memoryless. It is the existence of memory that allows filters to work. Bias-shifting is not instantaneous. It can't be.

Capacitors cannot instantaneously change voltage, and inductors cannot instantaneously change current, hence reactive circuits have memory. Guitar amps have reactive circuits (and reactive parasitics) so have memory.

Don't really want this go on for ever: it's not the argument of the topic.
Nevertheless, I guess language barrier doesn't allow me to better communicate my "vision of things". So far so good.

Low pass filters, hi pass filters (and so on...) have memory. Because they have capacitors and or inductors in them. So what? What audible effect do you think a capacitor memory has on the sound that pass trough the latter?
I'll try (last time, I swear! ) again with an example.Let's try:

Take a "soft" signal, think of it like the guitarist caressing the guitar strings, call it A, and pass it trough a low pass filter; record the output coming out of this low pass filter, depurated from its DC content, hence leaving only the signal.
Do you think you'll obtain a different output passing again this same "caress" signal in the filter, let's say 20 seconds later during which you "maltreated" the poor low pass filter with an hard rock riff? I don't think so. What will be different? Only DC levels on the capacitor of the low pass. Good. DC levels in guitar amp affect bias point and THAT has an effect on "compression", distortion and whatever adjective guitarists all around the world use to define the sound of an amp when pushed, because it influences the way active elements, the only one that are non linear (I know I know...capacitors are not ideal etc etc etc...but please...don't start on this too...).
Bias shift take time to move? Agree! Never said the contrary! I was only trying to give a "different point of view" on the "memory" argument. But I give up, I promise.

4) You say that phase discussions are usually philosophical or usually something like a philosophical discussion.

However, when a temporary signal has more than one frequency, shifting the relative phases can audibly change the sound. Music is composed of many temporary sounds so can be audibly changed by shifting the phases of the constituent frequencies relative to one another.

Whitin the limits already exposed in my previous answer, I repeat that I do not agree.
I'm not able to hear 360°, 530°, 110° phase shift of one frequency component of a complex signal.
Maybe it's a limit of mine.
When phase imply longest enough TIME DELAY, I think it can become audible.

This is what I discovered from my own tests. I do not pretend others agree with my own perception.

Thanks once again for your discussion. I'll let you know if I get some time to look at this myself, but I was planning to rewrite my plugins for CPU efficiency and clean them up a bit. (Lots of trial code is in all of them.)

Regards,
Dave Clark

Hope you'll find some time to post here your own results.

[electro]

TH1 and TH2 have the same algorithm for Twin Blackface and TH1 came out in 2009. Does anybody know if they updated it for TH3?

Amplitube Fender 2 is very recent and much better modeled than Fender 1 which also dates back to about 2009.

[alteregoxxx]

Ok, here again. Let's recap using some images (In the meantime I've added Guitar Rig5, PeaveyRevalver4 and also LG2Amp, from ExeConsulting developer, to the "battle"). The following are comparison on the TwinReverb 65 amp model (exception is LGAmp2, that is not, from my knoledge, specifically modeled on any amp; nevertheless it has in the various tone stack filters ad the user disposal, also one that is "Fender like", so I decided to see how it behaves).

Fender Twin Reverb '65 comparison:








JCM800 comparison:

[alteregoxxx]

Just some quick consideration on the reasons about this small analysis I performed.

First of all, it hasn't to be considered an exhaustive and complete analysis. My intention was only to perform a quick check, to verify if at least the frequency response of the simulated amp model, that is the simplest part to model, was right. From my point of view, if the amp sim is not able to mimic not even the frequency response of the original, there are few chances that it will reproduce the more complex aspects of the original amp accurately, for example the dynamic behaviour of the latter.

This doesn't mean that such an amp model can't sound good on its own, instead, you could like it even better than the original! Nevertheless, its modeling can not be defined accurate. It can only be considered a a good sounding amp on its own, but, despite marketing claims, not an accurate model of this or that amp, sorry....

I've to add that I've also checked for accuarcy in modeling the dynamic behaviour of these amp sims, but I've not posted any screenshot to avoid overcomplicating an already complicated topic...
It has not been a super-accurate check in testing THD in LTSPice versus all the amp sims, because such a test, would require accurate matching on the signal level feeding the amps, and this is a complicated matter when you are comparing a "real" simulation in LTSpice, where you can use "real" signal voltage, versus an amp sim that lives in the VST world, where the levels are confined in a -1 to +1 range...you know...
Nevertheless I exploited a characteristic of the JCM800 to conduct a "reasonably significative" analysis: the JCM800 circuit exhibit a "shift" its frequency response when subjected to low input signals levels versus high signals levels. In other words, I've simulated the JCM800 circuit in LTSPice, stimulating it with noise at 5-6 different voltage levels, from really low to really high. Did the same with amp sims, and looked for "similarity" in the frequency response "shift".


With this considerations in mind, I'll give my point of view on the amp modeler I've tested, both in frequency response accuracy terms (more objective evaluation) and dynamic behaviour terms (more rough, semi-scientific evaluation):

AMPLITUBE:

I'm a bit puzzled by the Amplitube modeling of the Twin Reverb 65, especially beacuse that is an "Officially approved by Fender" specimen. Frequency response seems to be really off.

Their generic "British tube lead 1", not officially approved by Marshall, seems instead, paradoxically a lot more accurate in modeling the behaviour of a JCM800. Dynamic behaviour seems accurate too. Good!

To the people that were suggesting that the Twin Reverb 65 is part of their "old" Fender1 collection, and so that it would have been better to investigate the "new" Fender2 collection, I reply saying that, unfortunately, in this second collection, they did not model a Twin Reverb65; They focalized more on the "tweed" models. Probably they think their first collection already did a good enough job in emulating Fender amps they modeled at that time. Who knows....

OVERLOUD TH3:

They seems to model accurately all the amps I've tested, included the nuances that leads to a frequency response alteration on certain amp models, for example, when having the volume pot at really low or really high values. Dynamic behaviour it's good too. Really good Job!

DIGITECH RP360:

Despite not being as meticulous as Overloud TH3 people in modeling "frequency response volume pot nuances", they did a discrete to very good job (depending on the specific amp model) in modeling the amps I've tested, both, in frequency response and dynamic behaviour. Considering the price of this unit, this is a really really really good sounding solution, except for their cab simulation "system", that completely sucks.
For example, their Twin Reverb 65 2x12 cab, has a frequency response that have some peaks and deeps, like real guitar cabs must have, but problem is that it extends "almost flat" up to 20Khz: no guitar speaker behaves this way, that's sure.
I've discovered, however, that it is this way, because, when you connect headphones to the unit (and behave this way also when you set the unit in what they call "mixer mode", as opposed to the "amp mode") they "superimpose" a strange low pass filter to the the cab sim selected by the user, and their conjuncted work act to "resemble", finally, the frequency response of a guitar speaker...well...a bad one, I mean...
In other words, try the Digitech RP360 with external guitar speakers impulses and this unit simply SHINES! Try it using its internal guitar cab simulation system and you'll fall in a spiky mosquito fuzz world.
An alternative solution, is to use the equalizer included in its internal effects modules and cut the hell around 4Khz (8-12dB), boost 4-5 dB at 125Hz, cut a couple of dB at 8Khz. It works more or less, from good to really good, with all the cab simulations included in the unit.

PEAVEY REVALVER 4:

Similarly to Overloud TH3, they have really accurate modeling going on here. The possibility of tweaking the amp sims circuits in a "SPICE like accurate way" it's simply mind-blowing!!
They also seems to have paid particular attention to the power amp-speaker interaction: It's more than a suspect that this is the reason for that "particular" peak in the really low region, around 50 Hz, for the TwinReverb65 emulation, and the "light" boost in the high freq region in comparsiono to the LTSpice simulation.
Revalver 4 has in fact a setting for choosing the power amp-speaker interaction, where the user can choose between 4 possible "kind of interaction". Moreover there is also the possibility to tweak the "fspeaker interaction" parameter, read the amount of interaction between the speaker and the power amp.
In tube power amps,in fact, being the latter naturally functioning as quasi-current drivers, their interation with guitar speakers (that have a peak impendance in the region between 40-100Hz, usually, and a rising impedance starting from 300-400 Hz) cause a "boost" at "low" freq and "high" freq.
Considering that for my simulation in LTSpice I've used as load for the output transformer a 4 Ohm resistor, hence not a guitar speaker, I should have set this "fspeaker interaction" parameter to the value 0: I've discovered this after I had already posted the screenshots, sorry.

Nevertheless, you can see that the power amp-speaker interaction, can't cause such big differences in frequency response as the ones visible for the Amplitube case. This, in conjunction with my previous LTSpice graph, should also turn away the suspects that the user @Dave Clark had about this point.
Hence, a part from that peak caused by my fault, Revalver 4 results seems to be spot on with LTSpice results. Outstanding.

GUITAR RIG 5:

It seems to be sufficiently accurate in terms of frequency response. However, on the TwinReverb 65 analysys, it seems to deviate from the LTSpice results when analyzing the Twin Reverb 65 with the switch "bright" active (this analysis is not shown in my screenshots for none of the amp sims, nevertheless I've tested also that ).
JCM 800 seems sufficiently well emulated.


LG2AMP (from exe consulting, read the user @Dave Clark):

This is not intented to emulate any specific amp, at least from my knowledge, so there is no sense in evaluating its accuracy in emulating a TwinReverb65.
It has to be intented, as from my understanding, as a general purpose clean amp.
Nevertheless, if it's supposed to emulate a generic amp, in the meaning of a generic TUBE guitar amp, judging from its frequency response in the high frequency, I'm humbly dubious on how (and if) the effects of grid stoppers resistors/non-zero impedance source stage feeding another following stage/miller capacitance have been modeled: in all the guitar amps, emulating such aspects of the real tubes, should lead to a subtle (or less subtle, in function of the circuit topology) hi-freq attenuation. Here I can see no such effect.
Still, this is a really good sounding amp sim! Bravo!

[Ichad.c]

Thank you for the detailed analysis, was a very interesting read.

[oxxyyd]

Most interesting read. Have you done any comparison of distortion (harmonic contents etc) at some levels, say half and full gain?

[alteregoxxx]

Most interesting read. Have you done any comparison of distortion (harmonic contents etc) at some levels, say half and full gain?

No, as I already said, this would require taking some time and thinking on matching levels of test signals in LTSPice vs VST world. It's doable for sure, but requires a little bit of thinking and test on my side. As soon I'll have some spare time maybe I'll give it a try

[DaveClark]

Hi alteregoxxx,

I have just started doing some simulations much like what you have done. More later.

Questions for you:

1) What were pot values in LTspice simulation for Volume, Bass, Mids, Treble?

2) What did you use to create white noise signal?


Answers and responses to your questions and comments:

Improper terminations: Wooden spoon, metal fork, car battery, wall socket (AC power), missing cabinet.

Proper terminations for guitar amp and sims: Same cabinet for all; matching cabinet for all.

There is a story in English about a patient going to the doctor. The patient says, “Whenever I bend my thumb way back like this (OUCH!), it hurts!” The doctor says, “Then stop doing that.”

There is a similar story about a guy on the KVR Audio forum who removed the cabinets from amp simulations then complained that the simulations don't behave correctly. Another guy says, “Then stop doing that.”

You assured me that the frequency response of the output of LTspice is 99% the same between pre-amp and output when using small signals and white noise. I assure you with 100% certainty that there are differences in simulations when you change the load and that you can hear them when you have higher gain and real DI guitar signals.

Output transformer for LG2Amp: For software amps, just like solid state amps, no output transformer is necessary. Not so “really strange” at all.

No attenuation of highs with cabinet turned off for LG2Amp: That's the way I want the amp to behave so that it can be used as a channel strip when the cabinet is disconnected. Really nice for acoustic guitar, also voice, piano, etc.

You posted, “Nope, that feedback is only...feedback. it's not memory because it happens instantaneously, the moment next to the one it happened it has no memory of what it did before...” AND: “Bias shift take time to move? Agree! Never said the contrary!” During active operation, bias shifting takes place primarily due to feedback (and this includes variable bias caused by “that feedback” under discussion) so you are contradicting yourself. Your error is that no physical feedback is instantaneous.

You said “Capacitors store DC values.” No, they don't. They store charge. The amount of charge depends on history: How much went in and how much went out.

Audible effect of capacitor memory: Easy examples are low-pass and high-pass filters and tonestack. Another is bucket-brigade delay pedal.

Memory and memoryless:

http://www.engr.iupui.edu/~skoskie/ECE3 ... ln_f06.pdf

Audibility of phase response – this was also meant to be entertainment as well as educational, so I hope you enjoy:

http://www.execonsultingtx.com/download ... ucational/


Regards,
Dave Clark

[DaveClark]

Hi alteregoxxx,

LTspice, TH3, A4, and LG2Amp comparisons were made. Some differences between what I did and what you did:

1. I included TH3 DarkFace cabinet in all simulations to provide proper load for each amp sim, including LTspice.

2. Used 0.1, 0.1, and 0.3 times the values of the bass, mids, and treble potentiometers, respectively, in both LTspice and LG2Amp because these have linear pots, not log A taper as with real, physical '65 Twin Reverb as shown in the reissue schematic. (It appears that you erroneously used 0.5 for LTspice and LG2Amp as well as probably correctly for A4 and TH3.)

3. Used a linear sweep, not white noise nor band-limited white noise.

4. Set the input signal level to -36 dBFS, not -12 dBFS, to obtain a response more representative of a linear circuit.

5. After obtaining initial responses, calibrated each amp simulation to LTspice using only the treble knob of each simulator so that the dominant low- and high-frequency peaks of the responses were in the same ratio.

6. Compared these calibrated simulations, including phase response.

7. Created clean simulations of DI guitar sample for each simulator using calibrated setup.

8. Created pre-distorted simulations (that is, using stomp box before amp sim) of DI guitar sample for each simulator using calibrated setup, in other words, distorted samples through clean amp sims.

9. Used very large FFT block sizes for analysis (262,144 samples and 441,000 samples) and rectangular windows with no overlap to obtain detailed information about spectra without averaging, smoothing, etc.


Findings:

1. TH3 required the most treble-knob tweaking to get the dominant peaks in the same ratio; A4 required significantly less; LG2 somewhat less than A4.

2. LTspice, A4, and LG2 were the most alike for the entire spectrum except for the extreme highs. TH3 matched LTspice falloff of the extreme highs best. It appears that something is off with the TH3 tonestack, something that did not appear to be adjustable through simple knob-setting changes.

3. After about 20 Hz, the phase response of A4 was virtually dead on with LTspice. Both TH3 and LG2 generally behaved the same, but the cycling through two pi radians was significantly quicker for both of these than for LTspice and A4. The phase difference between a fundamental and its harmonics will be the same for a struck guitar string, for example, for both A4 and LTspice. It will generally be something other than the LTspice and A4 value for both TH3 and LG2. For two different strings that are harmonically related, the timing won't be the same, therefore this phase difference discrepancy doesn't matter in cases like the latter.

4. LTspice, A4, and LG2 sounded very similar for the clean guitar sample played through them, in contrast to your claim that A4 is nothing like a '65 Twin Reverb because it contained a 20-dB discrepancy. TH3 sounds more like a Marshall to me, in contrast to your claim that TH3 was very close to LTspice simulation of the '65 Twin Reverb.

5. All four simulations sounded different but still quite similar for the pre-distorted guitar sample played through them. It's quite possible that these differences could be reduced by a different calibration procedure. The TH3 did still sound quite a bit more like a Marshall to me.


Plots:

1. Linear-sweep magnitude frequency response before calibration step (log Freq / dB):

Before Calibration


2. Linear-sweep magnitude frequency response after calibration step (log Freq / dB):

After Calibration


3. Detail of linear-sweep magnitude frequency response with linear frequency axis (with 10x magnification of values, that is, 5646 = 546.6 Hz):

Note that both LTspice and LG2 show small resonances while neither TH3 nor A4 do.

Magnitude Detail


4. Detail of linear-sweep phase frequency response with linear frequency axis (again with 10x, 1410= 141.0 Hz) and dB response axis (sorry about that, but I had to use my own analysis then display in wave editor program as WAV rather than FFT):

Early Phase Detail


Clean Tracks zip contains MP3 files with:

1. LTspice Clean.

2. TH3 Clean.

3. A4 Clean.

4. LG2 Amp Clean.

Clean Tracks


Pre-Distorted Tracks zip contains MP3 files of distortion through clean amp:

1. LTspice Pre-Distorted.

2. TH3 Pre-Distorted.

3. A4 Pre-Distorted.

4. LG2 Amp Pre-Distorted.

I would have preferred to use WAV files, but these are good enough to verify spectral analysis.

Pre-Distorted Tracks

After all this, I believe that I could more easily get A4 and LG2 to match LTspice for this particular '65 Twin Reverb amplifier than could get TH3 to match. In other words, with a bit more tweaking, calibration, knob-fiddling, I think I could achieve a closely matching sound with the first two.

Regards,
Dave Clark

[alteregoxxx]

Hi Dave,
I'll reply for the last time, because it seems all you want is to be right at all costs. You keep insisting I'm doing something wrong because I do not understand trivial things like loading effect, feedback, charge storage etc ect. I don't want to be rude, but man....I already took my degree, don't need to be examinated again, thanks.

1)Passing the output signal of LTSpice or any other software amp sim trough a third party cab sim, like you have done using TH3 cab sim, DO NOT REPRESENT A SIMULATION OF A REAL "LOADING EFFECT" on the amp sim circuit: the loading effect you keep insisting on, HAPPENS ONLY IF THERE IS "TALK-BACK", in the meaning of INTERACTION, BETWEEN THE LOAD AND THE PRECEDING CIRCUIT; THE CAB SIM WOULD NEED TO BE ABLE TO CHANGE, INFLUENCE, THE ELECTRICAL BEHAVIOUR OF THE PRECEDING CIRCUIT TO HAVE THAT KIND OF EFFECT, and this is clearly not the case when you use a third party cab sim. The cab sim or impulse response or any other third party plugin that you can use as plugin that follows the amp sim, is a black box that can only do one thing: transform the signal that its receiving at its input based on its own transfer function. The transfer functiomn this black box perform DOES NOT SEE AT ALL the "circuit" that feeds it: it could be a guitar circuit, a banana, a truck, this black box will perform its transfer function always the same way. Period. If you use two vastly different speaker impulse responses as "load" for the same amp sim, the signal that will reach their input will be ALWAYS THE SAME, and that is not what happens when loading real circuits with real loads. Period. You can for sure simulate the loading effect in LTSpice, BUT ONLY USING A CIRCUIT MODEL FOR THE SPEAKER IN LTSpice itself, not apllying a third party cab sim to the LTSpice simulation output wave file obtained performing the simulation without the cab sim in LTSPice itself. Is it clear now?

2)...it seems you don't know that in LTSpice you can simulate Audio, LOG, Inverse LOG potentiometers....I do...and in fact I used audio pot laws for the pots that were audio and linear for the linear...Please, stop supposing that I "erroneously" did this or that...please...
3)Linear sweep, white noise...you and I know that this really means nothing...you know, science is not racist...

4) As, I've already said more times, I've also tried levels up to -40dBfs ... please, read what I write ... in the case of a circuit like the one of a Twin Reverb, with great "clean" headroom ", there are insignificant differences

5) Calibrating?!? Are you kidding me?? You are tweaking pots values of the Amp A differently to that of an Amp B, with the goal of matching them and then to conclude that Amp A is a 1:1 emulation of the Amp B?? Are you serious?? If I start from my FIAT Punto diyng it in vivid red, change its weels, engine, bodywork, steer, maybe after a lot of tweaking it can become a Ferrari...So I can conclude from this my FIAT Punto is a 1:1 Ferrari emulation...right?

6)7)8)9)...nothing of that has influence on my findings...AT4 has the worst matching frequency response, TH3, Revalver 3, Digitech RP360 has better matching with LTSpice simulation...Period.

P.S. In the meantime I've also analyzed some of the Zoom G3xN amp sim models...well...unfortunately they do not match the LTSpice simulation not even remotely...