Ooooh... Charts and Graphs

Status
Not open for further replies.
Hmmm...this strikes me as a pretty flawed "study", because it glosses over some hugely important facts.

- Where's the control? I mean...you're right that analogue amps don't alias, but they *do* produce noise outside the the desired response to the input signal (background hiss, hum etc), especially at high gain settings, and it's most definitely audible.

- What were the settings for the two amp models (ie was it apples-to-apples)? Were they functionally identical (ie sound the same), nominally identical (the same number on the dial) or neither? This is relevant, of course, because TrueRTA doesn't just measure aliasing - as evidenced by the presence of the ground loop hum in both of the Fractal units' response graphs.

- The setup was clearly not apples-to-apples for all units, since both of the Fractal units appear to have a ground loop. Unless we're supposed to draw the conclusion that Fractal gear is more susceptible to ground loops than the competitor in question...?

- The threshold of hearing is not constant for all human beings. Women have a significantly lower threshold than men, and the threshold can increase by up to 60dB with age.

- The threshold of hearing is somewhat variable even when controlling for a single human being...for example, where it's around 60dB for a 30Hz tone (young folk in their early 20s - I picked these because they're the ones who'd be able to hear that "important" 18kHz spike...unlike most residents of musicians' forums, myself included). Up in the 10kHz range it's closer to 18dB, which means all three units should have audible aliasing. The reason they don't is that tone duration makes a massive difference (the shorter the duration, the higher the threshold), so sweeping the input wave muddies the results significantly.
 
Last edited:
Up in the 10kHz range it's closer to 18dB, which means all three units should have audible aliasing.

How do you even know that you see aliasing? In the case of the III, there’s broadband noise across the whole spectrum with a 180 Hz spike. It looks like “regular” noise with some gain applied.
 
How do you even know that you see aliasing? In the case of the III, there’s broadband noise across the whole spectrum with a 180 Hz spike. It looks like “regular” noise with some gain applied.

True enough - in that statement, I was just taking Cliff's assumption (ie that it's all aliasing, apart from the bits that he says aren't) and running with it....
 
...you're right that analogue amps don't alias, but they *do* produce noise outside the the desired response to the input signal (background hiss, hum etc), especially at high gain settings, and it's most definitely audible.
The Axe's background hiss is 90 dB below 0 dBu — well below the plot. Hum is present only in multiples of 60 Hz.


What were the settings for the two amp models?
Don't know.


The setup was clearly not apples-to-apples for all units, since both of the Fractal units appear to have a ground loop. Unless we're supposed to draw the conclusion that Fractal gear is more susceptible to ground loops than the competitor in question...?
Ground loops are a function of equipment interconnect, not equipment internals.


The threshold of hearing is not constant for all human beings.
Irrelevant to the test.


The threshold of hearing is somewhat variable even when controlling for a single human being...Up in the 10kHz range it's closer to 18dB, which means all three units should have audible aliasing.
You can't compare dB SPL to dBu. They're different things. SPL is dependent on many other things (power amplification, loudspeaker, room...)


...tone duration makes a massive difference (the shorter the duration, the higher the threshold), so sweeping the input wave muddies the results significantly.
Duration and frequency sweep are different things. Sweeping the input clarifies the results. Without sweeping, you would wind up with individual aliasing spikes that are too narrow to show up on the graph.
 
- What were the settings for the two amp models (ie was it apples-to-apples)? Were they functionally identical (ie sound the same), nominally identical (the same number on the dial) or neither? This is relevant, of course, because TrueRTA doesn't just measure aliasing - as evidenced by the presence of the ground loop hum in both of the Fractal units' response graphs.
I'd think the distortion settings are fairly comparable, given that 1)10 kHz waveform looks same and 2)second harmonic level is similar (actually 8 dB greater on AF III).

But I'm confused too about how to distinguish aliasing noise from other types of noise from the presented measurement graphs.

Unless we're supposed to draw the conclusion that Fractal gear is more susceptible to ground loops than the competitor in question...?
Confused about this one too.

The threshold of hearing is... around 60dB for a 30Hz tone. Up in the 10kHz range it's closer to 18dB, which means all three units should have audible aliasing.
Cliff used the term "threshold of audibility," which is perhaps a bad word choice. But his point still stands, at around 60 dB signal to noise ratio, you can't hear the noise.

The numbers you posted is of absolute threshold of hearing, which is defined "minimum audible SPL at a given frequency without any other sound present." It's not relevant at all considering the context. Rather, think relative threshold of hearing if you will.

If you want to hear it, a test at minute 32:
 
Last edited:
  • Like
Reactions: Rex
Irrelevant to the test.

I'm just going to pick on this one because a) I'm a little pushed for time, and b) it's the most relevant to everything else I said.

TrueRTA isn't just measuring aliasing - there is nothing done here which isolates the effect of the aliasing, which is my entire point. Those graphs, therefore, aren't relevant to the point he's trying to make - as evidenced by the presence of the ground loop signal. And, because we don't have a control signal - the amp that the Axe units and the "other" unit are trying to emulate, at comparable settings - we don't know whether the increased noise is aliasing or a faithful reproduction of the noise that the original amp would've made in the first place. The 2kHz and 4kHz spikes on the green plot, for example, strike me as exactly the kind of thing you'd expect from background noise in a real amp.

I'm not saying that the conclusion is wrong, I'm just saying that the information provided is wholly insufficient to support it.

EDIT: OK, I can play a little more...

The Axe's background hiss is 90 dB below 0 dBu — well below the plot. Hum is present only in multiples of 60 Hz.

Mains hum, yes (depending on where you are, of course).

Ground loops are a function of equipment interconnect, not equipment internals.

As above, nothing has been done to isolate this test to the equipment internals. My point stands - because there is no control, there's no way to know what other external factors are in effect here, given that the test environment for each unit was different.

You can't compare dB SPL to dBu. They're different things. SPL is dependent on many other things (power amplification, loudspeaker, room...)

Yep, OK.

Duration and frequency sweep are different things. Sweeping the input clarifies the results. Without sweeping, you would wind up with individual aliasing spikes that are too narrow to show up on the graph.

I'd say that individual aliasing spikes at different frequencies would be more beneficial to the test, because it'd give a much clearer picture.

In fact, what would've been really beneficial to the test is using an input signal which is representative of real-world usage given that the fundamental frequency range of a guitar sits in the 80Hz - 1.2kHz region; the ability of modellers to accurately respond to fundamentals an order of magnitude above any kind of real-world usage is completely irrelevant to any expected use case, wouldn't you say?
 
Last edited:
TrueRTA isn't just measuring aliasing - there is nothing done here which isolates the effect of the aliasing, which is my entire point.
You're right: TrueRTA measures everything it sees, not just aliasing. The test signal contains only a single frequency at any given moment, so we know there will be no intermodulation distortion. So it's going to measure the test signal, harmonic distortion of the test signal, the self-noise of the system, any ground loop present, and aliasing.

We can see the test signal — it's the most dominant thing on the plot. The frequency of the test signal is high enough that we know exactly where to look for harmonics, and we can see them clearly in the plot. We also know exactly where the ground loop will show up, and that's where we find it on the plot. The noise of current fractal gear is known to be at least 90 dB below 0 dBu. That places any self-noise so low that it is literally off the chart completely. What's left is aliasing.


I'd say that individual aliasing spikes at different frequencies would be more beneficial to the test, because it'd give a much clearer picture.
They're not of much help if they're too narrow to show up on the plot. By sweeping the test signal through a narrow band, you make the aliasing artifacts wide enough to be visible.


In fact, what would've been really beneficial to the test is using an input signal which is representative of real-world usage given that the fundamental frequency range of a guitar sits in the 80Hz - 1.2kHz region; the ability of modellers to accurately respond to fundamentals an order of magnitude above any kind of real-world usage is completely irrelevant to any expected use case, wouldn't you say?
It's not irrelevant at all. Sure, the highest fundamental of most guitars is 1.2 KHz. But that high-E note also produces harmonics at 2.4 KHz, 4.8 KHz, 9.6 KHz and beyond. And that's just the guitar, before the signal even hits the amp. And amplifier distortion generates more harmonics on top of that. The entire audio spectrum is relevant here.
 
You're right: TrueRTA measures everything it sees, not just aliasing. The test signal contains only a single frequency at any given moment, so we know there will be no intermodulation distortion. So it's going to measure the test signal, harmonic distortion of the test signal, the self-noise of the system, any ground loop present, and aliasing.

We can see the test signal — it's the most dominant thing on the plot. The frequency of the test signal is high enough that we know exactly where to look for harmonics, and we can see them clearly in the plot. We also know exactly where the ground loop will show up, and that's where we find it on the plot. The noise of current fractal gear is known to be at least 90 dB below 0 dBu. That places any self-noise so low that it is literally off the chart completely. What's left is aliasing.

As I said before - we have no control. If the noise in the "other" plot is actually representative of the real amp being modelled...well, we're down to an "accuracy vs improvement over the original" argument. The key word is "if", because we have no way of knowing and thus the provided info proves nothing on its own without further supposition and guesswork.

They're not of much help if they're too narrow to show up on the plot. By sweeping the test signal through a narrow band, you make the aliasing artifacts wide enough to be visible.

In that case, the testing tools are at fault for lack of resolution. Still makes it insufficient, IMO.

It's not irrelevant at all. Sure, the highest fundamental of most guitars is 1.2 KHz. But that high-E note also produces harmonics at 2.4 KHz, 4.8 KHz, 9.6 KHz and beyond. And that's just the guitar, before the signal even hits the amp. And amplifier distortion generates more harmonics on top of that. The entire audio spectrum is relevant here.

If what you're saying is true, then a representative input signal would be far more useful - say, a 1.2kHz fundamental overlaid with 2.4kHz/4.8kHz/9.6kHz at progressively lower amplitudes.
 
I'd think the distortion settings are fairly comparable, given that 1)10 kHz waveform looks same and 2)second harmonic level is similar (actually 8 dB greater on AF III).

But I'm confused too about how to distinguish aliasing noise from other types of noise from the presented measurement graphs.


Confused about this one too.


Cliff used the term "threshold of audibility," which is perhaps a bad word choice. But his point still stands, at around 60 dB signal to noise ratio, you can't hear the noise.

The numbers you posted is of absolute threshold of hearing, which is defined "minimum audible SPL at a given frequency without any other sound present." It's not relevant at all considering the context. Rather, think relative threshold of hearing if you will.

If you want to hear it, a test at minute 32:

In the video, I found "the nastiest noise he could muster" less annoying than the music it was mixed into!. lol
 
The noise of current fractal gear is known to be at least 90 dB below 0 dBu. That places any self-noise so low that it is literally off the chart completely. What's left is aliasing.

-90 dB wouldn’t be off the chart here. And if everything above -90 dB is aliasing, it seems like way too much of it. I just don’t see how it can be so broadband and so uniform for a relatively narrow sweep.
 
If what you're saying is true, then a representative input signal would be far more useful - say, a 1.2kHz fundamental overlaid with 2.4kHz/4.8kHz/9.6kHz at progressively lower amplitudes.

But wouldn't it also be true that, with harmonic content getting progressively lower, aliasing will be less prominent as well? This test is done for frequencies high enough to have a lot of energy in the relevant harmonics but still relevant for guitar, isn't it?
 
As I said before - we have no control. If the noise in the "other" plot is actually representative of the real amp being modelled...well, we're down to an "accuracy vs improvement over the original" argument. The key word is "if", because we have no way of knowing and thus the provided info proves nothing on its own without further supposition and guesswork.

There's not a modeler in existence that adds noise to match the noise of the amp being simulated. Noise is undesirable regardless of "realism". Hum and ghost notes are sometimes modeled but thermal noise never is primarily because the amount of thermal noise is partially dependent upon the resistance of the source (guitar pickup) which is a variable.

Furthermore any modeler exhibiting that much thermal noise would be roundly criticized for being excessively noisy.
 
The noise of current fractal gear is known to be at least 90 dB below 0 dBu. That places any self-noise so low that it is literally off the chart completely. What's left is aliasing.
Wouldn't the compression/distortion of the amp model in the signal chain bump up the base-noise level significantly? I don't expect the base-noise level to stay a constant 90 dB below 0 dBu, it's going to be a variable and rise with the amount of compression/distortion. So it seems to me that the graphs show the base-noise + aliasing-noise performance, and not enough info to separate what is base-noise from aliasing-noise.
As in, I don't know whether the higher noise from the green plot is entirely due to aliasing and a lack of oversampling. What if its base-noise floor was higher to begin with?

Either way, Axe-Fx III shows better performance and I'm happy about that. But I'm having trouble trying to decipher what is aliasing-noise and what is other noise from the given measurements (maybe that's for my lack of knowing what to look for, I'd be happy to receive input).
 
The scale of the plot goes down to -80 dBu. -90 would be literally off the chart.
Agreed.
(But again, within context, the -90 dB base-noise level would rise with the amp model in the signal chain, no?)
 
Status
Not open for further replies.
Back
Top Bottom