Unnatural sounding 'fizz' on sustained notes

[Cuiken]

I didn't want to get into this but here we go...

The plot below is a real Deluxe Reverb (green) vs. a QC (magenta) vs. an Axe-Fx III blue. This is a simple harmonic overtone test using a 1kHz sine wave (shifted up/down by one semi-tone to separate the traces) into the DUT.

If you look at the green trace (real amp) you'll see the overtones are strong all the way out to 20kHz. The Axe-Fx III's overtones are very similar in amplitude. The QC's overtones roll off starting before 10kHz. This makes the fizz less noticeable but it's less accurate.

As an aside you can see the significantly better SNR of the Axe-Fx III vs. the other two devices. Also note that there are no sidelobes on the QC at the fundamental frequency. This is due to the lack of power supply modeling.


View attachment 148580


If I were to guess I'd say the QC is using arctan or tanh to model the grid clipping. This results in smoother clipping with less overtones. We use a high-order function that more accurately models the forward-biased diode behavior of a grid.

'I didn't want to get into this but here we go...'

I love that you got into this. Especially that you took the time on Thanksgiving! Thanks.

The plots are fascinating. I took loads of similar onse when trying (and ultimately giving up on) building a JFET replica of a blues deluxe preamp stage. Amazing how close you/Fractal have got.

Only curious why a 1kHz tone was used vs something more representative of the fundamentals on a guitar? Forgive my question, just geeking out.

 

[Cuiken]

That may or may not help. There is a parameter to adjust how hard the grids are driven but it is not exposed to the user (FWIW that parameter is nearly at maximum for the Diamante Fire model. That amp absolutely obliterates the grids). There's also hidden code to use softer clipping models but, again, not exposed to the user.

Power tube grid clipping is nasty stuff but it helps cut through the mix. When listening in isolation using IRs, especially IRs made with dynamic mics (i.e. SM57), that fizz can be quite noticeable. As you can see from the plots the fizz is in the upper treble region.

There are several things you can try:
Use a different IR that doesn't accentuate this region.
Turn down the High-Cut Frequency on the IR.
Reduce MV.
Increase Power Tube Grid Bias.
Increase Transformer Matching and/or Speaker Impedance so that the plates clip before the grids (plate clipping is smoother).
Use a different amp model that relies more on preamp distortion.

I've contemplated for years a parameter that allows switching between "Authentic" and "Smooth" power tube clipping because this topic rears its ugly head now and then. But in the end I always stick to the "accuracy is paramount" philosophy.

Okay, this was transformative. From the list above, I had not tried changing the Power Tube Grid Bias. To my ear, this offered the biggest change in the distortion artefacts that I am hearing. It's not that they are dramatically reduced but more that they become less dissasociated with the harmonic guitar tones. This stuff is hugely subjective but I found this improved most of the models I had been struggling with and is exactly the guidance I needed. Thanks.

'I've contemplated for years a parameter that allows switching between "Authentic" and "Smooth" power tube clipping because this topic rears its ugly head now and then. But in the end I always stick to the "accuracy is paramount" philosophy.'

Personally, I'd love to see this feature. It's not that I disagree with the premise that "accuracy is paramount". I'd likely leave it set to 'accurate' for any actual live playing for all the reasons highlighted. It's just that a key feature of modellers for me is that they allow home practice at lower volumes or headphones. Being able to 'temper' some of the models characteristics for that use case would, to me, be really helpful.

Either way, the grid bias adjust is really helping me out already so, thanks for that tip.

 

[Dave Merrill]

Okay, this was transformative. From the list above, I had not tried changing the Power Tube Grid Bias. To my ear, this offered the biggest change in the distortion artefacts that I am hearing. It's not that they are dramatically reduced but more that they become less dissasociated with the harmonic guitar tones. This stuff is hugely subjective but I found this improved most of the models I had been struggling with and is exactly the guidance I needed. Thanks.

'I've contemplated for years a parameter that allows switching between "Authentic" and "Smooth" power tube clipping because this topic rears its ugly head now and then. But in the end I always stick to the "accuracy is paramount" philosophy.'

Personally, I'd love to see this feature. It's not that I disagree with the premise that "accuracy is paramount". I'd likely leave it set to 'accurate' for any actual live playing for all the reasons highlighted. It's just that a key feature of modellers for me is that they allow home practice at lower volumes or headphones. Being able to 'temper' some of the models characteristics for that use case would, to me, be really helpful.

Either way, the grid bias adjust is really helping me out already so, thanks for that tip.

Interesting.

Are you changing Power Tube Grid Bias by a lot or a little? Care to post a preset with before and after scenes, so we can hear the effect?

 

[Cuiken]

Sure, try the Deluxe Verb Preset and enable the Compressor (the one already selected in the Preset). Crank the gain and reduce the Master and Level to compensate. For me, this gets a bit 'Fizzy' (please forgive the term, just a short hand). As Cliff says, the high freq artefacts aren't super loud but they are (accurately) 'dissaociated'. At lower volumes through an FRFR or headphones, they seem to stand out to me. YMMV.







All I've then done is crank the Grid Bias as below and, to my ear, the tone gets a bit more 'natural' at lower room/headphone listening levels:


I'll no doubt end up messing around with all the other suggestions (MV, IR choice, Cab rol off) but the Grid Bias setting seems pretty useful to me as a starting point.

For this particular preset, if you really want to adopt a zero tolerance approach, turning the Power Amp Modelling off gets rid any last vestiges of 'fizz'. Goes a bit far for me though:

 

[Smilzo]

Crank the gain and reduce the Master and Level to compensate. For me, this gets a bit 'Fizzy'

Crancking the gain means lots of nasty clipping on 2nd stage, C.follower and lastly poweramp tubes. Lower the gain till the distortion is pleasant to your ears... any guitar tube amp is a mix and blend of different distortions, in each stage.

 

[Rex]

Only curious why a 1kHz tone was used vs something more representative of the fundamentals on a guitar? Forgive my question, just geeking out.

1 KHz is within the range of a guitar’s fundamental frequencies.

 

[Cuiken]

1 KHz is within the range of a guitar’s fundamental frequencies.

Sure, but that’d be around the 20th fret on the high E string. For me at least, I don’t spend much time up there.

I’ve always used ~200Hz test tones for FFTs of guitar amps as that freq sits comfortably ‘mid neck’ and places the fundamental and more of the harmonics within the bandwidth of the negative feedback of the amp. Just curious why the 1kHz choice was made?

I guess it could be an interesting frequency with which to explore intermod of high order harmonics or possibly just a resolution issue with the true RTA analyser?

 

[skolacki]

Sure, but that’d be around the 20th fret on the high E string. For me at least, I don’t spend much time up there.

I’ve always used ~200Hz test tones for FFTs of guitar amps as that freq sits comfortably ‘mid neck’ and places the fundamental and more of the harmonics within the bandwidth of the negative feedback of the amp. Just curious why the 1kHz choice was made?

I guess it could be an interesting frequency with which to explore intermod of high order harmonics or possibly just a resolution issue with the true RTA analyser?

1khz is in the guitar range as pointed out. I used a 1khz oscillator for testing in my amp building days. Cliff just grabbed a frequency to illustrate the difference between the real amp, the Axe, and the QC. It’s valid. Not sure what your point is. Nothing wrong with using 1khz.

 

[Cuiken]

I guess my question might just be whether the gain bandwidth of the amplifier is wide enough for the distortion characteristics of a 1kHz test tone to be representative of those at lower frequencies? Like I say, just geeking out on the engineering really.

 

[Rex]

I guess my question might just be whether the gain bandwidth of the amplifier is wide enough for the distortion characteristics of a 1kHz test tone to be representative of those at lower frequencies? Like I say, just geeking out on the engineering really.

If a device has a large enough gain-bandwidth product to do a thing at 1 KHz, it has more than enough gain-bandwidth product to do that thing at a lower frequency.

 

[PatrickWD]

Sure, but that’d be around the 20th fret on the high E string. For me at least, I don’t spend much time up there.

I’ve always used ~200Hz test tones for FFTs of guitar amps as that freq sits comfortably ‘mid neck’ and places the fundamental and more of the harmonics within the bandwidth of the negative feedback of the amp. Just curious why the 1kHz choice was made?

I guess it could be an interesting frequency with which to explore intermod of high order harmonics or possibly just a resolution issue with the true RTA analyser?

Wouldn't the 20th fret be closer to 2khz?

1khz would be closer to the 8th fret on the high e string I believe.

 

[Rex]

Wouldn't the 20th fret be closer to 2khz?

The fundamental frequency of a note played on the 1st string at the 24th fret on a concert-tuned guitar is 1320 Hz. That's the highest fundamental than any guitar can achieve unless it has more than 24 frets (which is almost none of them ).

 

[PatrickWD]

The fundamental frequency of a note played on the 1st string at the 24th fret on a concert-tuned guitar is 1320 Hz. That's the highest fundamental than any guitar can achieve unless it has more than 24 frets (which is almost none of them ).

Seems like I was an octave off. I thought that a 440 A was on the second fret of the G string, but it appears that it is the fifth fret of the E string.

 

[Cuiken]

If a device has a large enough gain-bandwidth product to do a thing at 1 KHz, it has more than enough gain-bandwidth product to do that thing at a lower frequency.

Sorry if my point was unclear. I was actually making the opposite observation.

I was wondering whether the reducing open loop gain of the power amp, as frequency increases, would mean the amp has less ability to surpress distortion/harmonics as the fundamental note increases in frequency.

So, at 200Hz, for example, the original amp would have more open loop gain and so generat less distortion/harmonics than if tested with a 1kHz tone. I only mention it out of curiosity and because this was a behaviour demostrated in some of the circuits I built.

Perhaps the Fractal already includes estimates of gain bandwidth and so already accounts for this effect? Or, maybe the gain bandwidth of the power amp in guitar amps is wide enough for this not to be a factor?

Just curious about the plots presented and what can definitely be gleened from them.

 

[Rex]

Sorry if my point was unclear. I was actually making the opposite observation.

I was wondering whether the reducing open loop gain of the power amp, as frequency increases, would mean the amp has less ability to surpress distortion/harmonics as the fundamental note increases in frequency.

So, at 200Hz, for example, the original amp would have more open loop gain and so generat less distortion/harmonics than if tested with a 1kHz tone. I only mention it out of curiosity and because this was a behaviour demostrated in some of the circuits I built.

Perhaps the Fractal already includes estimates of gain bandwidth and so already accounts for this effect? Or, maybe the gain bandwidth of the power amp in guitar amps is wide enough for this not to be a factor?

Just curious about the plots presented and what can definitely be gleened from them.

I think you’ll find that most audio amps, if they’re dialed in flat, have pretty much the same gain at 200 Hz as they do at 1 KHz. If they don’t, it’s because somebody designed the EQ that way.

 

[Dave Merrill]

I think you’ll find that most audio amps, if they’re dialed in flat, have pretty much the same gain at 200 Hz as they do at 1 KHz. If they don’t, it’s because somebody designed the EQ that way.

Open loop gain isn't the same as "gain", meaning closed loop gain of the circuit as a whole, with corrective negative feedback applied.

I've only heard that term used for opamps, where decreasing open loop gain at higher frequencies can be a problem in components that aren't designed to optimize sufficiently for it.

I'm unclear if it even applies here at all.

If the original point was about the validity of distortion measurements made at 1kHz when you're talking about distorted guitar signals, I'd take Mr Chase at his word that the specs he measured are reasonable ones to compare in this context

 

[Rex]

I'm unclear if it even applies here at all.

That's my point. It doesn't apply here. Open- and closed-loop gain apply to any amplifier, but @Cuiken is approaching this as a gain-bandwidth product issue, which is pretty much an op-amp concern.

 

[Smilzo]

In electronics, gain is the increase in amplitude of the output signal (a ratio like 10;1, 100:1). The flatter the gain bandwidth, the better the component.
In guitar amp, gain is the increase in distortion of the output signal (there's no gain increase in a tube amp, only increase of the input level before a stage)! The "squarer" the gain bandwidth, the higher the distortion! Every parameter can alter different band gain. Measuring the freq response of a tube amp is very complex, input level depending and evolving in freq and time domain. I think Cliff gave us an example, we don't have to worry much if it's 1kHz or 2...

 

[Wolfenstein98k]

That may or may not help. There is a parameter to adjust how hard the grids are driven but it is not exposed to the user (FWIW that parameter is nearly at maximum for the Diamante Fire model. That amp absolutely obliterates the grids). There's also hidden code to use softer clipping models but, again, not exposed to the user.

Power tube grid clipping is nasty stuff but it helps cut through the mix. When listening in isolation using IRs, especially IRs made with dynamic mics (i.e. SM57), that fizz can be quite noticeable. As you can see from the plots the fizz is in the upper treble region.

There are several things you can try:
Use a different IR that doesn't accentuate this region.
Turn down the High-Cut Frequency on the IR.
Reduce MV.
Increase Power Tube Grid Bias.
Increase Transformer Matching and/or Speaker Impedance so that the plates clip before the grids (plate clipping is smoother).
Use a different amp model that relies more on preamp distortion.

I've contemplated for years a parameter that allows switching between "Authentic" and "Smooth" power tube clipping because this topic rears its ugly head now and then. But in the end I always stick to the "accuracy is paramount" philosophy.

I really think giving those options to the user is not a bad idea. Since you have the capability, it's not a case of "buy the other product if you want a different sound" - this is purely a choice!

Perhaps labelling them more about context might help - e.g. "Mix" vs "Solo" or "Authentic (Mix)" vs "Smooth (Solo)" or something.

That's how I think of these authentic artefacts... everyone has heard the solo stems from famous albums, like EVH, and noticed how trebly and spitty they are. But you don't hear that at all in the mix.

 

[Rex]

In electronics, gain is the increase in amplitude of the output signal (a ratio like 10;1, 100:1). The flatter the gain bandwidth, the better the component.
In guitar amp, gain is the increase in distortion of the output signal (there's no gain increase in a tube amp, only increase of the input level before a stage)! The "squarer" the gain bandwidth, the higher the distortion! Every parameter can alter different band gain. Measuring the freq response of a tube amp is very complex, input level depending and evolving in freq and time domain. I think Cliff gave us an example, we don't have to worry much if it's 1kHz or 2...

And there's one of my pet peeves. Guitarists use the word gain as if it means distortion. But increasing gain is only one way to increase distortion.