Express Output Waveform Shots

[matt h]

(deleted)

[jazbo8]

the "volume above two" issue is a dead-end assuming one is not feeding fixed and infinitely sustained-as-fixed signal voltages. just because the knob of an amp is set to "2" doesn't mean every note coming out of it as "at 2"... *sigh*

Your point is well taken, volume at 2 has nothing to do with how much signal is coming into the amp, but Craig used nominal signal level (albeit continuous) to simulate a typical guiitar pickup's output, and as mentioned in my last post, most people leave the volume setting well pass "2", so the interesting waveforms are volume 5-8, am I still off my rocker?

[matt h]

(deleted)

[katopan]

The interesting thing about relative levels is how much output does the guitar put out with the volume rolled back to where we typically use it for a clean tone when playing this amp. OK, I've added that to the list for the next testing session along with PI tail resistor changes and B+ sag.

[cxx]

I've been following this with keen interest and thanks for all this work and massive data.

I'm not sure I understand how the NFB loop is defeated as distortion increases. I can see that it does its job as the power tubes start to distort by adding the negative signal back into the PI. What happens at higher levels to make it ineffective?

Also, I'm not sure that past a certain amount of squareness in the wave form that any more is at all desirable. A square wave never sounds that interesting. It would be nice to see what the wave form looks like at the best sounding distorted levels. I guess it is a matter of taste but I'm sure we could agree that Glen's mean tone still has character and is plenty mean.

[Tillydog]

...I'm not sure I understand how the NFB loop is defeated as distortion increases. I can see that it does its job as the power tubes start to distort by adding the negative signal back into the PI. What happens at higher levels to make it ineffective?...

I wouldn't say it was defeated, as such, more that it spirals out of control...

Once the power stage grids are driven to conduction (the start of serious distortion), the speaker output waveform is missing the peaks, so this part of the signal is missing from the NFB signal also. The peaks are still present in the signal coming into the PI from the preamp. Since the PI amplifies the difference between these two, the output of the PI increases markedly (due to spikes on the negative half of the PI output- where the peaks are missing in the output signal - pushing the average output up), which, of course isn't reflected in the NFB signal (due to being clamped by power stage grid conduction). This affects the response of the PI, and makes it behave as if there is no NFB. It may result in the PI hitting saturation shortly after the power stage distorts if the PI bias and supply voltages are suitably chosen. (IMHO, *that* is the magic in the Express - YMMV )

[katopan]

Thanks cxx! Tillydog has covered it. Just to add my own words about the same thing, once the power grids are clipping the feedback can't get any bigger and so stops having its normal effect.

In the fairly wide sweet spot, the duty cycle and mixed mode distortion mean that this amp's output is anything but just a square wave. It does finally transition to a square wave when the 3rd stage reaches full cut off. The volume setting this amp is normally used at doesn't reach that point with a real dynamic guitar signal (maybe just at the absolute peak of attack), which I haven't posted but have observed with my CRO. If you crank it and either have hot pickups or a booster you can hear what it sounds like, and the tone does get a bit like other amps that become square wave generators when very heavily overdriven. The shots I've posted with the volume around 6.5 for a 50mV test signal (anywhere either side of 5 depending on the pickups for a real guitar being played) where the most asymmetry and mixed mode distortion is produced does line up with what my ears tell me is the best mean tone of this amp.

It may result in the PI hitting saturation shortly after the power stage distorts if the PI bias and supply voltages are suitably chosen.

That certainly happens! The PI hitting the point of clipping (itself, not power grid clipping of the PI plate signals) happens very soon after power grid clipping.

[cxx]

...I'm not sure I understand how the NFB loop is defeated as distortion increases. I can see that it does its job as the power tubes start to distort by adding the negative signal back into the PI. What happens at higher levels to make it ineffective?...

I wouldn't say it was defeated, as such, more that it spirals out of control...

Once the power stage grids are driven to conduction (the start of serious distortion), the speaker output waveform is missing the peaks, so this part of the signal is missing from the NFB signal also. The peaks are still present in the signal coming into the PI from the preamp. Since the PI amplifies the difference between these two, the output of the PI increases markedly (due to spikes on the negative half of the PI output- where the peaks are missing in the output signal - pushing the average output up), which, of course isn't reflected in the NFB signal (due to being clamped by power stage grid conduction). This affects the response of the PI, and makes it behave as if there is no NFB. It may result in the PI hitting saturation shortly after the power stage distorts if the PI bias and supply voltages are suitably chosen. (IMHO, *that* is the magic in the Express - YMMV )

Seems like in this description the magic depends on power tube grid conduction. Doesn't that require an input that exceeds the bias on the positive side of the signal? Where are we at with signal levels at the different stages of the amp when that happens? Might we already be in buzzsaw territory with input to the power tubes being twice the bias voltage peak to peak or thereabouts?. I don't remember there being any evidence in the scope plots that show this.

I know it is difficult to get in there between the PI and the power tubes and measure because the signal is split into two sides at this point and is only recombined in the transformer. But this is the area that is the leading candidate.

The other thing is that a typical setup has the amp volume at a bit more than half and the guitar volume between very little (clean) and enough (mean). I wish that we knew what this represents in signal levels, otherwise we might attribute the phenomenon to something that happens way out on the extremes rather than the levels where it does happen.

[katopan]

The amp stays in AB1 and doesn't drive hard enough to go into AB2. This is shown on the power grid waveforms by the flat grid clipping when the signal reaches the 0V line. It doesn't go above 0V.

The waveforms I've published show every node of the amp all the way through, for the full range of the volume knob with a 50mVpeak input signal. You can happily see what is happening at every point in the amp for when power grid conduction comes in (or any other point before and after). Just not sure I understand what else you're looking for?

Your last paragraph has a great point, and is something I haven't measured. Can't remember if this came up before (*checks* ahh, yes it was added to the list in an earlier post at the top of this page). So now I have remaining tests as: behaviour with larger tail resistor values, B+ sag with varying signal, I've had a PM request to include measuring the HT line current with varying signal, and clean to mean input level from an actual guitar with the amp at typical played volume setting.

I have looked at the output with my CRO while just playing the amp, chords as well as single note soloing. What you see is exactly the behaviour shown in these sinewave test waveforms, but applied to the complex input of a real guitar. I've also observed what's coming from the guitar itself in different pickup selections and different tone knob positions. The magic area of this amp is centred around what corresponds to Volume 6.5 in my posted waveforms, where duty cycle shift and mixed mode distortion are both at their maximum. Play the guitar and watch the CRO screen, and you see that same manipulation of the signal where just after clean has the very tops clipped evenly, and then it transitions to shift the duty cycle and cut off the wider bottom trailing corner. Push it really hard and everything squares up again and gets fizzy. You still see it all even with a harmonically complex and very dynamic guitar signal. You can also capture the output of the amp by recording into a DAW, and then zoom in on sections of the waveform to witness all the same stuff.

[jazbo8]

There are so many concepts being thrown together again So it might help to remove some again...

1. The amp operates in AB1, so forget about power tube grid conduction.
2. when the output clips - the tubes are driven to cut-off - there is no (extra) gain left to correct the error signal, so the NFB is basically out of the circuit (collapsed - that word again), see p.182 of Merlin's book for the details.

Obviously, condition 2 represents the static case with sinewave input, as in Craig CRO shots. Under dynamic condition, when the input level drops, the output tubes become un-clipped then the NFB loop gets restored. So instead of sharpening the transition from clean to overdriven when the NFB is applied in typical amplifiers, the TWE in fact does the opposite - the transition from clean to overdriven is widened, because its NFB is "out of the loop" so much, that it pretty much runs at the full open loop gain all the time. The TWE is never "clean", ever. It's just more or less distortion depending on how the guitar is played and/or where the guitar volume is set, i.e., completely dependent on the player.

[cxx]

I can understand the defeat of the NFB loop if the PI is clipping, or out of gain, as it is the one to combine the input from stage 3 and the negative reduced output from the power tubes. Distortions in the PI plate waveforms before clipping should reflect the signal corrections needed to produce power tube waveforms that more closely reflect the PI input.

I don't readily see how this transition to distortion in the feedback loop takes place. I don't have Merlin's book unfortunately. Is it addressed there? Is there a scope plot here that I've missed?

It does seem that in the express schematics that I have, and I think I got them here, the PI and the rest of the power section is a direct lift of the Marshall 1987, with a different transformer.

[Jackie Treehorn]

Just wanted to chime in and express some gratitude to those who have contributed. It's a very dense thread and I haven't kept up very well, but wanted to add some encouragement.

I think with the grid conduction terminology we were just describing the initial clip.

One of my ongoing projects has been a Class AB2 Express based on KT88's which are good candidates/victims for AB2. So, I've spent a lot of time hearing this type of amp without the diode clamping of AB1. In that situation, the initial clip is the power tube plate, and one must configure the driver so there isn't too much signal swing. Anyway, I should be able to get some traces if I ever finish that one.

Thanks, again, great thread.

[Tillydog]

Seems like in this description the magic depends on power tube grid conduction. Doesn't that require an input that exceeds the bias on the positive side of the signal?

Yes, that's what *I* mean (there are several different discussions crossing! ).

Where are we at with signal levels at the different stages of the amp when that happens?

IME, everything else is clean at this point, with a little headroom left in the PI and loads of headroom before that. YMMV

[Tillydog]

2. when the output clips - the tubes are driven to cut-off - there is no (extra) gain left to correct the error signal, so the NFB is basically out of the circuit

This isn't my experience, but yours may differ.

IME the output valves(tubes) clip first due to grid conduction (not cut-off). At this point, the PI still has headroom left. The PI output therefore increases (due to the increased error between input signal and NFB due to NFB being pretty much clamped by the power valve grid conduction - as described in Merlin's book). The PI output will go as high as the B+ rail / biasing allows it - again, IMHO these are key choices in the TWE (which is why I expressed (small 'E'!) concern at the amount of sag on the PI B+ in interpreting the scope shots).

So, IME, the NFB loop does collapse / spiral out of control, or however you want to describe it, but this has consequences and produces a discontinuity in the response of the amp - a tiny change in input produces a very marked change in output. The NFB certainly doesn't just stop or 'disappear' gracefully.

Again my experience, but my Express does a lovely clean.

Just relating *my* experiences - they may or may not tally with those of others, but that doesn't make anyone wrong

[katopan]

Distortions in the PI plate waveforms before clipping should reflect the signal corrections needed to produce power tube waveforms that more closely reflect the PI input.

That could be the case if the PI had the grunt to drive the power grids into AB2. But they hit a ceiling and can't go any further. That effectively breaks the feedback loop before the PI itself clips, and the feedback return signal can't get any bigger.

Also remember in all this that from a feedback/control theory point of view, the open loop gain PI input to speaker output is relatively low, and the amount of feedback is relatively small. So the amount of correction provided by the feedback is only partial.

It does seem that in the express schematics that I have, and I think I got them here, the PI and the rest of the power section is a direct lift of the Marshall 1987, with a different transformer.

You're correct, except the screen power chain resistor in place of the choke makes all the difference between the two amps. It creates the sag of the PI power node that allows one side of the PI clipping to sneak in under the power grid clipping threshold. In a Marshall it has the same PI input grid clipping creating asymmetry (although not as much as an Express) but the choke screen power chain keeps voltages higher and both sides remain power grid clipping dominant when the amp is heavily overdriven.

Just relating *my* experiences - they may or may not tally with those of others, but that doesn't make anyone wrong

Your experiences agree with mine.

This amp kinda flips and flops between two operating modes, and you can see it do that dynamically if you hook up the CRO on the output while playing it with a real guitar. Turn your volume down for clean, and it is clean, like a Fender but with lovely screen compression added, which the 6V6 & 6L6 amps just don't have. The CRO display shows a complex, dynamic waveform that has the same headroom on either side and if you're just on the verge of clipping, it's symmetrical and just the peaks, providing a little compression (just like any other amp run up to just before clipping where you get a beautiful slightly compressed full volume clean). When it flips over into mean mode, the asymmetry kicks in, and headroom is lost on the wider bottom side, just like in the test waveforms I've posted. With the slight (about 1/3rd sec time constant) delay in flipping between modes caused by the 18K2/20uF PI supply rail filtering, the changeover has dynamics to it, and you can see on the CRO a visualisation of the feel you hear from the amp as you play and adjust the guitar volume or even you picking. It's really cool when you get the levels all right you can see this flip/flop strumming open full chords lightly vs. strumming heavy power chords without adjusting the volume knob. Maybe I need to capture this somehow on my DAW instead of the CRO so I can show you all what it looks like.