Express Output Waveform Shots

[Tillydog]

Hi,

I haven't posted here very much, but have been a lurker for a while, and have messed around with the TW Express type circuit in the form of a 4-4-0:

My 0.02p - you need to include the NFB to make sense of what's happening in an Express PI. When the power valve grids go into conduction, the NFB loop breaks down, resulting in an increase in PI output. I believe this happens a *long* time before the PI is driven to grid conduction.

I posted some scope shots & more rambling on AX84 a while ago:

http://www.ax84.com/bbs/dm.php?id=436562

Happy to be proved wrong

Andy

[katopan]

Hi TillyDog. Thanks for reminding me about your AX84 post with 4-4-0 waveforms. I had seen it before but can't see the photobucket links from work, and had forgotten to check them out at home. Looking forward to finally getting a look at them!

You are right that the NFB loop breaks down with power clipping, and this happens well before the PI starts clipping. When you get to the point of PI clipping and then 3rd stage cutoff the NFB isn't doing anything and the PI output will be increased as you say. The waveform pics I've posted are representative of that operating point.

[Tillydog]

.....When you get to the point of PI clipping and then 3rd stage cutoff the NFB isn't doing anything...

Not quite...

The NFB attempts to compensate for the effect of grid conduction. As a consequence, the waveform from the PI becomes *very* asymmetric (top to bottom), rapidly running out of headroom for just a small increase in input signal, so I don't think it's true to say that the NFB "isn't doing anything" - true, the NFB isn't having the conventional effect, but it's there.

This situation will persist as the drive is increased, so IMHO, I don't think that the waveforms posted would be representative of actual operation (they would be for the PI in isolation).

YMMV

Andy

[katopan]

You're probably right Andy. I checked out the waveform shots from your AX84 post. Fantastic stuff! Thanks.

[Tillydog]

No worries

Having played with the full circuit, it's my personal view that the 'clean-mean' happens abruptly due to the effect that the NFB has on the PI once the power valves start clipping.

(With the low drive requirements of the 6SN7 and EL84s that I was playing with, and the low gain / high headroom of the 12AY7 PI in the stock 4-4-0 it was easer to see this, but I'm sure that the effect is still there in an Express.)

Andy

[katopan]

I'd love to get more of your opinion on clean to mean, because you've done so much looking into how your amp works. I agree that the feedback will bring on an abruptness of transition from clean to mean, but if there's no amplitude difference in the clipping on the output, the distorted output will be louder. Jackie's output waveforms from earlier don't show any sag in output clipping level, only the reduction in the negative side of the waveform with the assymetry. For clean to mean to work with no or little change in the volume, then there must be a reduction somehow in the peak to peak output when heavily overdriven. That assymetry effect might be enough to even up the volume between clean and mean with a real guitar signal, but in the waveform shots it doesn't look like it reduces enough. What are your thoughts about that given what you have seen in your testing? Really appreciate your comments so far. Thanks, Craig.

[martin manning]

When the power amp clips hard, the feedback signal is a smaller percentage of the drive signal coming in to the PI, which reduces the feedback effect. When heavily clipped, a whole-amp simulation shows that disconnecting the NFB has virtually no effect on the waveforms at the power tube grids or the output, neither in p-p amplitude nor in the degree of asymmetry. This might be part of the clean-to-mean, where the NFB is effective at lower signal levels, transitioning to not effective at all when overdriven.

It it necessarily true that a square wave sounds noticably louder than a sine wave of the same p-p voltage? Perhaps a simple listening test with a function generator would give some insight.

[rooster]

Alright, let me get my facts straight.

Andy - In your scope pics, you are showing an Express circuit using EL84s?

Craig, you started this post with pics from some other site made by someone else of an Express? And then you moved on to shots of your own amp that is not an Express?

Jackie - Your pics are of a modified Express, and not a stock Version?

[katopan]

A square wave sounds louder than a sine wave of the same amplitude. You can test that for yourself fairly easily. Then add that a clean guitar sitting underneath power amp clipping is lower still than a clean sine wave even if it's driven into a bit of compression, and there's a volume difference unless you get some serious sag or other dynamic reduction in the clipped amplitude.

Rooster - the original link was John Polstra's site which was of an Express with EL34s but with a parallel first preamp stage. In my first post I mentioned that I am working on an Express circuit based around 6P1P or 6BM8 output valves. But for the waveform shots I posted I boosted the PI and preamp voltages to match a 'real' Express as per the voltages table in the Trainwreck Files section. With the power section disconnected it's identical to what's in an Express if you disconnected it's signal onto the power section. Exception is that I'm using 6N2P valves instead of 12AX7s but my experience and the datasheet is that they have the same characteristics. I've seen just as much variation in bias and gain between different 12AX7s than I have between 6N2Ps and 12AX7s. But I'm trying to learn about how the Express with EL34s does what it does so I can get my 6P1P version working just as well. As per the first post, that's my motivation behind all of this. One of the big problems is that no-one has ever posted these sort of tests on a real Express for everyone to see and understand. Also all the discussion on clean to mean I've seen in reading as many threads as I could find on it is based on guesses rather than testing. People have tested how much sag the 1k screen resistor causes, etc, but not solidly related how any of it relates to a change in the output amplitude that results in the mean being similar volume to the clean. I'm just trying to dig a little deeper.

[Tillydog]

Alright, let me get my facts straight.

Andy - In your scope pics, you are showing an Express circuit using EL84s?

Oh for a straight answer....

There is a schematic in that AX84 post, I think:

It started as a 'standard' 4-4-0 which is an Express preamp with a low er gain 12AY7 PI driving a 6SN7 dual triode fixed bias at ~2W (with some changes to NFB loop & PI resistors).

I added cathode biased EL84s to the same PI (with appropriate OT tap changes & NFB loop components.

The early scope pics on AX84 (with the big negative peaks on the PI output) reflect this cofiguration - EL84 or 6SN7 as per the note in the pics.

I didn't like it, so changed back to an Express PI, and used split load anode resistors to scale the PI output down to suit the EL84/6SN7 drive requirements (which are very similar). The later scope shots relate to this configuration. From memory, the output from the split load resistors is 40% of the full 'Express' output. I chose this to reflect the ~12v bias point of the 6SN7 and EL84 vs the ~30V of EL34s.

(As an aside, I don't think the EL84s were a good idea).


What I think is happening:
=================

1) I don't belive the output is lower 'mean' than 'clean'. I haven't see this on the scope.

2)The output of the PI is proportional to the difference between the signal coming in and the signal coming back through the NFB loop. (If we don't agree on this, you may as well stop reading now! )

3) Up until the output valves are driven to grid conduction, the output is clean & symmetrical - "clean".

4) Once there *starts* to be grid conduction in the power valves (by a fractional increase in the amplitude of the signal coming in) the NFB loop becomes unstable, as follows:

I'm working 1 signal waveform cycle at a time here, holding the input constant:

Because the output has clipped slightly, the NFB signal back to the PI is less than is should be through this part of the waveform, so the *difference* between the NFB signal and the signal coming in is greater than before clipping - so the PI output gets greater through this part of the cycle (in keeping with 2 above). Because the NFB is 180 degrees out of phase, the extra PI output caused by clipping on the +ve part of one PI output appears on the bottom of the opposite side of the PI - these are the 'nipples' (I can think of no better description, sorry!) that appear on the bottom of the PI output.

I think that much is standard.

The output of the PI is AC coupled to the power valves, so when the input signal moves on 180 degrees, the power valve grid doesn't 'see' a large negative excursion with a standard sized positive signal on top, they see a signal equally balanced about their bias point, so the added signal on the negative half of the PI output looks to the power valve like the whole PI signal has increased in amplitude (with the bottom half distorted).

As we were on the verge of clipping on the last cycle, and now have what appears to be a bigger signal, there will be more clipping on the next cycle, so a bigger difference between the signal in and the NFB signal, a bigger nipple on the -ve side of the PI output, so what appears to the power valves to be an even bigger signal, even more clipping, etc. etc.

This will continue until the PI runs out of available voltage swing, where the system becomes stable again, as the drive to the power valves becomes limited by the available voltage swing from the PI.

The nub:

Within a few cycles of the signal, the magnitude of the PI output has been increased from 'barely clipping' at the power valves until it has used up all the available PI headroom. All this *at a constant input signal*, at a level *just* sufficient to cause the onset of grid conduction in the power valves. - We are now "mean" !

I don't know how sim software would handle this unstable condition.

Any increase in the input signal will have only a small effect, driving the PI output a little further into compression.

5) I have my doubts about ever getting enough signal to cause grid conduction in the PI, but well before this, the input signal alone would be sufficient to drive the PI into distotion. Once this happens, the NFB has no effect - it's the 'window' between grid conduction in the power valves and saturation of the PI due to input signal alone that I'm working in.

This is just my best shot at explaining my take on it - Other opinions are tolerated

If somebody wanted to verify this (or otherwise), they could see if disconnecting the NFB changed the clean-mean behavior. It does on my amp, but it's not an Express. It's close enough that I'm happy with writing all this, though.

As ever: YMMV.

Andy

[martin manning]

The lower part of the signal at the power tube grid never gets to the output, and therefore doesn't get into the GNFB, because the valve will be in cut-off.

[katopan]

Andy, thankyou so much for explaining what you have seen. Great information from someone who has actually had a decent look at what's going on (admittedly in an amp that's not an Express, but exhibits similar characteristics - same boat as me!).

My 2 cents on your points:

1) Great to have this confirmed. But on both John Polstra's and Jackie Treehorn's output waveforms when the assymetry kicks in there is a discharging slope to the negative half of the waveform that means it's average amplitude on that side is reduced. One side reduced = reduced peak-peak = lower output power at this level of clipping. I thought I saw this on your scope pics too, but am at work again (can't access) and didn't copy your pics to a USB key so they're still on my computer at home. I'll be taking another look.

2) Agree! I kept reading.

3) Yep.

4) What you're saying is very well described and in great detail. I agree with everything you've described, but wonder about the effect of power grid clipping drawing grid current through the PI to power stage coupling caps. Your description and the balancing effect of the AC coupling is driving the signal at the power grids up in DC bias terms. But drawing grid current increases the bias voltage drop across across these coupling caps driving the signal at the power grids down, a cooling off of the bias, as evidenced by the appearance of crossover distortion in the output. Don't know how that fits in to the picture. Oh, and the sim software doesn't show bias changes across coupling caps very well at all, as I found out.

5) There is definitely grid conduction into the PI from the 3rd stage plate. The Express PI is warm biased with it's 470 ohm shared cathode resistor and sees grid conduction before the input gets large enough to go into cutoff for the input half side of the PI. Hence the more rounded top of the 3rd stage plate signal compared to the flat top seen in a Marshall style circuit. Also you can see where that point goes from PI grid clipped to 3rd stage cutoff when it suddenly goes completely flat.

I've hit an interesting point with my 6P1P based amp. I put in a dual gang pot on the PI plates (100k but with 470k in parallel on the normally 82k side to give the expected 82k plate load) so I could sweep through the full range of drive from the PI into power valve grids.

Firstly set for full PI signal into the power amp, and the lower half of the output waveform doesn't reduce by much with a real guitar signal. To the ear (what really matters) it sounds like other amps with a lot of power stage distortion. There's still some assymetry warming things up, but volume drops as it cleans up. Turned down so it's just PI distortion it looses the smoothness you get from power stage distortion (doesn't sound good), the distortion is thinner or less complex or something else missing, and likewise drops the volume when cleaning up. But somewhere in the middle there is a small band where you get the smoothness of power stage distortion, a heap of warming assymetry and clean to mean with very little volume change. Like with the 'tuning' the AX84 guys went through with the 4-4-0, for me so far the magic is all about adapting and getting right the level of PI output vs the power grid clipping threshold. Given that real Express' were tuned by ear by valve swaps, I think this PI output level could even be an area Express cloners could use to fine tune the clean to mean nature of the clone amps to whatever 12AX7 they are using. Of course you would assume a clone to be close enough to maybe only need a tweak in the PI supply voltage rather than split plate resistors, etc.

Thanks again!

[Tillydog]

The lower part of the signal at the power tube grid never gets to the output, and therefore doesn't get into the GNFB, because the valve will be in cut-off.

It's the output clamping due to grid conduction that gets into the NFB loop.

The lower part of the PI output doesn't get in directly, but I believe it has a significant effect: The extra NFB 'nipple' on the lower part has the effect of pushing up the peak to peak voltage of the PI. Looking directly at the anode of the PI, all of the increase in P-P voltage happens on the bottom half of the PI output, so would have no effect on the power valves, as you say.

If the DC level is lost (as happens due to the AC coupling between the PI and the Power valve grid), the AC signal simply looks to have a greater P-P voltage, centred around whatever the average level of the signal is. So the effect of the NFB on the lower part of the signal causes the power valve to see *increased* drive, even though the power valve is cut off during that part of the signal.

This can only happen if the PI has significant headroom left at the onset of grid conduction in the power valves.

Happy to be proved wrong

Andy

[Tillydog]

My 0.02p on your 0.02p.....

1) ...I thought I saw this on your scope pics too,....

Maybe - I haven't noticed it, but I wasn't looking for it. I think there are greater forces at work

4) ...drawing grid current increases the bias voltage drop ...

Agreed - I don't know what determines whether this effect, or the effect that I describe wins out. I can see that either could dominate, depending on the circuit design.

(Glad you could follow the description - it was a bit of a monster )

5) There is definitely grid conduction into the PI...

I know it is possible for it to happen, but wonder if it would ever happen in use - the guitar signal needed is much, much higher than that needed to drive the output into full distortion (if my measurements are to be believed!).

I've hit an interesting point with my 6P1P based amp.....

Interesting stuff - do you have NFB connected up yet?

...this PI output level could even be an area Express cloners could use to fine tune the clean to mean nature of the clone amps to whatever 12AX7 they are using.

I harbour a quiet suspicion that this is the case, but would never dare suggest it

Andy

[katopan]

You can put your CRO probes on either side of the 3rd stage to PI coupling cap and watch when it starts charging up from PI grid conduction current. I've seen it with a guitar signal as well as test signals. The full headroom of the PI input signal is only 3-4V so a couple of volts change across the coupling cap shifts the centre of the signal going into the PI by a lot relative to its input.

NFB has always been connected in my amp. It was only broken for the "3rd stage & PI only" tests I did because the signals into the power grids were disconnected. I'm seeing the effects of the NFB same as you with the power stage connected.

I probably shouldn't have mentioned it either. But I have full respect for the design of this amp and am trying to share my understandings of it with, hopefully, other open minded forum members. It seems discussions on this sort of stuff haven't gone on to any real conclusions before.