Strange solution to ugly distortion

[Gaz]

I had a BF Fender build I converted to have a post-phase inverter master ("Lar-Mar" type that replaces the power tube grid leak resistors). I observed that at lower settings (allowing the PI to distort first) there was static-like "undertone" that would appear at the attack of the fade away. It was only really noticeable at high treble settings and bright switch on, but very annoying and unmusical. It sounded kind of similar a cap discharging, but under the tone and would stop before the note fully decayed.

I thought maybe it was a bad tube or connection, but everything checked out. I then thought maybe a parasitic oscillation, but scope traces were totally clean at all frequencies and amp settings...

I don't know why I even thought of it, but I tried a larger PI input coupling cap, and the noise vanished. It was originally .001uf, but when changed to .1uF the noise was gone. I made sure the .001uF wasn't faulty and was able to repeat the results with different caps.

Why would a larger PI input cap resolve ugly distortion at the onset of clipping? The only thing I can think of is that is somehow improved the PI balance or something to do with there being less NFB with the PPIMV turned down...

[nickb]

It changed the time constant so it didn't happen on the short attack time. When the grid to cathode voltage goes sufficiently positive you get significant grid current, but not when negative so the coupling capacitor charges up. It's similar to the blocking distortion mechanism in the power tubes.

You could try a resistor, say 100K, in series with the coupling cap instead.

[pdf64]

there was static-like "undertone" that would appear at the attack of the fade away

I’m not clear which part of the attack-decay-sustain-release note envelope the above refers to?

Whatever, yes, having such different time constants in the grid circuits may be upsetting the respective operating points under heavy overdrive conditions.

Merlin mentions that LTPs might exhibit blocking distortion due to overly long time constants, and suggests 10nF-470k for both grid circuits http://www.valvewizard.co.uk/acltp.html

So it may be worth considering reducing the 0.1uF values of the grid caps, as it seems far larger than necessary?

I wonder if it might even be blips of parasitic oscillation?
I can’t recall the exact mechanism, but paraphase type phase splitters with a shared, unbypassed cathode can do that, and an LTP isn’t that dissimilar.

[pdf64]

It changed the time constant so it didn't happen on the short attack time. When the grid to cathode voltage goes sufficiently positive you get significant grid current, but not when negative so the coupling capacitor charges up...

Is it feasible for blocking distortion to manifest on the ‘other’ decoupled grid?

[nickb]

Is it feasible for blocking distortion to manifest on the ‘other’ decoupled grid?

The feedback, if large enough could do it too.

Some sims showing the reality of the effect for 10nF, 100nF and 10nF in series with 100K. If the OP tries the series resistor then it will be informative. There is significant modulation of the duty cycle in the output waveform between start and end.

Green is the output. Blue is the current in the input cap

10nF:



100nF:



10nF with 100K in series:

[Gaz]

Just to clarify, the noise disappeared when the cap was made larger. From 1nF to 100nF! I forgot to mention that I had already tried a a grid stopper which made no difference to the sound. I did leave a 100K in there for good measure.

The noise sounded not too dissimilar from rubbing your hand on a mic - a muffled distortion under the note. The noise would begin on the attack of the note then stop before the played note fully decayed. It was easiest to hear at the edge of breakup and with the treble up/bright switch on, which is why I thought it was maybe a HF oscillation.

I couldn't reduce the distortion with snubbers or grid stoppers anywhere. I remembered why I tried the large input cap. Running out of ideas, I thought maybe to switch to silverface PI with 330K grid leaks and 10n input cap. I mis-remembered the values and used 100K grid leaks with 100n input cap and the noise disappeared. I later realized it was just the input cap that made a difference.

Really curious as to why this helped!

[nickb]

If the 100K didn't address it, then it's the change in frequency response that's making it sound better and nothing to do with the input cap charging.

[Gaz]

Yeah, I should have also mentioned that besides being more noticeable with treble turned up, it really only affected the high strings. That was clue to me that it wasn't due to blocking, even before I tried the grid stopper. It's perplexing that making the cap larger in value would affect the high frequencies. That's why I thought maybe it has to do with PI balance somehow.

[uneumann]

Not sure if this accounts for what you hear, but I've found that using the same cap values at the two LTPI inputs can have benefits when the PI is driven hard. As already said, similar cap values make the two time constants similar. The benefit of similar time constants is that transient balance is improved. That means as a high signal hits the PI, both grids shift by similar amounts preserving their current ratios during the shift period. (The shift becomes a common-mode signal.) This also means the tube operating points remain more stable during the transient. As Nickb showed, a grid stopper doesn't really impact the time constant much, so it doesn't fix the mismatch. The simplest approach is just to use similar cap values.

[Gaz]

Not sure if this accounts for what you hear, but I've found that using the same cap values at the two LTPI inputs can have benefits when the PI is driven hard. As already said, similar cap values make the two time constants similar. The benefit of similar time constants is that transient balance is improved. That means as a high signal hits the PI, both grids shift by similar amounts preserving their current ratios during the shift period. (The shift becomes a common-mode signal.) This also means the tube operating points remain more stable during the transient. As Nickb showed, a grid stopper doesn't really impact the time constant much, so it doesn't fix the mismatch. The simplest approach is just to use similar cap values.

I think you nailed it! I'm going to try lower the 2nd cap on the inverting side to match the input side to see if I get the same result.