LTPI balance

[uneumann]

Sorry, but since the very beginning of this thread I have been asking about the open loop balance of the circuit.

Perhaps we're using some terms differently. In post #26 I'm showing a NON-NFB example. Maybe there's confusion since the spkr signal still comes back to create the tail signal. Yes, that's a fraction of the spkr signal getting fed into the tail, but it's not negative feedback (NFB) since that signal only goes to the tail and it has no impact on the gain of the circuit. The circuit is essentially open loop. To have negative feedback (NFB) you have to connect some version of the spkr signal into the V2 input. That's grounded in this example, so there is no global negative feedback in that circuit.

As for the open loop uncorrected balance, that's shown when R2 is replaced by a wire (in post 26). That's grounding the tail as a standard PI would.

[Helmholtz]

Perhaps we're using some terms differently. In post #26 I'm showing a NON-NFB example. Maybe there's confusion since the spkr signal still comes back to create the tail signal. Yes, that's a fraction of the spkr signal getting fed into the tail, but it's not negative feedback (NFB) since that signal only goes to the tail and it has no impact on the gain of the circuit. The circuit is essentially open loop. To have negative feedback (NFB) you have to connect some version of the spkr signal into the V2 input. That's grounded in this example, so there is no global negative feedback in that circuit.

As for the open loop uncorrected balance, that's shown when R2 is replaced by a wire (in post 26). That's grounding the tail as a standard PI would.

OK. I see its not NFB. So the circuit needs a common mode speaker signal at the bottom of the tail resistor to improve balance, right?
And it wouldn't have a benefit without the speaker signal?

[uneumann]

OK. I see its not NFB. So the circuit needs a common mode speaker signal to improve balance, right?

Bingo - you got it - that's a good way to phrase it.
Technically, the common mode signal could come from other sources too, but the spkr signal is pretty handy and works well for this purpose.

[Helmholtz]

Bingo - you got it - that's a good way to phrase it.
Technically, the common mode signal could come from other sources too, but the spkr signal is pretty handy and works well for this purpose.

Thanks a lot.
And congratulations on your circuit idea!

[Helmholtz]

Bingo - you got it - that's a good way to phrase it.
Technically, the common mode signal could come from other sources too, but the spkr signal is pretty handy and works well for this purpose.

Just one more remark:
While the common mode feedback signal produces no NFB, its phase still matters.

[SoulFetish]

Given how cheap Ds and ECs are these days, just add a negative supply and use a large resistor from the cathodes. Then it is easy to add in two channels, or use the second input for negative feedback.

I like this approach as well. Wouldn't this give comparable performance to an active source? Would you need to add some protection for the BJT in the event of a Tube Failure as well?

[R.G.]

Given how cheap Ds and ECs are these days, just add a negative supply and use a large resistor from the cathodes. Then it is easy to add in two channels, or use the second input for negative feedback.

I was thinking much the same, excepting that I was thinking that given how cheap high voltage MOSFETs are these days, why not put in a MOSFET constant current source for the tail current and have it be really constant? The advantage that a current source has in a long tailed pair is that you get the performance of a long tail with a very short (voltaged, at least) actual tail.

[SoulFetish]

Perhaps we're using some terms differently. In post #26 I'm showing a NON-NFB example. Maybe there's confusion since the spkr signal still comes back to create the tail signal. Yes, that's a fraction of the spkr signal getting fed into the tail, but it's not negative feedback (NFB) since that signal only goes to the tail and it has no impact on the gain of the circuit. The circuit is essentially open loop. To have negative feedback (NFB) you have to connect some version of the spkr signal into the V2 input. That's grounded in this example, so there is no global negative feedback in that circuit.

As for the open loop uncorrected balance, that's shown when R2 is replaced by a wire (in post 26). That's grounding the tail as a standard PI would.

Okay, I was confused about this as well but I see what you mean... I think. In order to implement Global Feedback in this case the grid of V2 would have to be AC grounded to the junction of R39 & R2. I just realized that I've misunderstood how feedback was integrated in a LTP! Ha... ugh It seems so obvious now I feel dumb. I've been mistakenly thinking that the tail was the input for the NFB at the junction of the tail/feedback resistors. (I guess I thought that since this was the source of the non-inverting input, GFB was coupled in via the cathode voltages).
So, am I correct in understanding that this location (node of R39 &R2), is only really used because it is a convenient location to AC couple the grid of V2 with the feedback dividers, without really changing the DC operating point?
Some questions still remain; doesn't the speaker signal have some sort of modulating effect on the cathodes? how does this not affect the signal at the grid of V2? ...asking for a friend

[Merlinb]

So, am I correct in understanding that this location (node of R39 &R2), is only really used because it is a convenient location to AC couple the grid of V2 with the feedback dividers, without really changing the DC operating point?

Injecting the NFB into the tail as well as the second grid is a historical artefact from Fender. Really the NFB 'ought' to go to the second grid only. The small portion injected into the tail basically just subtracts a little of the overall NFB from the grid, but not enough to crow about. From my book:

Early editions of the Fender Twin Amp and Bassman used cathodyne phase inverters driving 6L6Gs, with feedback applied to the gain stage preceding the phase inverter. In 1957 the designs underwent a fairly radical change with the phase inverter being converted into a long-tailed pair to drive the new 5881 power valves (although they were not much more difficult to drive, really), and feedback was applied to the tone stack. It is likely that this arrangement –with so many capacitors in the feedback loop– resulted in questionable stability and possibly parasitic oscillation because, in the following year, the designs were modified again by moving the point of feedback injection to the phase inverter. However, adopting the logical scheme of fig. 9.10 would have required either changing the layout completely, or mounting the second grid capacitor Cg2 directly across some other components, whereas only one wire needed to be moved to create the [familiar Fender arrangement] circuit in fig. 9.11. Other amps in the Fender range adopted the same scheme shortly after, presumably in the interests of standardisation.

[SoulFetish]

Injecting the NFB into the tail as well as the second grid is a historical artefact from Fender. Really the NFB 'ought' to go to the second grid only. The small portion injected into the tail basically just subtracts a little of the overall NFB from the grid, but not enough to crow about. From my book:

Early editions of the Fender Twin Amp and Bassman used cathodyne phase inverters driving 6L6Gs, with feedback applied to the gain stage preceding the phase inverter. In 1957 the designs underwent a fairly radical change with the phase inverter being converted into a long-tailed pair to drive the new 5881 power valves (although they were not much more difficult to drive, really), and feedback was applied to the tone stack. It is likely that this arrangement –with so many capacitors in the feedback loop– resulted in questionable stability and possibly parasitic oscillation because, in the following year, the designs were modified again by moving the point of feedback injection to the phase inverter. However, adopting the logical scheme of fig. 9.10 would have required either changing the layout completely, or mounting the second grid capacitor Cg2 directly across some other components, whereas only one wire needed to be moved to create the [familiar Fender arrangement] circuit in fig. 9.11. Other amps in the Fender range adopted the same scheme shortly after, presumably in the interests of standardisation.

Funny you should mention that. Just so happens I was just reading...

[Helmholtz]

To study the effect of frequency dependent speaker impedance on the "enhanced tail" LTPI I suggest to replace the load resistor in the sim with a circuit like this:


as proposed by Mike Sulzer.
Details ae discussed here: https://music-electronics-forum.com/...ad.php?t=47723

[uneumann]

To study the effect of frequency dependent speaker impedance on the "enhanced tail" LTPI I suggest to replace the load resistor in the sim with a circuit like this:
[ATTACH=CONFIG]52679[/ATTACH]

as proposed by Mike Sulzer.
Details ae discussed here: https://music-electronics-forum.com/...ad.php?t=47723

Right - My spkr model does include some inductance/capacitance, but I don't recall where these models came from (maybe the Duncan site?) or exactly what the model is. I do know the model behaves more realistically (and differently) that a simple 8 ohm resistor.

Model's aside, the real test would be an implementation, and I have not yet built/tested a version.

This all started as an observation that the tail in the Fender LTPI was not optimally bootstrapped and a separate divider for NFB and tail bootstrapping could improve things. Then there was the observation that better bootstrapping could lead to a lower tail resistor and thus more headroom (post 26 showed good balance with a 100 ohm tail). Pushing that to it's limit begs the question, could one build a zero-tail PI (ZTPI)?
Well - it seems one can. If we remove the tail altogether and split the bias resistor to allow for bootstrapping, the circuit looks like this.



The 470 ohm bias resistor is split into 30 and 440 resistors. The grids are now ground referenced, so no dc blocking caps are needed so the NFB grid input is simply grounded. The signal input can feed the V1 grid directly (if it's ground referenced, as say from a Master Vol pot). This example has no NFB, so it's an open loop output stage, and the spkr feedback is only used to bootstrap the lower 30 ohms of the bias resistor. The traces show good balance between the two PI outputs.

If you add NFB, you get the circuit below using the 820/47 ohm ratio for the NFB grid input. The R49 value for the bootstrap signal also changes to produce more signal. Again the plate signal traces are shown and they're well balanced.



Ok - so these are two extreme examples. The simulations show they're feasible, but I won't argue that they're practical. I just show them since its interesting to see how far you can push an idea. In both cases, tuning the tail signal requires high accuracy resistors. A change of a few ohms in R49/R39 has a noticeable balance effect.

Using even a small tail resistor (say 1K) only adds a few volts of elevation, but it makes all the resistor ratios far less sensitive and therefore practical.

[Mike Sulzer]

That you do. What you suggest has better performance. But I like the simplicity of connecting the grids to dc ground (makes me think of a simple differential amplifier). I also think about the current source misbehaving under extreme overdrive conditions. I do not see how this could happen, but guitar amps get subjected to some weird behavior sometimes..

I was thinking much the same, excepting that I was thinking that given how cheap high voltage MOSFETs are these days, why not put in a MOSFET constant current source for the tail current and have it be really constant? The advantage that a current source has in a long tailed pair is that you get the performance of a long tail with a very short (voltaged, at least) actual tail.

[SoulFetish]

uneumann, have you run any simulations for operation under overdriven conditions? Im curious what happens to the balance when the signal is clipping. But I like what you’ve done here. It’s simplicity is elegant, and seems quite effective. Hallmarks of good design

[Helmholtz]

Here is a different approach (tentative) to explain the observed results of „tail-enhancing“:

Injecting a speaker signal at the bottom of the tail means a feedback into a common-mode input of the LTPI. Part of this feedback voltage (it is attenuated by voltage dividing between the upper part of the tail and the 2 parallel internal cathode impedances) appears at the cathode junction and gets amplified by the triodes both acting as non-inverting common-grid amplifiers. If/as this voltage is in phase with the PI input, it makes the plate output voltage of the inverting (first) triode decrease and the output of the non-inverting triode increase. In other words, the method produces NFB in one triode and positive FB in the other. By adjusting the level of the injected speaker signal (via voltage divider) the method allows to balance the PI outputs.

The check the validity of my hypothesis, I suggest to invert the phase of the speaker signal (not using the „normal“ NFB loop for now) and watch PI output balance getting worse than without „enhancing“.

At least in theory it should be possible to make the non-inverting output signal larger than the inverting one by further increasing the speaker feedback signal. It would probably require the signal at the bottom of the tail to be larger than the signal at the cathodes.