NFB: Fender Princeton AA1164 vs Tweed Champ 5f1 vs Deluxe 5c3

[pnadora]

Hallo everyone,

I was looking at negative feedback loop options for my paraphase inverter 7591 build I am planning at the moment and found 3 different options that would be possible:
The Fender Princeton Reverb loops the signal to the "ground" of the last gain stage and seperates the "ground" with a 47Ohm resistor to the real ground (untechnically speaking). The Tweed NFB-loops go straight to the cathode of the last gain triode. The early Deluxe NFB goes to the grid of one of the output tubes. I am thining that going the Tweed champ way might be the easiest (together with removing the cathode bypass cap), but what are your opinions? Do you have experience with the different approaches?

Best regards!

[HotBluePlates]

They're all "6 of one, half-dozen of the other."

... The Fender Princeton Reverb loops the signal to the "ground" of the last gain stage and seperates the "ground" with a 47Ohm resistor to the real ground (untechnically speaking). ...

A different way to look at it is to realize the negative feedback loop on the Princeton Reverb is the 2.7kΩ resistor and the 47Ω resistor.

The loop from the output transformer (OT) secondary in guitar amp's is often returned to the phase inverter. However, the split-load inverter of the Princeton Reverb offers no gain of its own and the grid is the only convenient feedback-injection point. The feedback has to throw away some gain to linearize the output section, and the split-load plus output tubes just don't offer enough voltage gain.

To inject the feedback at the grid, whose voltage is elevated in that circuit, the designer would have to use a blocking cap to keep d.c. off the speaker. That requires one more part to buy (relevant if you building thousands), and introduces a low-frequency roll-off in the loop. We want to avoid that if possible to maintain stability, because there are already enough phase shifts inside the loop without the added cap. Without the added cap, the feedback reaches as low as the rest of the enclosed circuit will allow.

So instead, the loop is returned to the stage ahead of the split-load inverter. My habit is to always see that stage as part of the split-load inverter anyway. Fender stacked the gain stage on top of the loop/feedback injection point.

If you're looking to copy a feedback loop, you'd do best to copy a loop from another 7591 amp. At least for general values, etc.

The resistor values are determined by deciding where the injection point will be, and noting the available resistance-to-ground at the injection point (often a cathode resistor, or a resistor between cathode & ground). The Princeton Reverb uses 47Ω because it's only about 3% of the resistance of the 1.5kΩ resistor stacked on top of it, so it won't impact circuit operation any more than the ordinary 10% tolerance of that part.

Then Fender had to decide how much feedback voltage they wanted, to arrive at a given number of dB of feedback. They may have planned this, or determined it experimentally.

If we assume the Princeton Reverb makes 15w of output across its 8Ω speaker, then the voltage at the speaker terminals will be √(15w*8Ω) = ~11v RMS. The 47Ω/2.7kΩ divider reduces that speaker voltage to 11v RMS * 47Ω/(47Ω+2.7kΩ) = ~0.19v or 190mV RMS of feedback.

The speaker impedance isn't constant for all frequencies. Where it rises (bass resonant frequency and after ~400Hz), the speaker terminal voltage will rise, and so will the feedback voltage. The higher apparent load impedance tends to counteract that naturally to make the voltage rise less than it would otherwise be, and the remaining increase of feedback voltage further counteracts the effect. The net result is a more-controlled bass and more-even frequency response.

Too much feedback voltage will eventually make the amp unstable. Not enough sounds loose/raw but with added speaker-flap in an open-backed cabinet. I might be convincing myself that Fender arrived at the exact resistance value for the 2.7kΩ series resistor experimentally with a pot or resistance decade box... (and rounding the experimental result to the nearest standard value)

The different resistor values in the feedback loops of amps which otherwise seem similar will be due mostly to the differing output power of the amps, a different-impedance OT secondary tap from which the feedback is taken, and the fixed known resistance to ground of the feedback injection point.

[pnadora]

Thanks for the very informative answer! I feel that I learn a lot reading through the in depth comments you and the other experienced builders in this forum are writing and I highly appreciate that you are taking the time.

The only amps that I found that uses the 7591s and NFB are old ampegs. The reverberrocket I linked is an example, but the way they do it is not clear to me.
Also I am using a different type of phase inverter (paraphase). My plan at the moment is to go straight to the cathode of the last gain triode (right before the phase inverter triode) and start with a 20k resistor (as in the tweed). Or maybe a 10k resistor in series with a 25k or 50k poti. That way I could start at little NFB and find a sweetspot that I like.

http://schematicheaven.net/ampegamps/reverb_rocket_12r.pdf

[jjasilli]

An important distinction is the presence (Blackface) or absence (Tweed) of a cathode bypass capacitor at the injection point of the NFB.  If a cap IS present, then the split cathode resistor circuit must be used, in which the bypass cap is lifted from ground (Blackface).  Why?: If NFB is injected at the top of a simple cathode resistor/bypass cap circuit, then the NFB signal voltage will bleed to ground through the cap.  This would defeat the intended use of NFB.  (This is not a problem for the Tweed circuit, because it does not use a bypass cap.) 

With the split resistor circuit, NFB is injected at the bottom of the cathode resistor/bypass cap junction.  NFB is lifted from ground by the resistor at the bottom of the stack; then NFB passes up through the bypass cap (passing around the upper resistor), affecting the cathode inside the tube as intended.

Note: the bottom resistor also lifts the bypass cap from ground.  This could hamper the main bypass function of the cap.  So, the value of the bottom resistor must be small enough not to interfere with the cap's main bypass function.  This in turn requires matching component values, for proper voltage division, in the rest of the NFB circuit, per Hotblue's Reply.

[tubeswell]

The key advantage of the PR-type NFB, is that it enables you to get more gain than a tweed champ style NFB - because of PR NFB retains the cathode bypass cap on the preamp stage (for gain boost). Another advantage of this, is that you can adjust the cathode bypass capacitance to tweak the half boost point for this extra gain - so you can tailor the frequency response of the amp more.

[HotBluePlates]

The key advantage of the PR-type NFB, is that it enables you to get more gain than a tweed champ style NFB - because of PR NFB retains the cathode bypass cap on the preamp stage (for gain boost). ...

The tricky part is that "extra gain" is inside the feedback loop, so it really amounts to the ability to apply more feedback.

The feedback resistors could be viewed as setting the "gain of an opamp" where the opamp is everything inside the loop from the feedback-injected stage to the OT secondary. This "opamp gain" then specifies how much input signal to the 1st stage within the loop to get a particular output voltage (power output) across the speaker. You could also call this "power sensitivity". Most articles dealing just with feedback will call it "closed loop gain" but that might be too abstract for us to see what's happening.

If you were to remove the cathode bypass cap in that Princeton, the power sensitivity didn't change, and the closed-loop gain won't change. But the open-loop gain before feedback was applied did change (got smaller), so feedback was reduced.

[HotBluePlates]

... The only amps that I found that uses the 7591s and NFB are old ampegs. The reverberrocket I linked is an example, but the way they do it is not clear to me.
Also I am using a different type of phase inverter (paraphase). ...

The plan you linked has 6V6 output tubes, but does have a paraphase inverter.

The 470kΩ and 510kΩ resistors aren't the real plate load resistors for the paraphase stages; the 47kΩ resistors which connect to the power supply (at node C) are the plate load resistors. The bigger-value resistors are at once a voltage divider for getting the signal of the 1st paraphase stage knocked down and sent to the 2nd paraphase stage's grid, and are a self-balancing system because the output of the 2nd paraphase stage feeds the divider from the opposite side.

Since we can find Ampeg amps with a paraphase inverter, 7591 output tubes and a feedback loop (such as in the J-12-A and J-12-D Jet amps), could use simply copy the complete phase inverter, output stage, and power supply of those amps? That way a lot of the head-scratching over design considerations is already sorted out.

Otherwise, you'd need a full-up output stage & phase inverter design (to understand the open-loop gain from inverter input to speaker voltage), a survey of typical amount of feedback for these Ampeg 7591/paraphase amps (to know how much feedback is typically applied), then a re-working of the paraphase circuit specifics to determine the workable way to inject the feedback.

... I am using a different type of phase inverter (paraphase). My plan at the moment is to go straight to the cathode of the last gain triode (right before the phase inverter triode) and start with a 20k resistor (as in the tweed). Or maybe a 10k resistor in series with a 25k or 50k poti. That way I could start at little NFB and find a sweetspot that I like. ...

The way you proposed won't work well, IMO. The "stage ahead of the inverter" as the Princeton injected feedback only makes sense for the split-load inverter, which you won't be using.

Look back at the schematic you linked; ignore the fact of the 6V6's for now.

V4 pins 1-3 are the 1st stage of the paraphase inverter. It shares a 1kΩ cathode resistor (bypassed) with the 2nd stage of the paraphase, V4 pins 4-6. As noted before, the plate loads are the 47kΩ resistors on pins 1 and 5 of these stages.

The feedback loop comes from the OT secondary, and is made up of a series 10kΩ resistor, and a shunt 220Ω resistor inserted between the 1st paraphase stage's cathode & the top of the shared cathode resistor.

If it were done Princeton-style, with the feedback going from the bottom of the shared cathode resistor to ground, the earliest stage within the loop would be uncertain and the loop might oscillate (there is a different-polarity for injecting the signal at the 1st inverter stage's cathode as compared to injecting at the 2nd inverter stage's cathode).

And Ampeg has already appropriately-sized all the parts to work together, if you just copy their plan.

If you want less NFB, you could add resistance to the 10kΩ. I'd think by the time that resistor is raised to 100-200kΩ, there is no feedback actually being applied. You might use a 4.7kΩ resistor in series with a 50-100kΩ rheostat to determine experimentally what sounds best to you.

[pnadora]

My bad, I linked the wrong shcematic with the 6v6s (but it does not change from the plans you linked all that much).

It might sound stupid,  but I dont really see the voltage divider that brings down the output of the first half of the 6SL7 before entering the second one. That means I do not understand the self balancing system idea. (Therefor I dont know how to copy/modify the circuit to use a 12AX7 PI tube).

I would want to use a 12AX7 in place of the 6SL7 (because I have an old one). So I would probably want to change the load resistors to 100k from 47k and use 820ohm for the shared cathode reistor (bypassed by a 22u bipolar cap) if I use the amped PI layout. The voltage divider would have to bring down the signal by 1/45 before going to the grid of the 2nd tube in my case. Also what I find strange about the ampeg way is that they take the signal feed for the phase inverting tube before the coupling cap and add another capacitor instead of just taking the feed after the coupling cap.

I attached the way I planed to do my phase inverter with the 12AX7.

[pnadora]

To confuse things more here is an Univox schematic with 7591s and  NFB going to the cathode of a pre PI gain stage with what looks like a very confusingly wired split-load inverter

 

[HotBluePlates]

It might sound stupid,  but I dont really see the voltage divider that brings down the output of the first half of the 6SL7 before entering the second one. That means I do not understand the self balancing system idea. (Therefor I dont know how to copy/modify the circuit to use a 12AX7 PI tube). ...

I can't open .sch files right now (though you could save them as a picture within that program).

So let's just look at The other Jet schematic with the 6BK11. That tube is basically a 5751 triode and two 12AX7 triodes in one bottle. I presume Ampeg used the pair of 12AX7-like triode for the phase inverter. So you have no changes to make if you simply copy what they did.

Look again at the schematic: on this one the 120kΩ resistors are the plate load resistors (they were 47kΩ in the 6SL7-variant schematic). Now look at pin 5 of the 6BK11... Other than the 120kΩ plate load, there is a 470kΩ resistor before you get to the 0.02µF coupling cap. Trace through that resistor, and there is a line leading to another 470kΩ resistor and pin 11 (grid input of the 2nd paraphase stage). Consider the pair of 470kΩ resistors a voltage divider to reduce the signal going into the 2nd paraphase stage by 1/2.

But look at the output of the 2nd paraphase stage at pin 2: There is a 510kΩ resistor running back to the junction of the 470kΩ resistors. Two out-of-phase signals are meeting here. That junction of 3 resistors looks like it's 0v, but only if the two outputs from the halves of the paraphase inverter are equal & opposite. If they aren't, there's a see-saw effect which changes the input into the 2nd half of the paraphase, until the outputs are equal.

So it's (one version of) a self-balancing paraphase inverter.

[pnadora]

Amazing explanation! Thank you. I think the best thing to do is copy that layout :)
I attached the exported picture of my original plan.

[HotBluePlates]

... I attached the exported picture of my original plan.

Thanks for uploading that!

1st comment: yes, what you drew will work.

2nd comment:
Without drawing a loadline, I can only guess as to the gain of the 1st 12AX7 stage, but let's call it a gain of 63 (56-65 might be a typical range for a 12AX7 with a 100kΩ plate load, but the actual gain is dependent on the internal plate resistance, which is itself dependent on the particular operating point at the supply voltage used). Perfect balance would occur when the voltage divider knocks the signal down to 1/63rd of the output of the 1st gain stage. That's because we presume the 2nd gain stage will also have a gain of 63.

4.7kΩ/(4.7kΩ+220kΩ) = 0.0209 = ~1/48.

So this paraphase won't have perfect balance. Now it's time for some choices...

If you have identical 7591 output tubes, perfectly matched in every way, then any even harmonic distortion generated in the output stage will cancel due to how push-pull operation works. This is true for all push-pull output stages. But if the output tubes are unbalanced/unmatched to some degree and/or the drive signals to the output tubes are unbalanced to some degree, you will get some additional amount of even harmonic distortion in the output stage.

How much? Is it worth it? Should I balance or unbalance? I don't know, mainly because it's a judgement call. The sonic effect of unbalancing may be why some folks like paraphase inverters in the first place. And there are a lot of paraphase variants, even within the self-balancing types.

So you can try your plan, in which case I'd leave a space for a fixed NFB series resistor to go, but use a pot to dial in what sound right to you when the amp is actually built & running. Then you can replace the pot set at the sweet-spot for a fixed resistor of the same value.

Or you can steal Ampeg's plan as-drawn. You can always change overall negative feedback by adjusting the value of the 6.8kΩ series resistor, which is drawn from the OT secondary. If you count up the needed resistors for each plan & compare, you could even make sure to leave enough space to re-wire from one paraphase variant to the other to taste-test after the amp is built and you've had a chance to evaluate it.

To confuse things more here is an Univox schematic ...

I think you'll want to leave that beast alone.

I'd have to solicit PRR's help in deciphering that thing. What I think I see is an average split-load phase inverter, with overall NFB applied to the pre-gain stage's cathode.

However, the pre-gain stage also has NFB applied locally from its plate load to its cathode. That feedback path first has reduction compared to the tube's full output by using a tapped plate load at 1/2 the full resistance of the load (two 100kΩ resistors, so a total of 200kΩ plate load). The feedback signal is further reduced by the divider created out of the 330kΩ and the 1.5kΩ cathode resistor.

But! The pre-divider feedback signal also has the grid-reference resistor for the split load stacked on the feedback signal. Whatever other effect it has, it appeared to lower the effective value of the split-load's cathode load resistor (looks like 100kΩ, but the circuit sees something lower due to parallel path to ground for signals, plus whatever the feedback is doing). As a result, the plate load for the split-load is knocked down to ~43kΩ (50kΩ || 330kΩ).

I have no idea if all those parts & all that work was worth it. I'd have to breadboard it just to see what it does (guessing very clean, balanced drive to the output tubes... at least I hope so). And since the only gain would come from the pre-gain stage, which just had its gain reduced from ~63 to ~31 just by leaving off a cathode bypass cap, and then its gain is reduced further by local negative feedback... Well, I wonder why they didn't just use a 12AU7 in a standard split-load inverter & pre-gain circuit and be done with it (for a lot less cost in parts & engineering time).

[tubeswell]

I've built a couple of amps with NFB to the 1st inverting stage of a paraphase inverter. The last one I did used a 6k8 NFB resistor and 22R to ground ('underneath' the Rk||Ck for the 1st inverting stage - topology like a BF vibrochamp). Worked okay to tame the wild out of control sensitivity. I'll hunt down a schematic. Edit-  found it

[pnadora]

@tubeswell I think I will go that route as well.

I recently noticed that the output transformer I pulled out of the organ amplifier has a center tap on the secondary that is grounded. I have never seen one of those before and now I wonder how that influences my negative feedback design and the point I take the feed from. Since I use the two 12" speakers from the organ and the winding ratio of the transformer gives me pretty much perfect impedance for the 7591 I want to keep the CT grounded and use the normal green and black wires for the speaker in parallel as in the original amp.
I attached the organ schematic for reference.

[sluckey]

I recently noticed that the output transformer I pulled out of the organ amplifier has a center tap on the secondary that is grounded.

I ran across one of those in a Hammond AO-63 amp. Slightly different NFB than your circuit. I also left the center tap alone. Just for reference...

     http://sluckeyamps.com/hammond_2/AO-63.pdf