Came up with a simple idea today, wondered if it has ever been tried. The basic idea is having a differential driver (not a phase splitter) that can take feedback from the plate to linearise and lower the impedance of a pentode or tetrode (or further linearise a triode)
See image for the circuit I thought of
See image for the circuit I thought of
Last edited:
HollowState said:
Hmm, are you sure?
Input signal goes to first grid, positive going signal lowers anode voltage on first valve, causing the output valve to switch off more thus also having a positive direction on ITS anode, which is then coupled to the other side of the differential, turning on that side more and turning off the driving side to cancel it out
Is this right?
I think I've changed my mind, I'm pretty sure it's negative feedback
Starting at the diff pair
signal goes UP
left hand anode goes DOWN
causing OUTPUT tube anode to go UP
driving right hand of diff pair UP
causing ITs anode to go DOWN
thus making the left hand anode opposite go UP, against its downward... ness
Starting at the diff pair
signal goes UP
left hand anode goes DOWN
causing OUTPUT tube anode to go UP
driving right hand of diff pair UP
causing ITs anode to go DOWN
thus making the left hand anode opposite go UP, against its downward... ness
The general in band phasing looks right to me, but at some low frequency the double RC couplings will cause two sets of 90 degree phase shifts, turning it into an oscillator. So at least one set of Cs need to be removed.
Using a single diffl stage to correct another power stage is commonly called a buffer circuit and was championed by J. Ross Macdonald back in the 50's. (search on "augmented cathode follower" here to find a link to his paper, or just search for his website. He's still around!) The single diffl is often used in instrumentation circuits. Shows up often on Tubecad too.
Edward Cherry used such a double diff amp control stage for a SS amplifier back in the 70's in the same way you have it shown. Sometimes the double is used in modern SS amps where a large bank(s) of paralleled output transistors are controlled in separate groups for class AB. (makes for easier control of DC biasing for crossover when using matched diffl transistors)
The double diffl should be working fine once the Cs are fixed, but it does waste a set of triodes. There is only one output signal to control (sum of the two output tubes), so only one diffl stage (or driver) is more commonly used to control that.
The double diffl. does provide a high impedance signal input at the diffl. grids versus the usual single diffl driver with resistive feedbacks to the grids. But the feedbacks can be moved to the cathodes to avoid that if thats a problem. (see RCA RC-30 handbook, page 696, 50 Watt HI-Fi audio amp.)
Don
Using a single diffl stage to correct another power stage is commonly called a buffer circuit and was championed by J. Ross Macdonald back in the 50's. (search on "augmented cathode follower" here to find a link to his paper, or just search for his website. He's still around!) The single diffl is often used in instrumentation circuits. Shows up often on Tubecad too.
Edward Cherry used such a double diff amp control stage for a SS amplifier back in the 70's in the same way you have it shown. Sometimes the double is used in modern SS amps where a large bank(s) of paralleled output transistors are controlled in separate groups for class AB. (makes for easier control of DC biasing for crossover when using matched diffl transistors)
The double diffl should be working fine once the Cs are fixed, but it does waste a set of triodes. There is only one output signal to control (sum of the two output tubes), so only one diffl stage (or driver) is more commonly used to control that.
The double diffl. does provide a high impedance signal input at the diffl. grids versus the usual single diffl driver with resistive feedbacks to the grids. But the feedbacks can be moved to the cathodes to avoid that if thats a problem. (see RCA RC-30 handbook, page 696, 50 Watt HI-Fi audio amp.)
Don
but some issues yetFeedback to the cathode is usually more linear than a full differential pair. (the cathode to grid voltage is the same factor for both the input and the feedback, whereas a full differential has some residual odd order distortion between grids) The low cathode impedance Zk being OK as long as some attenuation of the feedback signal is intended.
However, one still has to consider the variability of the cathode impedance with driver current when deriving the attenuation factor. Usually the two driver tube cathodes are connected together, so their 1st order impedance variation cancels. With the separate cathodes here and a CCS on each one you will probably see some significant Zk modulation on the gain. I would include a resistor (adj. pot initially) between the cathodes and adjust for minimum distortion.
(I don't know if there will be a null point, or if it will just tilt all the way to 0 Ohms) Since the output pentodes generate some characteristic 3/2 power distortion, their may be some advantage to the driver having a little opposite dist. (Well, I guess it can't go all the way to 0 Ohms across the cathodes, or there would be no feedback. The RCA design has 390 + 500 pot + 390 Ohms across the cathodes and then no CCS, just a tail resistor of 47 Ohms)
I see the RCA 50 Watt amplifier also included some feedbacks to the driver plates. The plates have a roughly 3/2 power law effect on the driver current, whereas the g1s and cathodes each have more like 2 power law effect (due to grid wire "island effect") So one can also tweek for lowest distortion apparently by adjusting the neg. fdbk partitioning between them.
details, details
Don
OH, on the previous issue of the double cap couplings in the loop, I guess one could stagger the RC constants sufficiently to avoid getting two 90 degree phase shifts simultaneously. Same issue comes up in a standard C coupled amplifier with multi-stages.
Looking at the 2nd version here:
A center tapped inductor (CCS off the center tap optional) would work nicely (as long as its inductance is sufficiently high to not cause a phase problem) to replace the two driver tube CCSs. This would sum the two cathode Zk impedances to mostly null out their variation. Could also put a resistor across the inductor to set the feedback attenuation even more stably.
Its tempting to try the xfmr secondary (16 Ohm, 4 Ohm, 0 Ohm) for the CT inductor, but that introduces xfmr internal phase shift problems. Maybe if the secondary were a driven winding too (CFB using the secondary), the phase behavior would be a bit better. But likely would require a high performance (high bandwidth) xrfmr. Risky.
Don
A quick analysis (looking at one tube side only) of the gain modulation from the varying Zk's (if left uncorrected) appears to attenuate the neg. fdbk more when the driver tube is turned on and the output tube is turned off. So the loop will try to increase the output tube gain when its turning off, which is opposite to the natural tendency of the output tube. So this effect is causing similar even harmonic cancellation like an ordinary diffl. stage. But leaving an odd harmonic signature like an ordinary diffl. stage too.
A center tapped inductor (CCS off the center tap optional) would work nicely (as long as its inductance is sufficiently high to not cause a phase problem) to replace the two driver tube CCSs. This would sum the two cathode Zk impedances to mostly null out their variation. Could also put a resistor across the inductor to set the feedback attenuation even more stably.
Its tempting to try the xfmr secondary (16 Ohm, 4 Ohm, 0 Ohm) for the CT inductor, but that introduces xfmr internal phase shift problems. Maybe if the secondary were a driven winding too (CFB using the secondary), the phase behavior would be a bit better. But likely would require a high performance (high bandwidth) xrfmr. Risky.
Don
A quick analysis (looking at one tube side only) of the gain modulation from the varying Zk's (if left uncorrected) appears to attenuate the neg. fdbk more when the driver tube is turned on and the output tube is turned off. So the loop will try to increase the output tube gain when its turning off, which is opposite to the natural tendency of the output tube. So this effect is causing similar even harmonic cancellation like an ordinary diffl. stage. But leaving an odd harmonic signature like an ordinary diffl. stage too.
smoking-amp said:
Feedback to the cathode is usually more linear than a full differential pair. (the cathode to grid voltage is the same factor for both the input and the feedback, whereas a full differential has some residual odd order distortion between grids) The low cathode impedance Zk being OK as long as some attenuation of the feedback signal is intended.
The current comming down that FB resistor is current not seen going through the triode's plate load with a CCS in the cathode. Notice that in the RCA amp the driver was a pentode, and one running at about a quarter of the FB-delivered current.
I don't think the cathode Z has much to do with it; the current delivered through the FB resistor couldn't care much between being grounded or attached to the driver cathode I think.
cheers,
Douglas
Hi Douglas,
Hmm, yes, I see what you are saying. The CCS can draw the extra DC current from the feedback, and any variation of feedback current has to go thru the triode unaltered, showing up as a linear voltage variation on the driver plate load. Seems that this should be strictly linear. Now I'm trying to figure out where my previous analysis went wrong. ( I was assuming that the cathode would present 1/gm as its impedance, and this varies as the current thru the tube varies. Hmmm.)
edit:
I guess I have to take into account the additional impedance from the plate (cathode to plate), rp, which has a signal on it too.
Just to throw in another interesting variation, (of the 2nd version) suppose we put the CCSs in as the driver plate loads, and use a resistor for the cathode pull-down. With constant current thru the triode, its gm has to remain constant. So Zk would presumeably be constant too. The feedback attenuation factor being determined by the resistors and constant 1/gm.
edit:
Have to take into account the rp here too, but that should be constant as well. (1/gm and rp in parallel being constant)
Don
Hmm, yes, I see what you are saying. The CCS can draw the extra DC current from the feedback, and any variation of feedback current has to go thru the triode unaltered, showing up as a linear voltage variation on the driver plate load. Seems that this should be strictly linear. Now I'm trying to figure out where my previous analysis went wrong. ( I was assuming that the cathode would present 1/gm as its impedance, and this varies as the current thru the tube varies. Hmmm.)
edit:
I guess I have to take into account the additional impedance from the plate (cathode to plate), rp, which has a signal on it too.
Just to throw in another interesting variation, (of the 2nd version) suppose we put the CCSs in as the driver plate loads, and use a resistor for the cathode pull-down. With constant current thru the triode, its gm has to remain constant. So Zk would presumeably be constant too. The feedback attenuation factor being determined by the resistors and constant 1/gm.
edit:
Have to take into account the rp here too, but that should be constant as well. (1/gm and rp in parallel being constant)
Don
There is no main difference.
A differential pair for inverted and non-inverted inputs.
In this case the negative feedback to the cathode is buffered with another tube ( follower),
which will put the current into the cathode.
This is the inverted input tube doing buffering.
In the resistor to cathode case,
the negative feedback current is put to cathode without a buffer = directly.
A differential pair for inverted and non-inverted inputs.
In this case the negative feedback to the cathode is buffered with another tube ( follower),
which will put the current into the cathode.
This is the inverted input tube doing buffering.
In the resistor to cathode case,
the negative feedback current is put to cathode without a buffer = directly.
--------------------------
Re-analysis of cathode Z (assuming no grid currents in the driver):
Either variant of the 2nd version using cathodes for neg. feedbacks has a linear (or constant actually) impedance at the cathodes.
The CCS in the cathodes variant has 1/gm in parallel with rp. They sum to a constant Z because of the Mu relation between their magnitudes and the -1/Mu relation between the AC signals on the grid and plate. But there is still the DC impedance here. One can also put in a cathode to cathode resistor if one needs a bigger attenuation ratio for the feedbacks.
The CCS in the driver plates makes the cathodes look like a current source, or Hi-Z. The same Mu and -1/Mu relations above canceling AC signal wise again.
-----------------------------
This local neg. feedback to the driver cathodes approach looks quite nice to me now. And it appears to get rid of the residual odd harmonic signature of a full diffl. pair. I'll have to use it more often now!
-----------------------------
future possibilities:
Looking at Bigwill's original 1st version with the dual full diffls, one might consider increasing its duties rather than chopping out parts. Since the diffl pairs are taking the difference between input and output, one might consider computing an Error Correction signal as in Hawksford Error Correction. Some additional crosscouplings required I think.
Or, one might consider trying a Harmonic Equalizer with the driver stages acting as the balanced mixers for the output tubes. Considerable thinking required on this one yet I think. Common mode feedbacks likely required.
Don
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.