Shunt Feedback - effect on input impedance

[gingertube]

There is a thread below which talks about using "Partial feedback" and other shunt feedback schemes to lower amplifier output impedance (to improve damping factor) by reducing rp of the output tubes.

I've got my head around that - using shunt (voltage derived)feedback from the output tube anode will lower rp - great.

My question relates to how the shunt feedback is applied.

I've seen examples of it applied to the opposite grids in a diff amp driver (Parallel applied??) and I've seen it applied to the cathode of the driver on the same side when separate common cathode drivers are used (Series Applied??). Both consitute negative feedback (that is they subtract from the input signal).

My (admittedly limited) knowledge of feedback theory says that this will affect the input impedance of the driver stage (the stage where feedback is applied) in opposite fashion. One way should raise it and the other way should lower it. This needs to be understood to determine the loading effect on earlier stage(s).

Can anyone enlighten me?

Thanks,
Ian

[ThorstenL]

Konnichiwa,

Originally posted by gingertube
There is a thread below which talks about using "Partial feedback" and other shunt feedback schemes to lower amplifier output impedance (to improve damping factor) by reducing rp of the output tubes.

Actually, now it's my turn to nitpick. Shuntfeedback is applied around a device/circuit to an inverting input to which also the signal is applied and Series feedback is applied to the inverting input of a device/circuit with the signal applied to the non-inverting input.

In ideal shunt feedback conditions the inverting input becomes a "virtual ground", or a current to voltage converter.

Originally posted by gingertube
My (admittedly limited) knowledge of feedback theory says that this will affect the input impedance of the driver stage (the stage where feedback is applied) in opposite fashion. One way should raise it and the other way should lower it. This needs to be understood to determine the loading effect on earlier stage(s).

Absolutely. Here again my suggested circuit.



As said, my own circuit design uses in effect the inputs of the Driver stage as near zero impedance virtua; ground which in effect makes the Driver/Output stage combo a balanced current to voltage converter that takes any balanced or unbalanced input current and turns it into a balanced output voltage applied to the output transformer.

As this is a real stage neither the input impedance nor the output impedance is infinitly low, but they are low compared to usual conditions.

Thus the input stage operates in effect as voltage to current converter and operates similar in nature to the lower device in a cascode, thus with reduced miller input capacitance, better linearity etc.

Sayonara

[Sch3mat1c]

Yes, virtual ground is the idea.

What happens is the Zo (Rp || Rl, etc.) of the preceeding stage, or an additional series resistor, acts as the top of a voltage divider. You know there's some voltage at the virtual ground, because the tube in question is outputting a signal. (You need infinite gain, as op-amps approach, to get a true virtual ground.) And you know the voltage across the second resistor, between output and virtual ground.

I'll give you an example. Say you have a 12AX7 putting out 50V of swing. It has a gain of 100 (...rough example, you can find the exact values if you wish). Thus, the grid is swinging -0.5V (negative to indicate inverse phase). That puts 50.5V across the feedback resistor. Let's say it is 470kohms. Thus there is a signal current of 0.11mA flowing. This has to go somewhere, and that somewhere is the preceeding stage (or input in general). Let's say we want the stage to appear to have a gain of 25, or -12dB from open loop (i.e., 12dB local NFB). That means the input signal must be 1/4 the output, which is 50V; that's -12.5V input. Negative since I already made the odd convention of calling output positive... Thus, the resistor is [(-12.5) - (-0.5)] / 0.00011A = 109k. Since this is working into a virtual ground of reasonable accuracy (100 gain in the amplifier is 1% away from a true ground), that right there is your input impedance. Likewise, because the error is small, the feedback resistor itself appears to the plate as its face value.

The essence of what I did above is this: the two resistors act as a voltage divider from signal source to output; the output is determined by the voltage at the middle node; and the ratio R1/R2 sets the NFB -- if gain is high.

Tim

[ThorstenL]

Konnichiwa,

Originally posted by Sch3mat1c
Yes, virtual ground is the idea.

Nice analysis, but adding a series resistor in that case is rather counter productive....

If we connect the Anode of the preceeding stage directly into this sub-ideal virtual ground things start getting interesting.

Sayonara

[Sch3mat1c]

In that case, the Zo of the preceeding stage becomes the R1. Then you have to analyze it based on the basic circuit elements, taking the driving stage as a perfect voltage source in series with Zo, and so forth.

I suppose you can vary R2 exclusively to get the results, but the virtual ground principle kind of breaks down without a physical R1 and you have to calculate R2 as multiplied by gain (as g-p C is multiplied by gain (miller effect), for exactly the same reason in fact).

Tim

[ThorstenL]

Konnichiwa,

Originally posted by Sch3mat1c
In that case, the Zo of the preceeding stage becomes the R1. Then you have to analyze it based on the basic circuit elements, taking the driving stage as a perfect voltage source in series with Zo, and so forth.

Ahyes, of course, but what are the actual implications for the behaviour of the circuit?

First, if we take a given triode there are two quite linear operating conditions. One, if we load it with an "infinite" load, gives us a rather linear voltage amplifier, the other, if we load it with an infinitly low value and we have a pretty linear current amplifier. Only if our loadline has a significant slope will we run into larger nonlinearities.

In my case we have a significant value unbypassed cathode resistor in the first stage, which acts as local degeneration which linearises the Voltage/Current curves of the Input Valve.

At the current summing junction our input amplifier therefore generates a reasonably linear current for a given input voltage to the input stage. The differential "transimpedance" amplifier rather linearly converts that current modulation into a voltage modulation across the load (Speaker & Transformer). So, signal linearity is assured with short feedback loops.

Any more advantages?

Yup, bandwidth. Given that anodes of the input stage are basically shunted to a low impedance very little voltage develops there, so the miller amplification of the Input Capacitance is largely suppressed.

As a final note, as commented before, the resulting Amplifier has a much simpler harmonic profile (fewer harmonics overall and higher order ones in a proportionatly low level) compared with more conventional structures. Usually such a behaviour "sound s better".

Unfortunatly, despite being introduced in the HK Citation 2 Amplifier many years ahgo this circuit structure was probably fraught with too many complex interactions to be employed widely in an age before fast copmputers and P-Spice, so it fell into disuse and as in moderns everyone merely builds copies of Mullrad, Williamson or Dynaco Amplifiers as Push-Pull Amplifiers it reamins unknown.

Sayonara

[Sch3mat1c]

Originally posted by Kuei Yang Wang
First, if we take a given triode there are two quite linear operating conditions. One, if we load it with an "infinite" load, gives us a rather linear voltage amplifier, the other, if we load it with an infinitly low value and we have a pretty linear current amplifier. Only if our loadline has a significant slope will we run into larger nonlinearities.

Erm... sorry to say it, but when's the last time you took a look at transconductance curves?

In triodes, distortion is lowest as RL --> CCS. In pentodes, distortion is highest at RL = 0 due to Gm variation and again at RL --> CCS due to the very high gain naturally having distortions. Inbetween, almost exactly at Po max, coincidentially, is minimum distortion.

In my case we have a significant value unbypassed cathode resistor in the first stage, which acts as local degeneration which linearises the Voltage/Current curves of the Input Valve.

And also raises Zo, reducing the feedback "R2" ... erm. Increasing its value. ( )

So, signal linearity is assured with short feedback loops.

Until the OPT reduces your frequency response and damping factor.

Yup, bandwidth. Given that anodes of the input stage are basically shunted to a low impedance very little voltage develops there, so the miller amplification of the Input Capacitance is largely suppressed.

Interesting way of putting it - reducing the gain (in this case by the unfortunate process of loading it down, increasing distortion) thus reducing miller.

As a final note, as commented before, the resulting Amplifier has a much simpler harmonic profile

Do you have graphs? I find it hard to believe that light levels of NFB will improve the spectrum favorably due to IMD.

Unfortunatly, despite being introduced in the HK Citation 2 Amplifier many years ahgo this circuit structure was probably fraught with too many complex interactions to be employed widely in an age before fast copmputers and P-Spice

Nah, it can be calculated as any other feedback system. Most likely, GNFB is simply better and more effective.

so it fell into disuse and as in moderns everyone merely builds copies of Mullrad, Williamson or Dynaco Amplifiers as Push-Pull Amplifiers it reamins unknown.

There's a schematic in the RCA Recieving Tube book (I have an RC-26 I think), the 50W amp (uses PP 7027 aka rebased 6L6) has a Williamson topology IIRC, with pentode drivers and shunt FB from output plates to them.

Tim

[ThorstenL]

Konnichiwa,

Originally posted by Sch3mat1c
Erm... sorry to say it, but when's the last time you took a look at transconductance curves?

Earlier on today. Check yourself, With a vertical loadline anode current vs. input voltage is most linear, with a horizontal loadline anode voltage vs. input voltage is most linear.

Originally posted by Sch3mat1c
In triodes, distortion is lowest as RL --> CCS. In pentodes, distortion is highest at RL = 0 due to Gm variation and again at RL --> CCS due to the very high gain naturally having distortions. Inbetween, almost exactly at Po max, coincidentially, is minimum distortion.

Look again. I have noticed earlier that you make the same cardinal mistake which occludes understanding, you thin in the "voltage" domain. Igoring at the moment the fact that electrons are imaginary anyway, our signal is a mudulation of current and voltage tehds to be the result of using imperfect conductors.

Try thinking "current" for a while, treating voltages as the result of current flow.

Originally posted by Sch3mat1c
Until the OPT reduces your frequency response and damping factor.

That it does, but not that much, that discussion was carried on and out elsewhere. In addition, of course damping factor is another of those misunderstood things and in reality, as long as the internal impedance of the Amplifier is notably lower than the Voicecoil DCR a REAL speaker will be sufficiently damped.

Frequency response, well, in an Output transformer that allows sensible bandwidth when included in the feedback loop it will be just fine when outside....

Originally posted by Sch3mat1c
Interesting way of putting it - reducing the gain (in this case by the unfortunate process of loading it down, increasing distortion) thus reducing miller.

Actually, I REPEAT, as the input signal to the Driver/Output stage balanced transimpedance amplifier is (under ideal conditions) current only and given that the anode/output current vs. input voltage behaviour is most linear under such conditions, the circuit actually establishes best linearity and does NOT increase distortion, the reverse is true.

In some ways this may be seen as the "folded cascode" of solid state. The bahaviour of the input pair is similar to one that is cascoded in most ways.

Originally posted by Sch3mat1c
Do you have graphs? I find it hard to believe that light levels of NFB will improve the spectrum favorably due to IMD.

Who said that my circuit uses "light levels of feedback"? Actually, in the traditional sense I apply a lot more feedback than the conventional variants of the circuit with similar valve lineup.

Originally posted by Sch3mat1c
Nah, it can be calculated as any other feedback system.

Absolutely, but more variables interact.

Originally posted by Sch3mat1c
Most likely, GNFB is simply better and more effective.

Your ability to calculate probabilities is rather appaling, I might add....

If you need to, please refer again to the specific circuit I am referring to (in the post above).

Sayonara

[Sch3mat1c]

Well, I certainly have ample reason to give up this argument if you deny even the existence of electrons.

Say, you got a better subatomic theory? Oops...

Tim

[ThorstenL]

Konnichiwa,

Originally posted by Sch3mat1c
Well, I certainly have ample reason to give up this argument if you deny even the existence of electrons.

Why? Do you actually have any serious investment in irrational believes that suggest that Electrons are REAL, not merely convenient fictions that allow the practical use of certain theories in efficient ways for certain purposes?

Surely you do not suggest that our current science has come even close to penetrating ultimate reality?

Originally posted by Sch3mat1c
Say, you got a better subatomic theory?

I DONT NEED a subatomic theory. Unlike most others I can livge perfectly well without deluding myself that i know the causes of everything and can explain everything.

We have already observed that the various theories which include the particular electron as key building blocks fail to explain reality sufficiently, they leave holes in realms usually justly called esotheric, but if a theory is reliable only in general midfield but not under extreme conditions the theory is in my view worse than admitting ignorance, because it has you reacting to a fundamental misunderstanding of the facts.

Originally posted by Sch3mat1c
Oops...

Yes, very.

Sayonara