A way of using feedback with triodes

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

5v333

Well-known member
Joined
Jun 30, 2013
Messages
631
Location
Gothenburg
hi!

im playing around with feedback and compensation.

attached is an imaginary pic with three different tube setups. the first is just two triodes with global NF. plate resistor of first triode is 100K for lowest distortion.

the second uses some local feedback around the second triode.
the feeback improves thd of the second triode and seems to be able to compensate the amp or part of it without inserting extra caps.
to use as much feedback as possible without loading the first triode im thinking about using 200K on both the feedback resistor and the plate resistor. by doing so, i assume the first tube will keep its DC characteristics. right?

now i thought, can we have an even lower feedback resistor and still maintain the DC characteristisc within the first triode?
pic three shows that the plate resistor and feedback resistor has become one.
now taking its DC current through the OT while at the same time passing AC between the triodes for feedback within the same resistor.
would this work?

 

Attachments

  • feedback nets.pdf
    11.9 KB · Views: 45
First diagram is standard two triode pair with NFB except you have a transformerplate load for the second stage.

Second is pretty much identical to a REDD47 except the first stage is a triode instead of a pentode.

Third Idea is harder to evaluate by inspection. I suggest you simulate it first.

Cheers

Ian
 
5v333 said:
the second uses some local feedback around the second triode.
the feeback improves thd of the second triode and seems to be able to compensate the amp or part of it without inserting extra caps.
to use as much feedback as possible without loading the first triode im thinking about using 200K on both the feedback resistor and the plate resistor. by doing so, i assume the first tube will keep its DC characteristics. right?
You know that the input impedance of the second stage is 200k/Av. So your load line will change anyway.
 
Ah..!

i didnt realise that...

well if the AC load also changes then that would be problem if i want to maintain the initial sweetspot of the first tube.

will the new AC load be Rfeedback/AV in paralell of the grid resistor of the sec triode? (Rfeedback/AV)//Rgrid)
 
OK, I've done a Spice comparison of the three proposed topologies. The numbers speak for themselves.
It just demonstrates something we knew already. Global NFB is the most efficient way to improve THD.
As you can see, the 1st circuit is the one where global NFB regulates gain. The second one, I had to relax global NFB because open-loop gain was not high enough. And the third just can't achieve the same overall gain, even with very little global NFB.

Now it can be debated that local NFB, wisely applied, reduces the need for high global NFB, which some claim is good for TIM (transient intermodulation distortion). The problem with this theory is that its advocates use analogies that are similar to the paradox of Zenon of Elea.
 

Attachments

  • compare THD.jpg
    compare THD.jpg
    112 KB · Views: 40
abbey road d enfer said:
That's correct.

in that case the ac load will go down, and not by a little. the distortion of the first triode will prob go out the roof.
maybe adding a series resistor just after the first triode to make up the ac load. that would look even more like a redd47.


 
The problem with the second circuit as it stands is it heavily loads the anode of the first triode which is what causes the increase in distortion. EMI successfully used this circuit with a pentode stage. Their success I think is due to two factors. First (EMI claims the gains are arranged such that the distortion of the first stage is to some degree cancelled by the second. The second is that the pentode can be considered a current source which so it is less affected (produces less distortion) with the relatively low impedance load of the second stage.

Cheers

Ian
 
thanks for bringing up the load issue.

yesterday i made a wild test. added a series resistor just after the first triode in my circuit just like in the redd47, with the intent to make up the AC load of the first triode.
my initial circuit was
RP 100K
Rload 330K

now
RP 100K
Rseries 100K
Rfeed local 1M
Rload 1M

it worked great!!!!!!!!

freq response is more stright and smooth. thd and imd is down about 2db. perhaps even hum. and the sound is improved.

i think its not only about keeping loads for good thd but also about tweaking in the pole of the millerC and impedance of that stage so that stabilit is maintaind when the global loop is closed.
 
5v333 said:
thanks for bringing up the load issue.

yesterday i made a wild test. added a series resistor just after the first triode in my circuit just like in the redd47, with the intent to make up the AC load of the first triode.
my initial circuit was
RP 100K
Rload 330K

now
RP 100K
Rseries 100K
Rfeed local 1M
Rload 1M

it worked great!!!!!!!!

freq response is more stright and smooth. thd and imd is down about 2db. perhaps even hum. and the sound is improved.

i think its not only about keeping loads for good thd but also about tweaking in the pole of the millerC and impedance of that stage so that stabilit is maintaind when the global loop is closed.
Could you post the schemo of both versions? How much did you have to tweak global NFB?
 
5v333 said:
in that case the ac load will go down, and not by a little. the distortion of the first triode will prob go out the roof.
maybe adding a series resistor just after the first triode to make up the ac load. that would look even more like a redd47.
The ac load line will certainly go down, but the 1st stage will be required to predominantly generate current instead of voltage; indeed,  global NFB needs to be decreased for a given overall gain. It maybe there's a sweet spot to be found.
 
abbey road d enfer said:
but the 1st stage will be required to predominantly generate current instead of voltage;.

im not following you..? can you explain?

How much did you have to tweak global NFB?

in what sence?

attached is three circuits.
top is how i wanted it from the start before i started monkey araound with the local feedback. i did load lines for weeks only to find thd sweetspots. the global feedback is injected via two windings of the OT.

middle shows how i had it before my last move. the local feedback and the comp-caps around the cathodes made the freq response somewhat stable but had an almost resonance going on around 200Khz.
i had noticed looking on a scope that with just 470K local feedback, i could see an substantialy better symmetry on the output of one tube when looking at the overshooting. i guess this was due to even harmonics which cancels mostly but i liked the waveshape better and it actuall sounded better.

and the final. im thinking that there should not be very much feedback aound the sec triode. perhaps 3-6db will make a nice output stage.  right now im guessing (suprise!!) the gain of he sec triode is about 10. with 6db localfeedback its down to about 5.
1M/5 = 200K
paralell with 1Mload is 166K
add some series resistance of 100K is 266K. almost the initial load of 330K. with a 150K series resistor instead and we are back in the zone. im going to remove the comp caps now and see if they are needed at all.
 

Attachments

  • 3 styles.pdf
    23.4 KB · Views: 11
5v333 said:
im not following you..? can you explain?

in what sence?
Since the input impedance of the 2nd stage decreases significantly (from 330k without local NFB to about 10k with local current NFB), the 1st stage has to produce the same voltage in both cases (because the 2nd stage operation does not significantly change) but about 20-30 times more current.

attached is three circuits.
Do they all achieve the same overall gain? Because my sim showed that OLG (open loop gain) changes significantly between the 3 options.
 
abbey road d enfer said:
Do they all achieve the same overall gain? Because my sim showed that OLG (open loop gain) changes significantly between the 3 options.

i think so. everytime i hook up my adda and computer, send the same sinewave signal, i get about the same back...

i need to understand how the gain of the sec tube is set due to the local feedaback. im just guessing right now.

i think if there is less OLG, then there is less feedback voltage going to the cathodes and therefor more gain is maintaind.
so i think the GNF network is self regulating it self kind of.
 
5v333 said:
i need to understand how the gain of the sec tube is set due to the local feedaback. im just guessing right now.
The actual 2nd stage gain is roughly the ratio of the current FB resistor to the output Z of the 1st stage, although there is a limit, which is the amplification factor.

i think if there is less OLG, then there is less feedback voltage going to the cathodes and therefor more gain is maintaind.
That is correct. The fact that the overall gain does not change means the gain with NFB is much lower than the OLG, or in other words there is a high NFB factor.
The original schemos you submitted did not really show that possibility.
 
so about the first circuits i started out with.

the third circuit would prob not survive anything i guess. so lets forget about it.

but the sec we can see in some oldschool schems.
maihak v72
and the klangfilm rz062 EQ that we are discussing in the the filter section of the forum.

the redd47 variant seems great because it looks like we can move the poles away from eachother, making the circuit stable when closing the global loop.

as long as we dont take away too much gain from the last stage we maintain some miller C. we add some series resistance and  the first pole moves down.
the last stage reduces its outputZ so what ever C or leakageL it sees, that pole moves uppwards.

quite briliant.  :)


i also think this process  allows for more global nfb aswell.

i enjoyd the discussion alot on this!
 
Back
Top