a quicker link to Mile's cascoded LTP diagram:
http://i40.tinypic.com/2jaghdw.jpg
------------------------------
"This Kulish driver is not well thought out. It looks like a hack someone tossed together carelessly. ........ It should be stricken from the records and forgotten as so much tube folly."
Hehehe.....Thats why I like the circuit, tubies look at it and immediately analyze it as a Darlington and conclude it's Cuckoo. Sorry about the IGBT symbol, I just like it better since it is more readily understandable for analysis. (I did label it as a Mosfet)
You do need to read the earlier linked Kulish Circuit thread to fully understand its operation. It uses feedforward in an elegant way, and achieves around 100X better linearity than a conventional gain stage.
To briefly recap the modus operandi, (ignoring for the moment the effect of R2) R1 current is (Vgate-Vgs)/R1. Then putting R2 back in, R2 current is Vgs/R2. Since R2 is very nearly equal to R1, the combined current in R1 will now be (Vgate-Vgs+Vgs)/R1 or just Vgate/R1. This same current will appear in the load RL (currents from V1 and Q1 summed). So Vout is proportional to Vgate*RL/R1. The current thru R2 is "nearly" constant, since it is the Vthreshold voltage of Q1 plus a small variation for gm activation of Q1. Assuming a constant current thru V1 would just result in a constant voltage across V1's grid to cathode. We are almost there. If Vin is just a constant off from Vgate, then Voutput would be directly proportional to Vinput (AC).
The small variation of Vgs affecting V1 current is the remaining detail. This is where R2 "nearly" = to R1 comes in to play. By misadjusting R2 ever so slightly, we can arrange so that its earlier feedforward correction current to R1 can take into account the slight current variation thru V1 (this assumes a near constant gm of V1 over this limited range). This requires a tweek adjustment of R2 (using a dist. analyzer) until the distortion nulls. Practical results achieved with this circuit using all SS devices have achieved 100X linearity improvements over a simple one device stage. (by the way, there is no Darlington like product of gm's here due to R2's near constant current grossly swamping out V1's gm, it's not a Darlington)
Several topological variations of the circuit exist, including cascode. Device implementation flexibility exists, but factors like bipolar gate currents and finite Beta lead to some adjustments. This circuit uses feedforward correction in a remarkable way, rarely seen implemented in audio. These circuits were originally published in a Russian magazine by one of our members Mikhail Kulish. Last time I checked there were around a half dozen threads in the SS forum referencing this design (well, not the one with a tube it).
Don
http://i40.tinypic.com/2jaghdw.jpg
------------------------------
"This Kulish driver is not well thought out. It looks like a hack someone tossed together carelessly. ........ It should be stricken from the records and forgotten as so much tube folly."
Hehehe.....Thats why I like the circuit, tubies look at it and immediately analyze it as a Darlington and conclude it's Cuckoo. Sorry about the IGBT symbol, I just like it better since it is more readily understandable for analysis. (I did label it as a Mosfet)
You do need to read the earlier linked Kulish Circuit thread to fully understand its operation. It uses feedforward in an elegant way, and achieves around 100X better linearity than a conventional gain stage.
To briefly recap the modus operandi, (ignoring for the moment the effect of R2) R1 current is (Vgate-Vgs)/R1. Then putting R2 back in, R2 current is Vgs/R2. Since R2 is very nearly equal to R1, the combined current in R1 will now be (Vgate-Vgs+Vgs)/R1 or just Vgate/R1. This same current will appear in the load RL (currents from V1 and Q1 summed). So Vout is proportional to Vgate*RL/R1. The current thru R2 is "nearly" constant, since it is the Vthreshold voltage of Q1 plus a small variation for gm activation of Q1. Assuming a constant current thru V1 would just result in a constant voltage across V1's grid to cathode. We are almost there. If Vin is just a constant off from Vgate, then Voutput would be directly proportional to Vinput (AC).
The small variation of Vgs affecting V1 current is the remaining detail. This is where R2 "nearly" = to R1 comes in to play. By misadjusting R2 ever so slightly, we can arrange so that its earlier feedforward correction current to R1 can take into account the slight current variation thru V1 (this assumes a near constant gm of V1 over this limited range). This requires a tweek adjustment of R2 (using a dist. analyzer) until the distortion nulls. Practical results achieved with this circuit using all SS devices have achieved 100X linearity improvements over a simple one device stage. (by the way, there is no Darlington like product of gm's here due to R2's near constant current grossly swamping out V1's gm, it's not a Darlington)
Several topological variations of the circuit exist, including cascode. Device implementation flexibility exists, but factors like bipolar gate currents and finite Beta lead to some adjustments. This circuit uses feedforward correction in a remarkable way, rarely seen implemented in audio. These circuits were originally published in a Russian magazine by one of our members Mikhail Kulish. Last time I checked there were around a half dozen threads in the SS forum referencing this design (well, not the one with a tube it).
Don
Last edited:
Take a look at Joe Curcio's ST70 mod published in Glass Audio. Might also be on his website.
807 - status update
Today I used the pentodes in long tailed pair.
I first tried the EF184, with 22k anode resistors etc like yvesm suggested.
It worked and has plenty of gain (a little too much even) But I was a little short on voltage to run the EF184 optimally (preamp B++ being around 300V)
So I replaced it with EF80 tubes and used 170V on screens and anode. Works like a charm now! (altough I might still raise the B++ voltage a little)
Today I used the pentodes in long tailed pair.
I first tried the EF184, with 22k anode resistors etc like yvesm suggested.
It worked and has plenty of gain (a little too much even) But I was a little short on voltage to run the EF184 optimally (preamp B++ being around 300V)
So I replaced it with EF80 tubes and used 170V on screens and anode. Works like a charm now! (altough I might still raise the B++ voltage a little)
"@smoking-amp, can you post links please? thanks...."
http://www.diyaudio.com/forums/tubes-valves/174792-kulish-circuit.html
http://www.diyaudio.com/forums/solid-state/74861-single-darlington-line-preamp-3.html#post856982
The more general circuit uses an R3 with an R1 not equal to R2. This extension allows further flexibility to match the correction of the first device for not only its average gm slope but for the avg. curvature variation of its gm also, I believe. The RL load impedance would also affect the range of variation of Q1's gm versus V1's gm variation, so is another factor affecting the tracking/correction of V1.
For tube circuitry, the simplified and easier to use version presented above may be adequate for many cases since it allows cancellation of the Mosfet distortion, leaving some tube signature likely. (although tube and Mosfet devices both reasonably approx. square law curvature already due to the additional "Island effect" from tube grids, pushing the order from 3/2 to 4/2, so correction/ tracking of V1 could be decent too. Use of an RL that is more tube like (as opposed to 8 Ohms say) will likely improve the correction tracking for V1. A try it and see design approach.)
I can't read the Russian article, so would greatly appreciate any feedback on additional details or variations. I'm assuming the above simplified R1 "nearly" = R2 approach was used for the cascode design in the article. Further tube variant discussion of the "Kulish Circuit" would maybe best be continued in the "Kulish Circuit" thread, since we appear to be heading off topic here.
Don
http://www.diyaudio.com/forums/tubes-valves/174792-kulish-circuit.html
http://www.diyaudio.com/forums/solid-state/74861-single-darlington-line-preamp-3.html#post856982
The more general circuit uses an R3 with an R1 not equal to R2. This extension allows further flexibility to match the correction of the first device for not only its average gm slope but for the avg. curvature variation of its gm also, I believe. The RL load impedance would also affect the range of variation of Q1's gm versus V1's gm variation, so is another factor affecting the tracking/correction of V1.
For tube circuitry, the simplified and easier to use version presented above may be adequate for many cases since it allows cancellation of the Mosfet distortion, leaving some tube signature likely. (although tube and Mosfet devices both reasonably approx. square law curvature already due to the additional "Island effect" from tube grids, pushing the order from 3/2 to 4/2, so correction/ tracking of V1 could be decent too. Use of an RL that is more tube like (as opposed to 8 Ohms say) will likely improve the correction tracking for V1. A try it and see design approach.)
I can't read the Russian article, so would greatly appreciate any feedback on additional details or variations. I'm assuming the above simplified R1 "nearly" = R2 approach was used for the cascode design in the article. Further tube variant discussion of the "Kulish Circuit" would maybe best be continued in the "Kulish Circuit" thread, since we appear to be heading off topic here.
Don
Last edited:
It's always a good idea to use as hi B+ as you can, even for lo level stages !
The only problem is to "clean" it specially if you have choosen a configuration having poor PSRR.
Void cascodes and use pentodes ! Just their screens should be clean.
My half penny.
Yves.
Common mode rejection ratio..
And the ease of the circuit. If you know a better way, teach me
Why should cascodes be avoided by your opinion?
I think it works very well with the ECC88, and perhaps the ones with fets on bottom too (altough i still havent tried that yet)
pentodes add more (odd harmonics) distortion while a decent designed cascode with triodes adds very little distortion over the whole spectrum. (at least thats what the book says)
Don't bother too much, I was joking . . . somewhat . . . cos it's true that cascode has vy lo PSRR and asks for vy clean supplies !
Yes, pentodes add more odd harmonics when the load line enters in the curved region near the knee, but it's not so difficult to stay away in a lo level stage where efficiency is seldomly an issue.
Using more plate current and voltage along with a lower plate load resistor may help if you accept to have'em a bit hotter and to loose some gain.
This "pushes" the load line away from the knee.
However, like triodes, they produce even harmonics also that should be cancelled in a symetrical design as the LTP is.
You have , in theory, the maximum common mode rejecton because you are using a balanced signals. That's all.
You can use a simple two stage, one for gain and another to drive (with a little gain also or CF) the output stage.
At this link there is a simple circuit (input) with 6SL7 and 6SN7 to drive 4 pair of EL34:
http://www.multitask.it/bigamp/ingresso.bmp
the trafo in input is a 3575 from Sowter, but for use with balanced is omitted.
This is the ouput:
http://www.multitask.it/bigamp/uscita.bmp
the secondary is floating with a reference to ground with two resistors for feedback.
This project is based on the ouput trafo ( double C core); without feedback the BW is around 60 KHz, with the 20 Hz at less than 1 dB. With 16 dB of FB we reach about 120KHz (100kZ at 140 Wrms).
I test also a trafo with secondary with center tap; may be easy to get the FB signal
These are the photos for the article I wrote for Audioreview in Italy:
http://www.multitask.it/bigamp/apertura1.jpg
My suggestion is to spend time and money mainly for the output trafo instead to test different circuits where ( and this can be the case) you don't need complcations.
It is only my opinion
Ciao
Walter
You can use a simple two stage, one for gain and another to drive (with a little gain also or CF) the output stage.
At this link there is a simple circuit (input) with 6SL7 and 6SN7 to drive 4 pair of EL34:
http://www.multitask.it/bigamp/ingresso.bmp
the trafo in input is a 3575 from Sowter, but for use with balanced is omitted.
This is the ouput:
http://www.multitask.it/bigamp/uscita.bmp
the secondary is floating with a reference to ground with two resistors for feedback.
This project is based on the ouput trafo ( double C core); without feedback the BW is around 60 KHz, with the 20 Hz at less than 1 dB. With 16 dB of FB we reach about 120KHz (100kZ at 140 Wrms).
I test also a trafo with secondary with center tap; may be easy to get the FB signal
These are the photos for the article I wrote for Audioreview in Italy:
http://www.multitask.it/bigamp/apertura1.jpg
My suggestion is to spend time and money mainly for the output trafo instead to test different circuits where ( and this can be the case) you don't need complcations.
It is only my opinion
Ciao
Walter
I myself dont like the use of the input transformer
I do like to experiment with different circuits. The building of the amp and knowledge gained while doing so is more important to me right now then the date when it's finished or how it looks when it is. The sound offcourse also matters.
My initial question has been answered so I guess the mission of this thread has been accomplished
Hi smoking-amp, sorry to raise a very old thread, have you ever tried it in cascode configuration? Thanks, Roberto
This is a little hard to describe, but an often overlooked fact.
1. For a push pull amplifier, if there is common mode ripple in an early stage, it is amplified by all the later stages.
The problem is when the output stage gets that common mode ripple along with the signal to the control grids, the plate currents now have that same common mode ripple.
That drives the push pull output transformer.
Since the common mode ripple is the same phase from the output tube plates, that means the plates deliver that into a very low impedance, across the whole primary.
The common mode ripple is cancelled in the primary, and will not appear at the secondary. But the output tube plates are struggling to drive that across the DCR of the primary.
Drive a common mode signal across the primary, and the primary halves act as a short circuit, only the DCR of the windings is left.
A real waste of power.
And, if it is large enough, there may be intermodulation of the ripple with the music.
If I get the time and energy, I will draw a simple schematic that illustrates the problem.
The cure is to get rid of the common mode ripple in the early stages, before it is amplified all the way to the output tube grids.
2. The above problem, is totally different than ripple on the B+ that powers the primary center tap of the output tubes.
1. For a push pull amplifier, if there is common mode ripple in an early stage, it is amplified by all the later stages.
The problem is when the output stage gets that common mode ripple along with the signal to the control grids, the plate currents now have that same common mode ripple.
That drives the push pull output transformer.
Since the common mode ripple is the same phase from the output tube plates, that means the plates deliver that into a very low impedance, across the whole primary.
The common mode ripple is cancelled in the primary, and will not appear at the secondary. But the output tube plates are struggling to drive that across the DCR of the primary.
Drive a common mode signal across the primary, and the primary halves act as a short circuit, only the DCR of the windings is left.
A real waste of power.
And, if it is large enough, there may be intermodulation of the ripple with the music.
If I get the time and energy, I will draw a simple schematic that illustrates the problem.
The cure is to get rid of the common mode ripple in the early stages, before it is amplified all the way to the output tube grids.
2. The above problem, is totally different than ripple on the B+ that powers the primary center tap of the output tubes.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.