The Kleinschmidt 25A

[GK]

As alluded to early in the K10A thread, here are some teaser peeks of the big brother of my K10A hybrid amp (http://users.picknowl.com.au/~glenk/K10A.HTM).

This design uses On Semi RET devices for the output stage and, as in the K10A, a 12AX7 twin triode for the LTP. Split supply rails are used this time and the load is directly coupled.
Although not shown on the LTspice-drawn schematic attached below, the design incorporates a DC servo to null the offset voltage and has complete DC protection with a power up timer which allows the tube to warm up and the DC operating point to stabilise before closing the speaker relay.

The amplifier is rated at 25W into 8 ohms in class A and 50W into 4 ohms in class AB.
Quiescent current of the output stage is set at 1.3A and idle dissipation is approximately 65W.

Conservatively frequency compensated for a unity loop gain frequency of 1MHz, the simulated linearity is as follows:

25W into 8 ohms
THD-20kHz
0.001928%
THD-1kHz
0.000325%

1W into 8 ohms
THD-20kHz
0.000208%
THD-1kHz
0.000325%

Full power bandwidth is in excess of 200kHz and particular attention was payed to clipping performance.

Within the next 6 months I hope to have a complete prototype built and boards for this design for sale.
Just starting this thread to gauge interest.

Cheers,
Glen

[GK]

And here it is clipping, overdriven with a 20kHz sinewave.....

[Lazy Cat]

I'm impressed, nice design, looks very functional, I mean no nosense or excess components there.

I like the use of filters C1-R11 and C2-R12-R13 as frequency compensation and no Miller caps there. Maybe you should consider to implement current mirror transistors instead of R1 and R2 to have better drive control for the next stage. And at the moment my final remark about discrete voltage regulator for VAS; can not be done properly with some precision IC voltage regulator? Input-output voltage limitation of IC can be easily overcome.

I hope that this is planned to be also a growing project of this DIY thread and good remarks will be included inside.

[MJL21193]

Glen, you're killing me over here! Another one?

[CBS240]

Hi Glen

It appears you have too many ideas that the world needs you to produce. IMHO, the cascode differential VAS is the way to go. Just a couple of notes from my experiences, you want to have fast (matched) transistors for the source/mirror part of the Wilson mirror. Ft greater than 300MHz at least. There are some good quality matched arrays out there these days. This is important as the phase of the mirror current will delay as well as gain droop as frequency increases, limiting bandwidth. This is why I have headed in the symetric, fully complementary direction. Too bad there are no "P-type valves", eh?. In the mirror loaded, cascode differential VAS amp I made, I was able to achieve bandwidth over 250KHz. Although Q5 & 6 is cascode, personally I like them to have equal Pd and included a cascode at GND for the source side of the VAS. I don't think it would make a huge difference though because it is not the amplifying transistor, maybe I'm just a bit cynical. Anyway, nothing really wrong with this circuit that I see.



Should R15 be a pot or in series with a pot for output bias adjustment?


A little different than the way I did it, I used a constant power differential instead of a buffered diff, but to each his own.

Over-all, a solid looking design.

Good Luck

[GK]

Originally posted by Lazy Cat
I'm impressed, nice design, looks very functional, I mean no nosense or excess components there.

I like the use of filters C1-R11 and C2-R12-R13 as frequency compensation and no Miller caps there. Maybe you should consider to implement current mirror transistors instead of R1 and R2 to have better drive control for the next stage. And at the moment my final remark about discrete voltage regulator for VAS; can not be done properly with some precision IC voltage regulator? Input-output voltage limitation of IC can be easily overcome.

I hope that this is planned to be also a growing project of this DIY thread and good remarks will be included inside.

Originally posted by CBS240
Hi Glen

It appears you have too many ideas that the world needs you to produce. IMHO, the cascode differential VAS is the way to go. Just a couple of notes from my experiences, you want to have fast (matched) transistors for the source/mirror part of the Wilson mirror. Ft greater than 300MHz at least. There are some good quality matched arrays out there these days. This is important as the phase of the mirror current will delay as well as gain droop as frequency increases, limiting bandwidth. This is why I have headed in the symetric, fully complementary direction. Too bad there are no "P-type valves", eh?. In the mirror loaded, cascode differential VAS amp I made, I was able to achieve bandwidth over 250KHz. Although Q5 & 6 is cascode, personally I like them to have equal Pd and included a cascode at GND for the source side of the VAS. I don't think it would make a huge difference though because it is not the amplifying transistor, maybe I'm just a bit cynical. Anyway, nothing really wrong with this circuit that I see.

Should R15 be a pot or in series with a pot for output bias adjustment?

A little different than the way I did it, I used a constant power differential instead of a buffered diff, but to each his own.

Over-all, a solid looking design.

Good Luck

OK guys,

Attached below is a refined version. The I-source load for the LTP did virtually nothing for the overall linearity (they just made up for the little decrease in linearity incurred by the LTP cascodes). THD-20 at 25W into 8 ohms now simulates 0.0015%. However the cascodes will make this version more tollerant of a high impedance source, I guess.

I tried both feed-forward and shunt feedback. The shunt feed back compensation scheme that I ended up using (now slightly revised) actually gave 5 times less THD-20, albiet with a lower slew rate. However, the slew rate is still 75V/us, giving the amplifier a theoretical power bandwidth of nearly 600kHz, so that is hardly a problem.

The design doesn't really lend itself to conventional miller compensation around the VAS, so I didn't bother trying it.
However, like miller compensation, shunt compensation is lag compensation.

I could use a floating LM317 or similar for the +150V supply rail, but I prefer to roll my own.

R15 will be a trimpot (in series with a fixed R) to adjust the bias current.

I put a cascode transistor into the sim for Q5 just for something to do, but I will delete this as it really doesn't do anything usefull. The emitters of the VAS cascode transistors (Q5 & Q6) sit at ~130V, so the extra dissipation in Q5 incurred by the added ~25V won't really matter.

Cheers,
Glen


Cheers,
Glen

[CBS240]

Hmmm

It appears the part count has increased a bit.





Just a couple quick Q's...

1)Does the swap mirror work (simulate) that much better than the Wilson?

2)Have you had any thoughts on using MOS devices for Q5 & 6?


As a side note, when I made the constant power diff (PNP's with NPN followers, then cascode) I used a complementary matched array ($8 ). This allowed them to be lower Vce devices since there was no collector to GND connection as with the buffered VAS. I could not really tell you if there is much difference in performance, but a lower Vce allows you more component options.

[Bob Cordell]

Originally posted by G.Kleinschmidt





OK guys,

Attached below is a refined version. The I-source load for the LTP did virtually nothing for the overall linearity (they just made up for the little decrease in linearity incurred by the LTP cascodes). THD-20 at 25W into 8 ohms now simulates 0.0015%. However the cascodes will make this version more tollerant of a high impedance source, I guess.

I tried both feed-forward and shunt feedback. The shunt feed back compensation scheme that I ended up using (now slightly revised) actually gave 5 times less THD-20, albiet with a lower slew rate. However, the slew rate is still 75V/us, giving the amplifier a theoretical power bandwidth of nearly 600kHz, so that is hardly a problem.

The design doesn't really lend itself to conventional miller compensation around the VAS, so I didn't bother trying it.
However, like miller compensation, shunt compensation is lag compensation.

I could use a floating LM317 or similar for the +150V supply rail, but I prefer to roll my own.

R15 will be a trimpot (in series with a fixed R) to adjust the bias current.

I put a cascode transistor into the sim for Q5 just for something to do, but I will delete this as it really doesn't do anything usefull. The emitters of the VAS cascode transistors (Q5 & Q6) sit at ~130V, so the extra dissipation in Q5 incurred by the added ~25V won't really matter.

Cheers,
Glen


Cheers,
Glen

I like your input-load & VAS arrangement a lot. It looks a lot like the one I used in my original MOSFET power amp (different current mirror, though). If you have to go with a unipolar input LTP, I believe this is pretty much the best VAS arrangement. Of course, I used a FET pair instead of the 12AX7.

I had very little trouble with the current mirror phase delay, even in my amplifier where the closed loop bandwidth extended to 2 MHz.

Here's an interesting question. You've included a tube in the VAS and achieved very low THD numbers and wide bandwidth that is the equal or better of many solid state amps. Is there going to really be any discernable "tube sound" lent to this design? If so, why? This is not a criticism, but rather what I consider to be a fascinating question.

Cheers,
Bob

[emanuel_hr]

Wow what an amp! Very nice design, congrats! I have a question about it: what is the purpose of Q12 and Q13(along with D3 and D4)?

[GK]

Originally posted by CBS240
Hmmm

It appears the part count has increased a bit.





Just a couple quick Q's...

1)Does the swap mirror work (simulate) that much better than the Wilson?

2)Have you had any thoughts on using MOS devices for Q5 & 6?


As a side note, when I made the constant power diff (PNP's with NPN followers, then cascode) I used a complementary matched array ($8 ). This allowed them to be lower Vce devices since there was no collector to GND connection as with the buffered VAS. I could not really tell you if there is much difference in performance, but a lower Vce allows you more component options.

The 4-BJT mirror only gives a very small improvement in overall linearity. I have to simulate the design at very low output power(~1W) so that the output stage distortion does not mask the input stage / VAS distortion to see it.

I only like to use MOS devices as cascodes for high voltage VAS stages if the dissipation is too high for a high voltage CRT driver transistor. These BJT's still have the edge WRT to transconductance and junction capacitances.

Originally posted by Bob Cordell
I like your input-load & VAS arrangement a lot. It looks a lot like the one I used in my original MOSFET power amp (different current mirror, though). If you have to go with a unipolar input LTP, I believe this is pretty much the best VAS arrangement. Of course, I used a FET pair instead of the 12AX7.

I picked the biasing voltage off the VAS LTP tail resistor for a larger operating VCE (~10V) on the transistors (incidentally R6 and R7 should be 10k, not 47k). The tail attenuator also ensures less posibility of any AC signal being fed back into the current source transistor bases.

Originally posted by Bob Cordell

Here's an interesting question. You've included a tube in the VAS and achieved very low THD numbers and wide bandwidth that is the equal or better of many solid state amps. Is there going to really be any discernable "tube sound" lent to this design? If so, why? This is not a criticism, but rather what I consider to be a fascinating question.

I doubdt it. Maybe if you turn the volume down and tap the 12AX7 with a screwdriver.

Originally posted by emanuel_hr
Wow what an amp! Very nice design, congrats! I have a question about it: what is the purpose of Q12 and Q13(along with D3 and D4)?

They are a voltage clamp for the VAS to prevent the Wilson current mirror and the NPN predriver from saturating when the amplifier is driven into clipping.
This gives a cleaner clipping recovery.

Cheers,
Glen