High frequency roll-off from driver stage Miller Capacitance?

Guest · 1049

0 Members and 1 Guest are viewing this topic.

Deke609

  • Guest
I've been playing around with my SII-45 with the goal of getting it closer to the performance of my Kaiju.

A combination of new iron, tweaks to the parafeed cap values, 10 uF film cap bypasses of the last PSU 100 uF electrolytics, and 130 uF film cathode resistor bypass have gotten me pretty close. But there are two remaining and quite noticeable shortcomings as compared to my Kaiju: rolled-off highs and a narrow soundstage.  I don't care that much about the soundstage, but the rolled-off highs bug me.  And the roll-off issue isn't new - my "stock" SII-45 had the same issue.  The Kaiju has so much more treble and sparkle.

Could this be a function of Miller Capacitance of the 12AT7? I've read that MC can cause high frequency roll-off in triodes and that it is a function of two main things: the output impedance of the previous stage and the gain of the triode.  I measured the frequency response of the two amps a few weeks ago. I don't recall the values, but I remember that high freq. roll-off started considerably earlier (at a lower frequency) with the SII than with the Kaiju. 

Output impedance of preceding stage: I'm using the SII-45 in way that was not intended: as a power amp after the Beepre.  So instead of having the usual super low output impedance of, for example, a DAC feeding the SII-45, I have the Beepre with its (I think) 500 ohms output impedance. 

Gain of the triode driver: The 12AT7 has nearly double the gain of the 5760 - 60 versus 35.

Could one or a combination of the above two factors explain the roll-off I am seeing/hearing? If so, are there any tricks for dealing with this? Or is the only practical solution to switch to 5670s in the SII-45? The Beepre is staying in front of the SII-45.

many thanks in advance,  Derek



Offline Paul Birkeland

  • Global Moderator
  • Hero Member
  • *****
    • Posts: 19319
The gain of the 12AT7 is mitigated a little bit with the unbypassed cathode resistor, so its miller capacitance isn't as significant.  The BeePre is way more than capable of driving ten paralleled 12AT7s, so there isn't much of a concern there. 

I didn't design either amp, so take my opinion with a grain of salt, but it would be my opinion that the Kaiju output transformer is probably helping quite a bit. 

If you wanted to check on these things on your own, you could measure the frequency response of the driver stages of both amps.  Since they are cap coupled and auto biased, the AC signal at the output side of the coupling caps is a safe place to tap into with your scope. 

Paul "PB" Birkeland

Bottlehead Grunt & The Repro Man


Offline Paul Joppa

  • Global Moderator
  • Hero Member
  • *****
    • Posts: 5751
PB was posting at the same time, so there's some overlap. I'll leave my post anyhow, adding that you can measure the response across the OPT primary as well as secondary. That would tell you where it's falling down.

Do the calculation. Miller capacitance is ~gain times grid-plate capacitance, and -3dB is at 1/(2*pi*R*C)
if R is 800 ohms (600 Beepre, 200 grid stopper) and gain is 60 (12AT7) I get 2.2 megahertz.

Do the calc for the 45's capacitance. I get about 200kHz. Output of driver source resistance ~gain times cathode resistor, plus plate resistance (about 15kOhms at the chosen operating point). You can bypass the driver cathode resistor to eliminate that term.

There is some rolloff in the stock output transformer. I didn't measure one after the final design tweak, but calculations show an expected 20mH leakage inductance, which is -3dB at 44kHz, -1dB at 20kHz. In any case, you now have a different OPT.


Paul Joppa


Deke609

  • Guest
Many thanks both.

I'll do the measurements. I think I'll also fix up the SII-2A3 that's been collecting dust, measure it's frequency response, then convert it to 45, do another FR measurement, and then compare both FR measurements to what I'm getting with the 45 amp with the Lundahl headphone OPT. I still haven't tried the LL1679 OPT that is the actual intended OPT for the rebuild. I'll try that last.

cheers and thanks again, Derek