Playing with the DH-120 in simulation, the ziener stack across the gates is causing about 10dB of the output distortion. Two questions. I am not really sure what protection this provides as all three nodes will under normal conditions never exceed a couple volts difference, so why 20V clamp? Second, why is the center tied to the output? Is this a case of a little bit of protection is a bad thing? Lastly, its it just the capacitance that is causing the issue? The outputs already have the same network internally. So, redundant as well.
I thought I found another "euraka" in the base stoppers to the drivers, but what I really found was a lot more about how LTSpice FFT works. Kind of funny actually as I have seen posts that I know know clearly are not correct.
Tiny adjustments to the gate stoppers makes pretty big differences in balancing the standing bias current and in even order distortion. 420 and 330 seem to be the magic values in the sim. Higher "nominal" values greatly increase distortion. So do lower. Notice the ration is not what the OEM suggests. Until I test n the bench, all conjecture.
Removing the 27 Ohm resistor in the emitter of the VAS cascode as it is no longer part of a pole-zero helped by a couple of dB.
Tiny adjustments to the gate stoppers makes pretty big differences in balancing the standing bias current and in even order distortion. 420 and 330 seem to be the magic values in the sim. Higher "nominal" values greatly increase distortion. So do lower. Notice the ration is not what the OEM suggests. Until I test n the bench, all conjecture.
Removing the 27 Ohm resistor in the emitter of the VAS cascode as it is no longer part of a pole-zero helped by a couple of dB.
Cherry recommends the VAS degen R for improved stability in simailar BJT output amps
it is also necessary for the "correct" cascode bias AC return which bcarso has been advocating - hanging the cascode bias LED from the degenerated VAS emitter will likely give bigger distortion improvement at high frequencies - the Hawksford article shows the variations, measures distortion improvement - even if Hawksford didn't invent the circuits
it is also necessary for the "correct" cascode bias AC return which bcarso has been advocating - hanging the cascode bias LED from the degenerated VAS emitter will likely give bigger distortion improvement at high frequencies - the Hawksford article shows the variations, measures distortion improvement - even if Hawksford didn't invent the circuits
Gate stoppers shouldn't change the bias current at all. In fact I doubt that most simulators will give any useful information about gate stoppers, unless they include stray and package inductances that result in high frequency (~100's of MHz) resonances.
The return of the upper two bias LEDs to the 27 ohm, as mentioned by jcx, may have some benefit, but the effectiveness of this in reducing C at the output of the second stage cascode is mitigated somewhat by the ratio of the upper Q emitter impedance to the 27 ohms. Since we're running a little under 3mA, that emitter impedance is around 9 ohms, so the signal returned from the lower collector is divided between the 27 ohms and that 9 ohms. That output node will have voltage-dependent capacitances from the Ccb of the emitter followers, the reflected output Z of the output FETs, and the voltage-dependent C of the current source. So there will be some potential improvement from an "Aldridge" (aka Hawksford) connection, but it won't be tremendous.
Beware as well of zener models. My sim program doesn't model zeners well at all --- they are made to look like voltage sources regardless of bias! <searches for Smilie for dumbs#it>
BTW, given that the second stage's maximum positive swing is limited to of order three volts below the positive rail, there's little reason to have the lower current source swing any closer to the negative rail. And with that in mind, although the Jaeger JFET/bipolar cascode source used for the input diff pair takes up too much voltage burden, a hybrid using your existing "ring of two" source with a driven cascode (one LED bias sufficient) will work quite nicely to raise the current source impedance. Again, how much the following loads will make this less important I don't know offhand, and if one is hell-bent on improvement of everything, probably adding an emitter follower to the second stage to make it a better integrator is a better place to spend a transistor.
The nice thing about the cascoded ring-of-two, though, is that the ring of two's local feedback makes the impedance at the emitter-210 ohm R junction really low
. So virtually all of the cascode device's base current change gets pumped back, as opposed to the situation mentioned with the second stage amp, where such currents are split between the emitter Z and the 27 ohms. Aside from the temperature coefficient of the output current being substantial (which in some situations may even be an advantage), this is a three-transistor circuit with remarkably high performance.
Brad
The nice thing about the cascoded ring-of-two, though, is that the ring of two's local feedback makes the impedance at the emitter-210 ohm R junction really low
. So virtually all of the cascode device's base current change gets pumped back, as opposed to the situation mentioned with the second stage amp, where such currents are split between the emitter Z and the 27 ohms. Aside from the temperature coefficient of the output current being substantial (which in some situations may even be an advantage), this is a three-transistor circuit with remarkably high performance.Brad
I quite like the idea of simply adding a 3 Ohm power resistor in series with the output, since it also provides very effective protection for the output devices. Worst case, with a short across the speaker terminals, the output stage just has to drive the resistor, which should be no problem.
Overall efficiency is reduced a bit and maximum output is about 3dB less, but I don't consider that too much of a problem. If high efficiency and power savings were a top priority nobody would be using valves or class A.
YMMV, Different strokes for different folks etc.
Thanks! That helps no end.
The circuit looks like it will have very high output impedance, but lousy PSRR as shown, since any ripple current through R2 is fed straight through to the output. One could always add some filtering or regulation for R2's voltage though.
edit: Oops, crosspost with janneman
Attachments
Hi,
Me too, freedom from any dictatorships, be they of the Marketing Department, the Proletariat or the IMF...
Ciao T
Freedom from dictature of marketing departments, what to measure, and how to judge.
Me too, freedom from any dictatorships, be they of the Marketing Department, the Proletariat or the IMF...
Ciao T
Hi,
I was more thinking of adding 0.22R in the ground lead and using some clever feedback circuit to make it appear to be 3 Ohm...
Given that our VAS has a cascode as well, we can make both cascodes identical and use a 2-terminal CRD (current regulator diode) plus a suitable resistor to take up excess dissipation and massively increase the PSRR of the Cascode.
For the ring of 2 base voltage, same principle and as we have to bias some diode strings to provide clamping and limiting for best overload recovery we can use this "waste" current to bias the diode strings.
Then again, if we cascode the Load CCS, why not replace the resistor in the original CCS with a J-Fet CCS and use the original CCS Transistor as cascode? A BF862 should able to work OK at such low voltages...
Ciao T
I was more thinking of adding 0.22R in the ground lead and using some clever feedback circuit to make it appear to be 3 Ohm...
Given that our VAS has a cascode as well, we can make both cascodes identical and use a 2-terminal CRD (current regulator diode) plus a suitable resistor to take up excess dissipation and massively increase the PSRR of the Cascode.
For the ring of 2 base voltage, same principle and as we have to bias some diode strings to provide clamping and limiting for best overload recovery we can use this "waste" current to bias the diode strings.
Then again, if we cascode the Load CCS, why not replace the resistor in the original CCS with a J-Fet CCS and use the original CCS Transistor as cascode? A BF862 should able to work OK at such low voltages...
Ciao T
I lost track of who was it that did the testing of those CCS versions. Whoever it was, please test these two as well, the same way the others were tested.
One is a classic resistor-zener in parallel with capacitor-resistor; quite a few of those are still floating around.
The other is more interesting, and I for one am hoping it will return solid results.
Thank you in advance.
One is a classic resistor-zener in parallel with capacitor-resistor; quite a few of those are still floating around.
The other is more interesting, and I for one am hoping it will return solid results.
Thank you in advance.
Glad you asked
Actually, the ring of two is not very good for PSRR, true; but the Aldridge-cascoded version allows the tuning of the LED's (or other diodes') bias R and the "error amp" bias R to provide excellent PSRR. With high-beta ~2SC1815 parts for the example I gave, and with tvr's 210 ohms in the emitter, using 9.53k for the error amp and 210k for the LED is about optimal --- the residue is nearly pure 2nd harmonic of the rail signal at midband frequencies. I'll publish the examples and performance results in a bit. BTW, I've not used the ring-of-two that often, and tvr's developing design was the impetus for considering the Aldridge cascode version and a very pleasant surprise.Of course we still are focusing on high impedance, not on stability, which has the Vbe temperature coefficient of ~ -2mV/degree K, hence about -0.33%/degree K for a 600mV Vbe. And noise is not outstanding, since the "reference voltage" is Vbe and thus the emitter resistor is fairly small. However, at this point in the circuit it should hardly dominate.
Brad
Hi,
I would like to throw in a monkey wrench from the leftfield.
Why not keep a simple, straightforward ring of 2 CCS, but bootstrap it instead to output?
I can think of several options that could accomplish that, but why not a DC shifted PNP emitter follower and taking the current for the error amp transistor from the output?
I think the PSRR as well as effective impedance should be quite horrorshow*.
Ciao T
* Horrorshow - nadsat slang for "good" from "A Clockwork orange" - derived the cockney way from the russian хорошо (kha-ra-SHO)
I would like to throw in a monkey wrench from the leftfield.
Why not keep a simple, straightforward ring of 2 CCS, but bootstrap it instead to output?
I can think of several options that could accomplish that, but why not a DC shifted PNP emitter follower and taking the current for the error amp transistor from the output?
I think the PSRR as well as effective impedance should be quite horrorshow*.
Ciao T
* Horrorshow - nadsat slang for "good" from "A Clockwork orange" - derived the cockney way from the russian хорошо (kha-ra-SHO)
Hi,
Yes, remembering what little russian remains (I learned it in school and spend halve a year in Moskow in '87) I'm rather fond of using it.
Ciao T
Horrowshow is better transcription for хорошо than kha-ra-SHO, actually. However, oh is pronounced lke ah, it is right.
Yes, remembering what little russian remains (I learned it in school and spend halve a year in Moskow in '87) I'm rather fond of using it.
Ciao T
My daughter speaks German pretty well, she learns it in school. Tomorrow we are meeting an exchange student from Germany. 2 years after will be her turn to go to Germany...
- Status
- Not open for further replies.