drats, missed the edit window. hopefully this works and points to the link I described above: https://www.diyaudio.com/community/threads/bob-cordell-interview-bjt-vs-mosfet.101745/post-1268735
ok, that's not the original post from Hansen, but this one is: https://www.diyaudio.com/community/threads/bob-cordell-interview-bjt-vs-mosfet.101745/post-1268280
ok, that's not the original post from Hansen, but this one is: https://www.diyaudio.com/community/threads/bob-cordell-interview-bjt-vs-mosfet.101745/post-1268280
I only know neutralization (or neutrodynisatie in Dutch) as a technique invented by Louis Hazeltine in 1923 to compensate for the effect of anode-to-grid capacitance, so a common-cathode amplifier stage with LC tanks connected to its grid and anode would not turn into a tuned-plate, tuned-grid oscillator. It is more than a century old, but it is still used a lot in RF electronics. I don't know how to apply it to an emitter follower, though.
The idea of neutralization is that you add an extra capacitor to the grid, base or gate that gets driven by a signal that is in antiphase with the anode, collector or drain voltage, so it cancels out the current through the anode-to-grid, collector-to-base or drain-to-gate capacitance.
On the thread I linked to some posts ago, @keantoken made an interesting remark about the frequency-dependence of the conductance of an RC series network connected in parallel with the amplifier input being precisely what you need to correct for the amplifier's negative input conductance. I think that applies equally in this case, see the calculation.
Investigating Zobel network.
All below are simulating without the speaker load. Somehow a tiny capacitor is added to the output.
Putting RC on the amp side, not stable.
Putting RC on the speaker side, a little overshoot, not that bad.
Without RC, not stable.
Without the inductor, only RC, it is marginal stable.
People keep copying and pasting Zobel, but mostly are wrong, including me.
All below are simulating without the speaker load. Somehow a tiny capacitor is added to the output.
Putting RC on the amp side, not stable.
Putting RC on the speaker side, a little overshoot, not that bad.
Without RC, not stable.
Without the inductor, only RC, it is marginal stable.
People keep copying and pasting Zobel, but mostly are wrong, including me.
That’s the next thing I am going to investigate.
You have to separate out local circuit instability and associated mitigation from loop stability issues. They are separate but may interact with each other if not done carefully. A Zobel placed directly at the amplifier output, base stoppers and RC dampers as in MvG post above address local instability issues while and output coil is there to prevent pole migration in the OPS which is a loop stability issue.
The Zobel in normally placed before the output inductor and the associated loop areas around the OPS and thus inductances kept as small as possible.
I haven’t put global NFB yet. Most Miller compensated amps have very low output impedance before closing NFB. RC with 10 Ohm does little to nothing. If an amp requires RC before the coil. The R often is less than 10. You would see something like 2.2, 3.3, instead of a generic 10 Ohm.
The problem is without an output coupling coil, any capacitive load will cause the OPS HF pole to migrate downwards. This pole will usually sit at 4 or 5 MHz in the MC amp and the ULGF at 1 to 1.5 MHz so it doesn’t take much load capacitance. The Zobel must be placed before the output L and plays little role In stabilising the feedback loop - it’s primarily there as part of ensuring local OPS stability from HF parasitic oscillation.
Are you hoping to be able to draw conclusions from RF -> HF -> LF into the audio frequency range? Two completely different worlds and (n-pole) models.
Good luck and have fun with the simulations,
HBt.
🍿📽️
For the output stage, the inductor combine with RC with correct order (the 2nd case in #26) solve the issue. What's better is that it works almost regardless the size of the capacitor load.
However, the corner frequency of the LC network is only about hundreds KHz. It is not suitable to put this network between the stage.
In the case of the "Driver" stage, we can put a RC network at the base of the transistor. Please see the example below. 1MHz square wave into 600pF. I found C3 is not critical, you can think C3 is just a DC blocking cap. Anything over 100nF should be fine. As you see, the R1, 20 Ohm is not optimized in my case, as the wave form is overdamped. The optimal value of R1 depends on the size of the capacitive load (C1) and the transistor internal capacitance Cbe.
The R is essentially paralleled to the LC network.
If you know C1, C2, and L, you can calculate the maximum R allowed by the given target damping factor.
From the equation above, you see the more L you have, the more damped it gets. This property sometimes is more desirable.
Reference: https://en.wikipedia.org/wiki/RLC_circuit#Parallel_circuit
Nice to see that your simulator interprets 1F as 1 farad. Many SPICE-based simulators are not case sensitive and would interpret it as 1 femtofarad, or 0.001 pF.
I think you run into the same issue as I would with my sufficient condition for stability from post #24: the impedance of the RC network gets rather low, loading whatever has to drive the input. With a 600 pF load and fT = 30 MHz, my values would be at least 600 pF and at most 8.84194 ohm, yours are 1 F and 20 ohm.
I think you run into the same issue as I would with my sufficient condition for stability from post #24: the impedance of the RC network gets rather low, loading whatever has to drive the input. With a 600 pF load and fT = 30 MHz, my values would be at least 600 pF and at most 8.84194 ohm, yours are 1 F and 20 ohm.
Practically speaking, it is not a good idea to drive an EF with from a source impedance that raises with frequency, such as an inductor or an opamp output. Approximately, the output impedance of an EF is the source impedance divided by h21e. The latter falls with frequency at about 20dB/decade - here is 2N3904 as an example:
so with a capacitive source (such as a Miller-compensated common-emitter stage), an EF exhibits roughly stable output impedance over frequency; with a resistive source, the output impedance raises with frequency at 20dB/decade (that is, behaves as an inductor), and with an inductive source, goes up at 40dB/decade with the corresponding phase shift. It is possible, as discussed above, to damp the resulting oscillator, but usually it is easier to adjust the output impedance of a preceding stage. Often, a small capacitor to the ground or suppy rail is sufficient. BTW the problem becomes even more interesting with double and triple EF stages commonly found at the output of audio power amplfiiers.
so with a capacitive source (such as a Miller-compensated common-emitter stage), an EF exhibits roughly stable output impedance over frequency; with a resistive source, the output impedance raises with frequency at 20dB/decade (that is, behaves as an inductor), and with an inductive source, goes up at 40dB/decade with the corresponding phase shift. It is possible, as discussed above, to damp the resulting oscillator, but usually it is easier to adjust the output impedance of a preceding stage. Often, a small capacitor to the ground or suppy rail is sufficient. BTW the problem becomes even more interesting with double and triple EF stages commonly found at the output of audio power amplfiiers.
Here is the extreme example.
EF3 with 10 pairs output power transistors distributed on 4 separated boards. Imagine the driver stage has to drive into 10 Cbc in parallel. It is miracle that the original amp could be stable. That design along prevents anyone from cloning it.
Here are the schematics and information: GFA-565/585 SCH and more.
Mooly: If I short the probe tip to the ground, no unwanted signal appears.
The driver board components are based on the EBFA-565. The output driver stage operates in Class A, similar to the GFA-585.
Mooly: If I short the probe tip to the ground, no unwanted signal appears.
The driver board components are based on the EBFA-565. The output driver stage operates in Class A, similar to the GFA-585.
The output impedance of a Miller-compensated common-emitter stage has a rather complicated frequency dependence. Neglecting base resistance, you first have the output resistance of the common-emitter stage, then the Miller compensation kicks in and it becomes capacitive, then the current gain starts to drop and the impedance becomes resistive again, then the series connection of the Miller capacitance and the base-emitter capacitance starts to dominate and it becomes capacitive again. With base resistance, you can even get an inductive part.
Why do people use emitter degeneration on a Miller-compensated common-emitter stage? It only worsens the effect of the right-half-plane zero. Rudimentary type of current limiting?
Why do people use emitter degeneration on a Miller-compensated common-emitter stage? It only worsens the effect of the right-half-plane zero. Rudimentary type of current limiting?
The VAS is usually current-driven, so a resistor at the emitter of the VAS can be equivalent to a resistor at it's collector after the miller cap (impedances in series do not care about order). So the degeneration resistor can effectively be a base stopper for the EF drivers. Said another way, the VAS output resistance without degeneration is often too low to be optimal for the EF. Current limiting can be a beneficial side effect.
The VAS is usually current-driven, so the voltage gain reduction of degeneration has no effect (well it can increase distortion). The RHP zero occurs at frequencies far too high to matter in most cases I've seen.
Most VAS transistors will have low base resistance but if you use a KSC1845 or BC550C or something like that it will become inductive at a much lower frequency (~Ft*Rm/Rb).
The VAS is usually current-driven, so the voltage gain reduction of degeneration has no effect (well it can increase distortion). The RHP zero occurs at frequencies far too high to matter in most cases I've seen.
Most VAS transistors will have low base resistance but if you use a KSC1845 or BC550C or something like that it will become inductive at a much lower frequency (~Ft*Rm/Rb).