An idea I had while doing noise analysis on an amp design:
The idea is that capacitive divider provides a low-noise alternative to resistive feedback network at higher audio frequencies, yet capacitively loads the amp output - however a Zobel network is already present, so it can share that duty as part of a capacitive feedback network.
Note R1 is a major contribution to output noise which the capacitive divider bypasses at higher frequencies.
This relaxes the need to use low value feedback resistors for noise-performance (except perhaps for LF).
The idea is that capacitive divider provides a low-noise alternative to resistive feedback network at higher audio frequencies, yet capacitively loads the amp output - however a Zobel network is already present, so it can share that duty as part of a capacitive feedback network.
Note R1 is a major contribution to output noise which the capacitive divider bypasses at higher frequencies.
This relaxes the need to use low value feedback resistors for noise-performance (except perhaps for LF).
Surely you will try it in real world, but I'm skeptic. Zobel net is a kind of snubber, to equalize high frequency loading to those of the lower ones, mainly with loudspeaker loads that increase their impedance merced an inductive load. But this increasing voltage as frequency tends to infinite, inserted into the NFB loop don't seems to be a good idea, in my opinion. Think what happen, for example, in the amp must drive a line transformer, where the impedances are different (and more complex) than simple voice coils. And the behavior with frequency divider networks?
OK, C1 and C2 form a capacitive divider, but R3 and Ceq. create a major pole in the FB response, and this will impact the stability.
Some amplifiers might tolerate it, but even so the stability margin will be curtailed.
Placing R3 directly in series with C1 (= connecting R2 directly to the output) could solve this issue
Some amplifiers might tolerate it, but even so the stability margin will be curtailed.
Placing R3 directly in series with C1 (= connecting R2 directly to the output) could solve this issue
The snubber is to prevent oscillation - how does combining it with the feedback network change that? The amp oscillates the feedback network will couple that to the input stage, however the divider is implemented.
Huh? the time constant is 2.7ms, about 60Hz, marking the switch over from capacitive to resistive divider, which only affects the noise, assuming the IS isn't loading the network significantly.
R3 and Ceq are the Zobel network. The resistive divider can be ignored at the frequency of interest for the Zobel. It seems to simulate OK, but the phase shift might be an issue. The 10R prevents the capacitive divider from capacitively loading the OS so it seemed natural to use it.
Last edited:
I've realized there's a simple way to cancel the Zobel pole from the POV of the feedback, see the new R4 I've added to the circuit
You posted the wrong picture, but I now realize that my fix is not going to solve anything.
A correctly scaled resistor in series with C2 would do the job, and I imagine that this is the R4 you mention.
A correctly scaled resistor in series with C2 would do the job, and I imagine that this is the R4 you mention.
Novel idea, but having to rely on capacitor tolerances for gain accuracy, channel tracking and FR flatness sounds like a bit of a nightmare to me tbh. C2 sounds like an electrolytic, too, so aging may be another issue.
C2 may be "synthesized" from a 2.2uF and a 0.56uF or a 0.47uF. Probably a small-form polyester box-cap.
It would be much more convenient to massage your circuit to accept either a 3.3uF or a 2.2uF box-caps and be happy with +/-10% tolerances.
May the Spice be with you...
It would be much more convenient to massage your circuit to accept either a 3.3uF or a 2.2uF box-caps and be happy with +/-10% tolerances.
May the Spice be with you...
The exact values are not the point, its just to illustrate the concept. 0.09dB gain matching isn't usually that important, 1% resistors are often used in a feedback network I guess, though there's no deep reason to do so, the standard volume pot taper tracking accuracy is probably the limiting factor in most situations, 2.5% or 5% film cap matching is probably fine unless you are making a lab amplifier!
There are constraints on R4: it cannot be a WW type, because even a few hundreds of nH is going to cause problems in the FB.
Since the dissipation is negligible it is perfectly possible, but that's something to be aware of, because for this kind of low value the first type that comes to mind is wirewound
Since the dissipation is negligible it is perfectly possible, but that's something to be aware of, because for this kind of low value the first type that comes to mind is wirewound
On his NDFL amplifier, Ed Cherry connected the feedback network at the mid point of the Zobel RC :
It is an integral part of the NDFL scheme, and it might even be indispensable in this case, but if you try to do the same with any bog-standard amplifier that has never been designed for it, and already has a marginal stability margin to begin with, you will probably have problems, ranging from HF peaking to outright oscillation.
The Zobel pole could be compensated by a phase-advance cap across the feedback resistor, and that seems to be the case in this example
The Zobel pole could be compensated by a phase-advance cap across the feedback resistor, and that seems to be the case in this example