Just so we don't invade Mafioso's thread at http://www.avforums.co.za/index.php/topic,3265.0.html even further.
As suggested by Ampdog, a topic on high frequency loading of amps. I think one can also consider other stability issues (loop compensation) with this, as the same parts will burn if one gets it wrong, for somewhat similar reasons.
This is where I wish again that I had a real education. Lots of questions. But you expect long posts from me by now, so I'll sommer start with a whole lot of them. I hope it's not so many questions that nobody wants to answer any.
I never understood why amps need high frequency loading. As the speaker impedance rises to higher frequencies (though it's not pure inductance, yet another topic) obviously the load is getting easier to drive, so why?
Still, remove the Zobel from the amp and you have a fire after a short period of oscillation. Been there by accident, had to replace the output stage etc... :-[
So my assumption was that capacitance parallel to the inductive load is the killer. It forms a tank circuit with the accompanying ?phase inversion" around the response peak. If this is a high (ultrasonic) frequency, the amp has little gain/feedback/control there. On the peak, the load is very resistive and there is no phase shift. But next to the peak all hell breaks loose in terms of phase. Hence the chaos.
Adding the Zobel, lowers the impedance at high freq, so "spoiling/flattening" the high Q of the LC tank circuit formed by the inductive load and stray capacitance. It also lowers the frequency of the peak by adding more capacitance. Ideally, down to a frequency where the amp can control matters. So it would relate to things like open loop gain and phase margin as well.
Am I on the right track?
The next question. If we remove the Zobel RC network, amps can still burn without any load on their outputs. Is it stray capacitance on the board that does this by the same mechanism as above? Like voltage dependent capacitance in the semiconductor junctions combined with the components inductance? Or am I splitting hair now, while missing a much more important point?
Itty-bitty TO-92 amps don't misbehave like this. Due simply to less internal capacitance and much higher Ft in the devices? Thus no need for Zobel networks on the outputs of all our op-amp circuits? One hardly ever sees Zobel networks on opamp outputs, just resistive isolation from capacitive loads.
From the previous thread:
Ampdog, care to write a whole lot more about this? Firstly: Why is 45deg particularly significant in the HF loading? Secondly: How does it relate to the behaviour of C.dom?
Concerning C.dom, I understand that you prefer not to have phase shift in the open loop gain of the amp over the audio band, and it made sense to me. (In as far as that means anything!) I take it then that you use some other method than C.dom? Do you compensate with a cap from the VAS (or the output) to the inverting input rather, with perhaps a little something more in the input stage to make sure you get to below unity gain at some high freq point? Or is this quite another matter that should not be confused with the HF loading issue (by the Zobel network) at all?
Your mentioning low value resistors in the NFB loop has explained something to me of why my amp works. Mine is a slightly odd collection of hackery, and I never really thought about that part of it. Thanks! (My Zobel network is 8.2 Ohm and 10nF, cap on the ground side, for a corner at 2MHz, and NFB picked off from the junction of the two. But Rf & Rg are also low values, 2k2 and 47R.) Insanely enough, it appears to be stable.
I realise that you may not want to make your whole design public. Fair enough. But some pointers to straighten my thinking would be a great help.
As suggested by Ampdog, a topic on high frequency loading of amps. I think one can also consider other stability issues (loop compensation) with this, as the same parts will burn if one gets it wrong, for somewhat similar reasons.
This is where I wish again that I had a real education. Lots of questions. But you expect long posts from me by now, so I'll sommer start with a whole lot of them. I hope it's not so many questions that nobody wants to answer any.
I never understood why amps need high frequency loading. As the speaker impedance rises to higher frequencies (though it's not pure inductance, yet another topic) obviously the load is getting easier to drive, so why?
Still, remove the Zobel from the amp and you have a fire after a short period of oscillation. Been there by accident, had to replace the output stage etc... :-[
So my assumption was that capacitance parallel to the inductive load is the killer. It forms a tank circuit with the accompanying ?phase inversion" around the response peak. If this is a high (ultrasonic) frequency, the amp has little gain/feedback/control there. On the peak, the load is very resistive and there is no phase shift. But next to the peak all hell breaks loose in terms of phase. Hence the chaos.
Adding the Zobel, lowers the impedance at high freq, so "spoiling/flattening" the high Q of the LC tank circuit formed by the inductive load and stray capacitance. It also lowers the frequency of the peak by adding more capacitance. Ideally, down to a frequency where the amp can control matters. So it would relate to things like open loop gain and phase margin as well.
Am I on the right track?
The next question. If we remove the Zobel RC network, amps can still burn without any load on their outputs. Is it stray capacitance on the board that does this by the same mechanism as above? Like voltage dependent capacitance in the semiconductor junctions combined with the components inductance? Or am I splitting hair now, while missing a much more important point?
Itty-bitty TO-92 amps don't misbehave like this. Due simply to less internal capacitance and much higher Ft in the devices? Thus no need for Zobel networks on the outputs of all our op-amp circuits? One hardly ever sees Zobel networks on opamp outputs, just resistive isolation from capacitive loads.
From the previous thread:
Ampdog said:
Ampdog, care to write a whole lot more about this? Firstly: Why is 45deg particularly significant in the HF loading? Secondly: How does it relate to the behaviour of C.dom?
Concerning C.dom, I understand that you prefer not to have phase shift in the open loop gain of the amp over the audio band, and it made sense to me. (In as far as that means anything!) I take it then that you use some other method than C.dom? Do you compensate with a cap from the VAS (or the output) to the inverting input rather, with perhaps a little something more in the input stage to make sure you get to below unity gain at some high freq point? Or is this quite another matter that should not be confused with the HF loading issue (by the Zobel network) at all?
Your mentioning low value resistors in the NFB loop has explained something to me of why my amp works. Mine is a slightly odd collection of hackery, and I never really thought about that part of it. Thanks! (My Zobel network is 8.2 Ohm and 10nF, cap on the ground side, for a corner at 2MHz, and NFB picked off from the junction of the two. But Rf & Rg are also low values, 2k2 and 47R.) Insanely enough, it appears to be stable.
I realise that you may not want to make your whole design public. Fair enough. But some pointers to straighten my thinking would be a great help.
