Thank you.
I'm just pleased it works. It does go to show that simulating first in detail is a worth while activity.
Now its on to making a second channel. Can't really appreciate the sound properly with only half the music (and on "disposable" kef coda 7s.)
Once funds allow, will get some test equipment to allow distortion measurements to be taken.
Edit: and once I can get decent pics of the scope outputs I will post square wave and sine waves.
I'm just pleased it works. It does go to show that simulating first in detail is a worth while activity.
Now its on to making a second channel. Can't really appreciate the sound properly with only half the music (and on "disposable" kef coda 7s.)
Once funds allow, will get some test equipment to allow distortion measurements to be taken.
Edit: and once I can get decent pics of the scope outputs I will post square wave and sine waves.
Last edited:
Now that I have a working prototype amp. It's maybe time to experiment a little. Regarding PM I have noticed that the output devices are the main contributor to the "excess" phase. It it the pole formed by the gate capacitance and gate resistors that is the problem. However, reducing the gate resistors results in oscillation in the MHz. Would ferrite beads on the MOSFET gates be an option?
You cannot fix the gate stopper phase using ferrite beads. Beads act like L/R circuits, and to prevent gate ringing the LR corner needs to be below the RC corner of the gate capacitance. Therefore you end up gaining phase only below the corner.
The only way I know of to speed up the Mfets is to reduce the gate stoppers. Instead of swamping out parasitics, making the frequencies in question unavailable for increasing loop gain, you decrease parasitics using ground plane and snubbing techniques.
The only way I know of to speed up the Mfets is to reduce the gate stoppers. Instead of swamping out parasitics, making the frequencies in question unavailable for increasing loop gain, you decrease parasitics using ground plane and snubbing techniques.
Ground planes, I'm scared of those at present. Not enough understanding of them. It's a local problem as with one pair of FETs I could use much smaller gate resistances.
What sort of snubbing techniques could I use? This is pushing my understanding again.
I understand the need for reducing the gate resistors. This would raise the RC corner. But how to deal with the oscillations is a problem. they were high frequency somehwere around 10Mhz. They were 4V peak to peak.
Edit: If i can speed up the MFETs then I could maybe push the amplifier compensation a little further.
What sort of snubbing techniques could I use? This is pushing my understanding again.
I understand the need for reducing the gate resistors. This would raise the RC corner. But how to deal with the oscillations is a problem. they were high frequency somehwere around 10Mhz. They were 4V peak to peak.
Edit: If i can speed up the MFETs then I could maybe push the amplifier compensation a little further.
Last edited:
10MHz sounds like a feedback loop thing. Evidently your feedback loop is fast enough it's trying to correct the RC corner of the Mfets. Thus when you change the RC corner the compensation goes out of whack. Even if you could get it stable, the varying capacitance of the Mfets would make the amp conditionally stable across a range of operating points.
Lowering gate stoppers moves this problem further away from the problem frequencies, but makes local oscillation more likely. Increasing gate stoppers actually makes the first problem worse however, so decreasing local instability is the only way to really speed up the amp all things considered.
Lowering gate stoppers moves this problem further away from the problem frequencies, but makes local oscillation more likely. Increasing gate stoppers actually makes the first problem worse however, so decreasing local instability is the only way to really speed up the amp all things considered.
Really not sure how to proceed with this. Not happy with my current base stopper values. Would retuning the gate snubbers to the frequency seen may be a way forward.
Or maybe its counter intuitive but if I lower the gate resistors instead of raising them from their original values I shift the problem frequency higher and maybe away from the feedback problems. Then I may hit true local oscillation problems which could be dealt with by the gate snubbers.
It's just strange how the problem surfaced when the 2nd pair of MFETs were added
The power rails were not oscillating. I did check them.
Edit: I'm going to have a think about this while at work tomorrow. One's mind wanders and generally comes to some solutions (or at least mine does
)...
Or maybe its counter intuitive but if I lower the gate resistors instead of raising them from their original values I shift the problem frequency higher and maybe away from the feedback problems. Then I may hit true local oscillation problems which could be dealt with by the gate snubbers.
It's just strange how the problem surfaced when the 2nd pair of MFETs were added
The power rails were not oscillating. I did check them.
Edit: I'm going to have a think about this while at work tomorrow. One's mind wanders and generally comes to some solutions (or at least mine does
)...
Last edited:
Really not sure how to proceed with this. Not happy with my current base stopper values. Would retuning the gate snubbers to the frequency seen may be a way forward.
Or maybe its counter intuitive but if I lower the gate resistors instead of raising them from their original values I shift the problem frequency higher and maybe away from the feedback problems. Then I may hit true local oscillation problems which could be dealt with by the gate snubbers.
It's just strange how the problem surfaced when the 2nd pair of MFETs were added
The power rails were not oscillating. I did check them.
Edit: I'm going to have a think about this while at work tomorrow. One's mind wanders and generally comes to some solutions (or at least mine does
What are you trying to achieve? Why not just compensate it on lower frequency, it will be high enough.
Or stick to one pair of outputs...
Dealing with these problems is where the learning happens. Also, want to push this design as far as I can.
Compensating at a lower ULGF would be a last resort. Agreed, its high enough already for listening purposes but its an engineering challenge.
Or stick to one pair of outputs...
Dealing with these problems is where the learning happens. Also, want to push this design as far as I can.
Compensating at a lower ULGF would be a last resort. Agreed, its high enough already for listening purposes but its an engineering challenge.
If so, other techniques should be applied. In RF it is layout, short leads (SMD), shielding, ground and components designed for RF. And it is not guaranty of AF quality. But who knows? The way you started maybe you succeed that task?
How did you arrive at the snubber resistance values? Calculation or experiment?
Was the oscillation just from some LC resonance, and not really "instability" in the control system sense?
For optimizing snubbers, it's easy if you have an existing system and a scope. Here's one method: http://www.diyaudio.com/forums/powe...lm-caps-electrolytic-caps-30.html#post2828689
Was the oscillation just from some LC resonance, and not really "instability" in the control system sense?
For optimizing snubbers, it's easy if you have an existing system and a scope. Here's one method: http://www.diyaudio.com/forums/powe...lm-caps-electrolytic-caps-30.html#post2828689
If you want to have a fast MOSFET outputstage that deals with local oscillation you'll have to do this:
Use your desired low gate stopper in the 10 - 22 Ohms region (I use 15Ohms). To shut up the local oscillation, you have to mount a ceramic cap directly to the soldering joints of the Drain and Gate. Eventhough it looks like you are just loading the input with the power supply, it is in effect a local feedback at high frequencies. (Consider the drain the inverting output to the non-inverting gate input, it then starts to look like the capacitive feedback as you see with opamps, from output to inverting input).
In my MF80 project I silenced all parasitic oscillations in the FETs by putting a 47pF ceramic directly across the Drain and Gate pins while using a 15R gatestopper. Without the cap, I would have a 2V p-p 85MHz (!!) local oscillation.
So no, don't use gate snubbers. Use the concept of local capacitive feedback by viewing the FET as an inverting amplifier at the drain node.
I must note though, that in the simulator the feedback caps don't make any difference and that's mostly because the inductive parasitics are not modeled well or at all. I've been developing this point of view on the real thing: my MF80 amp.
Last edited:
Gootee - values were arrived at through experimentation and some logic. Not convinced it is not instability within the control system. It's just strange that with one pair it is very stable but as soon as a second pair is added problems start.
Keantoken - I have snubbers on all power rails as per your suggestion when I was doing the parasistics. The rails have been scoped and I cannot see any oscillating.
Magic - I shall try your idea. I have some ceramic smds that will fit between the pins of the mosfets without any problem.
One thing that has just come to mind is the bootstrapped drivers. I'm wondering whether they are having problems at high frequencies. The bootstrap capacitor is only a 680u Panasonic FC. Wondering if at high frequencies it is unable to pass much current. Wondering whether they should be bypassed with a film / ceramic cap say 330n.
Keantoken - I have snubbers on all power rails as per your suggestion when I was doing the parasistics. The rails have been scoped and I cannot see any oscillating.
Magic - I shall try your idea. I have some ceramic smds that will fit between the pins of the mosfets without any problem.
One thing that has just come to mind is the bootstrapped drivers. I'm wondering whether they are having problems at high frequencies. The bootstrap capacitor is only a 680u Panasonic FC. Wondering if at high frequencies it is unable to pass much current. Wondering whether they should be bypassed with a film / ceramic cap say 330n.
Last edited:
If you can't see ringing on the rails while doing square wave testing, then that's perfect. Injecting the square wave directly into the rails through a 1k resistor or so would be final and would take care of frequencies above the feedback loop.
Bypassing the bootstrap lytics might cause a resonance that really could upset the feedback loop (or twist the PM higher while adding a source of ringing). Personally, I would replace Q5 and Q6 with BC5x0C and if that didn't change anything, I would feel pretty sure that the drivers were not it.
Bypassing the bootstrap lytics might cause a resonance that really could upset the feedback loop (or twist the PM higher while adding a source of ringing). Personally, I would replace Q5 and Q6 with BC5x0C and if that didn't change anything, I would feel pretty sure that the drivers were not it.
You may have to empirically determine the proper value. For the Exicon 10N/P20s 47pF was ideal. 39pF and lower was just too low to silence quickly enough when I attempted to induce oscillation through step injecting.
Practical values would probably be between this 47pF and 150pF.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.