GODZILLA Revisited
Today i decided to work on the godzilla again.
I now have this hooked up on the bench:
http://i.imgur.com/ynkfK.jpg
Only problem though that won't show up in simulation is that near clipping on the negative half, the output makes a sharp peak toward 0V that in the speakers sound like a snap.
I know this is some form of instability but no matter where i put a small ceramic cap, the same thing is still there.
Today i decided to work on the godzilla again.
I now have this hooked up on the bench:
http://i.imgur.com/ynkfK.jpg
Only problem though that won't show up in simulation is that near clipping on the negative half, the output makes a sharp peak toward 0V that in the speakers sound like a snap.
I know this is some form of instability but no matter where i put a small ceramic cap, the same thing is still there.
R21/R14 should be simply connected in serial;
not connected to the output, as this increase
distorsion badly.
Q10/Q9 are undersized, use a pair that has more
TDP capability..
Guesses from afar...
I have some guesses about your clipping problem...
1. not the issue probably, but it still makes me nervous...do you have any high frequency bypasses on the supply rails on the real board? e.g. some 0.1 uF or 0.01 uF's...maybe a few here and there?
2. The main thing going on with negative rail drive is something called servo windup. Driver drives as hard as it can, it's not enough, so unsatisfied feedback loop keeps pushing...then when you come away from the rail, it has to undo all that windup before things straighten out. It's very slow doing that too because Mosfet Cgs gets huge quickly when Vds drops. Then as Vds of P gets bigger, Cgs drops so fast that the servo (feedback loop) gets fooled.
3. Depending on your green LED, when Q7 saturates, it may be stealing enough R18 current to drop the green LED voltage, playing with the tail current of the first stage...that can hurt, too...R17 can kind of keep that from happening, but then it makes Q7 slow to normalize again after being in saturation. A separate bias source for 2nd stage current source might be a good way to eliminate this as an issue.
4. Best answer is not to drive amp so hard that it clips.
5. Why is the Pside worse? Data sheet shows Ciss goes from 1200 pF to 2700 pF for Pchannel. Only changes from 2000 to 2700 pF for N channel.
6. Why doesn't simulator show this? I doubt the models you have model the nonlinear capacitances in the FETs...I haven't seen many models that do.
7. you can add a nonlinear feedback loop that only comes into play near clipping...that's sometimes a good way to keep things sane. If any of this seems reasonable...we'll see if we can find something workable...
I have some guesses about your clipping problem...
1. not the issue probably, but it still makes me nervous...do you have any high frequency bypasses on the supply rails on the real board? e.g. some 0.1 uF or 0.01 uF's...maybe a few here and there?
2. The main thing going on with negative rail drive is something called servo windup. Driver drives as hard as it can, it's not enough, so unsatisfied feedback loop keeps pushing...then when you come away from the rail, it has to undo all that windup before things straighten out. It's very slow doing that too because Mosfet Cgs gets huge quickly when Vds drops. Then as Vds of P gets bigger, Cgs drops so fast that the servo (feedback loop) gets fooled.
3. Depending on your green LED, when Q7 saturates, it may be stealing enough R18 current to drop the green LED voltage, playing with the tail current of the first stage...that can hurt, too...R17 can kind of keep that from happening, but then it makes Q7 slow to normalize again after being in saturation. A separate bias source for 2nd stage current source might be a good way to eliminate this as an issue.
4. Best answer is not to drive amp so hard that it clips.
5. Why is the Pside worse? Data sheet shows Ciss goes from 1200 pF to 2700 pF for Pchannel. Only changes from 2000 to 2700 pF for N channel.
6. Why doesn't simulator show this? I doubt the models you have model the nonlinear capacitances in the FETs...I haven't seen many models that do.
7. you can add a nonlinear feedback loop that only comes into play near clipping...that's sometimes a good way to keep things sane. If any of this seems reasonable...we'll see if we can find something workable...
No oscillation, but a lot of distorsion, for sure..
Try to increase the drivers quiescent current by reducing the 680R resistors,
and if possible, to not connect them to the output.
Try 470R, then 390 and so on...
At most, 15 to 20mA drivers current should be enough to correctly drive this cohort of hexfets.
Though, it would be better to mount the BD139/140 on little heatsinks..
No oscillations, but some measurements yesterday showed that one pair was taking the brunt of the bias current while another pair barely broke a sweat, the others were scattered in between.
One fet had even melted its tin coating on the heatspreader.
This amp was abused yesterday too, regularly brough to clipping into 2 ohms and consuming 1A of bias from the power supply which got so hot it started melting into the plastic i have on my table.
One fet had even melted its tin coating on the heatspreader.
This amp was abused yesterday too, regularly brough to clipping into 2 ohms and consuming 1A of bias from the power supply which got so hot it started melting into the plastic i have on my table.
0.1 ohm source resistors are probably too low unless the MOSFETs are matched. 0.33 or 0.47 ohms is better, and in addition will make bias easier to set and control. Q8 should be in contact with the heatsink - you might find using a TO-126 device such as a BD139 or an MJE340 better for this task.
The gate stopper resistors should probably be at least 1K too, to prevent oscillation. If you are using a lot of mosfets, you'll want to have good decoupling at both HF and LF (basically - 100-470uF electrolytic bypassed by 100nF ceramic) close to the MOSFETs to avoid oscillation.
If you're driving 6 pairs of MOSFETs something with higher fT and current handling than BD139/140 might be a good idea.
The gate stopper resistors should probably be at least 1K too, to prevent oscillation. If you are using a lot of mosfets, you'll want to have good decoupling at both HF and LF (basically - 100-470uF electrolytic bypassed by 100nF ceramic) close to the MOSFETs to avoid oscillation.
If you're driving 6 pairs of MOSFETs something with higher fT and current handling than BD139/140 might be a good idea.
Q8 was on the heatsink. The heatsink was also fan cooled so the bias was at the same value as yesterdays run.
I don't have any other resistors tha 0.1 ohm at home at the moment.
Anyways this project had too many problems to be worth continuing with so i've decided to scrap it.
The 2N3773 amp will take its place once i get ahold of some good 2N3773's or similar.
I don't have any other resistors tha 0.1 ohm at home at the moment.
Anyways this project had too many problems to be worth continuing with so i've decided to scrap it.
The 2N3773 amp will take its place once i get ahold of some good 2N3773's or similar.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.