Try adding one RC from the switching node to each rail. C can be around 470p-1n for IRFB4227. R would be a few ohms (3.3-22 or so).
Try adding various types of capacitors between supply rails and GND or directly from +V to -V. Since you seem to have already films/ceramics and big electrolytics (low ESR?) which usually create a big resonant peak at a few Mhz, adding standard medium-high ESR electrolytics may help a lot to damp it.
There are methods for finding out what is required for optimum damping, but it's difficult at first.
If you show us the progress of your new PCB we can give some hints.
Try adding various types of capacitors between supply rails and GND or directly from +V to -V. Since you seem to have already films/ceramics and big electrolytics (low ESR?) which usually create a big resonant peak at a few Mhz, adding standard medium-high ESR electrolytics may help a lot to damp it.
There are methods for finding out what is required for optimum damping, but it's difficult at first.
If you show us the progress of your new PCB we can give some hints.
Last edited:
Tried the amp on full power tonight with a 1000pf across the inductor input to ground.
The amp still resets when in hard clipping.
I had another pcb that doesnt show the fault so I swapped out chips and the pcb that previously reset no longer did it with the other chip in it !
The 1000pf does significantly reduced overshoot and ringing so I have left it in the design.
If I add more than 1nf then the 2092 blows up.
Still waiting for a class d inductor to arrive.
The current inductors manufacturer told me it is no good for class d as it has soft clipping.
Anyway it gets too hot, around 100-120 degrees C.
The amp still resets when in hard clipping.
I had another pcb that doesnt show the fault so I swapped out chips and the pcb that previously reset no longer did it with the other chip in it !
The 1000pf does significantly reduced overshoot and ringing so I have left it in the design.
If I add more than 1nf then the 2092 blows up.
Still waiting for a class d inductor to arrive.
The current inductors manufacturer told me it is no good for class d as it has soft clipping.
Anyway it gets too hot, around 100-120 degrees C.
Last edited:
There are two sets of resonances, one coming from power supply capacitors and the related parasitic inductances (usually in the 1Mhz to 10Mhz range), and the other from output stage parasitics (PCB tracks, MOSFET leads, die capacitance) usually in the 10Mhz to 100Mhz range. Both need damping.
Placing 1nF from switching node to ground without series resistor means doing two things wrong. Ground has nothing to do with the currents that originate the ringing, and a series resistor is required to provide actual damping, otherwise you just shift the resonance from one frequency to another and alter the Q.
So if you want to get closer to what you want, remove the 1nF capacitor and do as I explained.
The RC snubbers from switching node to the rails are intended to damp output stage resonances (I use an inductive resistor on purpose to create a tuned RLC and get more damping with less dissipation in the resistors, but this is quite advanced).
The lossy electrolytic trick I mentioned is intended to damp supply rail resonances. Another trick to damp supply rails is to add one or more film capacitors with a specific resonant frequencies (accounting for PCB inductances, which will move it to a lower freq). In one SMPS I'm testing now I use this trick with several equally spaced "sharp" resonances to damp a "wide" resonance, but this is quite advanced too.
You should always find the frequency components of the ringing and investigate the origin of each one. Getting a module ready for being sold requires some hard work...
Placing 1nF from switching node to ground without series resistor means doing two things wrong. Ground has nothing to do with the currents that originate the ringing, and a series resistor is required to provide actual damping, otherwise you just shift the resonance from one frequency to another and alter the Q.
So if you want to get closer to what you want, remove the 1nF capacitor and do as I explained.
The RC snubbers from switching node to the rails are intended to damp output stage resonances (I use an inductive resistor on purpose to create a tuned RLC and get more damping with less dissipation in the resistors, but this is quite advanced).
The lossy electrolytic trick I mentioned is intended to damp supply rail resonances. Another trick to damp supply rails is to add one or more film capacitors with a specific resonant frequencies (accounting for PCB inductances, which will move it to a lower freq). In one SMPS I'm testing now I use this trick with several equally spaced "sharp" resonances to damp a "wide" resonance, but this is quite advanced too.
You should always find the frequency components of the ringing and investigate the origin of each one. Getting a module ready for being sold requires some hard work...
Last edited:
I tried a snubber on the input to the inductor VS to the 3 supply lines.
For some reason I get better results if it goes from VS to ground rather than to +/-45 volts. I can get 10 more volts out of the output before clipping occurs with just a 1nf from VS to ground clsoe to the inductor.
Looking forward to trying a proper class d inductor.
I guess the power supply inductor I am using for the class d inductor is saturating and causing a short circuit above a certain current threshold. I found if I put two 22uh inductors in series the resetting goes away suggesting the inductor I am using is losing inductance at a certain current threshold.
For some reason I get better results if it goes from VS to ground rather than to +/-45 volts. I can get 10 more volts out of the output before clipping occurs with just a 1nf from VS to ground clsoe to the inductor.
Looking forward to trying a proper class d inductor.
I guess the power supply inductor I am using for the class d inductor is saturating and causing a short circuit above a certain current threshold. I found if I put two 22uh inductors in series the resetting goes away suggesting the inductor I am using is losing inductance at a certain current threshold.
But the saturation current of two inductors in series is the same as in one inductor alone, only ripple current is reduced. On the other hand, leakage magnetic flux from the inductor (induces AC potentials across PCB tracks) is halved too. Did you know that output inductor orientation matters? It's particularly important for non toroid cores, which should be wound across one of the two still "clean" spatial planes perpendicular to PCB track plane.
If the capacitor to ground produces less ringing than optimum RC, then there is something else going on. I mean, it's probably making switching slower and avoiding cross-conduction due to weak gate drive. Did you use turn-off diodes on the gates? High PCB inductances in the output stage can lead to cross-conduction too.
If the capacitor to ground produces less ringing than optimum RC, then there is something else going on. I mean, it's probably making switching slower and avoiding cross-conduction due to weak gate drive. Did you use turn-off diodes on the gates? High PCB inductances in the output stage can lead to cross-conduction too.
I ended up with a 1000pf capacitor in series with a 4R7 resistor.
I have ordered a 100R pot so I can try different series resistors more easily.
I have reduced the overshoot from 35 volts down to 8 volts with the RC. I am not quite sure what an acceptable overshoot voltage would be.
The gate resistors are 10 ohms with turn off diodes, I used max deadtime instead.
The ringing looks around 10MHz which is the same as the inductor resonant frequency.
10Mhz looks like resonance between big electrolytics and small films/ceramics, it's not coming from the output stage. I already told you how to damp that (non low-ESR small/medium size electrolytic).
Optimum overshoot is a single glitch of a few volts, without subsequent ringing.
The resonance of the output inductor is the frequency at which maximum impedance is reached, it's harmless, maximum impedance implies lowest current, it does not have anything to do with the ringing in the rails.
10 ohms gate resistors are probably too low for your PCB (too high di/dt), what is the gate supply voltage? IRFB4227 are fully on at 9V.
Optimum overshoot is a single glitch of a few volts, without subsequent ringing.
The resonance of the output inductor is the frequency at which maximum impedance is reached, it's harmless, maximum impedance implies lowest current, it does not have anything to do with the ringing in the rails.
10 ohms gate resistors are probably too low for your PCB (too high di/dt), what is the gate supply voltage? IRFB4227 are fully on at 9V.
Dead time is not the whole story about cross-conduction, it can still happen due to PCB parasitics in series with gate drive current loop and drain-gate capacitance. The gate can be pulled high very fast through this capacitance, sometimes faster than PCB inductances allow the gate driver to pull it low. As a result of bad PCB layout, voltage spikes can also appear at the gates when the MOSFET is supposed to be off (this happens when part of the gate turn-off current loop is shared with main switching current loop).
Ouch! Then you are right, the 10Mhz resonance comes from the power stage... The frequency is really low, which is probably the result of long PCB tracks and decoupling capacitors placed far from the output stage.
What is the ringing frequency with the RC temporarily disconnected? I hope the real resonance (without being shifted down by the RC) is higher.
To get decent switching you should aim for 50Mhz or so output stage resonance, which would theoretically require reducing all PCB inductances by a factor of 25 (with respect to 10Mhz). You can consider output stage ringing frequency an indicator of the quality of the layout.
In my best layout I get over 100Mhz resonance with IRFB4227, which is damped very easily (everything is more resistive than expected at 100Mhz due to skin effect) and gives no visible overshoot on oscilloscope. In these working conditions the full voltage and current capability of the IRFB4227 can be reliably used.
btw: 55Mhz with TO-247 is almost as good as 100Mhz with TO-220. Bigger package -> higher lead inductances -> lower resonance -> slower switching
Last edited:
Tried from 22pf to 100pf and it makes no difference.
Still waiting for the correct inductor.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.