problems measuring overcurrent in UcD output stage

[zilog]

I get problems with triggering of my overcurrent protection triggering no matter what my output power level is. I suspect the problem is from the violent current spikes that occur each polarity change of the output stage is triggering the transistors through some parasitic capacitance or so. I measure the current over a 47 mOhm resistor inserted in each output device leg as shown in the schematic.

When I measure the voltage across these 47 mOhm resistors, I get a "real" drop of ~300mV at full load, which is as expected, while I get spikes of 5V at the beginning and end of each switch cycle, which I wish for the overcurrent protection to ignore. See attached waveform.

Anyone have a solution to my problem?

[zilog]

Waveform

[mirlo]

Maybe:

Logic AND the output of the sensor with a blanking pulse that is logic 0 for a short time at the start of each switching to prevent the transients from triggering anything ...

[zilog]

I have thought of that as well, but how do I get a lookahead signal in a fully discrete UcD? The output stage is driven by current from a differential pair, so its hard to insert anything there.

Here is the output stage of the UcD, please dont be nasty about my paralelling of small/large capacitors in some places - that is already fixed in the prototype

[Eva]

Those spikes are caused both by MOSFET body-diode reverse recovery phenomena and by a possible cross-conduction issue.

Furthermore, sense-resistor inductance causes the frequency content of these spikes to be boosted at 6dB/oct above a certain frequency, altough that effect may be easily nulled with a low-pass filter (increase the value of C13 as needed to lower the pole until it matches the zero from resistor inductance).

You may both slightly increase dead time and make the reverse recovery process much smoother by slowing down gate turn-on (increase the value of R6 and R7, 0 ohms is not a good idea). That will also produce less EMI at the expense of a bit more heat.

Those MOSFETs (FDP3682) feature very low Trr and Qrr values (55ns and 90nC respectively). That should allow for a quite smooth turn-on process provided that you chose a high enough value for gate turn-on resistor.

BTW: Do you understand how a (body) diode behaves when you try to switch over it while it's conducting?

Concerning capacitor paralelling, I recommend checking for ringing in each particular case. Some combinations will ring while others doesn't.

Also, the current across your sense resistor may be either positive or negative depending on whether the amplifier is sinking or sourcing current to the load. You should account for both overcurrent cases in your circuit.

[zilog]

Originally posted by Eva
Those spikes are caused both by MOSFET body-diode reverse recovery phenomena and by a possible cross-conduction issue.

Furthermore, sense-resistor inductance causes the frequency content of these spikes to be boosted at 6dB/oct above a certain frequency, altough that effect may be easily nulled with a low-pass filter (increase the value of C13 as needed to lower the pole until it matches the zero from resistor inductance).

You may both slightly increase dead time and make the reverse recovery process much smoother by slowing down gate turn-on (increase the value of R6 and R7, 0 ohms is not a good idea). That will also produce less EMI at the expense of a bit more heat.

Those MOSFETs (FDP3682) feature very low Trr and Qrr values (55ns and 90nC respectively). That should allow for a quite smooth turn-on process provided that you chose a high enough value for gate turn-on resistor.

BTW: Do you understand how a (body) diode behaves when you try to switch over it while it's conducting?

Concerning capacitor paralelling, I recommend checking for ringing in each particular case. Some combinations will ring while others doesn't.

Also, the current across your sense resistor may be either positive or negative depending on whether the amplifier is sinking or sourcing current to the load. You should account for both overcurrent cases in your circuit.

I will try to experiment with dead time, but as it is right now with the given resistors, dead time is set to a mamimum, more and it stops oscillating.

I have also tried different values of C13, from 10n to 110n, no difference at all. I have also tried values between 82 and 470 ohm for R14, doesnt do anything either :/

I try to take the safe approach to capacitor ringing as I lack good enough equipment for measuring, for example I lack good probes, and in particular differential probes.

Concerning overcurrent when sinking/sourcing - I have one current sensing circuit on high side as well as on the low side - just showed the low side here for convenience.

[Eva]

Well, if the filter does make very little difference, then that seems more like an EMI issue. If you are using attenuated probes, I assume you have them well calibrated. I also assume that you are referencing the ground of the sense circuit and filter directly to one lead of the sense resistor in a star fashion, and not to another place in the supply rail track, because that will cause spikes due to ground loops.

Was the previous oscilloscope picture taken with the probe connected directly across the leads of the sense resistor? Try connecting both probe tip and probe ground to the same end of the resistor and tell me if the spikes also appear. Do the spikes show equal amplitude in both cases?

If you have some ferrite toroids of enough diameter (or other suitable ferrite cores), try wrapping the probe wires across them as the following picture shows. That will form a common-mode filter and will improve high-frequency non-floating measurements a lot, as the probe will turn into AC-floating (maximum AC voltage and minimum frequency are limited by core type and the amount of turns). Always ensure that there is no DC path between the ground of the oscilloscope and the place where you are connecting probe ground. With the right core and enough turns you can even connect probe ground to the drain of the floating MOSFET without trouble (that requires placing the ferrite very close to the probe to avoid radiating dV/dt).




Concerning capacitor ringing, you can actually build a nice capacitor combination tester around the good old 555 timer and little more stuff. It has been discussed in the power supply section of the forum. Check these threads:
Power Supply Decoupling
Capacitor (filter) phase question (pic attached)

And G4OEP page:
http://g4oep.atspace.com/caps/caps.htm

LAST EDIT:
I have just noticed that you forgot to put a base bleeder resistor in T4. That will cause false (if not permanent) protection triggering.

[zilog]

T4 is not used right now, the reason is that the entire high side protection curcuit is smoked, and R12 has been cut out. But I will install a base bleeder when I get the low side worked out.

I have tried using 47 ohm gate resistors (will try others soon), that halved the shoot through at low levels, but I still get the same shoot through at high levels. My scope seems to show 1-2 volts common mode when I measure at the same place on the resistor, I have tried with a toroid, but that doesnt cause any measurable improvement. The probe is a 10X probe, and has been calibrated.

What really worries me is that the probe constitutes a quite large antenna loop from prope tip to ground clamp, this antenna seems to pick upp all sorts of switching noise - and I dont want to mutilate my probe to reduce this loop either :/

[Eva]

That common mode component is spoiling your measurements. You have to find a way to remove it. It's very strange that the common-mode filters didn't make any change. What kind of core did you use? (mine are 36x23x15mm, 3E25 material)

Might the oscilloscope preamplifier or the probe be directly picking up high dV/dt or dI/dt radiated switching transients from the amplifier? Try to place the probe (with tip and ground shorted) very close to the sense resistor (and other high dI/dt or dV/dt places), but without physically touching the circuit. Does the oscilloscope show any spikes? (it shouldn't!).

Also, do you have a signal generator, or more probes or any other thing connected to the amplifier, that could be closing a common-mode AC ground loop? Then adding common-mode filtering to either the supply lines, the mains cords or the interconnects may help.

Concerning the ground lead of the probe, you can just bent it in zig-zag and hold it that way with electrical tape, it's not as efficient as cutting it but should reduce antenna effect (and undesired length). Anyway, you have to be radiating a lot of energy in order to be able to pick 2V spikes with that piece of wire. How does your layout look like?

[zilog]

I dont know what material the core is (scrapped inductor), but its size is 50x34x16. If I ground the tip of the probe, I can pick up noise from the circuit by just being adjacent, but I guess the magnetic fields around this beast are quite high.

I havent had the other probe connected when conducting these measurements, the only thing I have is a signal generator that doesnt share powerline ground with my scope that feeds the amplifier with a 1 kHz signal.

The layout is probably suboptimal since I have pieces of groundplane everywhere and the curcuit is divided into two boards (one for power, one for modulator) in order to overcome Eagle's 80x100mm limitation. It is also two layer, and I couldnt keep circuitry on only one side to meet size limitations. Nevertheless, my output waveforms show almost no ringing, and I cant pick up any disturbance to any radio/TV equipment.