Maker Pro
Maker Pro

Minimizing destructive HV spikes on square wave push-pull MOSFET drive to transformer

J

Joerg

Jan 1, 1970
0
P said:
wrote in message


Fred, thanks for the idea with the diode clamp. I had tried something
like that before, but it seems to work quite well. So I have added a
linear current limiter to the battery, set at about 80 amps, but I also
changed the output capacitors to what they actually are, 3300 uF each,
with 0.016 ohms ESR. I found that it will take about 300 mSec for the
output voltage to come up to 300V, and during that time the capacitors
seem to be dissipating about 300 watts each!

Actually, it will work without those capacitors, since the square wave
has only about 1 uSec during which the transformer is not being driven.
I needed them previously when I had them connected in a doubler
configuration to work on 12 VDC. So I should be able to use something
much smaller, and/or use the capacitors in the VF drive. But I think I
may need an inductor to ease the peak current in them as well.

In some ways it's amazing that the components lasted as long as they
did. The capacitors have 0.5*300*300*6600= 297 W-Sec so at 24V that
would be 12 amp-seconds, but since the original simulation showed the
peak voltage being reached at about 50 mSec it would be 240 amp-seconds.
If I limit the current to 80 amps it should reach the desired voltage in
about 297/(24*80) = 154 mSec. But if the capacitors are also dissipating
300 watts each during charge, that would be added, and there would be
more like 900 watt-seconds, so the peak might be reached at 450 mSec. Of
course real components do not always match the simulator (or
vice-versa), and the more accurate the simulation the longer it takes.
I'm running it now and it seems to be stabilizing at about 270 mSec, but
it took probably 10-15 minutes. I'll see what it looks like when it
reaches 350 mSec at which point the PWM stops.

Wow, the RAW file is 1 GB! Looking at the last 20 mSec, the input power
is 783W, output is 276W, the MOSFETs are 9.7W, the two series pass are
23W each, and there is still 214W in each of the output capacitors! At
this point the input current has dropped to 33A.

"Where has all the power gone, long time passing..." Bob Dylan, PP&M

Actually, though, this is still "reactive" power, because the capacitors
are still charging. The actual ripple current is about 1 amp, so the
real power is minimal. I needed to continue the simulation until full
charge was reached. Looking at the first 20 mSec of startup, the
capacitor current is just 3.7A. The series pass current limiters are
dissipating about 900W, so that might be a problem. But they will only
do so for a short time, and should gradually drop to the 23 watts at 250
mSec or so.

250msec is not short in the eyes of a transistor. When it comes to the
SOA that counts almost as "DC". You aren't really planning on those li'l
ZTX849, are you? That would be one quick *PHUT* and they'll turn into a
puff of black smoke :)

At least the linear regulator reduces the transients during turn-on,
since the MOSFETs only see about 3 volts at first. It may be better to
design a switching current limiter, but I'm trying to reduce complexity
and enhance reliability, so a brute force linear circuit might be
acceptable.

The way to handle this is via your controller. It should issue very
short pulses during start-up, slow repetition rate. If you can't do
short pulses then leave out lots of cycles until the caps are at nominal
voltage, then throw the "virtual clutch" and go full throttle. This
requires no extra parts.

Also, C10 could gradually charge up despite R6. Might want to zener that.

[...]
 
P

P E Schoen

Jan 1, 1970
0
"Joerg" wrote in message
250msec is not short in the eyes of a transistor. When it comes
to the SOA that counts almost as "DC". You aren't really planning
on those li'l ZTX849, are you? That would be one quick *PHUT*
and they'll turn into a puff of black smoke :)

I was going to use some MJ11029s that I have in some quantity. But I did not
look carefully enough at the SOA, so it should be limited to about 15A at
20V which it will see under start-up conditions. Then I considered using a
P-MOSFET, such as IRF4905, which is rated 55V and 74A, and 20 mOhms so it
will dissipate "only" 18 watts at 30 amps, but its SOA limits it to 20A at
20V for a 10mSec pulse. So I'd need four of them to be safe. And then the
power will be only about 5 watts total.

Since I already have a lot of these, maybe I can just cobb together a brute
force current limiter just so I can proceed with other parts of the design.
It might be a handy unit for other projects where I want to use batteries
protected with a circuit breaker but also current limited to reduce the
total "destructive" power available. I could also probably add some smarts
that would latch the series pass transistors off if there were a persistent
short and not just a big capacitor bank.

Actually, I don't really need the big 3300 uF capacitors with a square wave
and FWB. But I will have to consider the capacitors inside the VF drive.
With 440 uF they charge up within 25 mSec with a current limit of 70A. But
it's still 1200W for the first 10 mSec, and 660W over 30 mSec. So I'm
beginning to see the magnitude of the problem.
The way to handle this is via your controller. It should issue
very short pulses during start-up, slow repetition rate. If you
can't do short pulses then leave out lots of cycles until the
caps are at nominal voltage, then throw the "virtual clutch"
and go full throttle. This requires no extra parts.

One problem is that the PIC16F684 does not work for push-pull PWM using its
PWM module. I'm using it as a half-bridge with 50% duty cycle. I tried a
start-up by using a timer ISR to get 12% PWM with 250 uSec pulses and 4 kHz,
but the waveform was rather crappy. However, I could try even much shorter
pulses and maybe add some inductance to limit the rate of rise somewhat.
Then maybe measure the drain currents and use an interrupt to turn off the
pulse for a while. I'll have to think about that and maybe try fiddling the
PIC code a bit more. I really should be monitoring the drain current anyway,
as well as the battery voltage so it doesn't try to start before I have
enough voltage to drive the gates properly.

If I want to use the iron-core toroid, then I may have to live with its
leakage inductance and lower frequency limit, and use a switching
pre-regulator. It would be essentially a buck converter, and I could use
something like 100 kHz so I could get 50A peak at 5 uSec and 24V with a 2.5
uH choke. Something like this:
http://www.digikey.com/product-detail/en/AIRD-03-3R3M/535-11313-ND/2660604.
Also, C10 could gradually charge up despite R6. Might want
to zener that.

Good idea, but if the voltage builds up, R6 will pull more current and
power. Unless the spikes have a lot of power and high repetition rate, I
don't see a problem. But a zener wouldn't hurt. Maybe a TVS.

Thanks,

Paul
 
J

Joerg

Jan 1, 1970
0
P said:
"Joerg" wrote in message
[...]
The way to handle this is via your controller. It should issue
very short pulses during start-up, slow repetition rate. If you
can't do short pulses then leave out lots of cycles until the
caps are at nominal voltage, then throw the "virtual clutch"
and go full throttle. This requires no extra parts.

One problem is that the PIC16F684 does not work for push-pull PWM using
its PWM module. ...


One of the many reasons why I never warmed up to using uCs for such
jobs. A dedicated switcher chip will do a much better job and most are
only a buck fifty.

... I'm using it as a half-bridge with 50% duty cycle. I
tried a start-up by using a timer ISR to get 12% PWM with 250 uSec
pulses and 4 kHz, but the waveform was rather crappy. However, I could
try even much shorter pulses and maybe add some inductance to limit the
rate of rise somewhat. ...


Maybe you could disable the port for many cycle, let one through,
suppress again, and so on?

... Then maybe measure the drain currents and use an
interrupt to turn off the pulse for a while. I'll have to think about
that and maybe try fiddling the PIC code a bit more. I really should be
monitoring the drain current anyway, as well as the battery voltage so
it doesn't try to start before I have enough voltage to drive the gates
properly.

Monitoring is one of the best insurances against a major kablouie.

If I want to use the iron-core toroid, then I may have to live with its
leakage inductance and lower frequency limit, and use a switching
pre-regulator. It would be essentially a buck converter, and I could use
something like 100 kHz so I could get 50A peak at 5 uSec and 24V with a
2.5 uH choke. Something like this:
http://www.digikey.com/product-detail/en/AIRD-03-3R3M/535-11313-ND/2660604.

Looks a little wimpy but might survive :)

Good idea, but if the voltage builds up, R6 will pull more current and
power. Unless the spikes have a lot of power and high repetition rate, I
don't see a problem. But a zener wouldn't hurt. Maybe a TVS.

It's cheap insurance, considering the molten solder and plastic shrapnel
that could be splattering about if one of the big FETs avalanches.
 
J

John S

Jan 1, 1970
0
wrote in message


Fred, thanks for the idea with the diode clamp. I had tried something
like that before, but it seems to work quite well. So I have added a
linear current limiter to the battery, set at about 80 amps, but I also
changed the output capacitors to what they actually are, 3300 uF each,
with 0.016 ohms ESR. I found that it will take about 300 mSec for the
output voltage to come up to 300V, and during that time the capacitors
seem to be dissipating about 300 watts each!

Actually, it will work without those capacitors, since the square wave
has only about 1 uSec during which the transformer is not being driven.
I needed them previously when I had them connected in a doubler
configuration to work on 12 VDC. So I should be able to use something
much smaller, and/or use the capacitors in the VF drive. But I think I
may need an inductor to ease the peak current in them as well.

In some ways it's amazing that the components lasted as long as they
did. The capacitors have 0.5*300*300*6600= 297 W-Sec so at 24V that
would be 12 amp-seconds, but since the original simulation showed the
peak voltage being reached at about 50 mSec it would be 240 amp-seconds.
If I limit the current to 80 amps it should reach the desired voltage in
about 297/(24*80) = 154 mSec. But if the capacitors are also dissipating
300 watts each during charge, that would be added, and there would be
more like 900 watt-seconds, so the peak might be reached at 450 mSec. Of
course real components do not always match the simulator (or
vice-versa), and the more accurate the simulation the longer it takes.
I'm running it now and it seems to be stabilizing at about 270 mSec, but
it took probably 10-15 minutes. I'll see what it looks like when it
reaches 350 mSec at which point the PWM stops.

Wow, the RAW file is 1 GB! Looking at the last 20 mSec, the input power
is 783W, output is 276W, the MOSFETs are 9.7W, the two series pass are
23W each, and there is still 214W in each of the output capacitors! At
this point the input current has dropped to 33A.

"Where has all the power gone, long time passing..." Bob Dylan, PP&M

Actually, though, this is still "reactive" power, because the capacitors
are still charging. The actual ripple current is about 1 amp, so the
real power is minimal. I needed to continue the simulation until full
charge was reached. Looking at the first 20 mSec of startup, the
capacitor current is just 3.7A. The series pass current limiters are
dissipating about 900W, so that might be a problem. But they will only
do so for a short time, and should gradually drop to the 23 watts at 250
mSec or so.

At least the linear regulator reduces the transients during turn-on,
since the MOSFETs only see about 3 volts at first. It may be better to
design a switching current limiter, but I'm trying to reduce complexity
and enhance reliability, so a brute force linear circuit might be
acceptable.

Thanks,

Paul

=============== Here's my modified ASCII file ==================

(snipped)

Paul, the simulation is showing Q2 (2N2907) dissipating 20 watts.

John
 
P

P E Schoen

Jan 1, 1970
0
"John S" wrote in message
Paul, the simulation is showing Q2 (2N2907) dissipating 20 watts.

That was just to approximate the behavior of the darlington PNP I was going
to use. But there were other problems with the brute force approach and I
would need to use many in parallel to stay with their SOA. So I have
simulated a design using an inductor and a series MOSFET (and other
components) to current regulate the load with minimal losses. It seems to
work pretty well, finally. LTSpice doesn't have many good P-channel power
MOSFETs in their library. So I used a high side gate driver and a good size
NMOS. I'll post a separate thread with more details. It may be a handy
circuit to use when charging big capacitors from low impedance sources like
batteries.

Thanks,

Paul
 
J

Joerg

Jan 1, 1970
0
P said:
"John S" wrote in message

That was just to approximate the behavior of the darlington PNP I was
going to use. But there were other problems with the brute force
approach and I would need to use many in parallel to stay with their
SOA. So I have simulated a design using an inductor and a series MOSFET
(and other components) to current regulate the load with minimal losses.
It seems to work pretty well, finally. LTSpice doesn't have many good
P-channel power MOSFETs in their library. So I used a high side gate
driver and a good size NMOS. I'll post a separate thread with more
details. It may be a handy circuit to use when charging big capacitors
from low impedance sources like batteries.

Check Fairchild, they've got tons of SPICE models. You may have to
become a "club member" there but it's free.
 
Top