Interupting xenon flash current ?

[JosephKK]

The RCD snubber seems like a good solution but the problem is the cost
of the diodes needed. I have three high current diodes on the way in
the post and I may be able to spare one of them for that. However
even with that I am having some difficulty getting the values right so
that the snubber can be used multiple times, for instance to have
three 100uS pulses spaced at 100uS. The snubber capacitor finds
itself still charged on the second current disconnection and will not
have much effect. However that is probably something that can be
solved with the right values.

Actually, you may want to consider commutating the snubbers them selves.
Say 10 us after turn off.

The capacitor self discharging is not an option because the pulse
widths and spacing need to settable to arbitrary values controlled by
a pic from some 50 uS all the way up to 5mS, so it is not possible to
tailor component values to fit the timings required. Also, I need the
pulses to have a very sharp cutoff, something that cannot be achieved
if one depends on the capacitor voltage decaying to below the holding
voltage to cut off.

As for current sharing I am currently not doing anything except
paralleling the emitters and collectors of all 24 transistors on two
thick copper busbars. The gates are separated into four groups of six
to reduce to some extent the interaction between the transistors on
turn off. In future I intend to have a 0.25 ohm resistor on the
collector of each transistor to help distribute the current. It will
cause some 15 volts drop at maximum load, but that is a small price to
pay if it does its job. The plans are also for each gate to have its
own optically isolated driver to completely eliminate any interaction
between the gates of the various transistors.

You actually want a few (probably no more than 4) milliohms between the
emitter terminals and the bus bar.


Wild think. The voltages and currents make FET totem pole stacks
difficult. IGBTs have speed and balance related issues. SCRs at that
power level cannot be reasonably commutated fast enough. Have you
considered radar modulator tubes? Or even more common high power UHF
transmitting tubes?

 

[[email protected]]

Actually, you may want to consider commutating the snubbers them selves.
Say 10 us after turn off.







You actually want a few (probably no more than 4) milliohms between the
emitter terminals and the bus bar.

Wild think.  The voltages and currents make FET totem pole stacks
difficult.  IGBTs have speed and balance related issues.  SCRs at that
power level cannot be reasonably commutated fast enough.  Have you
considered radar modulator tubes?  Or even more common high power UHF
transmitting tubes?- Hide quoted text -

- Show quoted text -- Hide quoted text -

- Show quoted text -

Not doing anything yet for current sharing, just connecting all 24 in
parallel. The gates are in four groups of 6 so at least I don't have
all 24 driven in parallel off one driver (to reduce the interaction).
Reason I am not using current sharing yet is that I am running them at
significanlty lower current than the current I am designing for, so it
is a low priority thing at this time so to speak. The plans are to
have a 0.25 ohm resistance on the collector side (not the emitter),
that is between each collector and the load busbar. That will be meant
to compensate somewhat for the differneces in VI characteristics of
the individual units.

An important differnece between what I am doing and more common
applications using paralleled IGBTs is that one of the main
difficulties is absent in this case - that of thermal equalisation.
Whereas in most applications (motor controls, inverters and such) the
IGBTS would be dissipating a large amount of power and heating up
considerably, in my case the average power dissipation is very low and
the IGBTs' case temperature will not increase noticeably, so I don't
really need to worry about differences in the temperature coefficients
between the devices. (They will in practice always be at 'room
temperature').

I had originally though of putting it on the emitter side, the idea
being that increased collector current will decrease the gate-emitter
voltage reducing drive and thus current. However from some (admittedly
superficial) research I've done I find that the emitter-busbar
resistance (and especially inductance) should be kept as low as
possible. I haven't fully understood why but I was sort of taking
their word for it.

(Eventually each individual IGBT will have its own opto couple gate
driver)


Regarding 'totem poles' - I'm not very familair with the term but I
thought it related to series connections for higher voltage handling
rather than paralleling. My original target was for a 1400V system
which would have two groups of IGBTs in series but eventually I
dropped that and settled for a less ambitious 700 volts target.

As far as using tubes of any type I haven't really considered that for
various reasons, but probably most importantly because I don;t have
any in my junk box

I was however considering the use of a spark gap to handle the turn on
and main duration of the pules and then using the IGBTs to quench the
spark gap and then turn off cleanly. That was meant to reduce the
I*I*T load on the IGBTS while keeping the fats and clean turnoff that
they should be able to give me. However I have more or less come to
the conclusion that the IGBTs are having no difficulty with I*I*T and
that the big problem is what is going on during the turn off so I will
probably concentrate on getting that out of the way before tackling
the other issues.

 

[[email protected]]

Regarding 'commutating the snubbers', I'm not sure what you mean.
Would it involve something like switching in a resistance at the
appropriate time to discharge the capacitor in preparation for the
next turn-off ?

 

[Jim Thompson]

Hi,
I've been working on a high energy flash system for some time now
(about 4000 Joules) and just come across this forum. Like some others
here I've been having difficulty with interrupting the current to
terminate the discharge current. I had originally intended to run the
tubes at 1400 volts but I have had to scale that down to 700 volts.
I am currently using 24 IGBTs in parallel each rated at 60 Amps 900
Volts (will increase to 50 eventually to handle the full energy). The
flash tube draws some 900 amps once the curent stabilises after the
initial surge.

Unfortunately the IGBTS are not surviving (I have blown 3 already) and
I have had to limit the voltage further to 500 volts, reducing the
steady current to about 600 amps. I'm still trying to find out why
I'm getting the failures at 700 volts. The current sharing scheme is
admittedly quite primitive (or perhaps I should say non existent) but
it does not seem to be the problem and neither does the initial 5000
amp current surge. What does seem to be the problem is some very ugly
oscillation when the IGBT is turned off which probably means I need to
work more on the gate drive circuitry.

I've been reading further back in this forum and noted some discussion
on the initial surge of current when a flashtube is first triggered.
I have also found that phenomenon and with a very low resistance path
between the capacitor bank and the flash tube I was getting an
estimated 30 KA surge at 1400 volts, the duration of the surge being
about 30 to 40 microseconds. I had initially suspected an inductance
problem, as was also suggested here, but to eliminate that possibility
I tried using different shunts, ranging from a flat piece of lead to a
short length of 10mm2 copper cable and got similar results in each
case.

My ancient patent, 3496411, works the other way around.

...Jim Thompson

 

[JosephKK]

Regarding 'commutating the snubbers', I'm not sure what you mean.
Would it involve something like switching in a resistance at the
appropriate time to discharge the capacitor in preparation for the
next turn-off ?

Actually shortly after the turn off was the idea, but that is the line
of thought.

 

[legg]

Actually, you can. IR wrote an AN somewhere that describes the process.
At low current, it's true, but presumably, at these current levels, the
paralleled devices are each rated for enough that there is no problem. At
higher current levels, the resistive component of the IGBTs comes into
play, and current shares to an increasing degree. What you get is a
roughly constant difference in current between transistors, and as long as
this difference is less than the rated current of each device, you're fine.

I agree that saturated switching can give results +/-20%, provided
VCEsat is matched, per the old harris and fairchild app notes.,
however linear operation or operation during switching (particularly
at turn-off) require better control of emitter current than is
provided by the resistive DS path of the insulated gate input mos
structure or beta of the output power structure.

The IR app note (990?)seems to concentrate on thermal limits produced
by two devices exhibiting a 40% current imbalance under saturated
conditions.

The OP doesn't appear to be dealing with a long-term thermal issue,
but an energy issue experienced under dynamic conditions - notably
turn-off, where the highest gain and most charge retentive device
suffer the bulk of turn-off losses and possibly severe dv/dt.

RL

 

[[email protected]]

I agree that saturated switching can give results +/-20%, provided
VCEsat is matched, per the old harris and fairchild app notes.,
however linear operation or operation during switching (particularly
at turn-off) require better control of emittercurrentthan is
provided by the resistive DS path of the insulated gate input mos
structure or beta of the output power structure.

The IR app note (990?)seems to concentrate on thermal limits produced
by two devices exhibiting a 40%currentimbalance under saturated
conditions.

The OP doesn't appear to be dealing with a long-term thermal issue,
but an energy issue experienced under dynamic conditions - notably
turn-off, where the highest gain and most charge retentive device
suffer the bulk of turn-off losses and possibly severe dv/dt.

RL

As you pointed out, my major difficulty is actually what goes on
during the turn off.
I have succesfully fired the tube many times without problem, provided
that I ensure the turnoff happens only after the capacitor voltage has
decayed substantially.

I would like to thank everyone for the good suggestions. I'm still
trying to figure out what course of action to take, but probably it
will be:
1. Insert a resistance of about 0.25 ohm between each collector and
the collector commoning busbar.
2. Feed each gate independently with an opto coupled gate driver
3. Add an RCD snubber across the emitter-busbar and the collector
busbar

I am hoping that the collector resistance should help significantly
even in dynamic curent balancing. If any of the transistors were to be
significantly slower than the others at turning off I am hoping that
the resistor would take the brunt of the V*I*T as it will generally
have a higher voltage drop across it than the IGBT. The loss of
energy in the resistor is not really a concern in this application
unlike it would be in motor drives, inverters and such.

With the independent feeds I am hoping to eliminate completely the
interaction between the IGBTs, which I think could be part of the
problem. Each driver will take the gate to about 20 volts positive and
5 volts negative for turn off. I'm still not sure what value resistor
to put in series with the gate. From what I understand, too small a
resistor could lead to oscillation, while too high a resistor will
allow the turnoff DV/DT to turn on the gate. I've been looking for
information on how to calculate gate resistance but I haven;t quite
figured it out yet.

I will also be applying an RCD snubber as suggested by one of the
posters. It would replace the current RC snubber. The diode will be
placed such that on turnoff the diode will conduct and most of the
tube current would go through the diode and charge the capacitor -
with no series resistor (the tube itself is already limiting the
current).
The capacitor will then discharge through the IGBTs and a series
resistor at turn-on through a limiting resistor placed across the
diode. I considered having a separate IGBT or something to discharge
the capacitor independently but I could not find a suitable way of
doing it.

All this will take me some time to do, even because I do not have all
the components yet, so I may be trying out some other things in the
meantime.

Thanks to all for the help and good advice.

 

[legg]

2. Feed each gate independently with an opto coupled gate driver

If device to device dynamic timing is considered a potential problem,
you should probably use one opt-ocoupler and n x buffers.

RL

 

[[email protected]]

If device to device dynamic timing is considered a potential problem,
you should probably use one opt-ocoupler and n x buffers.

RL

The opto coupler I was planning to use is the HCPL3120 - it is an opto
coupler and gate driver combined onto one integrated circuit. It is
meant to have a very predictable and constant propagation delay,
varying only by a few nS between different units. I haven't verified
this yet though as I have not yet received the components.

What I like about using the optocouplers is that it eliminates
problems of ground bounce due to the small but finite inductance (and
to a lesser extent) resistance in the emitter busbar. The output of
the drivers will be referenced directly to the emitters thus
eliminating the problem on the driver-gate side. Without the opto
coupler however that problem is merely be transferred to the input
side of the buffers. I was hoping that the opto isolator would
eliminate this problem by making the drive signal completely isolated
from the gate drive circuit.

 

[[email protected]]

As you pointed out, my major difficulty is actually what goes on
during the turn off.
I have succesfully fired the tube many times without problem, provided
that I ensure the turnoff happens only after the capacitor voltage has
decayed substantially.

I would like to thank everyone for the good suggestions.  I'm still
trying to figure out what course of action to take, but probably it
will be:
1. Insert a resistance of about 0.25 ohm between each collector and
the collector commoning busbar.
2. Feed each gate independently with an opto coupled gate driver
3. Add an RCD snubber across the emitter-busbar and the collector
busbar

I am hoping that the collector resistance should help significantly
even in dynamic curent balancing. If any of the transistors were to be
significantly slower than the others at turning off I am hoping that
the resistor would take the brunt of the V*I*T as it will generally
have a higher voltage drop across it than the IGBT.  The loss of
energy in the resistor is not really a concern in this application
unlike  it would be in motor drives, inverters and such.

With the independent feeds I am hoping to eliminate completely the
interaction between the IGBTs, which I think could be part of the
problem. Each driver will take the gate to about 20 volts positive and
5 volts negative for turn off. I'm still not sure what value resistor
to put in series with the gate. From what I understand, too small a
resistor could lead to oscillation, while too high a resistor will
allow the turnoff DV/DT to turn on the gate.  I've been looking for
information on how to calculate gate resistance but I haven;t quite
figured it out yet.

I will also be applying an RCD snubber as suggested by one of the
posters. It would replace the current RC snubber. The diode will be
placed such that on turnoff the diode will conduct and most of the
tube current would go through the diode and charge the capacitor -
with no series resistor (the tube itself is already limiting the
current).
The capacitor will then discharge through the IGBTs and a series
resistor at turn-on through a limiting resistor placed across the
diode.  I considered having a separate IGBT or something to discharge
the capacitor independently but I could not find a suitable way of
doing it.

All this will take me some time to do, even because I do not have all
the components yet, so I may be trying out some other things in the
meantime.

Thanks to all for the help and good advice.- Hide quoted text -

- Show quoted text -

Do you know what the failure mode is of your IGBT ?

I'm finding out the major failure mechanism in this application is
indeed during turn-off and the problem invariable is in the IGBT
driver. You need a push pull driver with separate resistor for on and
off function. If the resistor for the off function is too small, you
risk blowing up the gate during the transistion. If it's too large,
your turn-off will be slow and you'll run into potential thermal
issues. As for the "on" resistor, shorter is better but one has to
stay within the max gate current limit. I'm finding out that the
resistor in the "off" leg is about 3 x larger than the "on" function.

On another note, as I read this early on in this thread ...

As for the delay in reaching the max light output of a xenon flash
lamp, this is called the "ionization" delay. It is indeed in the 10us
neighborhood for most photo flash lamps, can be shortened to maybe 3us
if very low ESR caps are used and the ESL fo the circuit is low.

It is extremely difficult to make 1us flash pulses with xenon flash
lamps, short of economical not feasible. If one is happy with very
low output, then some of the analytical xenon flash lamps could be
tried.

This is exactly the reason why spark gaps are used for extreme high
speed photography.


YMMV,

 

[Robert Baer]

Do you know what the failure mode is of your IGBT ?

I'm finding out the major failure mechanism in this application is
indeed during turn-off and the problem invariable is in the IGBT
driver. You need a push pull driver with separate resistor for on and
off function. If the resistor for the off function is too small, you
risk blowing up the gate during the transistion. If it's too large,
your turn-off will be slow and you'll run into potential thermal
issues. As for the "on" resistor, shorter is better but one has to
stay within the max gate current limit. I'm finding out that the
resistor in the "off" leg is about 3 x larger than the "on" function.

On another note, as I read this early on in this thread ...

As for the delay in reaching the max light output of a xenon flash
lamp, this is called the "ionization" delay. It is indeed in the 10us
neighborhood for most photo flash lamps, can be shortened to maybe 3us
if very low ESR caps are used and the ESL fo the circuit is low.

It is extremely difficult to make 1us flash pulses with xenon flash
lamps, short of economical not feasible. If one is happy with very
low output, then some of the analytical xenon flash lamps could be
tried.

This is exactly the reason why spark gaps are used for extreme high
speed photography.


YMMV,

BUT...one could try firing a spark gap and using the UV to illuminate
the flashtube...

 

[[email protected]]

Do you know what the failure mode is of your IGBT ?

I'm finding out the major failure mechanism in this application is
indeed during turn-off and the problem invariable is in the IGBT
driver.  You need a push pull driver with separate resistor for on and
off function.  If the resistor for the off function is too small, you
risk blowing up the gate during the transistion.  If it's too large,
your turn-off will be slow and you'll run into potential thermal
issues.  As for the "on" resistor, shorter is better but one has to
stay within the max gatecurrentlimit.  I'm finding out that the
resistor in the "off" leg is about 3 x larger than the "on" function.

On another note, as I read this early on in this thread ...

As for the delay in reaching the max light output of axenonflash
lamp, this is called the "ionization" delay.  It is indeed in the 10us
neighborhood for most photo flash lamps, can be shortened to maybe 3us
if very low ESR caps are used and the ESL fo the circuit is low.

It is extremely difficult to make 1us flash pulses withxenonflash
lamps, short of economical not feasible.  If one is happy with very
low output, then some of the analyticalxenonflash lamps could be
tried.

This is exactly the reason why spark gaps are used for extreme high
speed photography.

YMMV,- Hide quoted text -

- Show quoted text -

What I have managed to find out until now is:
- The failure occurs during the turnoff
- The faulty device ends up being a dead short between all three
terminals (only one device fails each time).
- The device is not noticeably warmer immediately after the failure.
- There seems to be no appreciable V*I*T heating during the turn-on
and steady on state
- The turn on is very fast and very clean. The collector voltage goes
down from 700 volts to practically 0 (given the resolution limitation
of the scope at that scale) in less than a microsecond. This applies
both to the first pulse (which has a gradual current rise) as well as
the second pulse, where the tube is already hot and the current rises
to its full level in about 2 microseconds.
- The gate drive voltage appears to be sufficient as even during the
intitial pulse of several KA the collector voltage never rises
significantly
- The turnoff is very 'dirty', with many high amplitude oscillations
with a period of about 50 to 100nS.

I had captured one of the failures on the PC scope. Unfortunately I
could not save it though because the EMP or something caused the PC to
freeze so I could just look at the trace on the frozen display (with
hinsight I realised I could have taken a photo of the screen!). What
I saw was the same voltage rise initially as in other turn offs, but
the voltage never went all the way up, instead hanging around at about
half the capacitor voltage for a few microseconds and then went down
again as if the transistors had again turned on. It stayed this way
until the cap was fully discharged.

The gate drive is admittedly somewhat primitive and I am probably
asking for trouble and need something better, probably on the lines of
what you suggest. What I have some difficulty with that is regarding
how to determine the minimum value of the turn off resistor. The
problem is that in the datasheet I don't find any specification for
what is the maximum allowable gate current. Is that a value that can
be determined form the other specifications ?

Alternatively, do you think it would be OK if I were to take the
maximum rating of the gate driver IC as the gate current limit I
should aim for ? The drivers I intended to use have a pulse rating
of about 2 amps, both on charge and on discharge. Can I just divide
the gate drive voltage by that current to obtain the best value for
the resistor?

 

[Robert Baer]

What I have managed to find out until now is:
- The failure occurs during the turnoff
- The faulty device ends up being a dead short between all three
terminals (only one device fails each time).
- The device is not noticeably warmer immediately after the failure.
- There seems to be no appreciable V*I*T heating during the turn-on
and steady on state
- The turn on is very fast and very clean. The collector voltage goes
down from 700 volts to practically 0 (given the resolution limitation
of the scope at that scale) in less than a microsecond. This applies
both to the first pulse (which has a gradual current rise) as well as
the second pulse, where the tube is already hot and the current rises
to its full level in about 2 microseconds.
- The gate drive voltage appears to be sufficient as even during the
intitial pulse of several KA the collector voltage never rises
significantly
- The turnoff is very 'dirty', with many high amplitude oscillations
with a period of about 50 to 100nS.

I had captured one of the failures on the PC scope. Unfortunately I
could not save it though because the EMP or something caused the PC to
freeze so I could just look at the trace on the frozen display (with
hinsight I realised I could have taken a photo of the screen!). What
I saw was the same voltage rise initially as in other turn offs, but
the voltage never went all the way up, instead hanging around at about
half the capacitor voltage for a few microseconds and then went down
again as if the transistors had again turned on. It stayed this way
until the cap was fully discharged.

The gate drive is admittedly somewhat primitive and I am probably
asking for trouble and need something better, probably on the lines of
what you suggest. What I have some difficulty with that is regarding
how to determine the minimum value of the turn off resistor. The
problem is that in the datasheet I don't find any specification for
what is the maximum allowable gate current. Is that a value that can
be determined form the other specifications ?

Alternatively, do you think it would be OK if I were to take the
maximum rating of the gate driver IC as the gate current limit I
should aim for ? The drivers I intended to use have a pulse rating
of about 2 amps, both on charge and on discharge. Can I just divide
the gate drive voltage by that current to obtain the best value for
the resistor?

"Turnoff very dirty" suggests (small) inductance and capacitance
resonating fast enough to oscillate a number of cycles before the FET
internal charge dissipates.
But the fact turnon is so clean suggests different active circuitry
during those two different times.
So the fixed passive (wiring, resistors, etc) and semi-passive (maybe
the capacitor in nonlinear mode) parts can be neglected as culprits.
What is left? The FET which has some definitely non-linear gate
capacitance and charge VS voltage characteristics - which must be
different due to the internal stored charge (as a function of time).
You could try a scaled-down system using only one FET and try an
assymetrical gate drive: turnon as you have it now, and vary the turnoff
voltage and current.
Be prepared to use cheap FETs as you may need to do a lot of replacement.
I would start with both ends, using a scope to monitor that turnoff
characteristic.
One end is a minimum turnoff: zero volts and hi Z (500 ohms max,
maybe 100 ohms would be more preacitcal).
The other end is blast the damn thing: maximum negative gate voltage
at maximum gate current .or. maximum gate power (whichever comes first).
The high turnoff drive should give better results, but most likely
not be enough - at least this experiment will let you know where you stand.
So, assuming that to be the case...
Rewind the card reader, unpunch the mag tape and press restart on the
Hoover Dam.
Basic dumb circuit:
Capacitor gets charged somehow from power supply; + is high end, - is
ground; flash tube ties to + side of capacitor and top of turn-on FET.
So add a second FET from capacitor to ground .or. from capacitor to
first FET.
Turn off first FET in "optimal" manner while turning on second FET.
Nasty.

 

[[email protected]]

I tried the first suggestion you made, that is using just one single
IGBT with a proportionally smaller load and various different gate
resistances from 1 ohm to 10k.
As a load I used 15 ohm 5 watt resistor instead of the tube since I do
not have a tube that would draw a low enough curent for one IGBT to
handle. Surprisingly I was getting a reasonably clean turnoff lasting
jsut under a micro second for gate resistors up to about 1k ohm, only
the turn on and turn off time grew longer with increasing resistance.
I only use 0/12 volts drive in all cases. Over 1 K things started to
get a bit ugly, but very different from what i see with the multiple
IGBTs and flash tube. I blew up three load resistors (quite
spectacularly , aparently because the IGBT went into latch up and
the resistor could not really take the strain for more than a 100 uS
or so. However that only happened with unrealistically high gate
resistance.

I had a bigger surprise when trying something else - I went back to
the multiple IGBT setup and replaced the flash tube with a 0.6 ohm
load resistance (which is more or less what the tube shows when lit)
made up of many 1/4 watt 1 ohm resistors in a series parallel
configuration.
Oddly enough the turnoff was reasonably clean, more similar to what I
had with the single IGBT experiment than what I normally get with the
tube, while handling the same current (about 1000 A). There were still
oscillations, but only abou 20V p-p and the turnoff time was also
shorter - less than 2uS compared to the 6uS with the tube. There was
also very little overshoot.

This has got me very puzzled and I am now starting to wonder whether
it is the tube itself that is doing something funny. One of the odd
things (which I'm not sure I mentioned previously) is that apart from
the high frequency ringing the turn off waveform also shows a rather
clean overshoot (of the collector voltage) exceeding the capacitor
voltage by some 25%. The duration of the overshoot is between 1 and 2
microseconds - much longer than the period of the oscillations during
the turnoff and I never quite figured out what is causing it. I don't
think it can be inductance as that would show up as a slow current
rise during turn on and I don't think the IGBT's themselves have any
way of creating a voltage higher than that of the supply. Now that
with the tube replaced by a resistive load I do not have this
overshoot the only explanation I can think of is that the tube is
doing something strange, however to get such an overshoot it would
have to momentarily generate a 300 or so volt potential opposite to
what was applied to it.

The overshoot seems to contain quite a lot of energy as it manages to
charge up the 0.66uF capacitor in the RCD snubber to the peak voltage,
that is about 0.1 joules. It may not appear much but is certainly much
more than I would think could be stored in stray inductances which I
would have thought were the only potential source for the flyback
energy.

One thing I would love to do would be to get a trace of voltage vs
current through the tube, as that would tell me whether all the funny
behaviour is originating there, however my measurement capabilities
are too limited to achieve that.


I have not tried the 'hoover dam' experiment yet, partly because
I made this new 'discovery' (the different behaviour between flash
tube and resistor), but also because I will need to prepare a suitable
high-side driver for the second IGBT.

 

[Robert Baer]

I tried the first suggestion you made, that is using just one single
IGBT with a proportionally smaller load and various different gate
resistances from 1 ohm to 10k.
As a load I used 15 ohm 5 watt resistor instead of the tube since I do
not have a tube that would draw a low enough curent for one IGBT to
handle. Surprisingly I was getting a reasonably clean turnoff lasting
jsut under a micro second for gate resistors up to about 1k ohm, only
the turn on and turn off time grew longer with increasing resistance.
I only use 0/12 volts drive in all cases. Over 1 K things started to
get a bit ugly, but very different from what i see with the multiple
IGBTs and flash tube. I blew up three load resistors (quite
spectacularly , aparently because the IGBT went into latch up and
the resistor could not really take the strain for more than a 100 uS
or so. However that only happened with unrealistically high gate
resistance.

I had a bigger surprise when trying something else - I went back to
the multiple IGBT setup and replaced the flash tube with a 0.6 ohm
load resistance (which is more or less what the tube shows when lit)
made up of many 1/4 watt 1 ohm resistors in a series parallel
configuration.
Oddly enough the turnoff was reasonably clean, more similar to what I
had with the single IGBT experiment than what I normally get with the
tube, while handling the same current (about 1000 A). There were still
oscillations, but only abou 20V p-p and the turnoff time was also
shorter - less than 2uS compared to the 6uS with the tube. There was
also very little overshoot.

This has got me very puzzled and I am now starting to wonder whether
it is the tube itself that is doing something funny. One of the odd
things (which I'm not sure I mentioned previously) is that apart from
the high frequency ringing the turn off waveform also shows a rather
clean overshoot (of the collector voltage) exceeding the capacitor
voltage by some 25%. The duration of the overshoot is between 1 and 2
microseconds - much longer than the period of the oscillations during
the turnoff and I never quite figured out what is causing it. I don't
think it can be inductance as that would show up as a slow current
rise during turn on and I don't think the IGBT's themselves have any
way of creating a voltage higher than that of the supply. Now that
with the tube replaced by a resistive load I do not have this
overshoot the only explanation I can think of is that the tube is
doing something strange, however to get such an overshoot it would
have to momentarily generate a 300 or so volt potential opposite to
what was applied to it.

The overshoot seems to contain quite a lot of energy as it manages to
charge up the 0.66uF capacitor in the RCD snubber to the peak voltage,
that is about 0.1 joules. It may not appear much but is certainly much
more than I would think could be stored in stray inductances which I
would have thought were the only potential source for the flyback
energy.

One thing I would love to do would be to get a trace of voltage vs
current through the tube, as that would tell me whether all the funny
behaviour is originating there, however my measurement capabilities
are too limited to achieve that.


I have not tried the 'hoover dam' experiment yet, partly because
I made this new 'discovery' (the different behaviour between flash
tube and resistor), but also because I will need to prepare a suitable
high-side driver for the second IGBT.

You have done fabousouly!
The evidence points directly to the flash tube.
It is known that a gas arc has negative resistance and so that
characteristic explains (almost) all of what you have seen.
What is puzzling is that turn-on spike; maybe the arc is
transitioning thru a negative resistance region.
It should be fairly simple to see both the tube voltage and current;
make a curect transformer using a toroid (fairly small ferrite - no
larger than 1/4 inch diameter) and run the tube wire thru the center;
say 100 turns on the toroid with 1-10 ohm load to scope.
Tube voltage (if floating like i mentioned) would have to be done
differentially: 2 probes A-B either via crude inverting of B or
differential plugin like the old W or 7A13 or equivalent.
Naturally, this helps you to better understand the problem source but
does not help solve the turnoff problem.
Bypassing the tube with a second FET to discharge the capacitor
cannot solve the problem source; that shunt FET will also get sass
fromthe tube.
May i suggest a nasty solution?
Use an artifical transmission line!
"Perfect" for a fixed ON time, and modifiable for a "programmable"
lengths of time.

 

[Clifford Heath]

This has got me very puzzled and I am now starting to wonder whether
it is the tube itself that is doing something funny. One of the odd
things (which I'm not sure I mentioned previously) is that apart from
the high frequency ringing the turn off waveform also shows a rather
clean overshoot (of the collector voltage) exceeding the capacitor
voltage by some 25%. The duration of the overshoot is between 1 and 2
microseconds - much longer than the period of the oscillations during
the turnoff and I never quite figured out what is causing it.

The overshoot seems to contain quite a lot of energy as it manages to
charge up the 0.66uF capacitor in the RCD snubber to the peak voltage,

I haven't read this entire thread, just bits occasionally, but if you
consider the physics involved in the tube current, there could be quite
a few electrons in flight when you turn off. Those electrons are still
going to hit the end of the tube and impart their charge... except for
some that have just started the journey and get turned back by the -ve
voltage transient.

Someone whose physics is more recent than mine could calculate the size
of the charge imparted based on the known current and voltage prior to
turn-off.

Clifford Heath.

 

[Tim Williams]

Someone whose physics is more recent than mine could calculate the size
of the charge imparted based on the known current and voltage prior to
turn-off.

I'm no plasma physicist, but it seems to be the charges will neutralize
pretty readily. There may be some seperation, due to ion drift towards the
negative. Seems to me it would look capacitive, if anything.

Seems more likely to me that the tube's inductance would do it. I mean,
come on, a kiloamp? How big is your current path? It only takes about,
oh, 0.2uH to store 0.1J at that current -- about a 6" loop? How does the
tube's current path compare with the resistor loaded experiement?

Tim

 

[[email protected]]

Hi, thanks for all the feedback.

This thing is getting curioser and curioser.

There does seem to be some interesting things going on in the tube. I
never really gave much thought to how complex the process might be.
Thinking of the current in terms of electrons flying one way and
possibly xenon ions flying the other way with all the associated
momentum, KE and stuff seems to have some potential for odd things
happening. I mean like Clifford said somehow they have to stop at some
time after the voltage is removed. What the implications of that are
way beyond my knowledge of plasma physics, which is more or less on
par with what my cat knows. Is there enough momentum in there to have
any significant effect on the current voltage characteristics and is
it even relevant - I really don't know, maybe it is far too small to
be significant or maybe not. Also, in some other tests I had done
some time ago I was looking at how the light output decays at turnoff
and found that the light output only starts falling a few microseconds
after the turnoff and takes quite a while (several uS) to fall
appreciably, so I supose there is still quite a lot of chaos going on
inside the tube even after the supply voltage is removed. It seems
very plausible to me that all those hot ions and electrons would not
just let the tube terminals float peacfully to zero volts. However
that supposition could be simply based on my lack of knowledge of how
such stuff works rather than anything else.

As Tim said there is also the possibility it could be plain and simple
inductance at play though I do have my doubts since my load resistor
was placed right instead of the tube leaving all the wiring just as it
was. That leaves only the inductance of the tube itself being
different between the two cases. Also, I have made a simple
calculation on what an inductance of 0.1uH would need if made by
winding a toroid. It resulting (if I calculated right that is) that a
typical way of achieving that would be a toroid with 5 turns on a
100mm (4 inch) diameter air core. That is about 60 cm (2 feet) worth
of wire, a lot more than I have between the various parts of the high
energy circuit. I have also tried as much as possible to have all
cabling in such a way that they are always in antiparallel, so for
instance the pair of wires going to the tube are kept parallel to
eachother all the way to the tube and so on, hoping that that will
cancel out some of the effect of the inductance.

What I shall be trying to do to find out whether or not it is a matter
of tube inductance is to construct a 0.7 ohm resistor which will have
a shape very similar to the tube. This I will make out of a good
number of 1 ohm 1/4 watt resistors trying to mimic as much as possible
the shape of the tube. The tube is a U-shape with each leg about 140mm
long and a tube diameter of some 10mm. I will stick this weird right
instead of the tube and see what happens. Hopefully the resistor will
last long enough to allow me some meaningful measurements.

I'll also be making that current transformer to get some hopefully
more accurate current measurements. The differential voltage probe
however is something that I haven't yet figured out. I checked on
eBay and the cheapest ones cost a fortune so I'll have to bodge
something together myself.

 

[Robert Baer]

I haven't read this entire thread, just bits occasionally, but if you
consider the physics involved in the tube current, there could be quite
a few electrons in flight when you turn off. Those electrons are still
going to hit the end of the tube and impart their charge... except for
some that have just started the journey and get turned back by the -ve
voltage transient.

Someone whose physics is more recent than mine could calculate the size
of the charge imparted based on the known current and voltage prior to
turn-off.

Clifford Heath.

Not electrons, ions.
Time of flight is not relevant here.

 

[Robert Baer]

I'm no plasma physicist, but it seems to be the charges will neutralize
pretty readily. There may be some seperation, due to ion drift towards the
negative. Seems to me it would look capacitive, if anything.

Seems more likely to me that the tube's inductance would do it. I mean,
come on, a kiloamp? How big is your current path? It only takes about,
oh, 0.2uH to store 0.1J at that current -- about a 6" loop? How does the
tube's current path compare with the resistor loaded experiement?

Tim

The charges do not "neutralize pretty readily"; in a standard
fluroscent tube, the ions (after cessation of current) remain for
seconds - more than long enough to apply a current-limited voltage of
35V to ret-start (i have done this many times).