RE: 3 Phase Full rectification, doide protection, HV Inductor needed, advise urged.!

["Tesla list" <tesla@xxxxxxxxxx>]

Original poster: "Leigh Copp" <Leigh.Copp@xxxxxxxxxxx>



	Original poster: Jim Lux <jimlux@xxxxxxxxxxxxx>
	
	At 03:12 PM 9/21/2006, you wrote:
	>Original poster: "Leigh Copp" <Leigh.Copp@xxxxxxxxxxx>
	>
	>Jim (both of you I guess),
	>
	>A few thoughts on the HV rectifier game:
	>
	>I'll agree that while the tubes are a little more resilient, they are
	>pretty expensive to replace.
	>
	>Not putting voltage sharing networks across the individual diodes
	>however would not be something I would recommend, but your mileage may
	>vary ;)
	>
	>Unless you measure your diodes to determine their reverse breakdown
	>characteristic, and ensure that they are all at the same temperature,
	>they will not share voltage, and in a fault condition when the first
	>diode in the string fails, the little inductive snap and the step change
	>in voltage across the rest of the string will cause the spike to go
	>through the rest of your series devices like poop through a goose.
	
	These days, diodes have very consistent reverse breakdown, or, more
	important, reverse leakage properties, but that's actually not the
	problem.  If you over design, none of them will breakdown anyway,
	even with some imbalance.  All you care is that the least leaky diode
	(which will have the most voltage) is still below it's rating.
	
	<LC> - Completely agree that with a sufficient factor of safety in 
the design, snubber/sharing networks are not required. Bear in mind 
that the "old" practices I mention also come from environments 
(industrial manufacturing, commercial and military radio 
transmission) where efficiency (energy, and cost) are sometimes 
sacrificed for extra reliability. (what happens to that one diode 
that -was- out of spec?).
	
	>The internal junction capacitance and resistance of these guys is not
	>necessarily consistent either - yet another reason to balance the
	>unequal impedances with relatively low impedance parallel paths.
	
	I think they are more consistent these days than you think.
	
	<LC> I can't say that I have paid a lot of attention to the changes 
in the specs over the years (shame on me) and I will definitely have 
to look into the latest HV diode specification range - particularly 
the worst case specs. The real test would be the consistency of the "knee".
	
	
	>If my hastily rigged voltage divider and scope had survived at the time
	>(young, poor, and stupid at the time - enough said) I could show you the
	>waveform from one particular episode of the sort I am describing above.
	>
	>And speaking of transients, and reverse recovery time, an R-C snubber
	>circuit across the individual diodes also reduces dv/dt handily to help
	>protect against such things.
	
	
	
	
	>100k ohms in parallel with 0.1 uF is very common when we have a 6-40 amp
	>DC supply to work with (BTW: my frame of reference is 50-800kW plate
	>supplies). The resulting losses may be unacceptable with the size of
	>your dc supply (1.4 amp diodes were mentioned?), however, so you may
	>need to increase "R" and "C".
	
	Exactly.. Say you consider a 5 kW supply (1/10th-1/100th the size
	which you cite), so say you increase the impedances by the same
	factor.. 1-10 Meg, 0.01-0.001 uF.  The leakage current through the
	diodes will start to be more than the current through the equalizing 
resistors.
	
	Not that one can't design appropriate snubbers, but I think that for
	the vast majority of non-ragged edge applications, they might
	actually cause more problems than they solve.
	
	As I recall, the purpose of the C is to slow down the rise time of
	the reverse voltage as the diode starts to turn off.  Modern HV
	diodes have slow (but consistent) recovery (aka "soft") for just this
	reason.  If you do have the parallel C, it's just stored energy next
	to the diode to help cook it, if it does breakdown.
	

	<LC> Right. So then you put another faster, smaller diode in series 
with another resistor, in series with C, all in parallel with the 
original R. (BTW: I am not expecting that to be taken seriously for 
this app.) That method is typical for thyristors (SCRs in particular) 
where the snubber current is designed to be on the order of amps at 
several hundred volts, to provide sufficient damping. And of course 
your reliability is then a product of many time more parts, with 
their own attendant specification variances.


	The other thing is that modern HV diodes are designed to safely
	operate in the avalanche (Zener) mode for a short time (while the
	rest of the string turns off).  For normal sized power supplies (few
	kW max, I would imagine) you can calculate the amount of energy
	deposited in the junction during the turn off event, and it's small
	enough that the device won't fail.  In fact, putting resistors around
	every diode increases the current through the diodes that turn off
	early, actually making the overall string less reliable.
	
	<LC> This (safely operating in the avalanche region) I did -not- 
know; So the reverse breakdown knee is effectively less sharp? Or is 
it at a lower reverse leakage current?
	
	The other issue is that you should trade off using more diodes vs the
	cost of resistors and capacitors. After all, the resistors will have
	to dissipate the reverse current, and both R and C have to be rated
	for the voltage and current they'll see.. they'll probably be more
	expensive than the HV diodes, which are truly cheap.
	

	<LC> This is where the real difference at higher power levels 
occurs, because as we increase the current rating of the stacks, the 
diodes get more expensive than the passives. At the power levels we 
are referencing here however, you are absolutely correct. Moreover, 
speaking from the standpoint of the guy who had to solder hundreds of 
R-Cs on to hundreds of diodes during late nights (more like early 
mornings I think) the labour involved in putting the stack together 
goes up rather substantially. (depends on how you value your time 
there I guess).


	Just as an example, if you put 100k and 0.1 uF across each 1kV diode,
	you'll be dissipating 10W in the resistor (while reverse biased)...
	in reality, it will be more like a few watts average (half cycle, rms
	vs pk).  And, the impedance of 0.1uF at 60 Hz is about 26k ohms,
	which will appear like an extra load of about 30-40 mA.
	
	Just goes to show that this is something worth analyzing, rather than
	blindly following a suggestion from any of us.
	
	<LC> I didn't even quickly calculate it - but that was my guess - 
the passives would be sucking away most of your available current in 
this application.

	There was a discussion of this on the list some years ago.  The semi
	mfrs data book (Philips makes HV diodes intended for series strings)
	goes into some detail on it, as well.  The only thing to be careful
	about is to check the date on the mfr ap notes.  What was common
	practice in 1985 or 1990 may not be so common any more.
	
	<LC> Not to suggest that the guys designing these things are 
necessarily correct, but many still do it that way in 2006! Great 
discussion by the way. Definitely will make me revisit some accepted 
assumptions. Sometimes it is dangerous to accept old design rules as 
being carved in stone as we all know.