[Prev][Next][Index][Thread]

Re: Re : DC power supply again




From: 	Harri Suomalainen[SMTP:haba-at-cc.hut.fi]
Sent: 	Saturday, September 13, 1997 5:42 AM
To: 	Tesla List
Subject: 	Re: Re : DC power supply again

On Sun, 7 Sep 1997, Tesla List wrote:
> > Not nessessarily. Resonant topologies often have some LC-circuit in
> > the primary side. Sometimes it is a series LC network, sometimes
> > a series L with C in parallel with the transformer. Those do not
> > put huge voltage stress like a normal bridge with just a choke in
> > series with the transformer would.
> 
> Don't disagree. However, the primary choke in a resonant topology is 
> not there to boost output voltage.

Usually not. Boosting it could very well be done by it too but it makes
no sense to boost the voltage at the primary side.

> As I said, it designed for 2.3kW throughput in flyback mode. It would 
> doubtless be capable of a lot more in an H-bridge.

Oops, I had already forgotten that. It is a reasonable size for that
much power.

> > BTW, what's the diode stack about? The simples and most reliable thing
> > for fast hv diode stack could well be a lot of avalance diodes (like
> > BYV96E) in series. However, that is certainly not the cheapest one.
> > Designing a fast stack is not fun at all with normal diodes!
> 
> _No way_ can you use ordinary diodes in this application, no way! The 
> BYV96E is also a bit of a slug (250nS rev recovery). I have chose

It is not fast, agreed. However, it is *not* an ordinary diode. It is an
avalance diode. Avalance rated diode acts like a zener at too high
voltage.

Too high voltage can be generated either by unequal capacitances or
unequal reverse leakage of the series connected diodes. As the voltage
rises "too high" for some acalance diode it will start acting like
a 1000V zener and limit the voltage (dissipating some heat). Therefore
voltage is actually equally shared by the diodes! SGS-Thompson application
note AN443/0691 covers details fairly well. Available at their web-site.
It suggests using transils (~= fast zeners) across the diodes to limit
voltage.

The other choise is that the diode itself can act like a zener. BYV96E
is rated for 10mJ (nonrepetitive) avalance energy. As the capasitance
at 1kV is around 10pF (and generally max 75pF) the heat dissipated
will not be huge in proper cases. Total heat losses are the limit.

Try some calculations of power dissipation on series-connected BYV96E
diodes. Espesially with resonant converters with lower dv/dt the reverse
recovery time will be very acceptable (if not at very high frequency).
You'll be supprised to see the series-connection of avalance diodes with
*no* parallel R/C is indeed possible and withing the specs.

The same idea has been already used and is mentioned at least in one
article I happen to have around. Electronics world October 1996 has
article named "Stepping out". It mentions the idea for a 16kV diode
in a 160kV supply.

With normal diodes this certainly would *not* be possible. Normal diodes
need always cap+resistor equalizer network in parallel.

> UF5408 - 3A I(fwd), 50nS rev recovery time, 1kV holdoff. I was 
> careful to choose not only appropriate rectifiers for the original 
> application but some that I could use in a forward converter as well.

Flybacks etc with "square" output waveform do indeed need fairly fast
diodes, especially at higher frequency. Resonant topologies are much
better in that way too.

> > That's quite a high B. I'd assume core losses at B peak-to-peak of
> 
> I carefully checked the Hanna curves and bulk loss curves before 
> deciding on that figure, then built and ran a smaller one at the same 
> level. There is not a huge amount of difference between 200 and 250mT 
> but I wouldn't go beyond that. This seems fine for the core material I 
> am using. The core barely gets warm after running for several minutes.

OK. Tests and curves are good proof.

> This is for _the particular material_ I am using. One could easily be 

Naturally.

> deceived by this because most people typically run saturating core 
> inverters and then wonder why the core gets hot (we are talking 
> typically 400 - 450mT pk in this app for the N27 type material).

Yap. I've also seen cores get very hot at much lower B when the frequency
is higher. Usually core loss is related roughly to f^2. Running too much
at the high frequency and difficulty of getting *good* cores is probably
the reason I'm so pessimistic about the core losses..
--
Harri.Suomalainen-at-hut.fi - PGP key available by fingering haba-at-alpha.hut.fi