Need input on high voltage regulator design

[Joerg]

Hi Graham,

That's the sort of thing I meant.

One of the folks here on s.e.d., I believe it was Chris, once mentioned
that many higher end battery-to-mains inverters are run with a uC being
the PWM controller. Unfortunately there isn't much out there in terms of
application notes.

Regards, Joerg

 

[Joerg]

Hi Frank,

Could you replace the D10 diode with a transistor, emitter to ground?
Then you get a digital signal that tells if the voltage is above
or below 500V, which you can use to increase/decrease the pwm.

You'd also need a resistor from base to emitter, to create a real
voltage divider.

I assume you are trying to keep current consumption to a minimum,
running everything from a 9V battery?

Then R9 and R10 would need to be very large values. Unless I am wrong I
believe Geiger counters only draw current when radiation triggers them
so in dormant mode this sense circuitry might be the only significant
load on the HV side.

Regards, Joerg

 

[Joerg]

Hi Scott,

The neon lamp tubes? I've seen that done. I'm using 250v varistors
instead. Doesn't seem like it would be any more efficient - it's still a
shunt regulator, and will put a constant drain on the PSU even when the GM
tube's not conducting, right?

That shunt would drain your battery rather quickly. Q1 keeps pumping in
energy and the varistors just burn that off. Also, please note that the
metal oxide variety has a finite lifetime. They are like a bank account,
after so many joules they go kaputt.

Regards, Joerg

 

[Scott Miller]

Could you replace the D10 diode with a transistor, emitter to ground?

Then you get a digital signal that tells if the voltage is above
or below 500V, which you can use to increase/decrease the pwm.

One of the circuits I based this on had a similar setup, but it just kept
the primary driver disabled while it was on. I tested it and it didn't seem
to do much good. I may experiment with this some more, I'm just a bit
worried that having only the single reference might make it hard to maintain
regulation.

Seems like the shunt would still be burning off power even in that setup -
though I guess it would depend on what kind of current limiting resistor I
put inline. Is that right?

There's still the matter of the limited lifespan on the varistors... not
sure they should really be there at all.

I assume you are trying to keep current consumption to a minimum,
running everything from a 9V battery?

Right. The version I built for myself ran off two AAs with a charge pump to
drive the LCD and MCU, but going with 9 volts lets me replace that with a
much cheaper linear regulator and reduce the number of multiplier stages.

Thanks,

Scott

 

[Scott Miller]

Shunts are inefficient because they regulate by 'burning off' the excess.

I'd just use it just as a feedback. How to do that with a TLV431 is
explained in a TI app note about isolated flyback converters. I forgot the
number but you should be able to find it.

I'll take a look.

minimum 'cathode' current of 100uA. Since your application appears not to
be an isolated one this feedback circuitry will become much simpler than
in the TI note.

Isn't it? It's using a transformer with isolated windings. Doesn't that
make it an isolated flyback converter? Or does that not count because it
feeds back into the +9v rail through the shunt?

Scott

 

[Scott Miller]

Well.... for next to no current draw you could choose a simple inverter.
Simply
choose the turns ratios to suit the supply and take into account Vce sat
on the
primary driver. Wind it so the primary takes next to no magnetisation
current.

Should do the trick. And you could still tickle it with your MCU !

Gah... I think that went a bit over my head. You're saying to just use an
appropriate transformer to get the required output voltage without the
doubler? Not something I really want to do for cost and time reasons - the
CCFL transformer is a cheap, off-the-shelf part.

As for the primary driver, I replaced the one originally specified in the
design (no longer made) with a BS170 MOSFET set up as a low-side switch.
Not sure how that would affect what you're suggesting.

Scott

 

[Scott Miller]

That shunt would drain your battery rather quickly. Q1 keeps pumping in

energy and the varistors just burn that off. Also, please note that the
metal oxide variety has a finite lifetime. They are like a bank account,
after so many joules they go kaputt.

Indeed it does... draws about 40 or 50 ma, with < 10 of that being the MCU
and display. Didn't know that about MOVs - one more reason to find another
design.

Thanks,

Scott

 

[Scott Miller]

The TL082 is still available from Radio Shack- as well as 10Meg 1/4W
I'm luck if my local RS has ANY parts. No problem, I order from Digi-Key
almost every week, and I have 10Meg 1/4W resistors by the hundreds. Use 'em
for feedback in MCU clock oscillators.

If you go with the TL082 and don't need the buffer, then a good use for it
would be as low-pass filter and buffer for the 5V shown in the
schematic...this will eliminate your 5V switching noise from the MCU and
other digital switching you have going on there.

I'll probably try out the direct version and see how much noise it's got.
Would a simple RC network make an acceptable filter if there's a lot of
noise on the op amp output?

Thanks,

Scott

 

[Scott Miller]

What does the ADC use as a reference?

The MCU's supply, which is provided by a 5-volt linear regulator in this
case. It's fixed internally, can't select a different reference.

How good does the regulator have to be?

500 volts, -50/+150 according to the GM tube's spec sheet. So not very. =]

You could sample the primary windings voltage some time after the MOSFET
is switched off. The waveform will look like this:

The voltage you want for this sort of feedback is the one at (C). The one
at (B) is almost as good. From (B) to (C) is almost a straight line.

You should be able to get fairly good regulation. The tricky bit is that
the micro needs to fiddle with the ADC timing to get the right point for
the feedback. The nice thing is that the load doesn't change much so the
software can take a bit of time.

Sounds kind of tricky. With the voltage doubler stage after the secondary,
how sure can I be that what I see on the primary will reflect the actual
output voltage?

The ADC will give a zero output up to some voltage and then go up towards
full scale as the volatge increases above that point. This allows the
ADCs bits to be put to better use since you know that the voltage must be
near 500V

This is more like what I was looking for. Fred Bloggs posted something
similar. Between the two examples, I ought to have enough to keep me busy
experimenting for a bit. Just need to check my parts bins for an
appropriate op amp, or go pick one up.

Thanks...

Scott

 

[Joerg]

Hi Scott,

Indeed it does... draws about 40 or 50 ma, with < 10 of that being the MCU
and display. Didn't know that about MOVs - one more reason to find another
design.

I'd just go with a resistive divider and feedback. BTW, 10mA for uC and
LCD display sound a bit high as well. That alone would empty a regular
9V battery in under 24 hours. I guess it's too late now but next time I
would look into lower power uCs like the MSP430.

Regards, Joerg

 

[Scott Miller]

I'd just go with a resistive divider and feedback. BTW, 10mA for uC and

LCD display sound a bit high as well. That alone would empty a regular 9V
battery in under 24 hours. I guess it's too late now but next time I would
look into lower power uCs like the MSP430.

Yeah, it's way high. I can reduce that by 2 ma by replacing the regulator,
and drop the remaining draw by a huge amount just by setting it up to use
low-power wait modes and reducing the system clock to something more
reasonable. The only reason it's so high is that I wrote the code in a
hurry as a proof-of-concept and never got around to making it efficient.
Didn't seem worth bothering with when the rest of the circuit was drawing
far more juice.

The LCD draws around 1 ma I think, though it originally had its backlight on
permanently, drawing waaaay more. Had to desolder a couple of resistors to
turn it off. I've got more appropriate displays, I just happened to have a
drawer full of these that I picked up for like $1.50 each.

If I can find an appropriate case, I'd prefer to go with 6 AAs. Or even 3
CR123As, but that gets expensive. I don't like 9v batteries.

Scott

 

[Scott Miller]

If you go with the TL082 and don't need the buffer, then a good use for

I'll probably try out the direct version and see how much noise it's got.
Would a simple RC network make an acceptable filter if there's a lot of
noise on the op amp output?

Ok, so I checked my parts bin and found a TL082. And now I see that it's a
DUAL op-amp, so I get the part about using the other one as a filter. Duh.
Ok, time to go experiment a bit.

Scott

 

[Joerg]

Hi Scott,

The LCD draws around 1 ma I think, though it originally had its backlight on
permanently, drawing waaaay more. Had to desolder a couple of resistors to
turn it off. I've got more appropriate displays, I just happened to have a
drawer full of these that I picked up for like $1.50 each.

With a serial controller on there? That is one heck of a deal. Much more
$$$ out here in Northern California :-(

If I can find an appropriate case, I'd prefer to go with 6 AAs. Or even 3
CR123As, but that gets expensive. I don't like 9v batteries.

Or dump the regulator and run it with three or four cells.

Regards, Joerg

 

[Jonathan Kirwan]

<snip>
The neon lamp tubes? I've seen that done. I'm using 250v varistors
instead. Doesn't seem like it would be any more efficient - it's still a
shunt regulator, and will put a constant drain on the PSU even when the GM
tube's not conducting, right?

Yeah, it used power. My supply for it, back then, was two 45V telephone battery
packs (bunch of 1.5V AA types in a package.) So I had 90V to work with and for
my needs the battery lifetime was just fine -- memory is thin and I've got it
packed away so I cannot easily verify, but I think it ran for about 8-10 hours.
I used it intermittently for testing rock samples (autunite, for example), so I
could be off by quite a bit.

Jon

 

[Scott Miller]

With a serial controller on there? That is one heck of a deal. Much more

$$$ out here in Northern California :-(

Not serial, standard HD44780 parallel. I've got it running in 4-bit mode,
write-only, so it takes 6 I/O lines. I just checked my old emails - it was
a special from Electronic Goldmine, part number G14023, $1.49 each, back in
January. No such deals at the moment that I can see. I think I bought 10 -
should have gotten a few more.

Scott

 

[Scott Miller]

Yeah, it used power. My supply for it, back then, was two 45V telephone

battery
packs (bunch of 1.5V AA types in a package.) So I had 90V to work with
and for
my needs the battery lifetime was just fine -- memory is thin and I've got
it
packed away so I cannot easily verify, but I think it ran for about 8-10
hours.
I used it intermittently for testing rock samples (autunite, for example),
so I
could be off by quite a bit.

That sounds about right. My first design used a tube from an old Civil
Defense meter. Those things usually took B or lots of D batteries and
didn't last too long. I see counters today running for thousands of hours
on a single battery. Just wish I could find a schematic for one of those
supplies.

I've got a couple suggestions and an op amp to play with, so I'll probably
see what I can do tonight.

Scott

 

[Fred Bloggs]

Ok, so I checked my parts bin and found a TL082. And now I see that it's a
DUAL op-amp, so I get the part about using the other one as a filter. Duh.
Ok, time to go experiment a bit.

Scott

You can clean up the 5V feed into the op-amp like so:

View in a fixed-width font such as Courier.

9V 5V
| 100K |
+------+---------+-/\/\-+
| | | |* |
/ / | === /
100K 100K | | 10K
/ / | --- /
\ \ | gnd \
| |TL081| |
10M 10M 10M 10M 10M | | |\| |
HV IN >-/\/\-/\/\-/\/\-/\/\-/\/\-+----------|-\ |
| | | >-|<|-----+--> V
5x 1/4 Watt | +---|+/ | out
| |/| |
| 1.2M | |
+----/\/\----|----------+
| |
| 620K | *
+----/\/\----+ 0.1u||0.01u
|
|

 

[Pooh Bear]

Gah... I think that went a bit over my head. You're saying to just use an
appropriate transformer to get the required output voltage without the
doubler? Not something I really want to do for cost and time reasons - the
CCFL transformer is a cheap, off-the-shelf part.

Which is where the compromise is.

As for the primary driver, I replaced the one originally specified in the
design (no longer made) with a BS170 MOSFET set up as a low-side switch.
Not sure how that would affect what you're suggesting.

Yeah, I forgot you had a mosfet there.

You're running the primary from +9V ( regulated ? ). You need a 55:1 ratio step
up TX. and the output will be 500 V.

There's a bit more to it than that but basically you don't need feedback
regulation if the supply voltage is regulated. Certainly not for next to no
load.

And the transformer needs to be in the opposite phase. Your current design is
flyback which can only regulate by feedback or ( as in your case ) shunt
regulation.

Google 'single transistor forward converter' and ignore anything about feedback
- the turns ratio sorts it.

You might google 'inverter' too for good measure.

CCFL backlight transformers are designed as inverter transformers btw.


Graham

 

[Scott Miller]

You're running the primary from +9V ( regulated ? ). You need a 55:1 ratio

step
up TX. and the output will be 500 V.

There's a bit more to it than that but basically you don't need feedback
regulation if the supply voltage is regulated. Certainly not for next to
no
load.

And the transformer needs to be in the opposite phase. Your current design
is
flyback which can only regulate by feedback or ( as in your case ) shunt
regulation.

Google 'single transistor forward converter' and ignore anything about
feedback
- the turns ratio sorts it.

Ok, thanks... I'll do some more research.

BTW, it turns out that using a 40Meg / 330K voltage divider, I was able to
get feedback directly to my MCU's ADC. Now there's the little matter of
figuring out the algorithm to best control the PWM. The point at which you
take the A/D sample does seem to be really important - I'll have to figure
out how to sync it up. Fortunately, I've got a dual-trace DSO that does a
pretty good job of showing me what's going on between the PWM input and the
voltage output.

A thoroughly educational experience all around...

Scott

 

[Frank Bemelman]

Hi Frank,


You'd also need a resistor from base to emitter, to create a real
voltage divider.

Would I? I was hoping that when below 500V the current passing the
two 500V vdr's is so small that it does not turn on the transistor.

In the current setup, the pwm needs to have a safe overdimensioning
large enough keep those VDR's regulating, and wasting power. With
the transistor, you could back off the PWM, keeping it on the edge,
with the transistor turning off/on etc. With all the soft knees of
the VDR and transistor, you'd still waste power, but perhaps not
as much.

Then R9 and R10 would need to be very large values. Unless I am wrong I
believe Geiger counters only draw current when radiation triggers them
so in dormant mode this sense circuitry might be the only significant
load on the HV side.

That seems the case. Perhaps it is better to use a comparator
or opamp + adc input and very high value resistors.