Notches in ground planes for multi-power multi-channel board

In our area BJT means BJ's "Tatonka Stout", from these guys:

Hi,

Is this npn-darlington solution good at rejecting HF better than a linear regulator like the LM317? Also where do the discharge diodes go in this circuit?

cheers, Jamie

Yes, for smaller transistors and small currents. Mostly I don't even use Darlington, just a transistor with a good guaranteed minimum beta. One diode goes from emitter to collector (cathode to collector which would be at the input rail), the other from base to collector (also cathode to collector). The diodes make sure the capacitors don't zap the transistor(s) in case someone drops the pliers and abruptly shorts out the input rail.

Remember that this drops 700-800mV or so, or twice that in case of a Darlington. And it is not too useful if the output of this stage must be very stable. Of course, then you could run a secondary loop back to the switcher but that becomes esoteric and is not for the faint of heart.

I am making an isolated bipolar powersupply that needs to be quite clean, for 12bit ADC/DAC's, here is what I have so far:

"

For the npn solution I put in a pnp on the negative rail, not sure if this is correct, or if it should still be an npn?

Thanks for any comments on the circuit. I would like to get as clean a supply as possible, maybe 0.5mV noise would be nice on the +-5V rails.

cheers, Jamie

"

It's correct, pnp for the negative rail. But you need to flip both D117 and D118 around.

I don't think you need those high PSRR regulators. The LM317 is actually pretty good. Don't know what you are using for the negative rail though, your schematic doesn't say. Also, you need ADJ pin dividers, else you'll only get 1.2V out of it. Bypass that ADJ pin to GND with a 10uF MLCC cap for even better noise muffling.

Add a 0.1uF each in parallel to C22 and C29 to make sure RF spikes won't sail through. When picking a negative regulator watch out. Many of those are LDOs and LDO usually means trouble. Make sure it's unconditionally stable.

I don't understand this. It looks like a pwm controlled forward converter, but the secondary circuit (D110, L5, C29) looks strange. There's usually another "catch" diode to keep the current circulating in L5 when the pwm is off. L5 might ring like a banshee when the pwm snaps off.

As far as 100 KHz noise is concerned, you have more normal-mode ripple rejection than you need. The bigger problem might be common-mode noise on the (floating?) common, coupled through the winding capacitances.

John

"

Thanks, I updated the circuit:

"

I used a LM337 negative voltage regulator on the negative rail. I saw an AC regulator circuit in the LM317 datasheet but am not sure if you can really use the LM317 for a negative regulator for this?

I left the second voltage regulators in the circuit in case 8V and 5V are both needed :)

cheers, Jamie

"

Thanks, I added those diodes.

Would a common mode filter be the solution for this I guess? I've seen some capacitors that can be used for this I think, they have 4 terminals. Also would a common mode choke from each rail to the floating ground be enough to get rid of the common mode noise or are there other techniques to use?

cheers, Jamie

Hi, Jamie,

I couldn't make suggestions without understanding the application. Is the circuit that this thing powers floating? What does it do?

We recently did a 16-channel thermocouple input board. Each channel has a dc/dc converter based on an ISDN line transformer, driven by some small mosfets at 60 KHz. The tc input channel is fully floating, with optoisolators for data i/o. The biggest noise problem is common-mode 60 KHz coupled into the floating circuits through transformer capacitance. It's only a problem in the sense that we don't want to poke a lot of 60 KHz into the customer's thermocouple leads. We added a 1 (or maybe 2?) nF cap from the floating ground to real ground as a compromise... too big a cap has problems of its own.

John

"

"

Not that way. I prefer the LM317 over the LM337 because the 337 has the architecture of an LDO and is not as stable. But it would require another transformer winding. Or maybe you could used the one at pins 9 and 10 with a bridge rectifier. Then you could build a separate +8V supply and just tie the positive side to GND. But you should be able to get the LM337 to work as well, it's a simpler circuit.

Ok. 3V ain't a lot of head room but for 100mA it's fine.

"

Good point, I had only looked at the DC side. Also, the primary driver side does not look right.

Jamie, the best way to prevent DC runaway of the core is to AC couple. Then second best is to use bridge rectifiers (and then you could actually use a LM317 for the neg rail as well).

Or at least connect the anode of D119 to xfmr pin 7, flip D120 around and connect its cathode to xfmr pin 4.

"

That's actually a pretty slick configuration. When the fets turn off, the flyback energy is dumped back into the power supply.

Looks like Q92 drives the gate of p-fet Q91, also slick. Interesting things happen over the pwm range from 0 to 50%, but I think it's OK.

Why didn't I think of that?

John

Hi John,

Yes its a floating supply for powering ADC/DAC and opamps that are interfaced to by opto from the microcontroller side.

The primary side is the isolated microcontroller supply, and the secondary side is 120VAC generated by a truesine inverter.

Thanks I'll add the 2nF cap across grounds. Would a common mode choke, maybe 3 coils in this bipolar supply case be effective as well?

What about these 4 terminal caps, they seem to be more effective than a larger common mode choke.

cheers, Jamie

"

Terry Given suggested that one in a thread earlier :)

cheers, Jamie

"

If Terry gave it to you then it should work. But make sure to obtain enough information regarding the core. A DC run-off situation or core saturation is no fun.

"

high praise indeed.

Thanks Joerg!

I came up with that circuit in 1994 - I did a 2MHz modulated smps/gatedriver circuit. initially I had a clamp zener, but I used the same zener for all 6 gatedriver/smps, and although the leakage energy was OK from one, 6 was too much. so I thunked a bit, and figured out the self-driven PNP diagonal half-bridge. the switching speed of the PNP/PFET is pretty much irrelevant too - as long as one device switches fast its ok - at least until you muck about for so long that the next switching cycle has been and gone.....

we built about 300,000 of these, with no problems. running from +24V, using FMMT491A and FMMT591A SOT23 BJTs. We could probably have patented it - though we did patent the toroid winding technique I came up with.

as far as the core is concerned, in theory all is well even at 50% duty cycle, as Vclamp = Vcc + 2*Vd, but Von = Vcc - 2*Vce. ware remanent flux though - that can bite quite a chunk out of the available flux excursion, rendering the typical "hey, Bsat < 300mT, no worries" approach somewhat fraught with peril

its really good for open-loop flybacks too, if you choose Vout = Vin*Ns/Np, as load reductions just mean more energy spins round in circles on the primary.

Cheers Terry

"

BJT's? With base resistors? It looks much nicer with fets.

Nice circuit, a few parts doing a lot of stuff. The only warning I'd give Jamie is that he may not want fast edges getting into his isolated analog stuff, and it would be tricky to slow this one down. Our thermocouple supply was a classic open-loop forward converter, dual nfets driving a center-tapped transformer, and we shaped the gate drives to soften things up. Efficiency didn't matter much here. And we got a nice clean square wave out of the secondaries, which also clocked the delta-sigma ADC!

John

"

Hey, you re-invented the modern electro-cardiograph :-)

Except that we (usually) feed back the analog signals and do the AD conversion on the system side.

they were cheaper. and the 6 npn base drives had to go a long way on a single layer (no 0V plane) to get from the CPLD to the gatedrivers, so Vbe being < Vt was a good thing.

Lovely!

Cheers Terry