small isolated converter

[Harvs]

My current project to go with the eload is a USB power DAQ.  Essentially an isolated multi-channel DAQ with current shunt on one channel.

To maximize versatility I want to isolate each of the channels (voltage from current measurement etc.)  That way it'll be easy to monitor various voltages in a DUT without concern for common mode voltages.

Previous designs I've just used off-the-shelf PCB mount dc-dc converters.  However, in this case I'm putting in a number of channels (at least 4), so I'm thinking I could build an array of small forward converters using the cheaply available coupled inductors.  As with the low end off the shelf items, there's no need for feedback, and they'd be followed by a 3.3V LDO.  As there's no feedback, a simple ~500kHz-1MHz square wave drive signal can be generated by the controlling micro, and could also simply interleave the channels to reduce input ripple.

Each channel only needs a few mA capacity to drive a low power op-amp, ADC, reference and isolator.

I'm not trying to achieve a safety isolation (i.e. it will be used on DC circuits with up to around 50V max common mode between any converter.)

Attached is the circuit idea, burns about 1.6mA per converter in magnetizing current, but that's fine. I've implemented the circuit and it behaves nicely so far.

I'm just wondering, obviously this isn't a new thing.  Are there any better ideas?

EDIT: changed attachment to include the ferrite beads I left off the first.

[jeroen74]

To be honest, I'd stick to the COTS DC/DC converters. Another approach is to use isolated amplifiers like the ISO122 from TI.

[Harvs]

Even with the isolated amps I'd still need a power supply across the isolation barrier, so I'd prefer to just locate the ADC on the isolated side and use a digital isolator, they're cheap and reliable.


Why wouldn't you bother building a small DC-DC converter?  Just because of added complexity?  Fair enough, the lowest cost in small quantities (10x) are only $3.14 ea of digikey. 


Still most of the circuitry below including the filters are still needed.  So by using a "roll your own" version saves about $2 per channel.  Not a lot I guess, but two diodes, an off the shelf coupled inductor and a 2c mosfet doesn't seem like an overly complex undertaking.  Am I missing something?

[jeroen74]

Oops, my bad. I thought these contained their own isolated power supply But the this one from Analog Devices does.

The reason I would use ready made stuff it that it saves time and possible headaches later on.

[Harvs]

Cool, didn't know such a thing existed.  However,  that one is >$70ea on digikey.

[jeroen74]

Quite (a bit too) expensive indeed

[jeroen74]

If noise is of any concern, consider the LT3439. I used it together with self wound transformers a few years ago.

[rbola35618]

If this is a flyback, then the transformer phasing is incorrect.

RB

[HackedFridgeMagnet]

The OP said it was a forward converter. So phasing is ok.

[rbola35618]

If it is a forward then you have an diode missing in the output filter . You would need an inductor on the output as well. Not sure a ferite bead would be good enough to use as an output inductor. Recheck the application note.

[jeroen74]

Doesn't a forward converter generally need a reset winding with a diode? It more looks like a boost converter.

[rbola35618]

Yep, a forward would need a reset winding for a single switch converter. A two switch forward does not need a reset winding. I think it supposed to be a flybacik. If it is, all you need is to reverse the secondary and you good to go.

Robert

[sorin]

Oops, my bad. I thought these contained their own isolated power supply But the this one from Analog Devices does.

The reason I would use ready made stuff it that it saves time and possible headaches later on.

try HCNR200

[jeroen74]

That's simply an optocoupler with an extra photodiode to compensate for any temperature effects. Does not solve the power supply issue.

[westfw]

Doesn't a forward converter generally need a reset winding with a diode? It more looks like a boost converter.

For the application, it sounds like you can skip most of the traditional switchmode topologies, and just generate AC to drive a traditional transformer/rectifier/filter/regulator arrangement.  By using a higher frequency (>> 60Hz), you get to use smaller magnetics and caps.

[Harvs]

It is a forward converter of sorts, but the power levels a so low that you can dispense with some of the stuff you would normally have in a forward converter.


The purpose of the reset winding on a forward converter is obviously to allow the core to return to zero flux.  Normally the reset winding is configured so that when the mosfet is off, the magnetizing current is returned to the input caps.  In this case, instead of having an extra winding, I'm simply burning the magnetizing current into a resistor (R4) via D2.  The resistor is sized to give a peak mosfet drain voltage of about 15V.  Driving a 1mH transformer at about 500kHz results in an idling loss of 1.6mA.  While that makes the converter stupidly in-efficient as a ratio of input to output for small currents, it's low enough to not matter for this application and certainly doesn't warrant the use of an extra winding.


As for the output inductor.  Again because current levels are so low, and the winding ESR is actually quite high (couple of orders of mag larger than the cap ESR) adding the usual output inductor doesn't make that much difference to output noise.  However I did just simulate adding a 10uH chip inductor, and the flying diode that goes with it. The reduction in peak currents in the circuit makes it worth it, so I'll add that.


As for control, I could add peak current limiting.  However, the ESR of the winding results in a short circuit current limit of about 190mA, further considering the output itself is contained within my own board (i.e. nothing external could be hooked up to it), I'm not sure it's worth the effort.  Certainly the RECOM converters I've bought before don't have any proper current limiting and actually die if you leave them with a shorted output for any length of time.


I'd certainly use a multi secondary transformer if there was a reasonable selection of very small low power items off the shelf for reasonable money.  There is one brand of 6 winding inductor/transformer on digikey for about $8 which is quite reasonable, and I may go that way.  Basically it'd be the same circuit though, just with multiple secondaries.

[Harvs]

Doesn't a forward converter generally need a reset winding with a diode? It more looks like a boost converter.

For the application, it sounds like you can skip most of the traditional switchmode topologies, and just generate AC to drive a traditional transformer/rectifier/filter/regulator arrangement.  By using a higher frequency (>> 60Hz), you get to use smaller magnetics and caps.

You can. Ive used small pulse transformers to generate mW isolated power. I've run them open loop clamped output with a micro. You could possibly if your so inclined also use the micro's on board comparator for primary side current limiting. I think I used a PIC10F.

Thanks I hadn't thought of pulse transformers, they're probably better suited than the coupled inductors, and you can easily get ones with 2 secondaries.

[rbola35618]

I am not sure about that the transformer will reset. The flux might rachet up and destroy the mosfet. The way I see it, when the mosfet is on, the top of the primary transformer is connected to VDD and the botton to essentially ground. In the simulation, you have equal ton and toff. Assuming 500KHz, that gives you 1us for both ton and toff. When the mosfet is turned off, the botton of the primary will fly to positive and the top of the transformer goes to negative. This means that the diode is turned on. That means that the primary is clampedf to -Vdiode drop (around 0.7volts).

For the transformer to reset, it must satifly the volts second rule. So at ton the primary is Vdd and at Toff is at -Vdiode

Therefore   volt-sec-rule 

Tp = period

Vdd*Ton = Vdiode*Toff =  Vdd*Duty*Tp = (1-Duty)*Vdiode 
 
Since the Ton and Toff are equal, then Vdd and Vdiode have to also be equal for the volt second rule to be satified.   

The only way to satifly the volt sec rule is to have a dutycycle of 12%.

And if that is the max dutycycle to ensure resetting, then that max voltage that you will get is Vdd*Duty = 600mV

Am I missing something and having a senior moment?

Robert

[Harvs]

I don't know about a seniors moment or not 


But that's what R4 is there for.  Actually in the original post it's only 2.7k, it's better off being about double that.  Because the duty cycle is fixed at 50%, the magnetizing current at the end of any one cycle (presuming it resets completely) is fixed.  Then you can sizing the resistor to give a peak voltage that's well within the Vds range of the mosfet yet still provide enough volt*secs to reset the core.


I've changed the output to include the inductor and changed the gate resistor and reset resistor to more appropriate values and attached the LTspice file in case anyone is interested.


I'll have to look into the PIC32's on-board comparator then for cycle by cycle current limiting.

[rbola35618]

Now I got it. Putting R4 in series with the diode will allow enough voltage to build up and reset the transformer. This is neat little circuit. I will have to remember this circuit for sure. Awsome design guys!!!

[jeroen74]

Have you already tried the circuit with real components? Always interesting to see how well an actual circuit matches the simulation

[Harvs]

Have you already tried the circuit with real components? Always interesting to see how well an actual circuit matches the simulation

I have actually.  Right at the start I built up the first circuit just without the linear reg and ferrite beads (as they're pretty predictable.)


I had to hand wind a small toroid I had laying around to get something close to the off the shelf inductors.


Actually scary how close it was to the simulation.  In this case I guess because the current is so low, stray inductance and like made very little impact on the real circuit.


Tonight I've just been rigging up some gear to play around with the PIC32's on-chip comparator and PWM fault shut-off timings.  Dam datasheets seems to overlook giving any figures on that (or maybe I've just missed it somewhere in the 3 or 4 datasheets/reference guides you need for 1 chip     


Anyway I thought it'd be cool to add a resistor to the mosfet source as the current feedback in the simulation, dump that plot into my AWG and play a "real" waveform to the PIC to watch how long it takes to turn the output off as the current ramps up.  Turns out that's almost a project in itself trying to work out all the foibles of the Rigol's waveform input.  There is something kind of cool about it though, to simulate a circuit in LTspice then replay a voltage node in "real life".

[rbola35618]

I agree with jeroen, seeing simulation and actual circuit are always good to have. I would also measure the efficiency. I thought that you would have a lot of power dissipated thru R4. But made some calculations and it turns out that R4 will dissipate about 10 mW. I would be interest to see what the efficient is of the forward converter first without the LDO.

RB

[Harvs]

I agree with jeroen, seeing simulation and actual circuit are always good to have. I would also measure the efficiency. I thought that you would have a lot of power dissipated thru R4. But made some calculations and it turns out that R4 will dissipate about 10 mW. I would be interest to see what the efficient is of the forward converter first without the LDO.

RB

Average idling input current (i.e. no load) on the simulation was about 1.7mA.  In real life I measure 1.56mA on my DMMs 200mA range (but who knows exactly how accurate it is down at that level.)  Given I only need 1-5mA out of it, efficiency as a ratio is rubbish because I'm not recycling the magnetizing current, but it doesn't matter for this application, as it's USB powered.


All the scope traces were pretty much spot on as per the simulation.


The next build up will include the PIC and actual parts, so I'm going to have to wait till I've got my Digikey order all sorted to submit.  Unfortunately here postage is very expensive from digikey if the order's less than $200.  So I tend to get it all together for a project and order at one time.