Voltage sampling in high power high voltage buck converter

[skn96]

Hi Folks
I have designed a DC/DC converter 3kW, 300V input /150V output buck converter that for now works in open loop fashion using a micro controller. Now I wanted to implement a voltage loop control. From my previous experience the easiest way is to use voltage divider for 150V output and then use an op-amp as buffer that goes to ADC pin on micro-controller. But I was told that since this is a high power and relatively high voltage design I better use an isolation amplifier to sample the output voltage. What I do not understand is that since this a buck converter with no need for isolation and the ground of power stages and micro-controller are the same then why should I use isolation amplifier to sample the voltage and feed it to the ADC?
Any comment?

 

[dick_freebird]

Some people just have to predict DIRE CONSEQUENCES any
time the volts can't be counted on fingers and toes. There's
that.

But there's not much margin for mishap when you rub 300V
up against a 5V-or-lower IC. Switching converters have a
lot of spikes, ringing, and so on that have to be handled.

A resistive divider you would think is the ticket. But a plain
R ladder that you can stand the current for, will add lag
and perhaps instability, while adding a feedforward cap can
magnify the ringing / overshoot as a fraction of sense voltage.

You want to spend some time in open loop, working on the
feedback network until you are sure that at least the normal
goings-on present a harmless and useful scaled voltage that
falls comfortably inside the ADC input range. Then you can
close the loop and find out about stability and what happens
to the uC when output overshoots by 100%.

Op amp buffer will only make lag worse. But it can also be a
cheap sacrificial element, to protect the uC. I'd still go with
a R divider, with some backstop clamps and a good combo of
feedforward and HF filtering, trying to stay faster than the
output unloaded risetime but slower than the edge ringing,
and do it all with passives (and diodes).

 

[KlausST]

Hi,

I'm designing industrial measurement devices.
Some of them measure precisely multiple channels voltages up to 2500V AC...without isolation and without Opamp.
They work without problems for more than 10 years now 7*24 h.
--> There is no general need for isolation
--> there is no general need for an Opamp

For sure you need proper design and careful PCB layout.
Especially the GND_node for the resistive divider needs to be clean. The same node need to be chosen for the ADC_GND_reference.
On high power switching applications this might be problematic. If so, then a difference amplifier may be useful.

In any case: use a suitable anti_aliasing_filter.

Klaus

 

[asdf44]

Whether you 'need' an opamp or not depends mostly on the input impedance of the ADC.

Personally I don't see a ton of reason to debate it when an opamp is about as cheap as the high voltage resistors you'll need for that feedback. In a recent 300V digital feedback path much like yours I don't think I even checked the input impedance of the ADC I simply added the opamp to make it a non-issue and give me more flexibility filtering the feedback.

Also my instinct is that the pole created by the resistor chain is a non-issue for the feedback loop (and in-fact can be used to create a nice pole below your switching frequency) unless you feel a need to go very high in input impedance (like >1meg) which I suspect you don't in a 3kw design unless you're pushing efficiency to the extreme. Though you need to factor this pole into your final compensation scheme: basically make it high enough its 'out of the way' or low enough to deliberately use as your final control pole.

On that note what's your compensation plan? Good buck performance typically demands TypeIII compensation - you have a complete plan for implementing that digitally? What output bandwidth and ADC sample rate are you targeting?


On the issue of isolation specifically you've already decided that by not isolating your micro-controller. If the micro is on the same ground as your output then use it. If you still have concerns you could use a differential amplfifier stage sensing the very output which would be next option before considering galvanic isolation.

 

[skn96]

On the issue of isolation specifically you've already decided that by not isolating your micro-controller. If the micro is on the same ground as your output then use it. If you still have concerns you could use a differential amplfifier stage sensing the very output which would be next option before considering galvanic isolation.

Yes. I have decided not to isolate my micro-controller from power stage. But I'm a little confused here. Why somebody should consider to isolate the micro-controller for a buck converter (which is non isolated power converter anyway) at all? Is there any application or circumstances that demands isolating digital micro-controller from power stage in non-isolated power converter?


Regarding my control design strategy:
The buck switching frequency is 50kHz. So I'm planing to design type three voltage controller with crossover frequency at 5khz and phase margin at 60 degree. Then I'm going to implement the type 3 controller in micro-controller after doing z transform. Also I was thinking to use an ADC sampling rate the same is the switching frequency. I appreciate your advice on this strategy.

 

[asdf44]

Well there are multiple reasons to use isolation. One is actual isolation where for example you might take two isolated 5V lab supplies and put them in series to make 10V. That's 'real' isolation.

Second is to break ground loops and prevent noise from moving from one place to another. I've seen fully isolated gate drivers where the gate driver ground is then connected back to the same ground as the controller driving it. This simply prevents ground bounce from getting into the control.

Along those lines, theoretically, and without a ton of trouble, you could have the micro sitting on an isolated island controlling the fets with isolated gate driver ICs.



The higher the sampling rate the better. I have a design on my desk right now running a step response and I start seeing negative effects when I reduce the sampling rate less than 10X the switching frequency and fairly noticeable effects at 4x the switching frequency. I have doubts a TypeIII compensator is going to work well with crossover at 5khz and sample rate at only 50k given that it requires poles and zeros between about 5 and 25 (too close to 50).

I'd suggest this: Set your analog pole to be the 'last' pole in your control which should be somewhere around 25khz. If you're using a biquad you need a 3rd pole anyway and this analog pole can do that for you. And I'd push the sampling rate as high as it can go. Synchronizing the ADC to the switching frequency may or may not be a good idea (I haven't done it). If they're in sync some duty cycle will result in an ADC sample happening at the worst time every time (if you have stable vin/vout perhaps you can just avoid this).

 

[fouwad]

It is always better to be on the safe side, add an isolation amplifier or any liner opto or galvanic isolator. This will not only reduce the current into your ADC but also save your design from any unwanted spikes or accidents.

 

[KlausST]

Hi,

I don't think it's always better.
Any additional circuitry will reduce linearity and precision. It at least adds noise and (drifting) offset.

My personal experience (may apply on your application or not)
Even in my highest power switching applications (up to 2500V and up to 6000 A AC) I don't use an isolated analog path.
These high power units need isolation (unavoidable), thus I did it on the digital side to avoid the said errors.
"Spikes and accidents" need to be suppressed by circuitry.
This especially applies for the measurement signal. If you expect switching spikes (and noise) to happen, then use appropriate analog filters, because they are much more difficult to filter ot on the digital side, because they will be shown as alias frequencies.
If you don't suppress them on the analog side, then they will cause some errors depending on sampling scheme:
* If the sampling clock is synchronous to the switching frequency, then it mainly will cause some very low frequency (DC) errors (1) at the output.
* if the sampling clock is not synchronous to the switching frequency, then you will see AC noise (alias frequencies) at the output.
For a precise and stable output voltage both errors should be avoided (suppressed) by appropriate analog filters.
(1) At least the digital regulation loop should be synchronous to the switching frequency.

If possible do the ADConversion just before the switching edge. This will avoid the biggest "error peak" in the digital signal. The remaining noise then mostly comes from some ringing...which should be suppressed by the analog filters (because it is much higher in frequency)

For sure you are free to use isolated analog signal ... I personally see more trouble than benefit.

Klaus

 

[Easy peasy]

As a power electronics designer of over 15 years I can say there is no problem dividing down 300V to 4v5 (say) for input to a uP controller.

Just use resistors that are rated for the voltage and power, and use say 4 in series to the bottom resistor, that way if one goes short, all the resistors will survive ( because you used 500mW resistors that were only dissipating 100mW) and the sensed volts go up and the Vout goes down correspondingly.

We have made military grade 4.5kV to 350VDC, 45A converters all with simple resistive dividers - still working after 10 years...

 

[treez]

You say you are already regulating it with the micro. So presumably you are using a fixed duty and kind of relying on that to give you 150V from 300V , (as long as your inductor is in CCM this is OK-ish).
So now you want to do the control with the micro….but presumably you are going to use current mode control?…..(Bucks are generally easier in current mode). So where is your high speed PWM comparator coming in to it?, does the micro have one internally? Remember, even if using voltage mode control, you still need a FET current sense to prevent possible disaster. If I was you I would use a high side NFET for the buck FET and put a current sense transformer up there for the source current sense.
Remember that the upper resistor of your output voltage divider is the input impedance of the error amplifier in your smps control loop…if it is high Ohmic value, then this could make for a slow feedback loop unless you compensate for it elsewhere.
Give me a shout if you want a PCB layout guide for smps…as this can ruin your day, and may well have been why the person wanted you to use an isolation amplifier..because they thought the pcb layout would be bad….in fact, I attach it anyway

- - - Updated - - -

Also, please find an article, which may put you off using a software control loop

- - - Updated - - -

ok, its 1444KB, and it wont upload, its called "the digital power supply revolution" by Dr Ray Ridley