How do I Design an Isolated High Voltage Analog Output controlled by a PWM input from 140V DC 3

[patelvijay12007]

Here are the requirements:

Supply 140V DC

Control Signal : 3.3V DC PWM Signal (16 bit)

Output Needed : 0 to 100% of Supply Voltage based on the control signal duty cycle

Need Good linearity and isolation upto 2KV

Need L/C Filter for smooth Output

Output current Required : 1Amp

I had started design with an isolated MOSFET Driver FOD3180 , and N-Channel MOSFET at the Low Side

The isolation was acheived, but the linearity is not so good, also observcing the peak to peak ripple noise of upto 28V

The 4th Order was LC Filter was used of value

L1 = 1.7mH

C1=47uF/450V

L2 = 1.7mH

C2=47uF/450V

I have no other pin left for MCU based feedback, but we are open to Any anlaog closed loop Feedback control , given that the isolation is maintained

Do write back with suggestion

 

[BrianG]

Your specification is unclear in a few areas:

"Control Signal : 3.3V DC PWM Signal (16 bit)" - 16 bit? PWM by definition is a single bit control signal at fixed frequency, but varying width. Are you producing a separate pulse train width from the value of a (variable) 16 bit word?

"Need L/C Filter for smooth Output" - why? what ripple level is accepatable? what is your switching freqency?

 

[OperaHouse]

If you convert that PWM to good old analog the rest is fairly standard if that accuracy is acceptable.

 

[patelvijay12007]

Dear BrianG,

16 bit PWM is here means the resolution, since the 16bit timer is used to generate the PWM signal.
The Analog Output is intended for electrical braking system, whose controller accepts analog signal(0 to 140V) at 130mA current.

To reduce the reliability issues with heat , and for suitable open-loop control beyond isolation, need to have a PWM driven solution.Need ripple to be low as possible(2-3Vpp Max), at present I am observing upto 28Vp-p ripple

I am attaching the present schematic and Test Result for your reference:
1) Schematic :

2) Test Result :

Regards,
Vijay Patel

 

[BrianG]

Vijay, soem questions:
How are you measuring ripple and have you analysed its frequency spectrum? I suspect least part of the ripple problem comes from the Vbat 110V supply - what generates this? This supply seems it may have a high level of residual ripple, as at 100% duty cycle you still have 4V of ripple present according to your results graph, but this is already greater than your target figure of 2-3V. You won't get any closer performance unless the control circuit can respond to and counteract this ripple frequency: this requires closed-loop voltage control.
Ripple may also depend on your switching freqency and the load inductance etc. What determined your choice of 5KHz? Does this cause any resonance with the load inductance? For output filtering, what is the expected response time of the braking system to a change in PWM?

 

[patelvijay12007]

Hi Brian,

Thanks for quick support,

I am measuring ripple using 200MHZ oscilloscope, by using trigger at 2ns range.This techtronix scope also has FFT function, but I do not know how to use it!Will try using it.

Yes I am supplying the system with a SMPS with 110V DC output from 230V mains (Reason : I do not have such battery in stock).
I also suspect that the SMPS is adding to the ripple.Can we delibrately subtract the same 4V ripple from the observed value? or it may increase at higher load value? (I shall make a small no load and full load measurement to get an idea)

The Load is a 100W lamp at present.(Again no big resistive load available with me for 110V 1A output testing (thus need 110W))

About the choice for 5KHz , were below reasons:
1) LC Filter size limitation made me go beyond 1KHz minimum
2) Linearity of the output improved as went to higher frequency

There's audible whisteling noise of the filter at present, so trying to making changes to operate the system at 20KHz and above (similar to buck converters!). That shall also help reducing the filter size and delay in output further!

The actual load inductance and expected response time are unknown to me ,and is analog input based controller as informed to me and shall be known once I test the present prototype with the engine.


For the closed loop idea, I do agree that that shall improve the performance, but I am not sure how can I implement the same, since I get only this PWM signal from the controller and the controller has no Analog input left for the feedback

 

[OperaHouse]

I see many problems with circuit.

What is the grey box. Assume this is a PWM chip. Why is there no capacitor on input signal to form RC filter to
make DC analog signal.

Lamp is a VERY poor choice for load. Lamps have 10 to 1 resistance change so power represented by voltage will
never be linear.

You need a high speed diode of appropriate current from FET output to power supply. Energy stored in inductors
needs a current path when the the FET turns off.

On filter board at the very least the teo capacitors should be connected together and in parallel with the brake
coil. The two inductors in series may not be enough inductance.

I doubt there is any reason for ripple spec anda filter board at all. It is common just to drive the brake
inductance with only a diode in parallel.

 

[MagicSmoker]

An "electrical braking system" that uses a variable DC current? That sounds a lot like an eddy current brake.

If so, this rarely requires such precise control of voltage/current. I mean, I've seen eddy current brakes and clutches on production lines controlled by what is essentially an incandescent lamp dimmer.

But assuming that you really do need 16b of resolution over a 0-140VDC output span at up to 1A of load current, I would separate the isolation function from the variable output voltage function. More specifically, buy (or design) an isolated semi-regulated 145V/1A supply to feed a synchronous buck converter. You want to use a synchronous buck (switch + switch/diode) rather than a conventional buck (switch + diode) to adjust the output voltage because it will remain in continuous conduction mode all the way down to zero current. This is important because the output voltage of the buck is directly and linearly proportional to duty cycle in CCM, but in discontinuous conduction mode (DCM) duty cycle is also proportional to load current. This complicates stabilizing the control loop, which is not usually an issue in general purpose power supplies supplying a fixed output voltage, but will almost certainly be a problem here, even without the 16b of resolution requirement.

I would use a fairly low switching frequency for the buck - probably 10kHz - because otherwise it will be difficult to reach the bottom end of the output voltage range (ie - below about 4V) without overheating the switch (from it barely turning on before being commanded to turn off again). For example, at 1V output with 145V input (and assume ideal switches) the duty cycle will be 1/145, or 0.69%. At a switching frequency of 10kHz that is an on time for the buck switch of 690ns. A fast MOSFET with a robust gate driver can easily be switched in about 20ns, but slewing 145V in 20ns can cause a lot of RFI and ringing so you generally try to stay below a dV/dt of 1-2V/ns; thus both the turn-on and turn-off time will each take 75-150ns, or as much as half of the supposed on time of the switch. The usual cure for this is for the PWM control IC to change over to constant on time, variable off-time control, but it can also be implemented in code on a microncontroller easily enough.

All that said, it will be extremely difficult to achieve 16b of resolution on the output voltage (ie - 2mV out of 140V) without resorting to exotic measures like an oven stabilized reference voltage, and if using a microcontroller to generate the PWM signal, the PWM timer will obviously need more than 16b of resolution as well (otherwise your quantization error and/or limit cycle oscillation error will dominate). All in all, this is an extremely difficult design task. I don't think I could pull it off in less than about 6 months and 4 prototype revisions and I've been designing switchmode converters for 20+ years now.

 

[OperaHouse]

From the cut and paste nature of the schematic building blocks this is obvious lt student lab work. Being such everything has been selected to work together. It will need to have a diode installed for the buck section to work. The OP Will need to refresh their knowledge of buck converter operation.

 

[patelvijay12007]

Dear Sirs,

The greybox is FOD3180 , an ISOLATED MOSFET Gate Driver for isolation of the MCU to the Output. The Non-isolated of the chip gets its supply from 15V Step Down regulator above.

For this application 8 bit resolution is also good enough(255 steps), linearity is a requirement, but as OperaHouse rightly said, it can not be verified by tieing the lamp as a load, hence will try a pure resistive load instead if its already doing fine.
I forgot to put the diode into the schematic, but have included in the prototype after having destroyed one MOSFET while testing!.Thank you for pointing this out! I have used 1N5408 presently since high voltage rated schottky diode at high current ratings are much difficult to find.
Though not much expert like you Sir, I am not a student , but working for this project , that has other high voltage I/O stuff which I and my team already tackled with successfully. This rough version of schematic is a cut paste of a whole schematic that contained various Topologies for testing , and this is the best one which have worked out. I believe that you and this forum is the best source from where I can get required support to get the Application working to the next level!

For the filters , in order to implement a 4th Order filter to achieve better Noise suppression at lower L and C values, I have tied 2 L and C one after other.


Please also suggest if there's an fully analog solution in ming, and excess heat can be dealt with reliably.

 

[OperaHouse]

Your filter is NOT a filter. It is what is storing the energy of the conversion. If those inductors are not sized to store about 10W, that will be your next problem. I can see you don't appreciate my help.

 

[patelvijay12007]

Dear Sir,

We do appreciate your valuable inputs, also Can you please put more insight on how can the filter create further issues?

 

[patelvijay12007]

Dear All,

I replaced the FOD3180 with VO3120, which has faster response to the MCU signal(upto 250KHz).
Increased the load frequency to 20KHz and results are quite optimistic..

The Graph:

The Readings :

The linearity is quite good.

The only issue is with peak-to-peak noise, at the switching frquency of 20KHz is about 8Vpp.
Even at 100% duty cycle there 4Vpp Ripple noise, which looks like the 110V SMPS is adding to the ripple in the output.

Could anyone suggest some method to reduce this noise further? Otherwise I feel the issue has been resolved.
Thank you all for your support!!

 

[OperaHouse]

Wisdom comes from knoweledge.
Knoweledge comes from experience.
Experience comes from bad decisions.

Now you have made a lateral move by replacing the FET driver. That is grasping at straws.

You still don't know what the circuit is doing. FFT won't help. Put a half ohm resistor
in the common lead of the FET and look at the current on a scope. This can tell a lot.

Want to see how mechanically fast a solinoid or relay is? Pulse it and look at the slope of the current.
When you see the slope change, that means the armature has pulled in. That shows a change in inductance
producing a change in slope. An inductor or a transformer is only that until the core saturates. After that
it is no longer an inductor, but a resistor. If you see a rapid change in slope of the FET current, your
inductance is too small to store the energy.

Be mindful of an inductor getting warm. Most inductors will half their inductance when they warm up.

If this is to run on batteries eventually, getting rid of power supply noise is moot.

The inductor before the brake serves no function. It is minimal compared to the inductance of the brake.
It might be better to place it in series with the other inductor.

FET and diode heat up with every transition. Now that you have gone from 5K to 20K there are four times as many
transitions.

I would consider lowering the frequency and driving the brake directly with the FET and diode.