Shyam,

Not an exact answer to your question but, its an additional thought.

Using a precision comparator that compare the attenuated output of the power supply with a standard cell (normally 10 PPM) and send feedback to PWM for due correction in output voltage.

Dear rakesh_semwa

Comparator having 10ppm common mode gain stability is a good idea. Let me see what can be dome for it. While absolute accuracy of 0.1% will be OK but stability for control at any given voltage setting, has to remain within 10ppm level irrespective to the amount of absolute error is the point here.

Your point is well taken and of course it is very important one in this design.

Shyam,

One of the difficulties is the precise measurement of high voltage and the stability of the components with which it's done. The standard cell suggestion is good, and they are typically stable if kept at a constant temperature. Voltage dividers with precision resistors should also be stable at constant temperatures also. The problem arises with voltage comparison! A wheatstone bridge made with precision resistors coupled to a high-gain op-amp comparator may do the trick, but who knows until it's tested. Of course, the impedance of the standard cell circuit must be comparable to the divider network.

I am trying to get 10ppm/C resistive dividers from Ohmite, Caddok and Vishay to see which performs better. I am also planning a constant temperature bath for calibration and this will have heavy thermal load and can not change temperature easily.

At that frequency, and considering the noise requirement, you might want to consider a resonant or quasi-resonant ZCS design instead of PWM.

Dear rcapper

You can give me little more details on this resonant ZCS design and let me know how I can control the output voltage from 0-Vmax? Usually resonant circuit have minimum working voltage to start the oscillator. What is Quasi resonant design in this case? If you have a link to let me understand these designs then I can look at the merits of these circuits. I thought they are only good for fixed frequency fixed voltage generation.

Switching frequency noise sure is a serious problem in such power supplies. I will consider all possible designs that can reduce noise in 1kHz to 20MHz range to <1mV peak to peak.

http://www.cpes.vt.edu/publications/volumes/01.php

google this, lots of references: High Frequency Resonant, quasi-resonant multi-resonant

I have looked into some of those designs and yes, the pulse energy shaping is a better idea. Feeding energy through inductor into transformer can reduce noise to some extent. Zero current switching is also a good idea. I will also use some filter design at the output of the voltage multiplier and need to see which is most suitable one for low current low noise power supplies. My designs are for under 10W maximum energy.

For 6W output, what should be the input peak energy for say 12V or 24V source? 12W or 18W or 24W or higher?

I hope the references turn out to be helpful. I don't currently have time to go into deeper design questions for free online but good luck with your design.

Regards-

I looked into reference from Vicor as they do design compact high power DC-DC converters and have lots of application notes online.

Zero current switching is also a good idea.

You might look into doing what I did once: Zero current switching is normally accomplished by using a silicon diode to turn on a device when it starts conducting in the forward direction. However, I once worked on a design for a piece of equipment whereby the noise from switching on at 0.7 volts was too high for stability ('cause the current was very high). I used a high frequency oscillator and a digital, programmable divider (mathematical) to back up the switching time to almost the exact instant that zero-voltage, zero-current occurred. Noise went to zero (or, at least nothing that we could see with an oscilloscope).

Dear Bill,

Switching circuit at high frequencies increases number of events in noise band but provides good average power. Suitable LC Filter and high frequency negative feedback can bring down noise to <1mV and that is the aim. Peak current is to be limited by feeder inductor to transformer looks much better choice for quasi sinusoidal shaping to reduce noise.

3000V output and you want 10ppm line and load regulation. That would be some 30 mV on top of the stick. You bring it down through a Caddoc voltage divider. I think that they guarantee them for higher voltages. Anyway, you will compare your voltage reference with a 10-12 uV signal. You must condition this very small signal and remember that very good op amps have an offset voltage in that area. A "chopped" ampli might do the right amplification for you.

And you didn't mention the temperature range. I know somebody who, right now is designing the same thing for 200 degrees C, with some size constraints and he, constantly, asks me for Advil.

Due to high resistances involved in high voltage sensing, small capacitor does a good filter job for feedback circuit. It is easy to get <1uV input noise level amplifiers these days. Our problems are greater in output noise due to high switching currents involved.

I have looked at these voltage dividers for measurement of high voltages 1000:1 ratio up to 5kV. There is one Model from Caddock HVD5-A50M-050-05 and looks very expensive. 100 parts for about $3500. These are 50M Ohms total resistance. Perhaps I will prefer 500M Ohms as leakage in this case looks much higher.5000V/50M is about 100uA which is real on high side. I want to limit measurement current to 10uA maximum.I think Vishay has 100M version but not sure if they can make 500M version also with high temperature stability and 0.01% accuracy.