RMS-to-DC Problem

[Christopher]

Hi all.

Going to go over this fairly quickly as it should make some sense and it's almost sleep time (been working on this for 8 hours  )

We are working on a High-voltage AC or DC to DC power supply, with digital monitoring.


I am using an LTC1966 RMS to DC converter potting down the 900Vac (50Hz single phase) to something sensible, 0.9V. I then have a 1uF averaging cap and a LTC2990 ADC digitising the result. I know the ADC works perfectly fine as I am using this all over my designs with the same code.


The problem I am having: When there is no load on the power supply, at 600Vac input, the output of the RMS to DC converter is all over the place, reading between 300Vac-900Vac. As soon as I put load on the power supply, the converter averages out to the correct voltage +- a few volts...

We have a big-ass filter network on the AC input consisting of more magnetics and caps than you can shake a rusty stick at.  I have a feeling this is causing us a few problems as we are measuring after the diode bridge & filtering. I have a seperate circuit measuring just after the diode bridge which is also doing the same thing.

The only thing I can think of doing is increasing the averaging cap (which is a poor 1uF jellybean ceramic) to a 10uF with a different dialectric.


I'll post some more information if I can't get it working tomorrow morning... But hopefully I am missing something very fundamental!

Simplified schematic:

[MK]

two thoughts:
1, the PSU is not going to zero under no load.

2, lots of filtering will mean that the RMS converter is not nesessarily seeing the correct waveform and reading the correct value.

you may need a 100:1 or 1000:1 frontend of a suitable rating to look at what is happening using an oscilloscope.

[tszaboo]

The 1966 has high output impedance. That means, you cannot draw current from it, even the capacitor should be a low leakage, and you should buffer it with an opamp before using the signal. Did you buffer the signal?
Other than that, did you AC couple the input? Maybe it is necessary.
IMHO with the LTC2990 you leave lot of potential in the 1966, meaning you can make a more accuate system with a slightly better ADC.

[NiHaoMike]

Have you thought about taking the easy route by measuring the DC bus voltage?