"John Larkin" wrote in message
All you do with fast ADC samples is square, average, root, display.
Autozero on the front end maybe. The only serious mistake you can
make is letting the ADC clip on peaks, and that's easily detected.
I'm doing that with a PIC, but for power line frequencies. I take the square
of each sample and add that to a running sum of squares, using integer math.
Then when I want to display the true RMS, I just shift the result to, 16
bits and then take a quick square root which is about 8 bits, which is still
0.4% accurate. My samples are typically 100 to 200 mSec. But I also
calculate the true RMS value of much longer pulses, such as are encountered
in circuit breaker testing for short and long time delays, which can be
anywhere from about 0.5 seconds to 200 seconds. So I still just maintain the
total of squares, which may be an unsigned long 32 bit number, which is
enough to hold 29 minutes of 10 bit data at 2400 samples per second.
This is needed because the current can vary during the test, due to heating
and other effects, so the breaker tends to react to the true RMS value of
the entire waveform, especially if it has a thermal delay element. Reclosers
are even more challenging, because their impedance changes by a factor as
high as 3, causing a huge difference in the RMS value of the current from
the beginning to the end of its stroke. This is an effect of low voltage
testing, and the change would be negligible on an actual fault with 14.4 kV.
Series resistance can be added to reduce the drop-off effect, but only so
much, so the RMS value must be computed for the entire waveform. This has
been shown to match the expected trip times very accurately.
The analog RMS converters are expensive and usually have mediocre
bandwidth. The wideband ones (like 100 KHz or so) wind up having
frequency compensation tweaks. All that's a lot of work to replace
some essentially free firmware.
There are lots of "True RMS" DVMs around, with analog converters
in the front end, that are good to 0.03% or whatever on DC but are
rotten on AC, both accuracy and frequency range... numbers like
1% and 1 KHz.
I chose my Fluke 45 multimeter as the best choice for true RMS reading at
the time I bought it about 20 years ago. It has an accuracy of 0.2% + 100
counts from 50 Hz to 10 kHz, but that means that on the 100 mV range,
specified for 15,000 to 99,999 counts, it may be an additional 100/15000 or
0.67%. It seems to be actually much better than that, although it has an
offset of about 0.240 mV with the input shorted. I have been using this to
calibrate my Ortmasters, which are rated at 1% from full scale ranges of 50A
to 10,000A, using a 1000A 100 mV shunt. So when I calibrate the 50A range, I
am reading about 5.000 mV, which could be as much as 100 counts off, or 2%.
The Ortmaster typically shows an offset of 0.10 A on that range, which is
only 10 uV. So I may need to get a better standard!
The Fluke 8520A has a 1 year accuracy of 0.15% reading plus 0.05% FS, which
in this case is 1.99999 V, so the true error is as much as 1 mV. So its
accuracy at 5 mV is only really 20%! No winner here.
The Fluke 289 has a true-RMS accuracy of 0.3% + 25 counts, 45-65 Hz, on the
50.000 mV range, which gives me an effective accuracy on a 5 mV reading of
0.8%. An improvement, but I'd really like 0.5% or better.
The Fluke 8808A is similar to the Fluke 45, but it does not seem to be true
RMS.
The Fluke 87 has a true-RMS accuracy of 0.7% +2 digits and 0.1 mV
resolution.
The HP 3478 has a true-RMS accuracy of 0.46% + 163 counts for the 300.000 mV
range.
I really have not found a meter that is significantly better than the Fluke
45. Any suggestions?
Paul
