Discussion in 'Embedded' started by Guy Macon, May 9, 2006.

1. ### Guy MaconGuest

I am doing some preliminary "back of the envelope" design work on
a high-end split phase / three phase 50 Hz. to 400 Hz. AC power

Generating the 2 or 3 phases is straightforward, but monitoring
the current and voltage is a bit more tricky -- especially when
monitoring the far end of a transformer, line filter, or power
factor compensator. I wish to determine the following values
for a single-cycle non-repeating transient:

Phase A/B/C to N (Neutral) voltage
Phase A-B/B-C/C-A voltage
Phase A/B/C/N current
Phase A/B/C apparent power (VA)
Phase A/B/C RMS power
Phase A-B/B-C/C-A voltage phase shift
Phase A-B/B-C/C-A current phase shift
Phase A/B/C THD+N

The first thing that comes to mind is to use some high-end 24-bit,
192 kHz audio ADCs, possibly interleaving if the sample rate needs
to be higher. I am leaning toward a sample rate that is divisible
by two and three (for split-phase 180 degree and three phase 120
rate.

Let's say that I wish to measure phase with 0.1 degree precision
at 400 Hz. That's a range of 3600 (0.0 degrees to 359.9 degrees).
So I need at least 3600 x 400 = 1.44 MSPS, right?

If I need that many samples per second using 192 kHz audio ADCs,
I would have to interleave 8 of them per measurement channel,
times 3 voltage measuring points and 4 current measuring points,
or 56 of them total (and that's assuming that I can make the
interleaving work right). Suddenly the low-cost audio parts
aren't looking so inviting...

Looking at faster parts, a quick Google search comes up with parts
(18-Bit, 2 MSPS) and the Linear Technology LTC2208 (16-bit, 130 MSPS).
Hmmmm... could a fast enough ADC allow me to measure multiple channels
with a single ADC? Or do I really need simultaneous sampling?

Or could it be that I am on the wrong track and there is a better
way than the brute-force lots-of-samples, lots-of-resolution lots-
of-processing-power solution that is the first thing to comes to mind?

Guy Macon, May 9, 2006

2. ### Paul KeinanenGuest

How are you actually going to measure the phase shift between phases
or between voltage and current ?

Trying to detect the time difference between the zero crossings would
work in an ideal situation with perfectly clean sinusoids, however, in
practice, the mains voltage is badly polluted by all kinds of noises.
With even order harmonic distortion, the half waves are different, so
it is difficult to even establish the zero-line.

Trying to detect the top of the waveform would even be worse.

So in practice, multiple mains cycles would be required to determine
the phase difference to such accuracy. As long as the sampling
frequency is not an _exact_ multiple of the mains frequency, a
repetitive waveform will be accurately reconstructed anyway.

Calculating the complex FFT for a sufficiently long sample should give
the phase and magnitude for the fundamental and for each harmonics in
the waveform.

Thus, I do not think that you would need such huge sampling frequency
for the phase measurement, but of course for transient recording the
high sampling rate is desirable.

There are a few problems with typical audio ADCs. The DC and low
frequency characteristics seem to be poor, sometimes artificially
limited at 3 Hz to avoid drift problems. This may be an issue if you
want to measure any DC component on the mains (which will increase
transformer noise and ultimately saturate the cores even with a few
volts of imbalance).

Even if some ADCs boost 192 kHz and 24 bits, the SNR quoted is at best
around 120 dB and this is usually specified for the 20 kHz audio
bandwidth only. The SNR for the full 90+ kHz bandwidth possible with
192 kHz sample rate can be much worse, corresponding to 18-20 bit
How did you intend the current measurement ? Have you looked at the
current transformer phase and frequency response or are you planning
to use shunting resistors in each phase and using separate floating
power supplies for each ADC at mains phase potential and bring down
the digital sample values in an optical fiber (and apparently feed the
ADC with a common clock source using an other optical fiber) ?

Paul

Paul Keinanen, May 9, 2006

3. ### Jim GranvilleGuest

Why do you need 0.1' / 24 bit ?
Most standards specs I've seen quote harmonics only to a certain
number, and if you are doing an analyser do you need to be 70x that ?

Many Audio DACS have poor impulse response - a 20KHz response is
gives out to the 50th harmonic of 400Hz.

-jg

Jim Granville, May 9, 2006
4. ### MKGuest

Hello Guy,

You are indeed on the wrong track,

To measure everything about a 400 Hz signal you need to sample at 800Hz and
a bit - in real life you need more samples than this to deal with non ideal
anti-alias filters etc. (and don't forget that uncertainty in the sample
time (jitter) will affect accuracy just as much as amplitude errors).

Most mains analysers allow you to measure up to 20th Harmonic - 1kHz for
50Hz mains, 8kHz for your 400Hz top limit. A 40kHz sample rate will do fine.
I wouldn't use audio parts for this - they are a pain to multiplex. There
are single chip parts with 4 and 8 way multiplexors that should work quite
well.
You get on much better measuring the phase if you use all the samples rather
than just the few near the zero crossing - Google for Goertzel - there is a
lot of 'noise' but you should find something.
You don't need simultaneous sampling - just to know when your samples were
sampled - you can calculate out any errors due to the exact sampling time.

Michael Kellett
www.mkesc.co.uk

MK, May 9, 2006
5. ### Guy MaconGuest

What I usually do is to calculate a Fourier transform on the
signal, discarded all harmonics leaving only the fundamental,
calculate a reverse Fourier transform to reconstruct the
signal, then measure the zero crossing.
Alas, my customers really do want to measure transient waveforms.
For example, when load X was suddenly connected to power, how
many cycles did it take to recover?
Indeed it does, and one of the requirements of an AC power source
is to shut down if there is more than a small amount of DC at the
output. I am likely to design in a seperate circuit for that.
Audio bandwidth is fine; my signal is in the 50 to 400 Hz. range.
24 bits is massive overkill given the noise and distortion typical
of real-world powerlines. I don't think I will end up using the
cheap audio ADCs, but it won't be because of SNR or LF issues.
Look here:
http://www.pearsonelectronics.com/current-monitor-products/standard-current-monitor.htm

Guy Macon, May 10, 2006
6. ### Guy MaconGuest

24 bits is a huge overkill, it's just what the 192KHz low-cost
audio ADCs put out. Measuring to a fraction of a degree is
real; the users of these things tend to drive large inductive
and they do that by measuring the phase between voltage and
current.

Guy Macon, May 10, 2006
7. ### Guy MaconGuest

<Big happy smile> Now *that's* the kind of advice that I was hoping
for! Far better to find out now when I am doing preliminary design
work...

A California Instrument 9003iX does waveform analysis and waveform
synthesis to the 51st harmonic and has a 16 Hz to 500 Hz Hz full
power bandwidth.

http://www.calinst.com/ixseries.html

A Pacific Power Source 320AMX does waveform analysis and waveform
synthesis to the 51st harmonic and has a 20-5000 Hz full power
bandwidth.

http://www.pacificpower.com/products/amx-series.aspx

A Voltech PM300 measures up to 250kHz. A PM3000A goes to 1 MHz,
and a PM6000 goes to 10MHz.

http://www.voltech.com/products/pwr_anl/index.htm
http://www.voltech.com/products/pwr_anl/pm3000/index.htm
http://www.voltech.com/products/pwr_anl/PM6000/pm6frame.htm

It seems that somebody out there seems to think that you have
to go past a few KHz.

I was under the impression that the Goertzel Algorithm was a way to
detect frequecies with less computation than a DFT or FFT. I usually
do an FFT, throw out all the harmonics, reconsruct a fundamental-only
version of the signal, and them meaure the zero crossing point.
Hmmm. That makes sense, but I have never done it. If it turns out
that I am wrong about bandwidth and can multiplex, I will revisit this.
I doubt that I will be multiplexing. Too much bandwidth loss.
If I end up using audio parts I may end up interleaving them!

Guy Macon, May 10, 2006
8. ### Jim GranvilleGuest

Could you not derive phase for power factor terms very accurately,
by using a whole-cycles samples. You do not need _each_ sample to
be 0,1', all you need is the phase of the 'best fit sine'.

Quite a low number of samples will give you that : If you know
the exact times of the samples. ~0,5us time delta is ~1 part in
5000, and 12 bit ADC is ~1 part in 4096 ad that's better than
0.1'

That leaves you with transient capture, which sounds partly real, and
partly 'bragging rights' stuff.

For the numbers game, look at the TMS320F28x, they have 12bit ADCs
with 12.5 Msps, and enough crunching you could do fancy compression on
the transient info : Store the delta from a ideal sine - so you don't
waste storage on clean waveforms....

a DDS core as well.

-jg

Jim Granville, May 10, 2006
9. ### Paul KeinanenGuest

It depends what you want to measure, if you are interested in
measuring lightning current waveforms, then at least 1 MHz sample rate
should be used, since the current slew rate is several kA/us and the
peak is reached in the order of 10-30 us.

For transients in typical man made electric systems, the transient
current is limited by the circuit inductances (including wire
inductances) and hence the current slew rate. Voltage transients are
limited by the stray capacitances. Thus you would have to analyse what
kind of voltage and current transients are expected in your
environment, before deciding the sample rate.

Paul

Paul Keinanen, May 10, 2006
10. ### Guy MaconGuest

I have been doing some more "back of the envelope" design work
on a split phase / three phase 50 Hz to 400 Hz AC power source.
been very helpful and are much appreciated.

I obtained a couple of different kinds of commercially available
AC Power sources and spent a couple of days running tests on
them in order to see what I am competing against. I found that
the real-world bandwidth limit isn't fixed at a particular
roll-off but is instead slew-rate limited, and thus the small
signal bandwidth -- and the number of useful harmonics that are
used in the signal synthesis function -- is considerably better
than the bandwidth at full voltage swing.

As a thought experiment, imagine a 400 Hz AC power source that
has just enough output-DAC bandwidth to generate a 400 Hz sine
wave, bumping right up against the Nyquist limit. This wouldn't
be enough in the real world, where one very common set of
waveforms that the users desire are sine waves with hard clipping
at 1%, 5%, 10% THD etc. Then again, some of them want square
waves or triangle waves... This requires more DAC bandwidth, but
the question is, how much? I am inclined to design in enough
DAC bandwidth so that the slew-rate-limiting of the output stage
dominates from 10% to 100% amplitude, and to give the waveform-