Power: IEC 61000-3-2 Current Harmonics Spec

[kerpal]

Hi guys,

I'm trying to figure out how to measure current harmonics of a power supply according to the spec.  It says:

For the 21st and higher odd order harmonics, the average values obtained for each individual odd
harmonic over the full observation period, calculated from the 1.5s smoothed r.m.s. values
according to 6.2.2 may exceed the applicable limits by 50% provided that the following conditions
are met:
? the measured partial odd harmonic current does not exceed the partial odd harmonic current
which can be calculate from the applicable limits;
? all 1.5s smoothed r.m.s. individual harmonic current values shall be less than or equal to 150%
of the applicable limits

What does "1.5s smoothed rms" mean?

What I am doing now is this:

• Acquire 10s of waveform data (Current and Voltage) from the oscilloscope.
• Perform FFT over the Current Waveform
• Obtain a table of current harmonics by looking at that FFT magnitude at multiples of the operating frequency (50Hz or 60Hz)
• Compare the obtained table with the IEC limits spec

What is the 1.5s filter about, and where does it fit in?  Do I first filter the Current waveform with a LPF (how to put in 1.5s?), and then only perform the FFT?  Or do I channel the FFT to a LPF instead?

[IconicPCB]

take samples and perform FFT
log harmonics values over a 1.5 second interval
calculate average for each harmonic over the 1.5 second period
alternatively write a program to download FFT values and run a moving average for each harmonic over 1.5 second interval

This is going to smooth ( average ) data over 1.5 second interval

[kerpal]

Sorry, I'm pretty new in DSP, so will need further questions.

take samples and perform FFT

Ok, I took 10 seconds worth of acquisition from the oscilloscope (1s/div)

log harmonics values over a 1.5 second interval
calculate average for each harmonic over the 1.5 second period

This is the part that I am confused.  If my FFT is based on data over 10s duration, does that already cover the 1.5 second interval?
Or do I have to perform [take samples + FFT] n-number of times until 1.5seconds?
Or I was already wrong from the start ... should I be taking 1.5s worth of waveform (150ms/div setting), do FFT, and repeat n-number of times and take the average result?

[IconicPCB]

specified averaging window is 1.5seconds...

You could try implementing running average over the 1.5 second interval for all harmonics in question

[T3sl4co1l]

Ideally, you'd use a sliding window FFT, that is...

Sample a screen of data (1.5s); FFT
Cut off the leftmost, say, 0.25s, and add the next 0.25s on the right; FFT again
Repeat for 0.5s, 0.75, 1, 1.25, 1.5... offsets

Note that setting your scope for 1s/div may not yield a buffer size of 10s.  Check the horizontal buffer size to be sure.

It's slightly worse than this, really, because scopes don't just chug an FFT, they apply a windowing function first.  Which kinda sorta throws away the beginning and end of the buffered range, but not completely.  So your buffer size should really be more like 2 or 3s, than 1.5.

Whether this will give you a number that's appropriate for the standard you're testing against, I have no idea.  This is more in regards to your "new in DSP" -- get a feel for the tradeoff of time versus frequency, and how much data / time is required for some frequency resolution and all that.

Something could also be said about the sidebands of the harmonics (variations of a given frequency component show up as modulation of that component), but that's tricky, and not usually very apparent from an FFT.

I'd be curious if "1.5s RMS smoothing" is defined elsewhere in the standard, or in related documents.  The way it's defined for EMC work -- spectrum analysis -- is quasi-peak, meaning, back in the old days a spec was simply a radio receiver constantly being swept across a frequency range.  The detector is simply a diode, resistor and capacitor: when RF is detected, it rises at a certain (attack) rate, and when not, it falls at some other (decay) rate.  If attack and decay are equal, it tends to filter out rapid changes in the signal, but if attack is shorter (which it is, for the "quasi-peak" condition), peaks will be emphasized in the readout.

There are algorithms to perform quasi-peak functions on FFT calculations, but what I'm getting at is, I wonder if they would use a similar, relatively simple to measure (in the analog sense) method in this case.  Namely, that the "1.5s average" is simply a receiver tuned to 50/60Hz times the 21st harmonic, and with attack/decay rates of 1.5s.  Actually implementing that digitally, again, not so fun.

Tim

[Neilm]

Are you doing this looking to do compliance testing? If so, remember that there will be noise on the mains supply unless you are using a stabalised supply or some high quality filters. If you are doing compliance testing, you will probably find it quicker to hire a tester. Most of these will measure what is on the mains and give the result straight away