Practical limits of extracting a known signal level from noise

Are you Sync-detecting this with an analog switch, of doing all the sync-detect action in software ?

Since you are looking for only resistance, you do not need all 360' of the sine : you can discard those areas with worse SNR, and concentrate on the peaks - this slicing discards more noise than signal, so you improve SNR.

You also do not really need a Sine drive, and may find you can drive with more current, with less FET losses, with a pulsed +/-/TriState drive.

Yes, more samples will always be better - who dictates the 1 second sample time ?

I'd make up a LC passive bandpass filter, and see if adding that improves your SNR and precision. If it does, then you probably have intermodulation / distortion effects in the chain, that need attention.

-jg

Filtering out the 50/60 Hz mains harmonics should be possible with some kind of comb filter, taking out all harmonics. Thus, a delay as long as the mains should be required.

Depending of the short time frequency stability of the UPS, the comb filter should be doable with some kind of phase locking to the mains fundamental, provided that the lock range is, say 45-65 Hz.

Paul

Sounds like an ideal application for a pseudorandom probe signal and matched filter. It will discriminate against any interfering signal that's not correlated with it -- and the only limitation on your ultimate discrimination is how long the battery resistance remains stable.

If you generate the probe signal with the processor, your probe and measured signal can be perfectly correlated, which means any sine wave (for instance) can be removed, even if it's exactly synchronized with your probe.

Google "matched filter", "correlation detector", etc. for information. I found several that could be useful, depending on your mathematical background. The academics dig deep into theory, but the practice is actually quite simple for the simple arrangement you need.

John Perry

Just saw this old thread. :-)

One hint I did not see is this: Make sure the test signal is _not_ a harmonic of the other frequencies in the system. If the mains supply is 50 Hz, 300 Hz in a harmonic, and thus a bad choice. Offsetting the frequency and doing syncroneous detection will eat harmonics of 50 Hz. ;-)

--
mdc at manbw dk  -  MAN B&W Diesel A/S, Copenhagen
www.manbw.com    -  Electronics & software dept.
      -  Speaking for myself only. -

There are few problems with the measurement method.

You injecting a known current, and measuring the rms value of the resulting voltage. Since this is AC , you will include reactances no matter how small since your rmeasurement is at 0.1% level, this will impact your accuracy.

If the current injected has harmonics, and you measuring the resultant voltage with a band pass filter, you not only rejecting noise but also the harmonics present in the driving current. This will directly effect your rms measurement.

Since you are using a large current, the heat dissipation will change the temprature and thus the resistance. Are you making sure the temprature of the device is within the accepatable margin, so as not to affect the accuracy.

apotheothenai

Charles Oram wrote:

Yes, I know that it will not only be measuring the resistance, but that is a compromise because to get less of the reactance we need to use a lower frequency, but that makes it harder to build an analogue filter (although this point may become moot since the digital filtering seems to do well enough without the hardware filter).

The current injected is a pure sinewave.

Mostly we are trying to get away with 1A, which will cause negligible heating in the battery. Also, the battery will already be heated by the ripple current, which can be quite large.

a

Try to beg/borrow/steal/buy/rent a real lock-in amplifier (warning: shameless plug for our products

)

You will see that with phase-synchronous signal recovery, your measurement requirement is easily met. You can choose to look at either the real or imaginary part independently. You can use any frequency from 5mHz (milli) to

100kHz, limited only by the parasitics of your fixture. It even comes with a manual that could help you design your own.

sorry for the plug, Bob

There is no need to do any detection for the synchronising, because the signal is generated by the software. The A to D conversion is triggered by the same timer interrupt that does the 1kHz signal generation.

We've done a bit of experimenting with square waves, but so far without as good results as we get with the sine.

The temperature of the FET is the main restriction, but we also don't want the complete measurement to take too long.

Thanks.

- Charles

You could experiment with a Sync detect using analog switch. This gives a DC result at the earliest possible stage, and allows you to use other digital voltmeters to 'double check' the results. It also uses the same area of the ADC for each sine half, so is more tolerant of ADC specs, plus you can easily add a filter pole ( or two) once you have a DC signal, which can also help the research...

-jg