2

I'm researching the possibilities to accurately register the capacitance changes of a big metal plate (area 10 m2) to the ground (GND, yes - earth, soil). It varies around 100-200 pF.

For now I'm considering using a "good" (thermocompensated) VCO like the old known SN74LS628 and measure its output frequency with an ESP32. The thing is that the '628 Cx input for a capacitor is NOT to GND on one side.

So I'm thinking of using two capacitors in series (C1 and C2) and connect the big metal plate (C?) in the middle. Here it is in a schematic:

enter image description here

I'm anxious if there would be some significant disturbances (not real capacitance changes but EMI noise) on the internal oscillator because of the big metal plate acting as an antenna.

If someone is familiar or has at least an idea about its internal structure please comment.

enter image description here

EDIT: Do you think a bridge approach would be better?

enter image description here

valioiv
  • 73
  • 7
  • 1
    How big is the plate, and why would its capacitance to ground vary? How accurate do you want the measurement to be? – Bruce Abbott Mar 08 '23 at 09:57
  • @BruceAbbott, the plate is ~10 square meters (~5x2meters dimensions). Capacitance varies because some animals are migrating between the plate and the Earth. I want it as accurate as possible - let's say 0.1% would be a good value, I guess. Just starting with some experiments and cannot tell exactly. – valioiv Mar 08 '23 at 10:04
  • 0.1% absolute accuracy is hard to achieve. However it may not be as hard to 'accurately' measure a _change_ in capacitance. Is the purpose of the plate to detect the animals 'migrating'? What types of animals are we talking about? – Bruce Abbott Mar 08 '23 at 10:15
  • @BruceAbbott - yes, detecting, counting, estimation of animal type during timeframes - AI is involved here. Hens, chicks, ducks, ducklings... etc. Still not 100% defined. Experiments are ongoing so need some hardware to start-up. I'm curious why do you think that - "0.1% absolute accuracy is hard to achieve."..? With what is hard to achieve? Okay, **change** in capacitance is still good. – valioiv Mar 08 '23 at 10:27
  • @BruceAbbott , yes I need just registration of some relative capacitance movement, don't need exact values in piccofarads. Sory for the confusion! – valioiv Mar 08 '23 at 14:24
  • Sensing is easy, just use a bridge; honestly I'm less worried about the detection method than environmental conditions. If this is outdoors, is it subject to induced lightning? What about losses, how purely capacitive is it? It should be linear based on the description, at least. If AI is involved, what's wrong with an overhead camera and object tagging/identifying algorithm? It'll be a heck of a lot more specific than a one-dimensional capacitance. – Tim Williams Mar 12 '23 at 07:48
  • A typical 555 multivibrator has a grounded capacitor. Why not use this device? – user69795 Mar 13 '23 at 22:37
  • @user69795 , because 555 has around 150ppm vs. 74LS628 which has 50ppm in terms of temperature stability. – valioiv Mar 14 '23 at 09:55
  • @TimWilliams, capacitive sensing is one of a couple methods of a fusion of sensor data streams. Have a requirement to not use cameras. By bridge do you have something like that in mind - updated the main post? – valioiv Mar 14 '23 at 09:55
  • No, not using that IC. A general impedance bridge. It can be a resistor divider on one side, and an R+C divider on the other. A reference AC voltage is applied, and the AC voltage drop is measured. This can be done with a few op-amps, an MCU, DAC and ADC for instance, or some ICs are available to do it as well. I did one a couple years ago with 16-bit precision at 100Hz test frequency. – Tim Williams Mar 14 '23 at 11:45
  • @valioiv, datasheet says cap is between CX1-CX2. However, in both your configurations, you provide a capacitive path to ground which is changing the topology of the VCO and hence not possible to predict the behavior. Is the earth ground and the VCO ground going to be electrically shorted? If they are not going to shorted & will have HiZ between them, can you connect the big plate to CX1 and the earth ground to CX2? The noise picked up by the plate will affect the jitter of the VCO. You'll need to average the frequency over a long period (based on noise freq) to nullify it's effect. – sai Mar 14 '23 at 16:31
  • @sai , actually you are 100% correct in your statements. It is up to me to decide where to put the earth. On the concrete field the earth is not galvanic connected to local power supply/signal ground. So connecting earth to CX2 seems like the best and easiest approach. Haven't think about that, stupid me :D I would suggest to add you statements as an answer and we will see, it more or less answers on what bothered me ;) – valioiv Mar 15 '23 at 14:53

2 Answers2

1

This is best done the way modern precision LCR meters do it: a sinusoidal excitation between the ground and the plate, and a measurement of the current waveform. The current waveform should be averaged over several cycles of excitation, to get rid of asynchronous external interference.

The AC source can be a DAC and an op-amp amplifier. The current shunt amplifier should have some selectivity, i.e. tuning to the AC frequency, to prevent the output from being overloaded by interference.

A "logic level" waveform on the plate will not be easy to work with. The higher the amplitude, the less sensitive the system will be to external interference. If you can, go for about 24VRMS. Due to very light loading, any 36V op-amp should handle that or a bit less (say 22VRMS). The op-amp load current will be <0.1mA, so the step-up power supply can be "tiny", with about 10mA output capacity.

The ADC samples the excitation voltage along with the current. That gets I-Q demodulated digitally. Both the voltage and current channels should pass through DC remover first, and optionally through a bandpass filter. Then multiply the current sample with the excitation sample giving I, and the current sample with excitation sample from 1/4 of the excitation cycle before, giving Q. The I and Q should be averaged over several excitation cycles.

Say we do this measurement with 1kHz excitation, and we're averaging 100 cycles of excitation. Every 100ms we get a 1ms long averaged I and Q waveforms. Suppose we sample at 10kHz - we have a 10 sample I record, and a 10 sample Q record, each obtained from averaging 100 such records.

We then convert I and Q to complex IQ waveform (I+i*Q), and get the magnitude and argument records \$|IQ|\$ and \$\arg(IQ)\$ respectively. The mean of the magnitude record is proportional to the RMS current, and the mean of the argument record is proportional to the phase of the current. From those we get ESR and the capacitance.

This sounds complicated, but is actually quite easy with just about any low-end ARM M0 or similar microcontroller with an external ADC. The excitation waveform can be generated using multi-pulse PWM, a.k.a. magic sinewaves. A low order low-pass filter will convert that pulse train to a relatively clean sinewave. For example, for a 48MHz timer clock and 1kHz output, or 12,000 cycles per quadrant, the THD of the pulse train up to 40kHz is only 0.05%, quantization error included.

(TODO: a full example in C)

Kuba hasn't forgotten Monica
  • 32,734
  • 1
  • 38
  • 103
  • Thanks for your detailed proposal - if it more generic it brings a lot of useful information! +1 from me and let's wait for more opinions! Thanks! – valioiv Mar 15 '23 at 14:56
1

The datasheet says cap should be connected between CX1-CX2. However, in both your configurations, you provide a capacitive path to ground which is changing the topology of the VCO and hence not possible to predict the behavior. Since you confirmed that the earth ground and VCO ground are electrically unconnected, you can consider connecting the big plate to CX1 and the earth ground to CX2.

The noise picked up by the plate will affect the jitter of the VCO. You'll need to average the VCO frequency over a long period (averaging period will depend on the noise freq that gets coupled) to nullify it's effect.

sai
  • 3,352
  • 1
  • 2
  • 13