AC Signal Conditioning Problem for Toroidal Conductivity Sensor

[digitalectron]

Hello everyone,
I am designing the analog circuit for a toroidal (inductive) conductivity sensor, and I am having trouble with the AC signal conditioning stage. I would appreciate any advice.
The sensor produces an AC signal  with 50khz and variable Voltage from 50mv to above whose amplitude changes depending on the conductivity of the liquid. My goal is to accurately measure this signal with an STM32 ADC. the initial part of my schematic is attached.
" alt="" class="bbc_img" />
The input passes through a band-pass filter. then  filtered signal is buffered/amplified by the first TL084 op-amp.The signal goes through another filter/amplifier stage like Channel1. Finally, I generate an inverted version of the signal using another TL084 channel and The outputs are labeled A11 and A00 , just for A00 to be the inverted version of A11.
The initial signal on pin 3 is :
" alt="" class="bbc_img" />
when I measure A00 signals with my oscilloscope:
" alt="" class="bbc_img" />
but , the signal is noisy and This makes me think that either:
I made a mistake in my analog design,
the TL084 is not operating correctly,
or the power supply is introducing a problem.
Currently I generate the op-amp supplies using two MC34063 switching regulators:
+12 V → +5 V then -5V
" alt="" class="bbc_img" />

Because I suspect switching noise may be affecting the analog circuit, I am considering replacing the supplies with a linear solution using:
LM317
TL431 (TL431BQDBZR, 0.5% reference)
in T431 datsheet
 LM317.PNG (43.73 kB. 883x570 - viewed 9 times.)" alt="" class="bbc_img" />
for the positive supply, and using a similar linear regulator for the negative supply.
now my questions is?
Does anything in my analog circuit look incorrect?
Is the TL084 a reasonable choice for this application?
Could the MC34063 switching supplies be causing this behavior?
Would replacing them with LM317/TL431 linear regulators likely improve the measurements?
Is there a better approach for conditioning a variable-amplitude AC signal from a toroidal conductivity sensor?

[MT4S301]

Could the MC34063 switching supplies be causing this behavior?

Yes. Please avoid powering analog front-end circuit directly with DC-DC regulator or charge pumps.

Is the TL084 a reasonable choice for this application?

50khz is a little too high for TL084 however the replacement depends on your design goals (power, precision, supply voltage....)

[Terry Bites]

Configure the first gain stage for high gain to maximise the SNR.
The TL084 is very noisy. 37nV/√Hz!  They only still exist because people buy them by mistake.
So 70's! Choose something more contemporary say, OPA1664.

LC filters need controlled impedances to work effectively. 1k drive and 10k receive puts you in the op-amp's comfort zone.
Deal with the filtering in one filter stage where, that another useful bit of noise and DC error reduction.

Design a proper band-pass filter.
markimicrowave.com/technical-resources/tools/lc-filter-design-tool/


If you want to DC couple the non-inverting stages a cap to ground is needed, C1 in my diagram. It sets the corner freq of the lowest freq of interest. ie high pass.
That makes the amplifier stages unity gain at DC so that op-amp offset voltages are not amplified.
You only need AC gain.
Inverting amplifier stages will work just as well if not better. AC coupling is not an issue in this project.

The MC34063 is operating just an octave above your coil frequency.
That makes it very hard to filter out. unshielded DC-DC converter inductors radiate EMI. Bad.
Use a DC-DC converter that runs in the in or up near the MHz eg a TPS54202
And try this tool https://webench.ti.com/power-designer/ read the datasheet and if you can lay it out like they say. fbp.jpg (414.16 kB. 1376x1289 - viewed 41 times.)
Or maybe cut your losses (not the power ones though) and just settle for a linear regulator.
I doubt it needs to be precise. How sensitive is your oscillator to its power supply?
Do small changes in frequency matter considering component value tolerances in the resonant circuits?

Poor PCB layout and wiring have a negative impact on noise both conducted and radiated
You need shielded cables for your coils. make sure to take into account the cable capacitance. So many pF per metre (or fahrenheit per cubit in freedom units).
On the receive side use a shielded twisted pair cable and ground tie shield to the amplifier input ground.
Do the same thing on the transmit side, tie its shield to the TX ground to stop radiating guff into to other instruments in your lab.
Shield the TX and RX coils, tie to their respective cable shields.
My understanding is that you need lock in detection and that the front end amplifier in these sensor systems measure the current induced in the coil, not the voltage across it.

I wish you good luck.

[Kleinstein]

The shown scope picture shows strong interference with strong ringing. This could well be the SMPS.

I would avoid LC type filtering in the amplifier. 50 kHz is a bit on the low side to make LC fitlering easy and there is the chance of interference. RC type filtering should be good enough, if there is not much external interference. A good part of the final filtering would come from the digital / software side.
The design of the actual sensor part also effects how sensitive it is to EMI pick up.

I would skip the SMPS part and run the analog part directly from 5 V or 12 V or even just 3.3 V.  One may want the last amplifier stage to be powerd with the same 3.3 V as the µC / ADC to simplify protection from excessibe voltage.
One can operate with a kind of virtal ground somewhere in the middle of the supply range. The TL084 is rather noisy - there are now better alternatives. For much of the amplifier part the noise leven may not even be a problem. Which type fits depends also on the supply used. Chances are one can get away with fewer amplifiers, like one amplifier with 3.3 V supply and maybe one with a higher supply to have more headroom.

[digitalectron]

Thank you very much for your valuable comments and suggestions. I really appreciate your time and experience.
I have attached some additional oscilloscope screenshots showing the noise on my power supply and the signal problem. After your comments, I think I need to redesign the analog front end.
+5 V output from the MC34063:
-5 V output from the MC34063:" alt="" class="bbc_img" />
24 V output from my MATRIX  power supply:" alt="" class="bbc_img" />
As you can see, the 24 V bench power supply is much cleaner than the outputs from the MC34063 regulators.

I also disconnected the toroidal sensor and applied an external signal generator instead. The signal is almost clean before it is connected to pin 3 of the connector. However, once I connect it to pin 3, it becomes noticeably noisier.
" alt="" class="bbc_img" />
" alt="" class="bbc_img" />

I understand now that these choices may not be ideal for a low-level AC measurement around 50–70 kHz.

I would like to redesign the circuit with:

A modern low-noise op-amp
Active RC filtering instead of LC filtering
A better power supply solution


I am considering:

single op-amp,
dual op-amp,
or quad op-amp.

I need several channels because I need:
signal amplification,
filtering,
and an inverted output for the rest of my project.

Would a quad op-amp such as OPA1664/OPA1644 be a good choice, or would separate dual op-amps give better performance?

[Kleinstein]

A quad OP-amp can be about as good as 2 x dual. A downside may be heat with high power consumption types. Separate amplifiers allows to used different types (like 30 V and 5 V supply). 2 dual op-amps don't need much more space and costs are often similar.

[MariuszD]

The second filter (post #1) doesn't work, the first one probably doesn't either because it depends on the source impedance.
MC34063 has a variable frequency, which is why it is the worst choice.

Active RC filtering instead of LC filtering

LC filters are not bad, but your implementation of the LC filter is bad. Making a good active RC filter at 50kHz requires amplifiers with large GBW. It is worth considering synchronous detection (lock-in amplifier), very simple, and provides a very narrow filter bandwidth. Sometimes it is said that this method allows detection below the noise level, although this is only partially true.

[Kleinstein]

Synchrous detection is defenitely a good idea. However this can be on the software side. The analog filter (RC or LC) is mainly to suppress large amplitude interference from SMPS and similar, not so much noise. So if possible definitely avoid a switched mode regulator - especially an old one with a frequency so close to the one used from measurement.
If one has a lot of interference, one could also build analog synchronous detection, but it is quite some extra effort.

A RC type active filter would need a really high GBW with high Q, but there is no real need for a high Q, unless there is a strong interfering source close to the frequency of interest. A low Q filter tends to be more stable.

[digitalectron]

Thank you again for all your suggestions.
I have one more question about the power supply.
Several of you recommended avoiding the MC34063, and I understand why. However, what would you recommend for generating the negative supply?

I have often heard that a dual supply (±5 V) is better than a single supply for analog signal conditioning because it provides more headroom and avoids biasing issues. Is that still the preferred approach for this kind of 50–70 kHz conductivity measurement, or would you recommend redesigning the analog front end to operate from a single supply (for example, 5 V or 3.3 V with a virtual ground)?

If a dual supply is still the better choice, what is a good way to generate a low-noise negative supply?

[Kleinstein]

50-70 kHz is very easy to AC couple. So there is no real need for a +- supply. Using just 5 V is a real option. 3.3 V is a good idea for the last stage, not to overdrive the ADC. There are now pretty good OP-amps for 5 V supply.

[MT4S301]

Several of you recommended avoiding the MC34063

Not only MC33063 but also most SMPS regulators we build by IC+Inductor+(optional) Schottky&MOSFETs on PCB. They emit deadly amount of H-field.

Is that still the preferred approach.....or would you recommend....a single supply (for example, 5 V or 3.3 V with a virtual ground)?

Single supply may be simpler and elegant if you don't need real zero-volt at OPAMP's output. However if your +5V comes directly from a AC-DC adapter or other switching regulator (e.g. USB bus) you might have to add a linear regulator for the analog frontend and accept at least 0.5V voltage dropout.
FYI, the TLV9062 is a cheap and available OPAMP for low-volt analog frontends with 5x higher slew-rate than your circuit requires.

what is a good way to generate a low-noise negative supply?

Please consider [charge-pumps + negative LDO] when powering sensitive frontends (e.g. LM2776+RP117Q for absolute output voltage <5V and load current <100mA).

[Terry Bites]

How much headroom do you need? There isn't much of dynamic component to your signals.
Presumably your signal goes into an ADC which likely has a unipolar input that will need protection from over-voltage.
Lock-in detection is an absolute must.
The attached should be right up your street. With your narrow band preamp of course.

[Kleinstein]

It depends on the enviroment how much interference is expected. If there is not too much interference, one can do the syncronous detection after the ADC. The analog synchronous detection makes sense if there are a lot of interference.