Architecture Review: Portable Electrochemical Gold Karat Sorter (High-Z, Microvo

[Bessous Fethi]

Hi everyone,
I would like your expert review on the system architecture for a project I am finalizing.
Context & Goal: I am building a portable, handheld electrochemical gold karat sorter (sorting 14K, 18K, 21K, etc. via a chemical gel pen).
Disclaimer: This device is NOT intended to replace an XRF analyzer. It is a quick field-estimation tool. Absolute accuracy is handled via statistical software filtering and a robust calibration database, so my main hardware goal is extreme stability, high repeatability, and ultra-low noise, rather than absolute DC precision.
The Architecture (V2.8 Design Freeze):
To isolate the sensitive High-Z analog front-end from RF and digital noise, I have split the system into two distinct boards:
1. Measurement Board (Fully Shielded inside a Hammond metal enclosure):
Analog Front-End: INA333 (Instrumentation Amp) -> ADS124S08 (24-bit ADC).
MCU: A local STM32G031 sits inside the shield to handle local SPI communication with the ADC (traces < 5mm).
Power: 2x18650 batteries -> LT3045 (5V for AVDD) and TPS7A20 (3.3V for DVDD/MCU).
Grounding: 4-Layer PCB. Layer 2 is a Solid Ground Plane (No splits, no star ground). Analog and digital components are strictly separated by physical layout partitioning.
Protection: 100kΩ series resistor on the IN+ input. No ESD diodes directly on the High-Z node to prevent leakage. GND Guard ring around the sense traces.
Temp Comp: A PT1000 sensor inside the probe is read by the ADC's internal current sources to compensate for gel/metal temperature drift.
2. Communication Board (Outside the shield):
An ESP32-C3 sits outside the metal box to handle Bluetooth to a tablet.
The Link: The STM32 sends processed data to the ESP32 via a slow, robust UART connection through a shielded M8 6-pin connector. (I actively avoided running SPI through the cable to eliminate clock jitter/EMI risks).
My Question to the Experts:
Knowing the limitations of electrochemistry, I am focusing purely on hardware robustness.
What are the biggest risks or weak points you see in this specific physical/isolation architecture?
Is there anything you would tweak in the shielding/grounding strategy to further improve immunity in a noisy workshop environment?
Thank you!

[thm_w]

I assume you are talking about something like this: https://www.aliexpress.com/item/1005010803309798.html
and not nitric acid gel pen right?

I don't see any difference in UART vs SPI in terms of "clock jitter" issues. Shouldn't matter. But if you want to use uart thats fine.

[Bessous Fethi]

thanks for the reply!
Yes, exactly. It relies on the electrochemical/galvanic potential generated between the gold alloy and a specific conductive gel, very similar in principle to the electronic testers you linked, not a visual acid color-change test.
Regarding UART vs SPI over the cable: You are totally right that properly terminated SPI wouldn't inherently suffer from jitter over a 10cm run. My choice for the local MCU + UART was actually driven by two other physical constraints:
1. Pin Count: I am using a rugged M8 circular connector which only has 6 pins. To run full SPI (MOSI, MISO, SCK, CS, DRDY) plus 5V, 3.3V, and GND, I would need 8 pins. Using UART lets me comfortably use 5 pins and leave one for the shield.
2. Complete RF Isolation: I wanted to keep the ESP32-C3 completely outside the Faraday cage to avoid any RF bursts corrupting the sensitive High-Z node during measurement. The internal STM32 allows me to handle the ADC polling/DRDY locally in a completely quiet environment.
Does this physical partitioning make sense to you for a noisy jewelry workshop environment?

[thm_w]

A little overkill to me, but no reason it shouldn't work.

Since the bandwidth is so low, you'll want to filter all that junk out regardless. eg. if someone places a cell phone beside the tool, or if there is a wifi router nearby.

There are also these tools: https://trustimetrix.com/en-global/products/sigma-metalytics-investor
But I think its more for coins and bullion.

Any LCD screen or it will be tablet only?

[Bessous Fethi]

Thank you, that makes perfect sense. I will gladly accept a conservative "overkill" design if it guarantees stability in a noisy workshop.
Regarding filtering, I agree with keeping it balanced to avoid slow settling times: a light hardware RC filter at the input, relying mainly on the ADC’s internal Sinc filter, and moderate software averaging. I want to avoid over-filtering a High-Z node.
Here is my most critical question for you regarding this architecture:
My ultimate target is to reliably detect a difference of 5 parts per thousand (0.5% variance) in the electrochemical potential to distinguish very close karat values (e.g., 18K 750 vs 755). Assuming the chemical gel and contact pressure are perfectly stable, do you see any hardware bottlenecks in this specific layout (like leakage currents, noise floor, or drift) that would prevent the hardware from resolving that 0.5% difference accurately?
Also, yes, the UI will be tablet-only via Bluetooth to keep the wand small and electrically quiet.

[thm_w]

I can't say for about real world, but for sure spec wise the ADS124S08 is capable of better than that.
Built in ref is +/-0.05% worst case, 8ppm/C drift, gain error 200ppm worst case, INR, etc.

What is probably more important is the variation you are talking about, if you press harder on the pen how much does that affect the measurement? If the gel is dry? If your hand is wet or dry? These are probably things you could test with a good DMM.
The unit linked above has a cal button and procedure, likely you'd need similar.
Maybe even a calibration using multiple gold reference samples, if you wanted to be super precise.

[temperance]

https://www.eevblog.com/forum/metrology/design-review-32-bit-ratiometric-ac-impedance-system-(0-05-accuracy-target)/msg6236157/#msg6236157

Edit: more specific:
https://www.eevblog.com/forum/metrology/design-review-32-bit-ratiometric-ac-impedance-system-(0-05-accuracy-target)/msg6232471/#msg6232471

When asked where the original method to measure phase differences came from, the question is ignored.

[Bessous Fethi]

@temperance

To clarify: the project initially explored an AC impedance approach. Following detailed technical discussions (especially with Kleinstein), the design was deliberately shifted to a DC electrochemical measurement method based on better stability and repeatability in this application.

So the current architecture is no longer related to the earlier AC concept.

If you have specific technical input on the present DC design (noise performance, layout, interface, or measurement repeatability), feel free to share it.

Otherwise, please keep the discussion focused and avoid adding unnecessary noise to the thread.

[Bessous Fethi]

thm_w
Thank you for the detailed insight. You hit the nail on the head regarding the physical variables (pressure, gel moisture). I will definitely implement a multi-point software calibration procedure using certified gold reference samples, and I will do the baseline physical testing with a high-end DMM as you suggested. It's reassuring to know the ADS124S08 specs are more than adequate.
Since you mentioned filtering out cell phones and Wi-Fi routers, I would love your opinion on our power and RF isolation strategy:
1. RF Isolation (ESP32 vs. Analog):
You mentioned the physical separation might be a bit "overkill". The reason we chose this route is that the ESP32 must actively transmit data via Bluetooth/Wi-Fi to a tablet during the measurement process. Since the ESP32 itself generates significant RF bursts, do you agree that keeping it completely outside the Faraday cage (with only UART, 5V, and GND crossing the shielded M8 cable) is the safest way to prevent its own RF from corrupting the microvolt readings?
2. Power Topology (2S Batteries):
To keep the analog supply absolutely pristine, we are planning a 2S battery setup (2x 18650 in series = 7.4V) located on the external ESP32 board. This 7.4V goes through a step-down buck converter to ~5.5V, travels through the M8 cable, and then feeds a low-noise LDO (LT3045) inside the analog shield to provide a perfectly clean 5.0V for the ADS124S08 and STM32.
Does this (2S -> Buck -> LDO) topology seem like the most robust approach for a microvolt-level system, avoiding the harsh switching noise we would get from a 1S Boost setup?
Thanks again for your time, your feedback is incredibly valuable to this architecture

[temperance]

Otherwise, please keep the discussion focused and avoid adding unnecessary noise to the thread.

The point is that you open a new topic to discuss the same thing.

[Bessous Fethi]

temperance
An AC impedance system and a direct DC electrochemical measurement system are fundamentally different architectures. They require entirely different analog front-ends, different ICs, and different PCB layouts. Reviewing a new V2.10 schematic requires a clean thread to avoid confusing two separate hardware approaches.
You claimed this is "the same thing." Any basic electronics engineer knows the massive difference between an AC and a DC measurement method. How can you confidently argue and derail this thread when you fail to distinguish between two completely different technical foundations?
I will ask you directly: Do you possess the actual technical knowledge to evaluate this new DC architecture? If yes, please bring your concrete engineering feedback to the table. If not, it is best to leave the discussion to professional engineers

[temperance]

You claimed this is "the same thing." Any basic electronics engineer knows the massive difference between an AC and a DC measurement method. How can you confidently argue and derail this thread when you fail to distinguish between two completely different technical foundations? I will ask you directly: Do you possess the actual technical knowledge to evaluate this new DC architecture? If yes, please bring your concrete engineering feedback to the table. If not, it is best to leave the discussion to professional engineers

1. Sure, I do new about the massive difference... But to be very blunt. You are the one who doesn't understand anything while you claim to have an engineering team... I questioned your impedance method and where you found this idea. Instead, you just beat around the bush.
2. In the other topic you've started, you were already discussing this new technique but people are not responding anymore in said topic. That's why you opened a new one. Any idea why you aren't getting much response? See number three.
3. You are supposed to be the engineer your product. Finding the limitations of the new method isn't that difficult. Solving those difficulties is something different and probably the reason why you aren't getting any responses in the other thread. But the main reason is probably because you and the engineering team you mentioned before don't even manage to list the limitations of this method nor propose ideas to solve them.

Anyhow, good luck. Anyone reading this has been informed now and knows where to look.

[Bessous Fethi]

@temperance

You keep asking why others are not responding.

The reason is simple: whenever you step in, the discussion shifts from engineering into personal and provocative remarks. Professional engineers generally avoid engaging in that kind of environment.

If you look carefully, none of the other engineers here resorted to personal attacks. Their contributions were technical and based on actual engineering. You are the only one turning this into something else.

You also repeatedly questioned the origin of the initial idea in a provocative way. In fact, other engineers already responded to you on that point more than once, yet you continued in the same direction instead of engaging technically.

As for your claim that I “don’t understand anything”, that is your opinion. I have no issue with that and I will not respond in the same tone.

If you have technical feedback on the current DC design, you are welcome to share it.

[temperance]

You also repeatedly questioned the origin of the initial idea in a provocative way. In fact, other engineers already responded to you on that point more than once, yet you continued in the same direction instead of engaging technically.

How is this provocative:

Can I be blunt and ask you where you got the below idea from?

This is the answer I obtained.

https://en.wikipedia.org/wiki/Polarography

But this method has nothing to do with the method you proposed. So nobody answered my question. My question was out of genuine interest and I wanted to know how you came to the original idea.

Is this provocative to you, quoting my own question from the other dead thread:

I must clarify my question. My question is not about tiny signals. My question is: where does the phase shift you want to measure came from? I don't think you will see any phase shift at all if piece of gold is 15 K or 24 K. Is this some new method?

And here is where I became provocative because you just prefer to beat around the bush instead of answering a simple question:

Never mind, this is not the answer I'm looking for. You first stated that you wanted to measure phase differences to distinguish between 15...24 K gold. You probably spend some time on that I think. My question is/was: where is the evidence of such? Did you misunderstand something? Or an AI whispered something ridiculous idea into your ears and there is no phase difference at all?

So, just stop putting words in my mouth such or insinuate things which didn't happen such as as:

You keep asking why others are not responding.

Keep asking= means asking the same question several times.

You also repeatedly questioned the origin of the initial idea in a provocative way. In fact, other engineers already responded to you on that point more than once, yet you continued in the same direction instead of engaging technically.

The parts in bold didn't happen.

All fine, if you see that as provocative, it is you seeing it that way. Perhaps because you don't like my questions. Asking the right questions is what engineering is probably all about.

[Bessous Fethi]

You asked for scientific specifics, so let’s address your questions directly and technically.
1. "Do you even know what phase shift is?"
Yes. In our previous AC approach, the phase shift (\theta) is the angle between the excitation voltage and the resulting AC current, allowing us to extract the double-layer capacitance (C_{dl}) and charge transfer resistance (R_{ct}). I had the entire AC framework defined. However, I pivoted away from it based on the excellent advice of respected forum member Kleinstein. He correctly pointed out that while an AC approach might work for pure gold, it becomes highly unreliable and unnecessarily complex when trying to measure mixed multi-metal alloys. He advised against overcomplicating the hardware, and I listened.
2. "What do you actually want to measure?"
In this V2.10 DC High-Z architecture, we are measuring the Open Circuit Potential (OCP) of a micro-galvanic cell. For context, I specialize in 18-karat gold electroforming and metallurgy. Managing these exact galvanic cells, electrolyte interactions, and distinct oxidation potentials of different copper/silver alloy ratios is my daily professional work, not just theory.
Why is it called "DC High-Z"?
Because the gel electrolyte has a high internal ionic resistance. If the analog front-end draws significant input bias current, it polarizes the electrodes, causing a voltage drop across the gel and collapsing the Nernstian potential. That is why we moved to a High-Z topology using the INA333—to measure the static galvanic DC voltage without drawing current.
That is the exact engineering difference. If you have any constructive feedback on optimizing this specific DC High-Z front-end, it is welcome. Otherwise, please allow the actual design review to proceed

[temperance]

And here is my technical answer.

1. "Do you even know what phase shift is?"
Yes. In our previous AC approach, the phase shift (\theta) is the angle between the excitation voltage and the resulting AC current, allowing us to extract the double-layer capacitance (C_{dl}) and charge transfer resistance (R_{ct}). I had the entire AC framework defined. However, I pivoted away from it based on the excellent advice of respected forum member Kleinstein. He correctly pointed out that while an AC approach might work for pure gold, it becomes highly unreliable and unnecessarily complex when trying to measure mixed multi-metal alloys. He advised against overcomplicating the hardware, and I listened.


2. "What do you actually want to measure?"
In this V2.10 DC High-Z architecture, we are measuring the Open Circuit Potential (OCP) of a micro-galvanic cell. For context, I specialize in 18-karat gold electroforming and metallurgy. Managing these exact galvanic cells, electrolyte interactions, and distinct oxidation potentials of different copper/silver alloy ratios is my daily professional work, not just theory.
Why is it called "DC High-Z"?
Because the gel electrolyte has a high internal ionic resistance. If the analog front-end draws significant input bias current, it polarizes the electrodes, causing a voltage drop across the gel and collapsing the Nernstian potential. That is why we moved to a High-Z topology using the INA333—to measure the static galvanic DC voltage without drawing current.
That is the exact engineering difference. If you have any constructive feedback on optimizing this specific DC High-Z front-end, it is welcome. Otherwise, please allow the actual design review to proceed

Thank you very much for the clarification.

1. double-layer capacitance. Are you refering to this? Not sure.
https://en.wikipedia.org/wiki/Double-layer_capacitance#:~:text=The%20other%20layer%2C%20with%20opposite,inner%20Helmholtz%20plane%20(IHP).

2. I think what you want to do from the start is to measure changes in conductivity at high frequencies caused by eddy currents.

https://en.wikipedia.org/wiki/Skin_effect

See "formula" on this page.

In this case, there is no capacitor, at least not at the frequencies you wanted to use. Only some parasitic capacitance in the measurement setup. Only the inductance varies. Then inductance is the sum of the internal and the external inductance of the material under test at some high frequency where the internal inudctance varies because of differences in Eddy currents between different materials.

This is explained on the wiki page under "Inductance".

Because you are actually only inetrested in the phase difference and not amplitude, it might work to sort gold. But I'm too lazy to plug some numbers into an excel sheet today.

[Bessous Fethi]

Hello

[Bessous Fethi]

Hi thm_w
Thanks again for the architecture feedback – especially your point about the ADS124S08's 8ppm/C drift being fine, and physical variables (pressure, gel) being the real limit. That's exactly where my Upwork hires failed.

I've tried a couple of freelancers, but they kept sending AI-generated schematics with no noise budget or thermal EMF analysis. None understood why a 5V→LDO matters for a 24-bit DC measurement.

Since you clearly get the low-noise DC challenges, would you be open to a paid 1-hour review of my final architecture? Or if you don't do freelance, could you point me to a community/platform where real precision analog engineers hang out?

No pressure at all – just value your judgment after reading your posts