Hello everyone,

We are currently developing a three-phase energy meter based on a non-isolated ("Hot MCU") architecture. In this design, the low-voltage section (MCU, AFE, and Ground) floats at line potential, with the measurement front-end referenced directly to the mains via high-value resistive dividers.

We are seeking feedback on the robustness of our current analog front-end implementation, specifically regarding our biasing strategy and signal conditioning for production-grade reliability.

System Architecture Overview:

• Power Supply: 5V SMPS-based supply (non-isolated).
• Voltage Sensing: High-impedance resistive divider (2 X 680KΩ + 2.2KΩ ). This is followed by an op-amp unity-gain buffer.
• Current Sensing: Internal Current Transformers (1000:1 ratio) with a 300Ω burden resistor.
• Biasing: We use an op-amp stage to generate a 2.5V mid-bias (Virtual Ground). Both the voltage buffers and the CT burden resistors are referenced to this single Vgnd point.
• Filtering: Standard RC low-pass stages before entering the MCU ADC pins.

Our Primary Concerns:

1. Global Virtual Ground (Vgnd): Is it acceptable to use a single op-amp-buffered virtual ground for all three voltage channels and three current channels? We are concerned about potential crosstalk or stability issues when multiple ADC sampling transients hit the same bias rail.
2. Differential vs. Single-Ended: While this single-ended approach (referenced to Vgnd) is cost-effective, would moving to a fully differential sensing approach provide significant gains in noise immunity for a "hot" design, or is it overkill given the lack of isolation-related common-mode noise?
3. Phase Shift & DC Offset: Given we are using unity-gain buffers and a shared bias, what are the best practices for compensating for op-amp input offset voltages and phase shifts introduced by the CTs and RC filters in firmware?
4. Production Robustness: For those with experience in industrial or utility-grade metering, are there any "gotchas" regarding long-term drift of the high-value dividers or protection against mains transients (4kV + bursts) in this specific floating-ground topology?

We would appreciate any insights on whether this implementation is suitable for a production-grade meter or if we should pivot toward a more decoupled architecture.

Thanks in advance for your expertise!

 

Allow me to be the first of many: s.c.h.e.m.a.t.i.c - ?

Paraphrasing Rear Admiral Joshua Painter,

"Engineers don't take a dump, son, without a schematic."

ak

 

I have always preferred to compensate the CT’s phase shift with a front end RC network. Of course, this only applies if you know beforehand the CT’s characteristics.

For the voltage offsets, if you are employing any of the newer ADI power metering devices, their offsets are already low enough to be considered within the device’s total error budget.

For the high voltage resistor divider network, I would suggest an integrated resistor divider available from Ohmite, Vishay and others. Their long term stability and temperature tracking will be significantly better than discrete devices, and they will have published surge ratings.

 

I have always preferred to compensate the CT’s phase shift with a front end RC network. Of course, this only applies if you know beforehand the CT’s characteristics.

For the voltage offsets, if you are employing any of the newer ADI power metering devices, their offsets are already low enough to be considered within the device’s total error budget.

For the high voltage resistor divider network, I would suggest an integrated resistor divider available from Ohmite, Vishay and others. Their long term stability and temperature tracking will be significantly better than discrete devices, and they will have published surge ratings.

Thank you for the detailed insights!

Regarding the phase shift, doing it in the hardware front-end with a tuned RC network is an interesting approach. We had initially planned to handle it in firmware to keep the BOM as cheap and generic as possible, but we will look into the feasibility of hardware compensation based on our CT's consistency.

For the offsets, we aren't using a dedicated ADI metering IC. The entire system runs on a CMS8S6990 MCU, and we are feeding the signals directly into its internal 12-bit ADC. Because we are relying on a general-purpose MCU rather than a specialized metering chip, dealing with the op-amp and ADC offsets in firmware (or minimizing them in the hardware design) remains a major priority for us.

The tip on integrated resistor dividers from Vishay/Ohmite is fantastic. We are using discretes for the prototype phase, but shifting to an integrated network for the production board makes complete sense for thermal tracking and surge compliance.

 

Allow me to be the first of many: s.c.h.e.m.a.t.i.c - ?

Paraphrasing Rear Admiral Joshua Painter,

"Engineers don't take a dump, son, without a schematic."

ak

Fair point!

I have attached a schematic snippet showing our 2.5V virtual ground generation, one of the voltage sensing channels, and one of the CT current sensing channels.

 

i was just curious about accuracy (in general)
https://www.google.com/search?chann...ve+watt+meter+3-phase+accuracy+vs+power+level
reveals a short reminder list https://cdn.tmi.yokogawa.com/Fundamentals_of_Power_Measurement_111412.pdf#page=88

 

Yokogawa is the Rolls Royce of power measurement instruments.

 

If you are isolating the current signal with a current transformer, why not isolate the voltage as well with a ZMPT101, or conversely, if you are running it mains-connected why bother isolating the current signal when you can use a shunt?
Whatever you do, DON’T use the current measurement circuit that you have drawn. When there is a big surge of current, you get a big voltage on the output of the CT at quite a low impedance. Enough voltage and enough current to do real damage.
The CT is best interfaced with a transimpedance amplifier (see the application circuits on www.micro-transformer.com) as it keeps the output voltage at almost zero which improves accuracy (because there is no magnetising current) and improves phase accuracy.
You can make a half-supply reference that will supply quite a few amplifiers, but don’t go running wires from it all over the place, otherwise you will upset the op-amp that is driving it and make it go unstable due to the capacitance on its output (And don’t connect a capacitor across its output, it will almost certainly oscillate)
Why do you have a connection between neutral and mains earth? They should be completely isolated (unless you are just using the wrong earth symbol)

 

If you are isolating the current signal with a current transformer, why not isolate the voltage as well with a ZMPT101, or conversely, if you are running it mains-connected why bother isolating the current signal when you can use a shunt?
Whatever you do, DON’T use the current measurement circuit that you have drawn. When there is a big surge of current, you get a big voltage on the output of the CT at quite a low impedance. Enough voltage and enough current to do real damage.
The CT is best interfaced with a transimpedance amplifier (see the application circuits on www.micro-transformer.com) as it keeps the output voltage at almost zero which improves accuracy (because there is no magnetising current) and improves phase accuracy.
You can make a half-supply reference that will supply quite a few amplifiers, but don’t go running wires from it all over the place, otherwise you will upset the op-amp that is driving it and make it go unstable due to the capacitance on its output (And don’t connect a capacitor across its output, it will almost certainly oscillate)
Why do you have a connection between neutral and mains earth? They should be completely isolated (unless you are just using the wrong earth symbol)

Thank you for the rigorous review! This is exactly the kind of feedback we are looking for. To address your points:

1. Architecture Choice (CTs vs. Shunts):

Since this is a three-phase meter, if we used shunts for current measurement, we couldn't tie them to a single non-isolated MCU without shorting the phases or adding expensive isolated ADCs/modulators for each channel. By floating the MCU at Neutral, we can use dirt-cheap resistive dividers for the three voltage channels, while the CTs provide the necessary phase-to-phase isolation for the currents. We avoided ZMPT101s simply to keep BOM cost and physical board size down.

2. CT Surges & The TIA:

Excellent point about the surge voltages across the burden resistor. In many cost-optimized commercial meters, we see the burden resistor kept low and the ADC pin clamped heavily (e.g., dual Schottky arrays to the rails + series resistors) to handle surges. I will definitely review the TIA app. notes on micro-transformer.com. In your experience, is a TIA strictly necessary to hit Class 1.0 accuracy, or can a heavily clamped burden resistor suffice for a cost-optimized design?

3. The VGND Capacitor:

I realized right after drawing it that putting 100nF directly on the op-amp output is a textbook oscillation trigger. We will be adding an isolation resistor R_iso to decouple it, or removing that cap entirely and keeping the VGND traces as short as possible to minimize parasitic capacitance.

4. Neutral/Earth Connection:

Good catch, that is just me using the wrong symbol. That line represents the local floating circuit ground, NOT Mains Earth/PE. There is no Earth connection in this system. The inductor/resistor network drawn there was an attempt at sketching the EMI filter on the neutral line entering the SMPS, but I drew the system ground reference in the wrong place. The MCU ground will be tied directly to Neutral.

Thanks again for taking the time to look at this!

 

Fair point about the position of the CTs. I’ve used shunt amplifier ICs to interface them to the A/D such as INA181. It keeps the burden low, and protects the A/D because the shunt amplifier will withstand a lot of voltage on their inputs. AD8418 is particularly good as you can bias the output to half-supply, but it is a little pricey.
You don’t need an op-amp on the voltage sense. This circuit does the job nicely, and the output will be at a low enough impedance to drive the A/D. (You will need two or three resistors in series for R1, as at worst case it could see 620V - loss of neutral at peak mains)

 

This is a very interesting post!! The TS is requesting a whole lot of very needed and complex engineering evaluations that also carry a fair amount of potential liability if they are not adequate.
In addition, I do not recall seeing any mention of the intended accuracy of the product.
Quite often, measurement accuracy is important.

One addition that I can offer is that with some three-phase loads, there is a fair amount of neutral circuit current. Not where the loads are 3-phase motors, but "systems", and so the phase currents will differ, and the neutral current will not be zero.

 

Fair point about the position of the CTs. I’ve used shunt amplifier ICs to interface them to the A/D such as INA181. It keeps the burden low, and protects the A/D because the shunt amplifier will withstand a lot of voltage on their inputs. AD8418 is particularly good as you can bias the output to half-supply, but it is a little pricey.
You don’t need an op-amp on the voltage sense. This circuit does the job nicely, and the output will be at a low enough impedance to drive the A/D. (You will need two or three resistors in series for R1, as at worst case it could see 620V - loss of neutral at peak mains)
View attachment 366767

Just to clarify our topology—we are actually using an isolated flyback SMPS to generate our 5V rail. However, because we are using resistive dividers to directly sense the mains voltage, we are intentionally bridging that isolation barrier. Our secondary system ground is referenced to the mains. Therefore, we are fully treating this as a "Hot MCU" architecture where the safety isolation is handled entirely by the physical meter enclosure, not the PCB.

 

Just to clarify our topology—we are actually using an isolated flyback SMPS to generate our 5V rail. However, because we are using resistive dividers to directly sense the mains voltage, we are intentionally bridging that isolation barrier. Our secondary system ground is referenced to the mains. Therefore, we are fully treating this as a "Hot MCU" architecture where the safety isolation is handled entirely by the physical meter enclosure, not the PCB.

You could turn that flyback into a SEPIC! Just add a capacitor.