Highly accurate high voltage sensor circuit

[Stiive]

Hi Guys,
I'm new on this forum, I am from Melbourne and did my undergrad at Swinburne in Robotics & Mechatronics Engineering.

I'm currently trying to design a 3-phase inverter for an induction motor.

I am trying to work out what would be the most accurate method to sense my ~400V phase leads. The sample time will be 5uS, and has to be an instantaneous representation of the phase voltage, which will either be high (close to DC bus voltage ~400V +- voltage sag/ripple/switching losses), or low (close to GND). The voltages will be driven by square pulses 5uS in width, and can therefore can go from 0V -> 400V, 400V->0V (or stay roughly the same) in 5uS. The pulse may also last a few seconds in duration. Since I will probably only be able to get 1 reading of each phase in that 5uS, i was thinking of sampling just before the end of the pulse to give it time to settle.

As these voltages are the most important part of my control code, what would be the most accurate method to sense these voltages with my 0-3.6V ADC? A transformer of known turns ratio? Voltage divider with ultra-precision foil resistors? Some precision IC chip?


Currently my design uses foil resistors with an instrumentation amp as a unity buffer for my STM32F4 DSP chip. I'm also using a small cap to filter out some of the noise on my otherwise digital signal. As the phase voltage can go negative a few volts, and to cover noise, i have added in an offset of 312mV (~40V before 141:1 divider). I also need a quick clamping method to stop the instrumentation amp from going into overload (takes too long to recover), and to protect the ADC.
I'm not sure about the zener on the input of the instrumentation amp, but i'm hoping i can find one which will have negligable (<pA) of leakage at 312mV. Is there a better clamp method I could use here?





I'm going to get this made soon and want to try make sure its right the first time, as it will be costly (time&money) if its not.
I think the filter cap needs to be a bit smaller, probably want a 5*RC<2.5uS to get me enough time to sample the signals. I also think i may need to decrease the output current of the voltage divider to give my clamps a bit more chance incase of large voltage spikes, but i don't want too small a current cus of any noise.

Any thoughts/improvements/suggestions would be greatly appreciated.

Stiive.