high-side current sensor

Instead of buying nine HV (instrument or op)-amp chips, use any-old-one, and a resistor-zener-capacitor to bias nine transistors with suitable ratings in cascode.

Actually, since this is average-current (heat, really) related, you can probably dawdle a bit; maybe just use that zener supply to do current/frequency converters, and then nine capacitors will do all the level translation required; it just takes a count-and-average bit of hardware at the logic-level to complete the sensing.

The c code can shut down supply modules as needed. Each power supply module has its own FPGA that the main controller can address. There's no need to actually remove 48v power from the baby boards. The master controller just needs to know the currents and have some reasonable rules.

tirsdag den 11. januar 2022 kl. 23.31.35 UTC+1 skrev John Larkin:

check.

depending on how good the ADC is, maybe just a resistivity divider on the 48V and after each shunt is good enough?

I've used the ina168 and ina219. Liked the results - easy to use. The ina168 seems to meet your specs.

The AD8479 handles up to 600V common mode. Couple this with a current shunt and you are good to go.

also much much more expensive than a current sense amplifier

If a data sheet looks clear, I wouldn't test the part.

We have room to put two different highside chips on the PCB and just stuff the one we can get. TI and ADI maybe.

Looks like we can use the XADC in the Zynq, the main controller FPGA. That's a 1-volt full-scale, pretty terrible ADC. The voltage drop across a shunt might be 50 mV with a common-mode of 48 volts. So we need some sort of high-side amp.

A flying capacitor mux would be fun, but too slow. And a dual SSR would cost more than a current sense chip.

On Tuesday, 11 January 2022 at 17:21:32 UTC-8, John Larkin wrote: ...

... I like the LTC6102.

It's not as cheap as some of the other solutions but has a very low offset voltage so the burden (and power dissipation in the shunt) can be very low. The architecture is the same as many others using a P-channel FET.

The vanilla one can tolerate 70v surges and there is a 100V part available as well. I'm using it in an automotive 48V design where surge tolerance is vital as well as automotive qualified.

kw

Cars use a lot of high-side current sensors

and

The semi shortage/glut cycle is swinging up

and

China is scary.

Of course five LM358As, five dual 2N3906es, one 24V zener dropper, and nine quad pack resistors would do it too.

Alternatively, a regular resistive diff amp using your garden variety op amp and a decent quad pack could run off +26V and ground, say. With unity gain, the output is near ground and the inputs are near +24V. You'd have some offset due to resistor mismatch, but nothing horrible.

Cheers

Phil Hobbs

onsdag den 12. januar 2022 kl. 02.21.32 UTC+1 skrev John Larkin:

yeh, at the moment it makes sense to practice "defensive design" with multiple options

On a sunny day (Wed, 12 Jan 2022 08:09:56 -0800) it happened snipped-for-privacy@highlandsniptechnology.com wrote in snipped-for-privacy@4ax.com:

Oh, and there is an other solution Make a nice front panel with some of these for example:

:-) Was just one of those wild ideas, 'lemme try this'. So simple..

A single current sense amp, but switched to the different sense resistors with a 50V ana mux. Syncronized to the ADC sampling

We could mux on the high side, with a cheap 4051 type mux. We'd need a highside power supply, just a zener, and some way to translate the mux address lines.

True; that's only the first step, though, it just makes a ground-referenced signal, and next step is an ADC. Trouble is, this is a power signal, with surges expected, so the sampling due to an ADC is problematic. Since low-power-using modules aren't the most important to get accurate readings on, I'd note that an eightpack of VCOs can deliver pulse streams that don't ignore any of the abrupt changes in their inputs, and at the ground-referenced end it just takes counter inputs to digitize the averages over any convenient period. Ten kilohertz center frequency, half-second average period, beats the precision of a 1% shunt resistor. Total the counts to get that ninth number...

A bunch of TLC555 with the INAxxx current doing the capacitor charge is the general scheme. You'd want the capacitors accurate, and/or adjust counter gate times after calibration, and a regulated Vdd for the '555.

Less elaborate, capacitors in parallel with the low-end receive resistors to do averaging.

the Zynq ADC is 1MSPS ...

And kinda nasty. But we can scan fast and do a bit of signal averaging in the FPGA. I only need the current measurements every millisecond, to do my shutdown protection logic.

Exponential smoothing (1st order lowpass) is easy

Vout = Vout + (Vin-Vout) * K with smallish K

but my FPGA kids will probably want to do something fancier.

Well, you might actually compare prices of a working circuit to the AD8479. I don't think you can.