The problem was shoot through.
Definition of an H bridge
short waiting to happen.
IGBT's have nasty current tails that do not switch off even when the drive gate is totally discharged do to charge acumulation transistor decay times.
Speed up the IGBT turn on...see definition above
1) air core in upper drain
2) more dead time (less dead time dead IGBT's) (not a problem exept in high accuracy aplications)
3)1st and 2nd reduce source problem another solution (sometimes you need them all!) is to use the same clock or sense the power suply clock and interleave sampling so to not be on when H bridge switching.
5) guard rings and ground planes around dif (or single ugg) signal.
6)Use that extra op amp to pic up RF, amplify, condition it and subtract it from signal in real time.
7) Two or more of above
I suspect a poor layout, ground loops, poor or ineffective grounding, separation of critical circuits or components, etc. Using high speed IBGTs magnified the problem due the t6he faster switching.
The use of opto isolators solved the problem by removing or reducing the effects of parasitic "antennas" caused by an apparent poor layout.
Re-engineering the IGBT drive circuitry using EMC reduction techniques,is the better solution.
The analog 4-20mA current loop does not make optical isolation easier.
It removes the necessity of taking the amplifier ground out to the field.
It was invented before optoisolators and the current range was higher.
You are confusing it with the digital adaptation, which switches between 4 and 20mA.
This story exemplifies why the 4-20 mA current loop analog interface was designed. A current loop interface gives you optical isolation combined with a low impedance, differential input for maximum noise immunity.
If you are ever faced with this sort of problem again, you can buy 4-20 mA converters from vendors like Phoenix contacts. They take a 0-10V single ended input and convert it to an isolated 4-20mA current loop. To convert a current loop back to single ended voltage at the receiver end only takes a single resistor.
Another advantage to a 4-20 mA current loop is you can tell the difference between a zero signal and a cut cable.
Whatever works. Reagan tried a number of valid hardware solution tests. With the cost involved and bosses no doubt breathing down necks, one goes with the first solution that solves the problem.
Good story, Reagan.
Isolation is indeed what it takes. I wound up using Analog Devices isolation amplifier modules back in the late 1970s for the four-quadrant drive used in engine test stands. The only difference is that I never even tried to go directly with the control and data wires. Isolation amps are sometimes woth their weight in gold. Thanks, AD.
Good Story of Success, this is the problem virtually every design engineer faces with in his life, and everyone finds some special solution, but unfortunately there is no standard defined solution to this problem, thats Electronics. That is the reason we love electronics.