Plug-out waveform traces

Standby, ground floating, some noise Standby, ground bonded, noise gone
10V/div     0.1 ms/div
Standby, no chassis connection
10V/div     0.1 ms/div
Standby, chassis bonded to ground
With the unit not running in "bypass" mode, there is a bit of noise on the output which completely vanishes once the transformer neutral and unit chassis are connected together.

Running, ground floating, serious noise Running, ground bonded, nice and clean
50V/div     ~ 1 ms/div
Running, unbonded
50V/div     ~ 1 ms/div
Running, bonded
When running in "invert" mode the noise actually gets rather severe, but again, the neutral/ground bond fixes it.  This is with no load.  Applying a load also tends to reduce the noise in the unbonded state, but not reliably.

Hash on battery leads from the car The noise most likely comes from the car.  This is one of the battery leads, AC coupled into the scope.  The battery inputs actually float above and below the unit's neutral reference by about 75V, with this hash riding on top.  When the engine is running to charge the battery, all of this wiggles up and down a bit.

This is about 0.1 ms/division again.  Remember how we observed that the Prius battery ground-fault detection uses about a 5 KHz signal?  Here it is. 

Most likely noise path from car to output The most likely coupling path for the noise.  Both battery input leads and both AC output phases have the blue RFI suppression capacitors connected to ... guess what, the pads where the standoffs support the board.  But if the entire chassis is floating, it simply becomes a conductor for all that!  This is why the neutral/chassis bond is important.

[This is actually a shot of the 5 KVA unit's inverter board]

High load, some flattening at top The output does start to flatten and clip a bit at very high load.

_H*   150512