It's all about the Software

test

Adaptive noise reduction is dependent upon history. Since you have no history for the demodulation in the new mode you might get unpleasant results. The signal processing chain changes with mode with regards to the number of decimation steps (or equivalent) so even the sample rate into the noise reduction module changes. So past history cannot be used.
{o.o}

This isn't an adaptive learning issue. Rather it is a predictable, reproducible, and undesirable condition triggered by certain mode changes that did not occur or happen prior to the NR4 changes/updates/improvements last summer.

Oh well, my effort here was worth a try.