I have a source for an MS beacon that generates MSK144 directly at 50MHz by programming a DDS chip.   It is straightforward to generate the modulation, switching between two RF 'tones' exactly 1000Hz apart at the symbol rate of 2kHz

The snag with doing that though is that there is no filtering of the sidelobes, those that are generated above the null at 3kHz separation from the centre frequency, on either side due to the abrupt frequency transition.   In conventional usage, upconversion in an SSB transmitter, these will clearly be removed by the SSB filter.

Would it be permissible in my case to add a small amount of Gaussian shaping to the waveform, to turn it into GMSK to lower those sidelobes a bit?   Will the decoder manage to track this?   Assuming there will be some degradation in decoder efficiency do you have any feel for how bad this may be and what sort of B value would be acceptable?

 Looking at the spectrum of the raw audio from WSJT-X when sending MSK144, it looks very similar to the right hand side of mine at RF, with the reappearance of audio above 3kHz.