This should indeed be possible, the baudrate is simply the buss clock divided by 16 then divided by an integer between 1 and 8192 for each SCI module. The ability to create a baudrate that is very close to the intended baudrate is then dependant on the clock rate used. If there is no other restriction simply create a busclock that assists with this outcome. The accuracy, long term stability etc is then totally dependant on your clock source as the baudrate is simply a division of this. Why the 2% requirement?
To reiterate what Mac pointed out, 18.432 MHz is higher than the AW32's 10 MHz upper limit for a crystal.
18.432 MHz is also higher than the AW32's 16 MHz upper limit for an oscillator with the FLL enabled.
18.432 MHz is fine for an oscillator with the FLL bypassed, but then you are only getting 9.216 MHz for the bus frequency, which is less than half what the AW32 offers.
My suggestion would be to either use a crystal with one of the frequencies that Mac listed, or use a 36.864 MHz canned oscillator. That frequency oscillator is not common, but I have used a Citizen CMX309HBC36.864MTR in non-cost-sensitive products.
Yes, I was incorrect about using the 9.8304 MHz crystal - I had overlooked that the MFD multiplying factor must be an even number.
The following is a list of possible crystal frequencies for FEE mode, and the setup required. In each case, the RFD division factor is 1.
Crystal MFD Bus freq.
MHz factor MHz
2.4576 12 14.7456
3.6864 10 18.4320
4.9152 6 14.7456
7.3728 4 14.7456
True, only the 3.6864MHz crystal produces the highest bus frequency. However, since the reduction for the other crystal frequencies is only 20 percent, these should also work for all but the most stringent applications. Perhaps it also depends on the crystals available from the "junk box".