You don't necessarily need to use floating point for most of the calculations if you scale the numbers. If you are using C then all the hard work is done for you anyway, even if you use an 8 bit PIC. The only thing you need to watch out for is how long the calculations take. From memory a 16*16 multiply with 32 bit result is done in a single instruction cycle and divides are around 18 cycles. If the numbers are 32 bit or floating point it will take longer. I am not sure if there are times given in the C compiler help but if not you will need to benchmark it. Provided you have some 'dead time' in which to do the calculations I doubt it will be a problem with a PIC24.
Keith
- - - Updated - - -
By the way, my PIC24 experience is mostly the E and H versions but I think the F versions are similar - just slower clocks,