Given a group of decimal numbers for example 10.121, 1.002 and 0.00003. How should I determine the mantissa and exponent for a custom floating point so that the distinct entries in the group are still distinct after converting from decimal to floating point binary?