normalisation of floating point numbers in binary