Hm.. now that Assmaster has suggested.. I don't know , first off, what the requirement is .

But if it is a voice waveform that you have , does squaring the each value be the right thing to do ?

I mean, if I consider the 8-bit value as 2 points in a waveform, then , according to this , the ratio between 2 amplitudes at two instants will not be the same in 8 bit and 16-bit.

Are you trying to scale the values ? In that case , won't a simple multiplier help ?