16 bit Fixed point FFT

Hello,
I have a query regarding 1024 point FFT implementation with 16 bits 
input data( real and imagainary) and the twiddle factors are also 16 
bits(real and imaginary). The algorithm i am using is Decimation in 
Frequency Radix-4 FFT. Hence the no. of stages in the FFT is 5. Please 
let me know how the scaling or rounding of data should be done to fit 
the final result also in 16 bits.

Thanks in advance,
Chetan