Why has the output bit width of the FFT core been reduced since Quartus II software version 11.0?

For variable streaming fixed point FFTs, the bit growth algorithm is changed starting in the Quartus® II software version 11.0.  For the new algorithm, the overall bit growth is less than or equal to Log2 (N) 1, where N is the length of the FFT.  This new algorithm allows us to reduce the output bit width by removing the 2 unused MSBs in the frequency domain.

As an example, take a 1024 point FFT, with data precision width of 18 bits.

In version 10.1

Output bit width = data precision width 2.5 ×(log4(MAX(N)) = 31 bits

In version 11.0 and above

Output bit width = data precision width log2(N) 1 = 29 bits