6.2.23. Variable-Size Low-Resource FFT
The FFT accepts 16-bit fixed-point inputs. The FFT produces block floating-point output using an 18-bit mantissa and a shared 6-bit exponent.
This FFT implementation is unusual because the FFT gain is 1. Therefore the sum-of-squares of the input values is equal (allowing for rounding errors) to the sum-of-squares of the output values.
This method of scaling gives two advantages:
- The exponent can be smaller.
- The output remains consistently scaled when the FFT dynamic size changes.
To configure the design example, edit any of the parameters in the setup file.
The model file is demo_servofft.mdl.