To combine radix-4 FFT with radix-2 FFT

I am trying to calculate a 2048 points(2^11) FFT. Normally only cooley-tukey radix-2 algorithm can be employed. But there are ways to combine them as shown by Xilinx's FFT solution.

I have coded both radix-2 and radix-4 Decimation-In-Time algorithm with binary reversed and quaternary reversed input repectively. Radix-4 is 20% faster than radix-2 algorithm but can't be applied to 2048 points fft. How to combine both algorithm?

[Bjorn Gustavsson]

Why not use the built-in fft-function? - It is using the fftw which should be the best there is (at the moment at least).

[庆乐 曾]

@Bjorn Gustavsson Thanks for your reply. Because it will be used for hardware implementation. I am working on a fast 2048 points fft algorithm. So it's more of a mathematical question here. I am currently reading fftw's doc to learn how it optimizes tho their practices might not be suitable for hardware.

[Walter Roberson]

Note: there is a Radix 2 for hardware use at https://www.mathworks.com/help/fixedpoint/ug/convert-fast-fourier-transform-fft-to-fixed-point.html and at https://www.mathworks.com/matlabcentral/fileexchange/22326-fixed-point-radix-4-fft?s_tid=FX_rc1_behav there is a Radix-4 for fixed-point use.