Hi,
I'm running 1024 point FFT with 2 port and comparing the output from the testbench with the output of FFTs from Numpy (which will also agree with MKL/FFTW/etc). What I see is that I can get a rough agreement if I take the output from the testbench and multiply the real part by minus 1 - but its not a great match - and its unfortunate that the real part of the input in the testbench was all set to zeros so I'm not sure how much of the actual FFT implementation is being tested. The output from the testbench is roughly equivalent to doing the standard FFT of the signal: {first_sample, time_reversed(all_other_samples)}. Unfortunately the documentation does not clearly state what the input or output data order is, so perhaps this is expected behaviour?
Thanks,
Matt
Hi,
I'm running 1024 point FFT with 2 port and comparing the output from the testbench with the output of FFTs from Numpy (which will also agree with MKL/FFTW/etc). What I see is that I can get a rough agreement if I take the output from the testbench and multiply the real part by minus 1 - but its not a great match - and its unfortunate that the real part of the input in the testbench was all set to zeros so I'm not sure how much of the actual FFT implementation is being tested. The output from the testbench is roughly equivalent to doing the standard FFT of the signal: {first_sample, time_reversed(all_other_samples)}. Unfortunately the documentation does not clearly state what the input or output data order is, so perhaps this is expected behaviour?
Thanks,
Matt