Autocorrelation function is calculated using
tmp = [ fft(abs2.(ifft(r))) for r in eachrow(μ)]
which seems to have ifft and fft the opposite way.
The resulting spectrum is correct, because it is corrected when frequency variable is taken out in spectrum-function.
But it migh be a good idea to correct this in future.