Improved approximation of bias in squared coherence estimates for weakly smooth spectra
Peter Lohnberg · IEEE Transactions on Acoustics Speech and Signal Processing · 1978
Bias in squared coherence estimates for normal processes is approximated by a function of coherence and second derivatives divided by values of spectral densities which may vary slightly over the window width. The result approximates more accurately than a known formula results from nonrefuted literature concerning very smooth spectra.