R.J. Marks II, "Noise sensitivity of bandlimited signal derivative interpolation", IEEE

each value of R, 100 independent realizations of u and w of 64 samples during T = 10 s each wzre used. From each resulting pair of realizations of u and y, R was estimated using a Papoulis correlation window [4] with maximal window lag-of 15 sam-ples, yielding a window bandwidth [ 1 1, [ 41 W = 1.896 X 54/ 1 5 T = 0.80 Hz forAF1 = 0.1 and F2 = 3.2; thereforeF = 3.1 Hz. Fig. 1 shows R ( R) with 95 percent confidence interval approximations caiculated according to (30)-(33). For the 10 000 values of R obtained, 34 of the actual R were over and 152 under the calculated 95 percent confidence limit approximations. IV. CONCLUSIONS Equation (21) shows khat the variance of relativelinear power contribution estimate RAis lower than that of squared coher-ence function estim9e C ( f). This iszccomplished at the cost of a larger bias in R compared to C ( f) , according to (28). Such bias might be corrected for. Fig. 1 shows that for white signals, confidence intervals derived according to (30)-(33) are reazonable bounds on the simulated R for resulting estimate R. The number of occurrences of R under the lower and over the upper 95 per-cent confidence limit did not reach the mathematical expecta-tion 250 each, however. These deviatiop may be due to approximation errors lik2 no linearity of 2 required for (29) %ver the wide range of R for 64 samples or nonnormality of 2. Similar analysis could be carried out for arbitrary spectra by using cocoloration factors according to (1 9). ACKNOWLEDGMEN