Confidence bounds for magnitude-squared coherence estimates

In underwater acoustics where signals are digitally processed at the outputs of two or more receiving sensors, it is desirable to estimate the coherence spectrum, both for detection and position estimation. A processing technique for computing arbitrary confidence bounds for stationary Gaussian signals is presented. New computationally difficult examples are given for 80 to 95% confidence with independent averages of 8, 16, 32, 64 and 128. A discussion of the computational difficulties together with algorithmic details are presented.