Some comments on bias/Misalignment effects in the magnitude squared coherence estimate

This correspondence presents a simple functional model for obtaining the magnitude squared coherence (MSC) estimate. The decrease in the approximated mean of the estimate with signal misalignment is shown to be a function of the signal spectrum and misalignment time, τ. The mean of the estimate in the absence of a signal common to both channels is shown to be a product of the bandwidth of the noise W and the averaging time T. An alternate process employing clipping for normalization is shown to provide a similar result and provides inherent constant false alarm rate (CFAR) properties.