Order Statistical Filtering: a Robust Method of Noise Estimation

The detection of a narrowband signal in a sonar environment with spatially uncorrelated white noise depends very much on the accuracy in estimating the noise power so that a threshold can be correctly fixed. The conventional MA method presents a large bias when operated in an environment where interfering signals exist in the neighborhood. Various types of nonlinear methods based on the use of order statistics are introduced and analyzed and are found to be much more robust.