Order statistical filtering: A robust method of noise estimation (Q1090656)

The paper is concerned with the detection of a narrow-band signal in a sonar system. The estimation of noise power is an essential part of this detection process. The author includes and analyses various types of nonlinear methods based upon the use of order statistics. After an accurate introduction, the chapter ''Methods of noise power estimation'' discusses the essential properties of: 1. Moving average filter (M.A.); 2. Median filter (M.D.); 3. Delta trimmed (\(\Delta\) T.) filter and 4. Generalized residual (G.R.) filter. The third part of the work offers a comparison of the various methods of noise power estimation. The probability density functions of the outputs of these estimators are developed. Their biases, their variances as well as their mean square errors are evaluated. The author concludes that all the nonlinear estimators are very much more robust than the conventional M.A. estimators. Among them, the best estimators come from the suggested generalized recursive (G.R.) filters which give the smallest mean squared errors. Compared to the M.A. filters, they yield a lower probability of false alarm and a higher probability of detection.