Robust detection of a weak signal with redescending M-estimators: A comparative study

摘要

On finite samples redescending M-estimators outperform linear bounded Hubers M-estimators. To provide stable detection of a weak signal of arbitrary shape, robust Neyman-Pearson detection rules based on redescending M-estimators of location are introduced and studied. It is shown that, on the whole, robust detectors based on redescending M-estimators outperform conventional Huber's linear bounded detectors rules under light- and heavy-tailed noise distributions both on large and small samples.