Trainable FIR-order statistic hybrid filters

In this paper, an optimization algorithm for FIR-order statistic hybrid (FIR-OS) filters is introduced. The algorithm minimizes the total cost function of the filter output by dividing the training set into subsets using soft order statistics criteria and then applying conjugate gradient search for the subfilters. The amplitude extraction of pulses acquired from high energy physics detectors is presented as an application example. The trained FIR-OS filter is shown to give a precise amplitude estimate in the presence of sample timing jitter.<
>