Skip to Main Content
This paper discusses a detection technique for impulsive noise in power line cables. Essentially, a reduced set of features (signature vector), which is selectively extracted from the power line signal feeds a detection technique. The features are higher-order statistics and the Fisher's discriminant ratio is the selection technique. The designed detectors are a multilayer perceptron neural network and a Bayes implemented according the maximum likelihood criterium. Simulation results indicate that improved performance can be attained if multilayer perceptron neural network is considered as a nonlinear detector. Also, the results reveal that higher-order statistics is a very interesting technique to extract a reduced and representative signature vector of impulsive noise.