Discriminative Dictionary Learning for Mixture Component Detection with Application to RF Signal Recognition
Hao Chen, Seung-Jun Kim, Thomas Chatt · 2018 52nd Asilomar Conference on Signals, Systems, and Computers · 2018
Pattern classification algorithms based on sparse dictionary learning are derived. After training a discriminative dictionary and a linear classifier using the samples of the individual classes, the aim is to apply the dictionary and classifier for recognizing the component signals in a mixture of different class signals. A key issue is to prevent the “leakage” of strong signal components to weaker components in the classifier. We tackled this issue by encouraging orthogonality among the discriminants during the training, applying the algorithms to RF signal recognition verified the efficacy of the approach.