An optimal generalized theory of signal representation
J. Scott Goldstein, J.R. Guerci, I.S. Reed · 1999
A new generalized statistical signal processing framework is introduced for optimal signal representation and compression. Previous work is extended by considering the multiple signal case, where a desired signal is observed only in the presence of other non-white signals. The solution to this multi-signal representation problem yields a generalization of the Karhunen-Loeve transform and generates a basis selection which is optimal for multiple signals and colored-noise random processes under the minimum mean-square error criterion. The important applications for which this model is valid include detection, prediction, estimation, compression, classification and recognition.