Filtering time series with topology

Following a brief discussion of the potential relevance of chaotic noise models, we consider the problem of separating a signal from an additive mixture with nonlinear noise. The approach we take exploits various properties of linear filters: their linearity is, of course, important when dealing with additive mixtures of signals, but we also need to understand their effect on nonlinear processes. We describe how FIR and IIR filters differ radically in this respect, and discuss the ways in which each can be used in conjunction with various nonlinear transformations for signal separation.