Classical methods of filtering time series use Fourier power spectral
analysis to separate signals from noise. Improved methods of signal se
paration can be developed by using projection techniques based on conc
epts of nonlinear dynamics. However, such methods are limited in their
ability to distinguish between dynamically independent signals. Here
we show how it is possible to combine Fourier projection with local no
nlinear prediction to provide a methodology which can, in principle, r
ecognise independent dynamical signals. We apply the methodology to a
variety of chaotic signals with superimposed sine waves, and show how
the sine wave frequency can be recognised dynamically (but not spectra
lly).