The Karhunen-Loeve expansion is applied to scalar signals and the effect of
window length (t(w)), time lag (tau) and embedding dimension (d) is analys
ed for periodic signals and for signals modeled by the Lorenz equations. Fo
r tau not equal k/2 f(i) (f(i) are characteristic frequencies of the signal
, ii is positive integer), we obtain 2m modes from an m-periodic signal. Fo
r a large set of parameters a finite number of modes was not obtained from
the Lorenz system. It is further shown that, on the time scale of a minute,
the peripheral blood flow signal contains oscillatory modes that occur in
pairs thereby confirming that the blood flow through the cardiovascular sys
tem is oscillatory. Some of the difficulties of applying Karhunen-Loeve exp
ansion to scalar signals are pointed out. (C) 2000 Elsevier Science B.V. Al
l rights reserved.