Figure 6.
(a) The four panels show the data matrix after fitting a linear combination of the first N left-singular vectors and the sample-dependent difference signals to the measured data and then subtracting the best-fit SVD contribution ΔISVD. The inclusion of the reference signals leads to slightly less effective background subtraction, but the background-subtracted signal is now insensitive to the number N of SVD components used. (b) Summed residuals (see main text for details) in the regions preceding and following t0 as a function of number of SVD components included in the background subtraction. The inset shows how the difference in Akaike information criterion values, ΔAICci, indicates N = 4−5 as optimal for background subtraction. (Online version in colour.)