User`s guide

Subtracting Trends from Signals (Detrend ing)
Forlinearsystemidentification, detrending steady-state data is u seful
because arbitrary differences between the input and output signal levels
cannot be explained by a linear model.
For nonlinear black-box system identification, detrending data might be
helpful when si g nals v ary arou nd a larg e signa l level, you m ight im prove
computational accuracy by first removing the means.
When to Subtract the Mean Values
You can subtract mean values from your data when you have steady-state
(not transient) data. If you have steady-state data, it is usually sufficient
to iden tify line ar models from signals measured relative to an equil ibrium.
Thus, you can avoid modeling the absolute levels in physical units.
Tip When you know the mean levels that correspond to the actual physical
equilibrium, remove the equilibrium values instead o f the mean value of
the s ignals for best results.
When to Subtract Linear Trends
When the mean levels drift during the exp eriment, you can eliminate this drift
by removing a linear trend or several piece-wise linear trends from the signals.
Signal drift is considered a low-fre qu ency disturbance. If you know the drift
rate, you can also build a custom high-pass filter and apply it as described in
“Filtering D ata” on page 1-107.
When Not to Detrend Data
Do not detrend your data when the physical levels are built into the underlying
model or when input integrators in the system require absolute signal levels.
In the cas e of estimating nonline ar O DE pa rameters (nonlinea r grey -box
models), do not detrend the data to make sure that the models represent the
actual physical levels.
1-95