An alternative to removing the simple common-mode is cleaning by way of principal component analysis (PCA). In this case, a new set of basis vectors (components) are calculated from the bolometer time-series such that they have 0 covariances. The amplitudes of these new correlated signals are then calculated, and the largest amplitude components can be removed. In this simplest case of white detectors noise $+$ a common atmospheric signal, this operation would be similar to using parameter "compreprocess" . However, PCA is capable of detecting multiple signals with different correlation patterns across the array.

It is $\ast$ESSENTIAL$\ast$ that the bolometer data first have their mean values removed (i.e., parameter "order" should be 0), as this assumption is used to speed up the calculation.

The main parameter for PCA cleaning is the threshold above which components will be removed from the bolometer time-series. For each component, the RMS amplitude across all bolometers is calculated – a single positive number related to the average strength of the component. An iterative sigma clipper is then used to flag components that are more than thresh$\ast$RMS away from the mean value.

This approach is quite arbitrary, but a value of about 4 seems reasonable for some test data. Be warned that this statistical black box will remove real correlated astronomical signals as well! As with common-mode removal, the actual impact on science will need to be calibrated with simulations.

If this parameter is set to 0 no pca cleaning will occur. [0]

SC2CLEAN, MAKEMAP, CALCQU