Signal-space projection (SSP)

A projection-based cleaning method most used in MEG. Empty-room or artifact-epoch data (for example blinks or heartbeats) are used to estimate the spatial pattern or patterns — a few PCA directions — that the artifact occupies in sensor space; the data are then projected onto the orthogonal complement of that subspace, removing anything living along the artifact directions. Because the projector is a fixed linear operator applied identically to all data, it is simple, causal and easy to account for in subsequent source modeling.

The cost is the flip side of its simplicity: projecting out an artifact subspace also removes any brain signal with a component along those directions and reduces the rank (degrees of freedom) of the data, which must be respected when computing the forward and inverse solution. SSP suits low-dimensional, spatially stable artifacts (external interference, cardiac, blinks); ICA is preferred when artifact topographies are more complex or must be identified data-drivenly.