Common Spatial Pattern
Common Spatial Pattern Filter · Also known as: CSP, Spatial filtering, CSP decomposition
Common Spatial Pattern (CSP) is a spatial filtering technique that identifies electrode combinations that maximize the variance difference between two classes of EEG activity, typically used in brain-computer interfaces to enhance motor imagery discrimination. Introduced by Ramoser and colleagues in 2000, CSP has become a standard feature extraction method in BCI research.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use CSP when designing motor imagery BCI systems and you need robust, interpretable spatial features. It is most effective for two-class problems with clear frequency-band differences. For multi-class (3+ classes), variants like Multi-class CSP or CSP-CV exist. Assumptions include stationarity within trials, sufficient data to estimate covariance reliably (typically 50+ trials per class), and known class labels during training.
Strengths & limitations
- Enhances signal-to-noise ratio of discriminative brain activity compared to raw EEG
- Computationally efficient; based on eigendecomposition with closed-form solution
- Interpretable spatial filters reveal where in the brain class differences originate
- Consistently improves BCI classification accuracy across many studies
- Requires sufficient training data (50–100+ trials per class) to estimate covariance reliably
- Assumes stationarity within trials; performance degrades with session-to-session drift
- Designed for two-class problems; extensions to multi-class are less elegant and require careful tuning
- May overfit if data is limited; regularization of covariance estimates is often needed
Frequently asked
What is the relationship between CSP and principal component analysis (PCA)?
PCA finds directions of maximum variance without considering class labels; CSP finds directions of maximum variance ratio between classes. CSP is supervised, PCA is unsupervised.
How many CSP filters should I use?
Typically 2–4 filters (using the top m eigenvalues and their negatives) for two-class problems. Use cross-validation to select m; more is not always better due to overfitting.
Can CSP be applied to single-trial real-time BCI?
Yes. Once trained offline, CSP filters apply to each single trial, projecting it onto the spatial patterns. The variance of filtered signals is computed and fed to a classifier (LDA typically). Runtime is milliseconds.
Sources
- Ramoser, H., Mueller-Gerking, J., & Pfurtscheller, G. (2000). Optimal spatial filtering of single trial EEG during imagined hand movement. IEEE Transactions on Rehabilitation Engineering, 8(4), 441-446. DOI: 10.1109/86.895946 ↗
- Koles, Z. J., Lazar, M. S., & Zhou, S. Z. (1991). Spatial patterns underlying population differences in the background EEG. Brain Topography, 2(4), 275-284. DOI: 10.1007/BF01129656 ↗
How to cite this page
ScholarGate. (2026, June 3). Common Spatial Pattern Filter. ScholarGate. https://scholargate.app/en/biomechanics/common-spatial-pattern
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- BCI Motor ImageryBiomechanics↔ compare
- EMG EnvelopeBiomechanics↔ compare
- Markerless Motion CaptureBiomechanics↔ compare