Skip to main content
. Author manuscript; available in PMC: 2010 May 15.
Published in final edited form as: Neuroimage. 2010 Jan 28;51(1):123–134. doi: 10.1016/j.neuroimage.2010.01.069

Table 1.

Joint BSS performance comparison via simulations.

Source index S1–S2 corr A1–A2 corr Absolute correlation between sources inference and the ground truth under PSNR=10dB
jICA mCCA sCCA CCA+ICA 2ICA
1 0.987 0.988 0.967±0.004 0.892±0.084 0.980±0.000 0.990±0.000 0.990±0.003
2 0.88 0.781 0.976±0.005 0.941 ±0.005 0.871 ±0.000 0.995±0.000 0.980±0.000
3 0.76 0.678 0.435±0.041 0.330±0.077 0.630±0.007 0.820±0.001 0.817±0.001
4 0.451 0.590 0.490±0.020 0.538±0.024 0.728±0.013 0.924±0.001 0.892±0.001
5 0.383 0.217 0.820±0.001 0.578±0.035 0.577±0.003 0.916±0.001 0.461±0.003
6 0.004 0.881 0.700±0.034 0.514±0.055 0.821 ±0.001 0.840±0.001 0.278±0.001
Inter-symbol Inference A1 0.184±0.019 0.251±0.021 0.199±0.002 0.051±0.001 0.034±0.005
Inter-symbol Inference A2 0.166±0.014 0.248±0.007 0.315±0.002 0.081 ±0.001 0.067±0.001
X1=A1-S1, X2=A2-S2 statistical assumption for every method A1 and A2 should be identical “A1–A2 corr” should be distinct “S1–S2 corr” should be distinct enough “S1–S2 corr” can be either common or distinct No assumption to link two data sets.

Inline graphic BS S performance degrades due to unsatisfaction of the requirements

Inline graphic Sources are linked correctly and performance is the best among the five methods

Inline graphic Sources are mis-linked via cross-correlation between two data sets.