Graphics Programs Reference
In-Depth Information
Separated Signals − PCA
Separated Signals − ICA
4
4
2
2
0
0
−2
−2
−4
−4
0
1000
2000
3000
4000
0
1000
2000
3000
4000
a
b
2
4
2
0
0
−2
−2
−4
−4
0
1000
2000
3000
4000
0
1000
2000
3000
4000
c
d
4
2
1
2
0
0
−1
−2
−2
0
1000
2000
3000
4000
0
1000
2000
3000
4000
e
f
Fig. 9.6
Output of the principal component analysis (
a
,
c
,
e
) compared with the output of
the independent component analysis (
b
,
d
,
f
). The PCA has not reliably separated the mixed
signals, whereas the ICA found the source signals almost perfectly.
subplot(3,2,1), plot(sPCA(:,1))
ylabel('s_{PCA1}'), title('Separated signals - PCA')
subplot(3,2,3), plot(sPCA(:,2)), ylabel('s_{PCA2}')
subplot(3,2,5), plot(sPCA(:,3)), ylabel('s_{PCA3}')
The mixing matrix
A
can be found with
A_PCA = E * sqrt (D);
Next, we separate the signals into independent components. We will do
this by using a FastICA algorithm which is based on a fi xed-point iteration
scheme in order to fi nd the maximum of the non-gaussianity of the indepen-
dent components
W
T
x
. As the nonlinearity function we use a power of three
function as an example.
