Biomedical Engineering Reference
In-Depth Information
generating fifty interference sources in the source space. The interference sources
had Gaussian random time courses, and the interference magnetic recordings were
generated by projecting these time courses onto the sensor space using the sensor
lead field at the interference source locations. The simulated MEG recordings were
generated by adding the interference (and a small amount of sensor noise) onto the
same simulated signal-only sensor data in the previous denoting experiment.
In the PFA application, the first half period was used as the control period and the
mixing matrix of the interference
B
and the diagonal noise precision matrix
were
estimated using this control period. We then estimated the signal mixing matrix
A
and the signal factor activity
u
k
using the second half period. The signal component
was estimated using
ʛ
y
k
in Eq. (
5.132
).
Results are shown in Fig.
5.4
. In this figure, the top panel shows the control period
(the first half period) of the original interference-added sensor data, and the middle
panel shows the target period (the second half period) of the original interference-
added sensor data. The bottom panel shows the interference-removed sensor data,
y
k
.
Here, the number of factors for the interference was set to 100, and that for the signal
of interest was set to 10. Note that the true number of independent interference activ-
ities is 50, and that for the signal activity is 3. The results in Fig.
5.4
demonstrate that
0.2
0
0.2
0.2
0
0.2
0.1
0
0.1
0
200
400
600
800
1000
time points
Fig. 5.4
Results of interference-removal experiments using the PFA algorithm. The
top panel
shows
the control period (the first 1,200 time points) of the interference-added simulated sensor data. The
middle panel
shows the target period (the second 1,200 time points) of the interference-added
simulated sensor data. The
bottom panel
shows the results of interference removal experiments.
Here, results of
y
k
in Eq. (
5.132
) are plotted. The selected three sensor time courses (channel #1,
channel #101, and channel #201) are shown in these three panels
