Digital Signal Processing Reference
In-Depth Information
•
Initial population: 10 Individuals
•
Number of clones (
N
c
): 7 Clones
•
β
:60
•
Suppression threshold: 2
On the other hand, for the particle swarm algorithm, the parameters used in
the experiment are
•
Number of particles: 60
•
AC
1
=
AC
2
= 2.05
•
V
max
=
0.1
•
V
min
=−
0.1
In Figure 8.7a, the joint distribution of the mixture signals is shown. For
this situation, we considered 2000 samples of the mixtures in the training stage.
Figure 8.7b depicts the joint distribution of the recovered signals using the artificial
immune network.
We can observe that a residual nonlinear distortion remains, as it is impossible
to invert the hyperbolic tangent using the chosen polynomial. Nevertheless, the
obtained distribution is approximately uniform, indicating that the separation task
was accomplished. Similar results are obtained with the particle swarm.
In Table 8.4, we depict the residual MSE between the estimated signal and
the corresponding source. It can be noted that both algorithms were able to build
the separating system quite well, yielding small residual errors.
4
1
3
2
0.5
1
0
0
−1
−2
−0.5
−3
−4
−1
−2
−1
0
1
2
−2
−1
0
1
2
(a)
(b)
FIGURE 8.7
Distributions of the (a) mixtures and of the (b) recovered sources.
TABLE 8.4
Residual MSE of the Estimated Sources Using an Artificial
Immune Network and a Particle Swarm Algorithm
Algorithm
MSE—Source 1
MSE—Source 2
10
−
2
10
−
2
×
×
Artificial immune network
0.11
0.65
10
−
2
10
−
2
Particle swarm
0.73
×
0.53
×
