Page 312 -

Appendix B - CD Tools

CD tools include both didactical Microsoft Excel files and installable programs for
assisting in PR design.

B.l Adaptive Filtering

The ECG 5OHz.xZs file allows one to perform adaptive filtering of a 50Hz noise
signal added to an electrocardiographic signal (ECG), using an LMS linear
discriminant (two-input linear neural network). The user can change the learning
rate and other parameters of the filter in order to get insight into the learning
properties of the gradient descent method.
Variables:

T Time in seconds ECG Original ECG signal
sin 50 Hz sinusoidal noise nG Noise gain
t Target signal: ECG + noise eta Learning factor
n 1, n2 Discriminant inputs r Discriminant output
wl, w2 Weights e Error signal
phi Phase angle of discriminant inputs
g Amplitude factor of discriminant inputs

Author: JP Marques de S5, Engineering Faculty, Oporto University.

B.2 Density Estimation

The Parzen.xls file allows one to perform experiments of density estimation on
data with log norm distribution, using the Parzen window method. The user can see
the effect of changing the window width and the number of training samples. A
worksheet containing the values of feature ARM for the first two classes of the
Cork Stoppers dataset allows one to perform a Bayesian classification based on
Parzen window estimation of the distributions.

Author: JP Marques de SB, Engineering Faculty, Oporto University.