Data Acquisition Process - Signal Processing of Random Physiological Signals - page 34

Considerations that must be taken into account when digitizing an analog signal will be
covered in two chapters: one on Sampling Theory and the second on Analog-to-Digital
Conversion.
Prepossessing may include formatting of the digital data (Binary, ASCII, etc.,
format), filtering to clean up the digital signal, removing any DC offset, obtaining the
derivative of the signal, etc. But before getting into digital conversion, let us continue
with the data-acquisition process by examining the fourth step of “Random Data Qual-
ification.”

5.4 RANDOM DATA QUALIFICATION
Qualification of data is an important and essential part in the statistical analysis of random
data. Quantification involves three distinct processes as shown in Fig. 5.4:

1. stationarity,

2. periodicity, and

3. normality.

It is well known in statistics that researchers should not use “parametric statistical
methods” unless the data are shown to be “Gaussian Distributed” and “Independent.” If
the data are not Gaussian Distributed and Independent, parametric statistics cannot be
used and the researcher must then revert to “nonparametric statistical methods.”
Likewise, necessary conditions for Power Spectral Estimation require that the
random data be tested for “Stationarity” and “Independence,” which may be tested para-
metrically by showing the mean and autocorrelation of the random data to be time
invariant or by using the nonparametric “RUNS” Test.
After showing the data to be stationary, researchers may examine the data
for “Periodicities” (cyclic variations with time). Periodicity may be analyzed via the

Stationarity Periodicity Normality

FIGURE 5.4 : Data qualification