Neural networks - Signals, Processes, and Systems - page 443

Sawtooth time domain FFT f-parame ter Parameter Li st

Illustration 286: Select and configure frequency parameters

This is a simple test system to learn more about the frequency parameter module. At the input, a periodic
sawtooth oscillation of 52 Hz is used. Certain frequency ranges have now to be selected in the menu.
Starting at 50 Hz, frequency intervals of constant bandwidth (50 Hz) are configured without gaps,
resulting in 18 frequency bands at 1000 Hz limiting frequency. Each frequency band therefore contains a
certain part of the signal. From this frequency band- amplitude- spectrum, an output option is now
selected, e.g. “integral” or “maximum value”.

Data reduction: In the above case, practically the entire frequency information is contained in the 18
discrete values, because with an interval width of 50 Hz only one frequency lies in each interval. The
normal spectrum would have 1024 discrete values. However, the frequency parameter module or the
integral for each of the 18 defined frequency ranges only provides 18 discrete values.

The frequency parameter module

To use this module, the signal first has to be transformed into the frequency range by a
FFT. Generally, when choosing frequency intervals, there is a distinction between
variable and constant interval width. With a variable interval width, the intervals broaden
from the bottom to the limiting frequency, their scaling takes place by “octave”, “quint”,
“quart” etc. The frequency parameter module provides one discrete value for each
interval.

Being virtually indispensable for the realization of neural networks, this module can be
used for a variety of other applications that include the selection of certain frequency
bands.

With the new “collect” module, parameters of time range and frequency range can be
summarized and displayed consecutively. Thus, neural networks can simultaneously use
parameters of the time- and the frequency range.