Campbell Scientific ID-2000W Software - Manual (Page 48)

SECTION 4. FFT VIEW
4-2
ID-2000 calculates FFTs based upon 64, 128,
256, 512, 1024, or 4096 data points. You select
the number of points to use in the FFT
Preferences dialog box. The number of data
points is important because it directly affects
how many frequency ranges can be determined
for the signal.
Let’s take an example where we specify a 1024
point section of data for a parameter where all
the points were recorded at a constant scanrate
of 1000/second. An FFT calculation on these
points would result in the amplitude (Amplitude
Spectrum plot) or power (Power Spectrum plot)
for each of 512 frequencies (half the number of
data points) ranging from 0Hz to 500Hz (half the
scanrate). The number of frequencies
determined by an FFT is always half the
number of data points in the FFT. The
minimum frequency is always 0Hz and the
maximum is always half the scanrate.
Therefore, an FFT using a small number of data
points results in a coarse frequency resolution
where an FFT using a large number of data
points produces a fine frequency resolution.
Figure 4-2 contains Amplitude Spectrum plots
of the exact same data using both the minimum
(64) and maximum (4096) number of data
points for the FFT calculation.
Both plots have an X-axis scale of 0Hz to
500Hz (half our scanrate) because this scale is
only based upon the rate at which the data was
recorded - not the number of data points used
for the FFT. However, this scale is divided up
into frequency ranges equal to exactly one half
of the number of points used in the FFT
calculation.
For our 64 point FFT plot on the left we have
only 32 frequencies ranges to represent the full
range from 0Hz to 500Hz. This means that
each one corresponds to a frequency range of
approximately 500 / 32 = 15.6 Hz (half the
recording rate / half the number of points in the
FFT).
The plot on the right used 4096 points for the
FFT calculation so we can break our 0Hz to
500Hz up into many more frequency ranges
with each corresponding to 500 / 2048 =
0.24Hz. You can visibly tell the difference by
the width of the bars in the plots. The left plot
has fairly wide bars while the plot on the left has
very thin bars that look like single lines.
FIGURE 4-2. Effects of FFT sizes