App Note 01 [Link].

com 8-Oct-19

MEscope Application Note 01

The FFT, Leakage, and Windowing
INTRODUCTION
The steps in this Application Note can be duplicated using any Package that includes the VES-3600 Advanced Signal Pro-
cessing Option.
Click here to download the MEscope Project file for this App Note.
MEscope makes it very convenient to look at signals in either the time or the frequency domain. You can transform you data
between these two domains at will without losing any signal characteristics.
Transform | FFT transforms a Data Block of time domain waveforms into their equivalent frequency domain spectra
Transform | Inverse FFT transforms a Data Block of spectra into their equivalent time domain waveforms
Another command in MEscope lets you create Fourier spectra, Auto & Cross spectra, Power Spectral Densities (PSDs), or
Energy Spectral Densities (ESDs).
Transform | Spectra creates a new Data Block of spectra from time domain waveforms or other spectra
Additionally, MEscope has a command with which you can synthesize a broad variety of sine, random, chirp, or impact sig-
nals, plus Auto spectra.
File | New | Data Block synthesizes time waveforms or Auto spectra, and puts them into a new Data Block window
You will use these commands to review some basic properties of the Fast Fourier Transform (FFT) and learn how common
FFT window leakage errors can be minimized.
BASIC TIME & FREQUENCY RELATIONSHIPS
The FFT is an efficient algorithm for computing the Discrete Fourier Transform (DFT) of a digital time domain signal, or
time-history. Three basic formulas govern all FFT calculations.

Time Domain Windowing & Sampling.

Page 1 of 12
App Note 01 [Link] 8-Oct-19

Time Domain Equation

Digital signal processing with an FFT assumes that a digitized time domain signal is represented by N uniformly spaced
samples of data, which satisfy the equation,
T = Nt (1)
T = total time of the sampled signal (in seconds)
t = time step (or increment) between samples of data
N = number of samples, also called the Block Size
The schematic figure above depicts the digital sampling of a continuous (analog) time signal. This consists of two steps:
1. Multiply the signal by a rectangular observation window (also called a boxcar or uniform window) to limit its dura-
tion to a practical observation time. This is termed time truncation.
2. Measure the amplitude of the truncated signal at a series of equally spaced discrete times.
Note that the total observation time (T) of the sampled signal includes the time up to but not including sample N+1, which is
the first sample point of a subsequent sampling window, or block of data.
Frequency Domain Equation
Secondly, the FFT assumes that the DFT of the signal is represented by its mean value (DC) plus N/2 uniformly spaced
complex-valued samples (magnitudes & phases), which satisfy the equation:
Fmax = f (N/2) (2)
Fmax = maximum frequency of the signal (in Hz), also called the Nyquist frequency
f = frequency step (or increment) between samples (or Lines) of data
N/2 = number of frequency domain samples, always one half of the Block Size
The resulting spectrum has a total of (N/2 + 1) samples in it; a DC value plus N/2 frequencies. Furthermore, the DC value
and the value at the Nyquist frequency (Fmax) contain no phase information. The samples (or Lines) at DC and Fmax are real
valued. All other Lines are complex-valued.
Nyquist Sampling
Thirdly, the maximum frequency of the spectrum (Fmax,) is related to the time domain sampling rate (fs), specifically:
Fmax = fs /2 (3)
fs = 1/t = the sample rate (samples/second) at which the time domain data is sampled.
Equation (3) is a statement of Shannon’s sampling theorem, namely: “The maximum frequency of a spectrum is equal to
one half the sampling rate of its corresponding time domain signal.”
Solving the frequency domain equation for f, and substituting for 2Fmax gives a fourth equation:
f = 2Fmax /N = 1/ Nt = 1/T (4)
Equation (4) tells us that the frequency resolution (f) of a digital spectrum is equal to the reciprocal of the length of the ob-
servation time (T). In other words, “to obtain finer frequency resolution (smaller f), a signal must be sampled over a
longer time period (T), not at a higher sampling rate (fs).”
SYNTHESIS OF A PERIODIC SIGNAL
Let’s start by synthesizing a periodic signal and analyzing its characteristics. The Fourier series of a square-wave of ampli-
tude ± 1 and frequency (f0) is given by:
K

 (2k + 1) Sin2 (2k + 1) f t 

4 1
y (t ) = (5)

0
k =0

The first three terms of this series are:

Page 2 of 12
App Note 01 [Link] 8-Oct-19

y(t ) = Sin (2f 0 t ) + Sin (6f 0 t ) + Sin (10f 0 t )

1 1
3 5 (6)
These terms approximate a square wave of amplitude (± π/4 = 0.785) with a fundamental frequency (f0), let’s say of 5 Hz.
• Execute File | New | Data Block and click on OK in the dialog box that opens
The following dialog box will open.

Synthesis Dialog Entries for a 5 Hz Square-Wave.

• In the Data Block Parameters section, enter the following entries:
Block Size (N) = 100 Samples
Ending Value (Fmax) = 50 Hertz
Note that entry of any two of the Data Block Parameters determines the remaining four parameters.
The remaining four parameters are:
T = 1 Second
ΔT = 0.01 Seconds
N/2 = 50 Samples
Δf = 1.00 Hertz
• Verify the following default values:
Pre-Trigger Delay = 0 Samples
Number of Averages = 1

Page 3 of 12
App Note 01 [Link] 8-Oct-19

• Click on the Sinusoidal tab to display it, and enter the following parameters:
Number of Frequencies = 3
Number of M#s = 1
Frequency (Hz) values of 5, 15 & 25 Hz
Magnitude values of 1, 1/3 & 1/5
• Verify that all Phase and Damping values remain at the default 0 value.
• Press the OK button at the bottom of the dialog box
A new Data Block (BLK) Window will open, displaying the 5 Hz square-wave approximated by three sine waves. Note that
the amplitude (in the center of the ringing) is approximately ±0.795 g, as anticipated. Notice that exactly five cycles of this
periodic waveform are captured in the 1 second duration of the window.

Synthesized 5 Hz Square-Wave.
To verify the parameters of this synthesized time waveform:
• Execute File | Properties to open the Data Block Properties dialog box

Data Block Properties of Synthesized 5 Hz Square-Wave.

Page 4 of 12
App Note 01 [Link] 8-Oct-19

Verify that the properties agree with the N, T, ΔT and Starting Time that you entered to synthesize the square-wave.
• Press OK to close this dialog.
Now let’s look at the frequency spectrum of the square-wave.
• Execute Transform | FFT

Spectrum of Synthesized 5 Hz Square-Wave.

The frequency spectrum of the signal is shown above.
• Execute File | Properties again to verify the number of frequency Lines (N/2) as well as Δf and the Fmax.
Note that peaks clearly appear at 5, 15 and 25 Hz. Use the cursor to display the magnitude & frequency at each peak.
• Execute Display | Cursor | Line Cursor
• Execute Display | Cursor | Cursor Values
• Drag the cursor to each of the three peaks in the spectrum
Note that the spectrum contains the 1, 1/3 and 1/5 magnitudes that were used to synthesize the square-wave.
• Execute File | Save As
• Enter “Periodic Square Wave” and save the Data Block.
LEAKAGE, A FUNDAMENTAL FFT PROBLEM
In the preceding steps, we synthesized a square wave using three sinusoidal components at 5, 15 and 25 Hz. The square-wave
was synthesized over a time period (T) of 1 second. Each of the square-wave components completed exactly an integer num-
ber of cycles (5, 15 & 25 respectively) during the 1-second time period. As a result, each resulting peak in the spectrum fell
exactly on an integer multiple of Δf (1.00 Hz).
When T coincides exactly with N cycles of a periodic signal, the signal is said to be periodic in the sampling window. Mag-
nitudes and frequencies of the waveform components can be identified with high precision, as we have just seen.
SYNTHESIS OF A NON-PERIODIC SIGNAL
Let’s repeat the square-wave synthesis, but with a small change. We will change the fundamental frequency (f0) from 5 Hz to
4 2/3 Hz and leave all of the other parameters as they were before.
• Execute File | New | Data Block, and make the following entries;
• Change the frequencies to 4.6667, 14 & 23.333 Hz, which are 1, 3 & 5 times 42/3 Hz respectively.

Page 5 of 12
App Note 01 [Link] 8-Oct-19

Synthesis Dialog Entries for a 4 2/3 Hz Square-Wave.

Notice that the third harmonic (14 Hz) remains an exact integer multiple of f, and that it will complete exactly 14 cycles
within the 1 second sampling window. However, the first and fifth harmonics are non-periodic in the sampling window.

Synthesized 4 2/3 Hz Square-Wave.

The 4 2/3 Hz square-wave synthesis produces no surprises. The waveform exhibits the same shape and peak amplitude as the
5 Hz case. The difference is that 4 2/3 cycles are contained within the sampling window instead of 5 cycles. This seems like
an insignificant difference!

Page 6 of 12
App Note 01 [Link] 8-Oct-19

Spectrum of Synthesized 4 2/3 Hz Square-wave.

• Execute Transform | FFT
Clearly, the resulting frequency spectrum (shown above) is quite different from the spectrum of the periodic signal. While
three peaks may still be seen, the rest of the spectrum is not intuitively obvious. There is significant energy at all frequencies,
even though only three discrete and well-separated tones were used to create the signal.
Note further that the magnitudes of the 1X and 5X peaks are less than the correct values of 1 & 0.2. Only the 14 Hz periodic
3X peak has a magnitude, which is close to the correct value (0.333). This phenomenon is known as spectral leakage.
WHAT CAUSES LEAKAGE?
Whenever the FFT is applied to a signal that is not exactly periodic in its sampling window, leakage will occur in its spec-
trum. It is the result of multiplying two signals together in one domain, which is equivalent to convolving their Fourier trans-
forms in the other domain. Convolution is a process of shifting and adding together two signals, so that the resultant signal
is a smeared version of the expected result. (See reference [1] for details.)
In the case of a truncated sine wave, its correct spectrum was convolved with the spectrum of the rectangular sampling win-
dow. The resultant frequency spectrum was the convolution of the spectrum of the sine wave with the spectrum of the rec-
tangular window, which smeared the peaks of expected square-wave spectrum.
Another Interpretation
Another way to understand leakage is to look at what happens to a sampled signal. One of the fundamental rules of the FFT
is that “sampling of a signal in one domain causes repetition of it in the other domain”. This phenomenon is illustrated
below.
The actual signal is a continuous sine wave. The finite duration captured version is truncated in the sampling window. The
FFT computes the spectrum of the assumed signal, which when repeated outside of the sampling window, is no longer a con-
tinuous sine wave.

Page 7 of 12
App Note 01 [Link] 8-Oct-19

Repetition of a Non-Periodic Signal.

Instead of getting the FFT of the continuous sine wave (the expected result), we get the FFT of the assumed signal, which is a
smeared spectrum.
From this example it should be also clear that “any type of signal that is completely contained within the sampling window
(like a transient that damps out within the window), is also periodic in the window.”
REDUCING LEAKAGE
If a periodic signal is truncated by its sampling window, leakage in its spectrum cannot be eliminated. However, it can be
significantly reduced so that the resulting spectrum contains a more accurate representation of the signal. This is accom-
plished by changing the shape of the sampling window, which is multiplied by the time waveform.
MEscope has three built-in time domain windows for windowing data; Rectangular, Hanning, and Flat Top. So far, you
have used the Rectangular window, which is always used by default unless another window is specified.
Rectangular Window
The Rectangular window is the proper choice for analyzing transients that are completely contained within the duration T of
the sampling window.

Window shapes: Rectangular, Hanning and Flat Top.

The Rectangular window is also the proper choice when analyzing periodic signals using order-normalized sampling, a pro-
cess by which the sampling rate (fs) is changed in proportion to the operating speed of a rotating machine.
In virtually all other circumstances, the Rectangular window choice is inappropriate, and the resulting spectrum will contain
leakage.

Page 8 of 12
App Note 01 [Link] 8-Oct-19

Hanning Window
The Hanning window is the best window choice when analyzing broad-band random signals, non-periodic cyclic signals, or
a mixture of the two. It provides excellent suppression of signal truncation artifacts while retaining the ability to separate
closely spaced tones or detect a tone buried in a noisy background.
Flat Top Window
The Flat Top window is specifically designed to give improved amplitude precision when measuring cyclic or narrow-band
signals such as sine waves. It excels in this capacity but is not as selective as the Hanning window for discerning closely
spaced peaks in a spectrum.
Both the Hanning and Flat Top windows are designed to optimize the spectrum analysis of non-transient signals. Both win-
dows symmetrically taper the leading and trailing portions of the sampled time domain signal to essentially zero. This has the
effect of giving the time signal the appearance of being periodic in the time period T.
While the Rectangular window merely acts as a time gate (multiplying the signal by 1 or 0), the Hanning or Flat Top win-
dow completely changes the time waveform shape. Even though these windows distort the time waveform shape, their ef-
fect on the resulting spectrum is highly beneficial for reducing the effects of leakage.
FFT Digital Filters
Leakage and the FFT can be viewed from yet another perspective. The FFT numerically implements a bank of parallel digital
filters. Each filter is of the same constant bandwidth (nominally Δf), and the filter center-frequencies are spaced Δf apart. In
an ideal world, the transfer function of each filter would be a “brick wall” rectangle of width, Δf.
However, such ideal filters do not exist, either physically or mathematically. Instead, the transfer function of each FFT filter
has a complicated shape. The filter shapes for the Rectangular, Hanning and Flat Top windows are shown below.

Magnitude of FFT Filter Transfer Functions.

Note that the FFT filter for each of these windows is far broader than one Δf. Each one exhibits a maximum gain (sensitivity)
at the center frequency (fcenter). Adjacent filter center frequencies are spaced Δf apart, so it is clear that considerable overlap
exists in a bank of filters.
The Rectangular filter has the narrowest center lobe, (a desirable attribute) spanning 2Δf’s between adjacent zeros. Outside
of the center lobe, the Rectangular window exhibits zero gain at every multiple of Δf. It also has the highest side lobe am-
plitudes and the greatest curvature across the center lobe, compared to the other two window filters. These are detracting
characteristics.
This window works well when applied to signals that are periodic in the sampling window. In this situation, each periodic
signal component exists at exactly the center frequency of one of the window’s filters. That signal component is coincident
with the most sensitive frequency of the Rectangular filter. More importantly, the frequency of a periodic signal component
corresponds to a zero-gain point for all other filters.
Alternatively, a truncated or non-periodic waveform has non-zero spectral components at all frequencies in the spectrum,
resulting in leakage of the signal into all of the filter sideband frequencies. When a single frequency tone moves slightly by
(±Δf / 2 or less) off of the center frequency of a filter, it no longer coincides with the zeros of the adjacent filters.

Page 9 of 12
App Note 01 [Link] 8-Oct-19

The Hanning window improves upon the Rectangular window because the height of its side lobes is much less in amplitude.
This reduced side lobe gain minimizing the leakage of a signal into other frequencies. However, the Hanning window has a
broader center lobe with 4 Δf‘s between its nearest bounding zeros. Thus, the effective frequency resolution in a spectrum is
reduced when a Hanning window is used.
The Flat Top window has an extremely wide center lobe, spanning more than 8 Δf’s. However, as the name implies, the top
of the center lobe is extremely flat (within ± 0.1 %), providing an accurate amplitude even when a tone varies as much as ±Δf
/ 2 from the center fcenter of the digital filter. The side lobes of the Flat Top filter are well suppressed, but they are all of near-
ly equal amplitude.
Some of the characteristics of these three windows are listed below. The noise bandwidth, Δn, is a standard measure of a
window’s broadband random noise performance. It is the bandwidth of the broadband power passed by the filter when it is
subjected to white (or bandwidth) noise.
WINDOW COMPARISON
The three windows will be compared by applying them to the 4 2/3 Hz non-periodic square-wave. Start by transforming the
Fourier spectrum of the synthesized 4 2/3 Hz square-wave back into the time domain.
• Execute Transform | Inverse FFT
Now we will create a new Data Block containing three different spectra, all from the same synthesized 4 2/3 Hz square-wave,
but with each different time window applied to it.
• Execute Transform | Spectra in the BLK: Non-Periodic Square Wave window
• Select Fourier spectrum in the Transform | Spectra dialog box
• Select the Rectangular window in the dialog box that opens, and click on Calculate

When the calculation is complete,

• Click on New File in the Data Block selection dialog box
• Enter a file name “window comparison” in the dialog box that opens, and click on OK
The new BLK: window comparison now holds the first spectrum, that was calculated using a Rectangular window.
• Execute Transform | Spectra again in the BLK: Non-Periodic Square Wave window
• Select the Hanning window in the dialog box, and click on Calculate
• When the calculation is complete, select BLK: window comparison in the dialog box that opens, and press the Add
To button

Page 10 of 12
App Note 01 [Link] 8-Oct-19

This adds the spectrum that was created using the Hanning window as M#2 to BLK: window comparison.
• Execute Transform | Spectra again in the BLK: Non-Periodic Square Wave window.
• Select the Flat Top window in the [Link], and click on Calculate
• When the calculation is complete, select BLK: window comparison in the dialog box that opens, and press the Add
To button
This adds the spectrum that was created using the Flat Top window as M#3 to BLK: window comparison.

4 2/3 Hz square-wave analyzed using 3 different windows.

• Execute Format | Overlaid
This comparison plot clearly illustrates that either the Hanning or Flat Top window will eliminate most of the leakage,
making the three peaks clearly evident in the spectrum.
The Flat Top window provides the best magnitude estimates compared to the other two windows, as shown in the following
table.
The periodic 14 Hz component did not contribute leakage to the 4.6667 Hz or 23.333 Hz peaks, but the magnitude at 14 Hz
was affected by the leakage from the other two frequency components.

Page 11 of 12
App Note 01 [Link] 8-Oct-19

Peak Rectangular Hanning Flat Top

4.66 Hz -19.50 -6.95 0.00

14 Hz 7.01 -0.07 0.00
23.33 Hz -13.85 -6.95 0.00
Percent Amplitude Error using 3 different windows.
The MEscope time window functions comply with ISO standard 18431-2
REFERENCES
Richardson, M., Fundamentals of the Discrete Fourier Transform, Sound and Vibration Magazine, March 1978.
Richardson, M., Modal Analysis Using Digital Test Systems, Seminar on Understanding Digital Control and Analysis in Vi-
bration Test Systems, Shock and Vibration Information Center Publication, Naval Research Laboratory, Washington,
D.C., May 1975.

Page 12 of 12