Professional Documents
Culture Documents
Effects of Windowing On The Spectral Content of A Signal by P. Wickramarachi
Effects of Windowing On The Spectral Content of A Signal by P. Wickramarachi
V, dB
(b) (c)
sured signal. When choosing the appro- lobes of all other filters overlap f with
–60
priate window, one needs to be aware of nonzero gain and these filters respond in
its advantages and pitfalls in order to fit proportion to this gain. This perverse ef-
the measurement situation. The follow- fect is maximum when f = f 0 – ∆ f/2 [case –80
(a)
ing deals with some practical consider- 1(c)], since the frequency f coincides with
–100
ations on the effects of windowing. 1 the peak of each side lobe. 250 500 800
The Fourier series assumes periodicity If side lobes could be reduced in am- Frequency, Hz
of the signal in the time domain. An FFT plitude, this error would decrease as Figure 1. Spectrum of a sine wave with the
is actually a Fourier Series performed well. This is why people have used a Rectangular window: (a) f = f 0; (b) f = f 0 – ∆ f/8;
upon an interval, T span = n ∆ t where n is number of windows to weight the trun- (c) f = f 0 – ∆ f/2.
the number of samples observed and ∆ t cated signal such that the starting value
∆f
is the constant time between samples. and the ending value are zero. This pro- 0
13 dB
Since the sampled time signal may not duces a signal that appears periodic in 31 dB
20 dB/
exactly contain an integer number of pe- Tspan, meeting the basic assumption of the (a) decade
riods, this assumption may not be truly Fourier Series. Weighting avoids the
V, dB
satisfied. sharp discontinuities induced by the 60 dB/
(b)
In effect, the truncation of the original Rectangular window and yields reduced- decade
signal corresponds to its multiplication amplitude side lobes as desired. Figure
with a Rectangular window of length 2(b) shows the spectrum of the well
T span. The Fourier series then assumes known Hanning2 window. Observe that –80
fi−2 fi−1 fi fi+1 fi+2
that the signal is the succession of ver- the amplitude of the first side-lobe is re-
sions of this truncated signal in the time duced from –13.2 dB to –32.2 dB. More Figure 2. Comparison of spectra for the Rect-
domain leading to a spectrum with har- importantly, notice that the amplitudes of angular window (a) and the Hanning window
(b).
monic components at frequencies equal subsequent side-lobes fall off at 60 dB/
to multiples of ∆ f = 1/T Span . decade as opposed to 20 dB/decade for
0
Let us examine the situation with a sine the Rectangular window.
wave of frequency f0. In theory the cor- These improvements come at a cost. –20
responding spectrum is a peak at f 0 . The width of the primary lobe essentially
When a noninteger number of periods is doubles, eliminating the first set of zero- –40
V, dB
acquired, this results in signal leakage, amplitude points. The primary lobe of the –60 (c)
characterized by the smearing of the Rectangular window has a –3 dB band-
spectrum. Figure 1 illustrates this phe- width of 0.85 ∆ f. That of the Hanning –80
(b)
300
tor 3 system provides the desired transfer
200
function for this example. The damping
(a)
100 (b) coefficient is adjusted to obtain a lightly
0
damped system. Then the averaged trans-
fer function is measured by 5 impact
–100
tests, with excellent frequency resolution
–200
0 20 40 60 80
(large Tspan). This constitutes our baseline
Time, msec [Figure 5(a)]. To exaggerate the different
windowing effects in this example, the
Figure 6. Time responses: (a) with Rectangular
window; (b) with Exponential window. sampling parameters are modified such
that T span = 98.304 msec. The natural fre-
window is a far better choice when try- quency of the baseline is set to ω =
ing to find a tone masked by random 162.125 Hz in order to be close to a fre-
background. The Flat-Top window lacks quency line. Hence the leakage observed
selectivity for two reasons. First, its pri- in case 1(a) is avoided. Then the impact
mary lobe is over twice as wide as that of test is performed with the Rectangular
the Hanning window suppressing addi- window [Figure 5(b)] and the Exponen-
tional zeros. Second, there is no roll-off tial window chosen with an appropriate
of the side-lobes with frequency. Both of decay rate [Figure 5(c)]. Although both
these characteristics render the Flat-top windows give a good estimate of ω, the
window sensitive to broadband noise, Rectangular window achieves a better es-
compromising its ability to ‘find’ a timate of the amplitude (error of 1 dB as
masked tone. opposed to 6 dB with the Exponential
A window’s sensitivity to broadband window). The time responses are dis-
random noise is standardly characterized played in Figure 6. Table 1 summarizes
by its equivalent noise bandwidth N b . the different characteristics of the win-
This one-number description of a compli- dows mentioned here.
cated shape may be found as follows.
Consider an ideal (unity gain) rectangu- References
1. Discrete-Time Signal Pr ocessing , A. V.
lar ‘brickwall’ bandpass filter and an FFT Oppenheim and R. W. Schafer, 1989.
filter that results from applying a window 2. Named after the Austrian meteorologist
in time domain. When both filters are Julius Von Hann. It is in fact a corruption of
excited with the same random signal, N b the name Hann Window, perhaps through
association with the Hamming Window
would simply be the frequency width at
named after the mathematitian R. W. Ham-
which the ‘brickwall’ filter passes the ming.
same power as the FFT filter. N b is the 3. The Mentor is a self-contained integrated
‘Hz’ reported in a g2 /Hz Power Spectral signal analysis training system, G. F. Lang,
Data Physics Corporation, 2001.
Density measurement.
While a Flat-Top window has a noise The author can be contacted at: wickram
bandwidth of 3.43 ∆f, a Hanning window arachi@dataphysics.com.