A NOVEL WINDOW FUNCTION YIELDING SUPPRESSED MAINLOBE WIDTH AND MINIMUM SIDELOBE PEAK

International Journal of Computer Science, Engineering and Information Technology (IJCSEIT), Vol.2, No.2, April 2012
DOI : 10.5121/ijcseit.2012.2209 91
A NOVEL WINDOW FUNCTION YIELDING
SUPPRESSED MAINLOBE WIDTH AND MINIMUM
SIDELOBE PEAK
Md Abdus Samad
Department of Computer and Communication Engineering
International Islamic University Chittagong, Chittagong-4203, Bangladesh
asamadece@gmail.com
ABSTRACT
In many applications like FIR filters, FFT, signal processing and measurements, we are required (~45 dB)
or less side lobes amplitudes. However, the problem is usual window based FIR filter design lies in its side
lobes amplitudes that are higher than the requirement of application. We propose a window function,
which has better performance like narrower main lobe width, minimum side lobe peak compared to the
several commonly used windows. The proposed window has slightly larger main lobe width of the
commonly used Hamming window, while featuring 6.2~22.62 dB smaller side lobe peak. The proposed
window maintains its maximum side lobe peak about -58.4~-52.6 dB compared to -35.8~-38.8 dB of
Hamming window for M=10~14, while offering roughly equal main lobe width. Our simulated results also
show significant performance upgrading of the proposed window compared to the Kaiser, Gaussian, and
Lanczos windows. The proposed window also shows better performance than Dolph-Chebyshev window.
Finally, the example of designed low pass FIR filter confirms the efficiency of the proposed window.
KEYWORDS
FIR filter, Hamming, Kaiser, Gaussian window, Dolph-Chebyshev.
1. INTRODUCTION
The ideal approach to the design of discrete-time infinite impulse responses (IIR) filters involves
the transformation of a continuous-time filter into a discrete-time filters meeting some prescribed
specifications. This is partly because continuous –time filter design was highly advanced art
before discrete-time filters were of interest [1]. In [2] different windows has been given with their
classification such as fixed, adjustable window functions, weight windows based on Atomic
functions, polynomial windows and, Z-window functions. In [3], three different window
functions Han window, Hamming window and Blackman window in a general format according
to evolutionary algorithm has proposed. However, it does not have any close loop formula. For
window method, frequency response of a filter obtained by a periodic continuous convolution of
the ideal filter in frequency domain with the Fourier transform of the window. The most
straightforward approach to obtain a causal finite impulse response (FIR) is to truncate the ideal
response. If hd[n] is the impulse response of the desired (ideal) IIR system, the simplest way to
obtain a casual FIR filter is to define a new system with impulse response h[n] of length M+1 as:
][][][ nwnhnh d= (1)

92


 ≤≤
=
Otherwise
Mnnf
nw
0
0)(
][
(2)
Where f(n) is a function of n and for different function, w[n] have different characteristics; some
of them are: Bartlett, Hanning, Hamming, Blackman, Lanczos, Kaiser, Gaussian, and Dolph-
Chebyshev windows [4]-[9]. In other words, we can represent h[n] as the product of the desired
response hd[n] and a finite-duration “window”, w[n]. Therefore, the Fourier transform (FT) of
h[n], )( ωj
eH is the periodic convolution of the desired frequency response, with FT of the
window, )( ωj
eW . As a result, the FT of h[n] will be smeared version of the FT of hd[n]. In the
application, it is desired for a window function to have characteristics of smaller ripple ratio and
narrower main lobe width. However, these two requirements are contradictory [1]. For the equal
length of M+1, Hamming window offers the smallest peak of side lobe as well as main lobe width
compared to Bartlett and Hanning window. The Blackman window has wider main lobe width but
smaller side lobe peak compared to the Hamming window. Lanczos window [10], shows different
characteristics in the main lobe depending by a positive integer. The Kaiser and Gaussian
windows are tunable functions, and there is a trade-off between side lobe peaks and main lobe
widths, and can be customized. The Kaiser window has the disadvantage of higher computational
complexity calculating the window coefficients. Dolph-Chebyshev window has all side lobes are
equal and the main lobe width is the minimum for a specific ripple ratio but it has high cost of
computation. There has been great interest into the design of new windows to meet the desired
specification for different applications [11]. In this paper, we present a proposed window function
which has at least 6.2~22.62 dB less side lobe peak compared to the Hamming window, while
offering smaller or equal or slightly larger main lobe width. We also show that the proposed
window is better than the other windows such as Hanning, Bartlett, and Gaussian, Kaiser,
Lanczos, Dolph-Chebyshev and recent proposed window in [9], [15]. We also design low pass
FIR filter with the proposed window to evaluate its efficiency.
2. PROPOSED WINDOW FUNCTION
We have derive the proposed window by comparing is followed by comparing )(cos nL
with
(n)L
sinc (where L is integer) at least within a particular range. By comparing mathematical
functions, new window was derived and proposed in [11], [12].
(a)

93
(b)
Figure 1. Shape of the proposed window for different lengths. (a) M=14, (b) M=200
(a) (n)(n) 2sincand2cos vs n
(b) (n)(n) 2.5sincand5.2cos vs n
(c) (n)(n) 5.3sincand3.5cos vs n
Figure 2. (a), (b), and (c) comparison of c(n)Lsin and (n)Lcos vs ).( Nn ∈

94
The same procedure has been followed here to derive a new window function. The optimum
value of window function for FIR filter design was calculated in [9] and a new window was
proposed based on their findings in terms of sinc function. The optimized values of windows
shows abrupt behaviour for some of these components when M is large (Figure 1.b.) around n=0
and M and for lower M the optimized components do not shown (Figure1.a) this abrupt behavior
[9]. Figure 2 shows the similarity between c(n)Lsin and (n)Lcos . Therefore, based on this
similarity, we have examined the Lanczos window function ( c(n)Lsin type) by (n)Lcos
function. In addition, an important behavior from these Figure’s is that as the values of L
increases the two functions presents become more similar property. Based on this of similarity a
new window proposed in terms (n)L
cos where the optimum values of window function has
generated according to [9]. Finally, suitable formula has derived. It was observed that, the
following window function fit with the optimized components for M<19.



 ≤≤−
=
otherwise
MnMn
nw
0
0)1/2(cos
][
3.5 (3)
But if M is large ( 20≥M ) then,









<<−
=
++
−+−
=
otherwise
MnMn
Mn
M
MEME
nw
0
0)1/2(cos
,0
036034.0007339.0
68899.196051.4
][ 3.5
23 (4)
For n=0, M a polynomial was proposed to fit the nonlinearity as shown in Figure 1.b. The new
window yields narrower main lobe width and minimum side lobe peak comparative to other
window.
The new window has the property that:


 <<−
−
otherwise
MnnMw
nw
0
0][
][
(5)
i.e. it is symmetric about the point M/2 and consequently has a generalized linear phase. Fig’s. 1.
shows the shape of the proposed window for different values of M. For 34≤M , w[0]=w[M] are
smaller than w[1]=w[M-1], respectively and the window is like a bell-shaped function. With
34≥M , w[0] and w[M] are larger than w[1] and w[M-1], respectively.
3. COMPARISON WITH OTHER WINDOWS
In this section, we compare the performance of the proposed window with several commonly
used windows with MATLAB [13], [14].
3.1. Hamming Window
It has the shape of:


 <<=−
=
otherwise
MnnMn
wH
0
0)/2cos(46.054.0 π (6)

95
Figure 3.a shows that the proposed window offers -13.76751 dB peak of side lobe with an
increased main lobe width ( ~2 .005π× ) with M=14. With M=50 (Figure 3.b) the proposed
window has some larger main lobe width but side is much smaller. Figure 6.c demonstrates that
the proposed and Hamming window has approximately equal main lobe width but the proposed
window offers -48.811dB-(-42.357dB) =-6.454dB less side lobe peak. The reduction in side lobe
peak is 23.08151dB with M=10 (-58.96933dB compared to -35.88782 dB). It shows that
for 20010 ≤≤ n , the side lobe peak of the proposed window compared to that of the Hamming
window is -23.081512dB~6.2dB smaller. Therefore, the proposed window offers slight larger
main lobe to that of the Hamming window while offering much less side lobe peak. It also reveals
that, in the case of side lobe peak, the proposed window is also better than Bartlett and Hanning
windows.
Table 1: Frequency response domain comparison of the proposed and Hamming window
Proposed Window Hamming Window
Main lobe
width(-3dB)
Normalized Side
lobe peak (dB)
Main lobe width(-3dB) Normalized Side
lobe peak (dB)
M=10 2×0.28906π -58.44563 2×0.27344π -35.82400
M=14 2×0.20313π -52.62333 2×0.1875π -38.85582
M=50 2×0.050781π -48.46017 2×0.05078π -42.47663
M=50 2×0.050781π -48.46017 2×0.050781π (M=47) -42.23333 (M=47)
(a) Proposed M=14, Hamming M=14
(b) Proposed M=50, Hamming M=50

96
(c) Proposed M=50, Hamming M=47
Figure 3. Fourier transforms of the proposed and Hamming windows for different lengths.
3.2. Kaiser Window
The Kaiser window has the following shape:





<<
−−
=
otherwise
Mn
I
MMnI
wK
0
0
)(
]])2//()2/[(1([
0
5.22
0
β
β (7)
where β is the tuning parameter of the window to it shows a trade-off between the desired “side
lobe peak- main lobe width,” and I0(.) is the zero order modified Bessel function of the first kind.
From simulated result, it is observed that for M=50 and β=6.55, the two windows have same side
lobe peak (~ -48.1 dB) while the proposed window gives less main lobe width.
Proposed M=50, Kaiser (β=6) M=50
Figure 4. Fourier transforms of proposed window and Kaiser window with M=50.

97
(a) Proposed M=50, Gaussian (σ=0.373) M=50
(b) Proposed M=50, Gaussian (σ=0.373) M=58
Figure 5. (a), (b) Fourier transforms of the proposed window and Gaussian (σ=0.373) windows
with M=50.
If we want the Kaiser window to have the same main lobe of the proposed window, then its
length should be M+1=53. Therefore, the proposed window offers the desired specifications with
lower length. By decreasing β to 6 it also shows less side lobe peak -48.50949dB compared to -
44.22365dB (-4.28584dB less), while maintaining the same main lobe width and M as shown in
Figure 4.
3.3. Gaussian Window
The Gaussian window is of the form




≤≤=
−
−
otherwise
Mnew
M
Mn
G
0
0
2
)
2/
2/
(
2
1
σ
(8)
where σ<0.5 is the tuning parameter of the window to have the desired “main lobe width – side
lobe peak” trade-off.
By setting σ=0.373 for this window, Figure 5.(a)-(b) depicts that the side lobe peak of the two
windows is -48.85843 dB (M=50), while the main lobe width of the proposed window is much
less than that of the Gaussian window. Our analysis show that, for the approximately equal main
lobe width, the Gaussian window need extra 8 point i.e. M=58.

98
3.4. Dolph-Chebyshev Window
This window can be expressed as a cosine series [5]:
]
1
2
cos
2/
1
)
1
cos0()0([
1
1
][
+
∑
= +
+
+
=
M
iM
i M
i
xMTxMT
M
nw
ππ (9)
where ( )M
T x is the Chebyshev polynomial of degree M and x0 is a function of side lobe peak and
M. Dolph-Chebyshev window has high cost of computation, but it’s important property is that all
side lobes are equal and the main lobe width is minimum that can be achieved for a given ripple
ratio. Figure 6, we observe that, the proposed window has a little better performance than Dolph-
Chebyshev window (0.5 dB higher side lobe peak), but with a greater main lobe width
of π01.02~ × . However, note that the proposed window coefficients can be computed easier than
Dolph-Chebyshev window.
Figure 6. Fourier transforms of proposed window and Dolph-Chebyshev window with M=50.
3.5. Window proposed in [9]
The proposed window [9] is of the form:






<<
−
=+++
=
−
otherwise
Mn
M
Mn
c
MnMM
nw
0
0)
654.0
2/
(sin
,0)502(001.002.0
][ 5.2
1 (10)
Proposed M=14, window proposed in [9] M=14
Figure 7. Fourier transforms of the proposed window (M=14) and window proposed in [9].
With M=14 the proposed window offers about -2.7 dB peak side lobe (Figure 7) reduction than
the proposed window in [9].

99
3.6. Lanczos Window
The Lanczos window has the form of:



 ≤≤−
=
otherwise
MnMnc
nw
L
L
0
0)1/2(sin
][
(11)
where L is a positive integer number. With M=50, Figs 8.a and 8.b compare the proposed and
Lanczos windows for L=1 and 2, respectively.
(a) Proposed M=50, Lanczos (L=1) M=50
(b) Proposed M=50, Lanczos (L=2) M=50
Figure 8. Fourier transforms of proposed window and Lanczos window with M=50.
Figure 8.a shows that the side lobe peak of the proposed Lanczos window is -21.1 dB less than
that of the Lanczos window, while its main lobe is just a little wider 2 0.03π× ; therefore for L=2,
the proposed window has both less main lobe width and side lobe peaks. Figure 8.b,
demonstrates that main lobe width increased with 2 0.004π× in case of Lanczos window compared
to proposed window and increased side lobe peak by -7.9 dB.
3.7. Window proposed in [15]
The form of the proposed window is:

100
MnMnMnnw ≤≤−−= 0)/6cos(003.0)/2cos(461.0536.0][ ππ (12)
For M=50 the Fourier transforms of the proposed window shows (Figure 9) about equal Mainlobe
width and but reduced Sidelobe peak than the window by the equation (12).
Proposed M=50, Window proposed in [15] M=50
Figure 9. Fourier transforms of the proposed window and window proposed in [15].
4. PERFORMANCE ANALYSIS
To study the efficiency of the proposed window we have compared the results by observing the
Fourier Transform of a low pass FIR filter designed by truncating of an ideal IIR low pass filter.
Having a cut off frequency of ωC, the impulse response of an ideal low pass filter is:
nnnh idealLPF ππ /)sin(][, = (13)
By windowing this IIR filter with the windows discussed in this paper, different FIR filters can be
obtained. With cut off frequency ωC=0.2π, Figs. 10.a-10.g show the frequency response of the
FIR filters designed by applying different windows of length M+1=51. Figure 10.a demonstrates
that the filter achieved by the proposed window has 17.44 dB less side lobe peak than the
Hamming window (M=50). This value is about 2.2 dB for Kaiser window function. This side
lobe reduction is almost zero for Gaussian window but it shows larger main lobe width as shown
in Figure 10.c. Figure 10.d-10.e demonstrate that the side lobe of the proposed window is smaller
but main lobe is larger than Dolph-Chebyshev and window proposed in [9]. However, in Figure
10.g the proposed window length is 50 but the window proposed in [15] is for M=47 and the
frequency shows that the Sidelobe of the proposed window is more attenuated than the window
proposed in [15].
5. CONCLUSION
The proposed window is symmetric and shows better equiripple property. Performance
comparison of the proposed window compared to that of the Hamming and Kaiser window shows
that the proposed window offers less side lobe with the same main lobe width. This value is
almost zero for Gaussian window but the proposed window offers suppressed main lobe width
than Gaussian window. It is obvious that the proposed window also outperforms the Hanning and
Bartlett windows because Hamming window shows better side lobe reduction than these windows
for the same main lobe width.

101
Table 2: Side lobe peak attenuations (dB) of the FIR filters πω 2.0=c obtained by windowing of
an ideal low pass filter with different window lengths.
M=50 M=70 M=100 M=200
Hamming -51.37 -52.09 -52.69 -53.55
Kaiser (β=6) -66.61 -61.95 -62.22 -62.75
Gaussian -66.70 -66.09 -65.63 -65.36
Dolf-Chebyshev -61.52 -61.39 -61.59 -63.88
window proposed in [9] -60.84 -61.37 -62.26 -62.11
window proposed in [15] -58.70 -59.10 -60.80 -60.68
Proposed -68.81 -67.97 -67.28 -66.56
(a) Proposed M=50, Hamming M=50
(b) Proposed M=50, Kaiser (β=6) M=50
(c) Proposed M=50, Gaussian (σ=0.373) M=50

102
(d) Proposed M=50, Gaussian (attenuation -48dB) M=50
(e) Proposed M=50, window proposed in [9] M=50
(f) Proposed M=50, Lanczos (L=2) M=50
(g) Proposed M=50, window proposed in [15] M=47
Figure 10. Performance of low pass FIR filter, M=50, obtained by windowing of an IIR filter with
different windows with cut of frequency ωC =0.2π.

103
The designed low pass FIR filter using the proposed window achieves less ripple ratio compared
to the above-mentioned window filters. Finally, for the same specification the proposed window
shows more side lobe reduction with slight increased main lobe width in comparison with
window proposed in [9] and with M<19 it shows better side lobe reduction and main lobe width
comparative to other these windows for M>20 .
REFERENCES
[1] A. Oppenheim, R. Schafer, and J. Buck, Discrete-Time Signal Processing, fifth printing , Prentice-
Hall, 1998 pp 444.
[2] Wang Yunlong, “Sinc Sum Function and Its Application on FIR Filter Design”, Acta Applicandae
Mathematicae Volume 110, Number 3, 1037-1056.
[3] Kangshun Li and Yuan Liu “The FIR Window Function Design based on Evolutionary Algorithm”
2011 International Conference on Mechatronic Science, Electric Engineering and Computer August
19-22, 2011, Jilin, China.
[4] A. Antoniou, Digital Signal processing: Signal, systems and filters, McGraw-Hill, 2005.
[5] J. Proakis and D. G. Manolakis, Digital Signal Processing fourth edition, Prentice-Hall, 2007.
[6] A. Antoniou, Digital Filters, McGraw-Hill Inc, N.Y., 1993.
[7] A. G. Deczky,“Unispherical windows ”, in Proceedings IEEE ISCS, vol. II, pp. 85-89, 2001.
[8] M. Jascula, “New windows family based on modified Legendre polynomials”, in Proceedings IEEE
IMTC, pp. 553-556, 2002.
[9] Mahrokh G. Shayesteh and Mahdi Mottaghi-Kashtiban, “FIR filter design using a new window
function”, The 16th International Conference on Digital Signal Processing (DSP 2009)
[10] C. M. Zierhofer,“Data window with tunable side lobe ripple decay”, IEEE Signal Processing Letters,
vol. 14, no. 11, Nov. 2007.
[11] Kemal Avci and Arif Nacaroğlu, “Cosine Hyperbolic Window Family with its Application to FIR
Filter Design”, Information and Communication Technologies: From Theory to Applications, 2008.
ICTTA 2008. 3rd International Conference.
[12] Kemal Avci and Arif Nacaroðlu, “Modification of Cosh Window Family” Information and
Communication Technologies: From Theory to Applications, 2008. ICTTA 2008. 3rd International
Conference.
[13] Vinay K. Ingle and John G. Proakis, Digital Signal Processing Using MATLAB V4 -1997.
[14] CHEN Min-ge “Design of Equiripple Linear-Phase FIR Filters Using MATLAB” Control,
Automation and Systems Engineering (CASE), 2011 International Conference.
[15] Mahdi Mottaghi-Kashtiban and Mahrokh G. Shayesteh, “A New Window Function for Signal
Spectrum Analysis and FIR Filter Design” Electrical Engineering (ICEE), 2010 18th Iranian
Conference.
Author
Md Abdus Samad received his B.Sc. degree in Electronics and Communication
Engineering from Khulna University, Khulna, Bangladesh, in 2007. He is
working as lecturer of the Faculty of Science and Engineering, International
Islamic University Chittagong, Bangladesh since August 10, 2008. His research
interest includes digital signal processing, wireless networks, optimization of
wireless networks.

A NOVEL WINDOW FUNCTION YIELDING SUPPRESSED MAINLOBE WIDTH AND MINIMUM SIDELOBE PEAK

More Related Content

What's hot (20)

Similar to A NOVEL WINDOW FUNCTION YIELDING SUPPRESSED MAINLOBE WIDTH AND MINIMUM SIDELOBE PEAK (20)

More from IJCSEIT Journal (20)

Recently uploaded (20)

A NOVEL WINDOW FUNCTION YIELDING SUPPRESSED MAINLOBE WIDTH AND MINIMUM SIDELOBE PEAK