Frequency Transformation in Digital Domain

Phuoc S. Nguyen
Melbourne, Victoria, Australia
Email: nguyensiphuoc@frequencytransform.com

Abstract This paper introduces a new algorithm to Bilinear z-transformation is a method for transforming
transform a digital low-pass filter into a digital low-pass, an analog filter to an equivalent digital filter [1]. It is
high-pass, band-pass, band-stop and narrow-band filter in defined by (1), where T is a sampling time period, z=esT
the digital domain. The technique of this method is based on
bilinear and inverse bilinear z-transformation with pre- and s=j. Let A be an angular frequency in the s-domain
warping frequency and the support of Pascals triangle. and D be an angular frequency in the digital domain:
From this, a matrix equation is derived, called Digital low- 2 1 z 1
pass to digital filter Pascal matrix equation. This equation s (1)
will make it easier for hand calculations and programming
T 1 z 1
when transformations are made from a digital low-pass Equation (1) can be rewritten as:
filter to another digital filter in the digital domain. 2 1 e jDT 2 T T
j A tanh( j D ) j tan( D )
Index Terms frequency transformation, bilinear z-
T 1 e jDT T 2 2
transformation, digital filters, analog filters, pre-warping 2 T
frequency, Pascals triangle. D tan 1 ( A ) (2)
T 2

I. INTRODUCTION From (2), the relationship between analog frequency

A and digital frequency D is a non-linear function
One method for designing a digital filter is to transform caused by the tan function. Pre-warping frequency can be
a given analog low-pass prototype into a desired digital used to overcome this.
filter by using frequency transformation in the s-domain 1 z 1
Let sk
1 z 1
where k is a design parameter and c be a
and bilinear z-transformation [2]. The main advantage of
this method is stability from mapping the poles and zeros T
wanted frequency c k tan c , solve for k:
2
from the stable region in the s-domain into a stable region
in the z-domain. However, when applying bilinear z- c
k
transformation, the warping frequency will give a non- T
tan c
linear relation between the analog frequency and the 2
digital frequency [3,4]. To solve this problem, a method The bilinear z-transformation with pre-warping
called bilinear z-transformation with pre-warping frequency is:
frequency is used. And also from this method, applying 1
T 1 z
inverse bilinear z-transformation [1], the inverse s c cot c 1
(3)
transformation from a digital filter to an analog low-pass 2 1 z
prototype can be found. Thus, it is called inverse bilinear A. Frequency transformation from the s-domain to the z-
z-transformation with pre-warping frequency. This paper domain with bilinear z-transformation with pre-warping
presents a new algorithm for frequency transformation in frequency
the digital domain using bilinear and inverse bilinear z-
The transfer function H(z) of a digital filter can be
transformation with pre-warping frequency, which can obtained from the transfer function H(s) of an analog
described by a mathematics equation called Digital low-
low-pass prototype filter by first transforming an analog
pass to digital filter Pascal matrix equation.
low-pass prototype to an analog filter that is the same
class of the digital filter using frequency transform in the
II. BILINEAR AND INVERSE BILINEAR Z-
s-domain, and then applying the bilinear z-transformation
TRANSFORMATION WITH PRE-WARPING
with pre-warping frequency[5]. Table I illustrates the
FREQUENCY
frequency transformation from the s-domain to the z-
domain.

This paper is submitted for review on June 10, .2014. This is self-
research by P. S. Nguyen.
P. S. Nguyen was a staff member at the Faculty of Electrical and
Electronic Engineering, Department of School of Communications and
Informatics, Victoria University of Technology, Melbourne, Australia
(Email: nguyensiphuoc@frequencytransform.com).
Let fs is the sampling frequency, fc is the cut-off
frequency for the low-pass and high-pass filter, and fU
and fL are the upper and lower frequency for the band-
pass, band-stop and narrow-band filter. The parameters of
c, t, U and L can be found as below.
 f f frequency, a digital low-pass is transformed into a digital
c cot c t tan c low-pass, high-pass, band-pass, band-stop and narrow-
fs fs
fU
band, as shown in Table II.
cU
cU cot U 1 c t
fs TABLE II. FREQUENCY TRANSFORMATION IN DIGITAL DOMAIN
U L

t tan f L
tL
L Converting
L fs 1 cU t L types
(z-1)

In the case of a narrow band, if given the center Low pass to c c N c c N z 1

frequency f0 and quality factor Q, the upper and lower low pass c c N c c N z 1
frequency can be calculated as below.
c t c t z 1
1 1
Low pass to
f L f0 1 high pass c t c t z 1
4Q 2
2Q
c U L 2 U L z 1 U L c z 2
f f 1 1 1 Low pass to
c U L 2 U L z 1 U L c z 2
0
U
4Q 2 2Q band pass

Low pass to cU cL 1 2c U L z 1 cU cL 1 z 2
TABLE I. FREQUENCY TRANSFORMATION FORM S-DOMAIN TO Z-
DOMAIN USING BILINEAR Z-TRANSFORMATION WITH PRE-WARPING
band stop
cU cL 1 2c (U L) z 1 ( cU cL 1) z 2
FREQUENCY

Frequency Bilinear z-transformation In Table II, to transform a digital low-pass with the
Converting types transformation in s- with pre-warping transfer function H(z) into a digital filter, z is substituted
domain frequency
by (z-1). Then, with the help of Pascals triangle, a
Low pass to low s 1 z 1 matrix equation is derived, called Digital low-pass to
c
pass c 1 z 1 digital filter Pascal matrix equation.
Low pass to high c 1 z 1
t IV. DIGITAL LOW-PASS TO DIGITAL FILTER PASCAL
pass
s 1 z 1
MATRIX EQUATION
Low pass to s U L
2
1 z 1 1 z 1
U L The digital low-pass to digital filter Pascal matrix
band pass U L s 1 z 1 1 z 1 equation is used to transform the coefficients alp and blp of
1 a digital low-pass filter to the coefficients ad and bd of a
Low pass to U L s
1 z 1 1 z 1 desired digital filter. This matrix equation is multiplied by
band stop s 2 U L U 1
L some matrices, which are found from the coefficient and
1 z 1 z 1
binomial expansion in Pascals triangle, and they are
introduced in the next section
B. Inverse bilinear z-transformation with pre-warping A. Diagonal matrix Dx
frequency
Matrix Dx is a diagonal matrix with a size of
Inverse bilinear z-transformation with pre-warping (n+1;n+1), where n is the given nth-ordered digital low-
frequency is a method used to transform a digital filter
pass filter, as shown below:
back to an analog low-pass prototype. In this paper, it
xn 0 0 0
shows the transformation from a digital low-pass to an 0 x n 1 0 0 0
analog low-pass prototype. Dx 0 0 x n 2 0 0

In Table I, to transform an analog low-pass prototype 0 0
0 0 0 1
with transfer function H(s) to a digital low-pass with
1 z 1 B. Matrix P
transfer function H(z), s in H(s) is replaced by c
1 z 1
. The
inverse bilinear z-transformation with pre-warping Matrix P has a size of (n+1;n+1) and contains the
frequency is then: positive and negative coefficients of Pascals triangle.
There are three different matrix P. Depending on the
1 z 1 cs
sc z 1 (4) conversion type, matrix P may be PLP, PHBS or PBS, as
1 z 1 cs shown below:

Equation (4) is an inverse bilinear z-transformation
with warping frequency for transforming a digital low- PLP i 1; j 1 n1 j n 1
pass to an analog low-pass prototype.


PLP i n1; j 1 n1 ( 1) j 1
j 1 n


III. FREQUENCY TRANSFORMATION IN THE DIGITAL PLP PLP i 1 n1; j 1 1

DOMAIN PLP i 1 n1; j n1 ( 1) i 1

This section introduces a new algorithm for frequency
P P
LP i ; j 1 PLP i 1; j 1 PLP i 1; j
transformation in the digital domain. From Table I and an LP i ; j

inverse bilinear z-transformation with pre-warping


PLP i 1; j 1 n1 ( 1) j 1 j n 1

H z
1 2 z 1 z 2
14.8246 16.9443 z 1 6.1196 z 2




PLP i n1; j 1 n1 j 1 n


fc 200 Hz , fcn 400 Hz , fs 2000 Hz

f f
PHBS PLP i 1 n1; j 1 1 c cot c 3.0777 , cn cot cn 1.3764
fs fs
PLP i 1 n1; j n1 ( 1) i 1 Apply (5)

P P
LP i ; j LP i ; j 1 P
LP i 1; j 1 P
LP i 1; j

ab ba
d
d LP
LP PLP DcTcn PLP
PLP DcTcn PLP

PLP i 1; j 1n1 j n 1

1 2 1 9.4721
ad 1 2 1 1 0 1 0
0 0
3.0777 0

1 2 1 0 0 1

PLP i n1; j 1 n1 ( 1) j 1
j 1 n
1
0 1.3764
0 0 1 2 1
0 1 0 1
0 0 1.8944 1 2 1
PBS PLP i 1n1; j 1 ( 1) i 1
37.8885 75.7771 37.8885 37.8885 1 2 1
1 2 1 9.4721 0 0
PLP i 1 n1; j n1 1 bd 14.8246 16.9443 6.1196 1 0 1 0 3.0777 0
1 2 1 0 0 1
P P P P 1 0 0 1 2 1
0 1.3764 0 1 0 1

LP i ; j LP i ; j 1 LP i 1; j 1 LP i 1; j
0 0 1.8944 1 2 1
183.4156 67.77771 35.9157 37.8885 4.8409 1.7889 0.9479
C. Matrix T 1 2 z 1 z 2
H LP new
Two kinds of matrix TTUL and Txare introduced. 4.8409 1.7889 z 1 0.9479 z 2
One useful application of Pascals triangle is the
expansion of binomial (U+L)n. Inserting zeroes will make
matrix TUL, with a size of (n+1;2n+1), and matrix T x,
with a size of (n+1+;n+1) as shown
below:
0 0 0 0 0 1 0 0 0 0 0
0 0 0 0 U 0 L 0 0 0 0
0 0 0 U2 0 2UL 0 L2 0 0 0
TUL
0 0 U3 0 3U 2 L 0 3UL2 0 L3 0 0
0 0 0 0
Un L
n
0 0 0 0 0
Fig. 1. Transforming a digital low-pass to a digital low-pass using the digital
1 0 0 0 0 low-pass to digital filter Pascal matrix equation.
0 x2 0 0 0
Tx 0 0 x3 0 0
B. Example 2: Transforming a digital low-pass to a
0 0
0 0 0 0 xn digital high-pass
The transformation from a digital low-pass to another Transforming a second-ordered Butterworth digital
digital filter is called the digital low-pass to digital filter low-pass with transfer function H(z) at corner frequency
Pascal matrix equation, as written below: 200 Hz to a digital high-pass with corner frequency
ad aLP PLP Dx T P 200 Hz at sampling frequency 2000 Hz:

(5)
bd bLP PLP Dx T P
Apply (5)

ad aLP PLP DcTt PHBS

V. EXAMPLES FOR DESIGN OF DIGITAL FILTER
bd bLP PLP DcTt PHBS
This section demonstrates four examples for f
fcn 200 Hz , fs 2000 Hz t tan cn 0.3249
transforming a digital low-pass to a digital filter using the fs
digital low-pass to digital filter Pascal matrix equation
with MATLAB programming. 1 2 1 9.4721 0 0
ad 1 2 1 1 0 1 0 3.0777 0
A. Example 1: Transforming a digital low-pass to a 1 2 1 0 0 1
digital low-pass 1 0 0 1 2 1
0 0.3249 0 1 0 1
Transforming a second-ordered Butterworth digital 0 0 0.1056 1 2 1
low-pass with the transfer function H(z) at corner
frequency 200 Hz to a digital low-pass with corner 37.8885 75.7771 37.8885 37.8885 1 2 1
frequency 400Hz at sampling frequency 2000 Hz:
 1 2 1 9.4721 0 0
bd 14.8246 16.9443 6.1196 1 0 1 0 3.0777 0
1 2 1 0 0 1
1 0 0 1 2 1
0 0.3249 0 1 0 1
0 0 0.1056 1 2 1

59.2986 67.7771 24.4785 37.8885 1.5651 1.7889 0.6461

1 2 z 1 z 2
H LP new
1.5651 1.7889 z 1 0.6461z 2
Fig. 3. Transforming a digital low-pass to a digital band-pass using the
digital low-pass to digital filter Pascal matrix equation.
200 Hz to a digital band-stop with upper frequency
400 Hz and lower frequency 200 Hz at sampling
frequency 2000 Hz:
Apply (5)

ad aLP PBS Dc _ invTUL PHBS

bd bLP PBS Dc _ invTUL PHBS

Fig. 2. Transforming a digital low-pass to a digital high-pass using the fu 400 Hz , fl 200 Hz , fs 2000 Hz
digital low-pass to digital filter Pascal matrix equation. fu cu
cu cot 1.3764 U 1 c t 2.4899
fs
u l

t tan fl 0.3249
tl
C. Example 3: Transforming a digital low-pass to a L 0.5878
l
fs 1 cu tl
digital band-pass
1 2 1 1 0 0
Transforming a second-ordered Butterworth digital ad 1 2 1 1 0 1 0 3.0777 0
1 2 1 0 0 9.4721
low-pass with transfer function H(z) at corner frequency
1 4 6 4 1
200 Hz to a digital band-pass with upper frequency 0 0 1 0 0 1 2 0 2 1
400 Hz and lower frequency 200 Hz at sampling 0 2.4899 0 0.5878 0 1 0 2 0 1
6.1996 0 2.9271 0 0.3455 1 2 0 2 1
frequency 2000 Hz: 1 4 6 4 1
Apply (5) 358.9 1 2.472 3.5277 2.472 1
1 2 1 1 0 0
ad aLP PLP DcTUL PHBS bd 14.8246 16.9443 6.1196 1 0 1 0 3.0777 0
1 2 1 0 0 9.4721
bd bLP PLP DcTUL PHBS 1 4 6 4 1
0 0 1 0 0 1 2 0 2 1
fu 400 Hz , fl 200 Hz , fs 2000 Hz 0 2.4899 0 0.5878 0 1 0 2 0 1
fu cu 6.1996 0 2.9271 0 0.3455 1 2 0 2 1
cu cot 1.3764 U 2.4899 1 1
fs 1 cutl
4 6 4
358.9 1.565 3.04 3.3166 1.9041 0.646
t tan fl 0.3249
tl
L 0.5878
l s
f 1 cu tl 1 2.472 z 1 3.5277 z 2 2.472 z 3 z 4
H BP z
1 2 1 9.4721 0 0 1.565 3.04 z 1 3.3166 z 2 1.9041z 3 0.646 z 4
ad 1 2 1 1 0 1 0 3.0777 0
1 2 1 0 0 1
1 4 6 4 1
0 0 1 0 0 1 2 0 2 1
0 2.4899 0 0.5878 0 1 0 2 0 1
6.1996 0 2.9271 0 0.3455 1 2 0 2 1
1 4 6 4 1
37.8885 1 0 2 0 1
1 2 1 9.4721 0 0
bd 14.8246 16.9443 6.1196 1 0 1 0 3.0777 0
1 2 1 0 0 1
1 4 6 4 1
0 0 1 0 0 1 2 0 2 1
0 2.4899 0 0.5878 0 1 0 2 0 1
6.1996 0 2.9271 0 0.3455 1 2 0 2 1 Fig. 4. Transforming a digital low-pass to a digital band-stop using
1 4 6 4 1 the digital low-pass to digital filter Pascal matrix equation.
37.8885 14.8247 28.7964 31.4164 18.0364 6.1196

1 2 z 2 z 4
H BP z E. Example 5: Transforming a digital low-pass to a
14.8247 28.7984 z 31.4164 z 2 18.0364 z 3 6.1196z 4
1

digital Notch filter

D. Example 4: Transforming a digital low-pass to a
digital band-stop
Transforming a second-ordered Butterworth digital
low-pass with transfer function H(z) at corner frequency
Transforming a second-ordered Butterworth digital
low-pass with transfer function H(z) at corner frequency
200 Hz to a digital Notch filter with center frequency f0 =
500Hz and quality factor Q=50 at sampling frequency
2000Hz.
Apply (5) ACKNOWLEDGMENT
ad aLP PBS Dc _ invTUL PHBS
This is my own research, and I would like to share my
bd bLP PBS Dc _ invTUL PHBS
idea of how to apply Pascals triangle in filters design. I
f0 500 Hz , Q 50 , fs 2000 Hz would appreciate feedback from readers to improve this

research.
1 1
fu f0 1 2 505 Hz
4Q 2Q

f f 1 1 1 495 Hz REFERENCES
0
l
4Q 2 2Q
fu cu Emmanuel, C., Ifeacor, B. & Jervis, W. Digital signal
cu cot 0.9843 U 1 c t 31.8259
[1]
fs processing, Addison-Wesley, 1993, ch 7, pp. 392-394.
u l

t tan fl 0.9845
tl [2] Konopacki J.: The frequency transformation by matrix operation
L 31.8309
l 1 cu tl and its application in IIR filters design, IEEE Signal Processing
fs Letters, 2005, 12, (1), pp. 58.
1 2 1 1 0 0
Garcia-Ugalde, F.J. Z-transform by Pascal matrix and its
ad 1 2 1 1 0 1 0 3.0777 0 [3]
applications in the design of IIR Filters, Journal of Applied
1 2 1 0 0 9.4721
1 4 6 4 1 Research of Technology, 2011, 9, (3), pp. 355366.
0 0 1 0 0 1 2 0 2 1 [4] Chivapreecha, S. Bilinear s-z frequency transformation using
0 31.8259 0 31.8309 0 1 0 2 0 1 matrix Pascal operation, IEEE Communications and Information
1012.9 0 2026.1 0 1013.2 1 2 0 2 1 Technology, 2005, 1, pp. 764767.
1 4 6 4 1 [5] Phuoc S.N. Bilinear z-transformation using Pascals triangle,
105 1.5353 0.0005 3.0706 0.0005 1.5353 International Journal of Scientific and Engineering Research,
1 2 1 1 0 0 2013, 4, pp. 19741978.
bd 14.8246 16.9443 6.1196 1 0 1 0 3.0777 0
1 2 1 0 0 9.4721
1 4 6 4 1 BIOGRAPHY
0 0 1 0 0 1 2 0 2 1
0 31.8259 0 31.8309 0 1 0 2 0 1 Phuoc Si Nguyen received a Bachelor's
1012.9 0 2026.1 0 1013.2 1 2 0 2 1 degree in Mathematics and Physics from the
1 4 6 4 1
105 1.5698 0.0005 3.0699 0.0005 1.5016 University of Can Tho , Vietnam and
experienced teaching Mathematics and
1.5353 0.0005 z 1 3.0706 z 2 0.0005 z 3 1.5353 z 4
H BP z Physics for highschool students in his
1.5698 0.0005 z 1 3.0699 z 2 0.0005 z 3 1.5016 z 4 hometown . A few years later , he received a
bachelor's degree in Electrical and Electronic
Engineering from Victoria University of
Technology in Melbourne , Australia . Because of the outstanding
graduation , he joined the research group in High Efficiency Power
Amplifier for Mobile Communications System and he was also a lab
assistant and tutor at Department of Communication and Informatics,
Victoria University of Technology. His interests are in the field of
mathematics and signal processing techniques . Currently, while Maths
tutoring for highschool students , he enjoys doing self- study on " How
Pascal 's triangle can be applied in design of digital filters" . This idea
Fig. 5. Transforming a digital low-pass to a digital Notch filter using the
digital low-pass to digital filter Pascal matrix equation. shows a lot of promise applications which he absorbed in them and hope
to share to whom have the same interests.

VI. CONCLUSION

A new method was developed for transforming a digital

low-pass filter with transfer function H(z) to a digital
filter (low-pass, high-pass, band-pass, band-stop and
narrow-band). The involvement of Pascals triangle in the
digital low-pass to digital filter Pascal matrix equation, as
presented and demonstrated in the examples, made the
work easier for hand calculations and computing when
transforming the digital filter in the z-domain. The
algorithm of this method is simple because all operations
imply the matrix multiplication; therefore, it is more
effective in programs and calculations.