2010 Second International Conference on Information Technology and Computer Science

Design of Analog Gaussian Filter Used in Roughness

Measuring Instrument
GUO Jian1 ZHU Jie2
Information School Information School
Beijing WUZI University Beijing WUZI University
Beijing, China Beijing, China
[email protected] [email protected]

Abstract—According to the rational function approximation 2

:2
method of Gaussian filter, the transfer function of an 8-order
G ( j:) e D (1)
approximation filter is obtained. Based on value of the quality Here, Į is a const related to the bandwidth of filter.
factor Q of the pole of the transfer function, analog cascade
circuits are determined. The optimal cascading sequence and the Formula (1) is a theoretical mathematical model. When
principle of components are given. When selecting the cut-off squaring the amplitude of analog filter, we can obtain [4-6]:
frequency of 400 Hz, the experiment of the circuit in frequency
2 2
:2 1
domain and time domain is done. The result shows that the G(j:) G ( j:)G ( j:) e 2D 2D 2: 2
circuit achieves the desired target. e (2)
Keywords-Gaussian Filter; Function Approximation; Analog When expanding the denominator of (2) with 16-order,
Filter; Quality Factor Taylor series, (3) can be obtained [7].
I. INTRODUCTION 2 1
G( j:) | 8
2i
Usually, in roughness measuring instrument based on 1 ¦ (D:)2i
inductance sensor, analog low-pass filters are used to separate i 1 i! (3)
roughness signal from carrier signal and high-frequency noise
of filter to limit the width of signals before sampling. At the Here, we make s j: ˈ : 2  s 2 , then (3) can be
same time, ideally, to extract the roughness signal, filter need turned to (4).
to own linear phase in pass band in order to avoid the signal
distortion resulted from non-linear phase[1-2]. 2 1
G ( s) | 2i
| G p ( s)G p ( s)
8
2i
Gaussian filter is a special, ideal, optimal filter [3] with 1  ¦ (1) i (Ds)
the minimum product of time duration and frequency i 1 i! (4)
bandwidth, which is much fit for emphasizing reducing signal
Here, Gp(s) is a system function of Gaussian
distortion when the signals pass filter and ensuring attenuating
approximation filter.
effectively when signal frequency is larger than the cutoff
frequency of filter . Therefore, Gaussian filter is an ideal filter Calculate the poles of (4) and only poles that are in
in roughness measurement application [4]. International left-half plane of the complex plane are used to constitute the
standard ISO11562 has already stipulated definitely that system function Gp(s). The poles are shown in TABLEĉ. The
Gaussian filter mean line as an assessment mean line in expression of Gp(s) is as following:
measuring surface roughness is employed to replace the
traditional 2RC filter median. k p1 k p2
G p (s)
( s  s1 )( s  s3 ) ( s  s5 )( s  s7 )
Based on rational function approximation of Gaussian filter,
an 8-order analog Gaussian filter is designed in this paper. k p3 k p4
According to the size of quality factor of pole of transfer ( s  s9 )( s  s11 ) ( s  s13 )( s  s15 ) (5)
function, the circuit is decided. The formulas of each
component value are given too. Here, kpi (i=1,2,3,4), a const, is the gain of system. We
substitute the poles to (5) and then according to the gain
II. GAUSSIAN FILTER AND RATIONAL FUNCTION G (s) G ( j:)
APPROXIMATION condition of p s 0 p : 0 , we can get the gain const

kpi. At last, Gp(s) can be expressed as:

Without loss of generality, we define the
amplitude-frequency response of Gaussian filter as:

Supported by Funding Project for Academic Human Resources Development in

Institutions of Higher Learning under the Jurisdiction of Beijing
Municipality(PHR200906210, PHR201007145), Funding Project for Science
and Technology Program of Beijing Municipal Commission of Education
(KM200810037001), Funding Project for youth fund under Beijing WUZI
University (2008XJQN009)

978-0-7695-4074-0/10 $26.00 © 2010 IEEE 247

DOI 10.1109/ITCS.2010.67
 2.57034429 1.92672538 From the formula (2), the relation of ¢ and cut-off
G p (s) D2 D2 frequency of logarithmic amplitude-frequency characteristics at
1.7610 2.57034429 2.1546 1.92672538 the-3db is as (8):
s2  s s2  s
D D2 D D2
1.64928817 1.53655112 ln 2
D
D2 D2 2: c (8)
2.3632 1.64928817 2.4572 1.53655112
s2  s s2  s
D D2 D D2 (6) By deciding the value of ¡c, the value of ¢can be
gotten and then the transfer function of low-pass filter for
Formula (6) is the transfer function of 8-order Gaussian specified cut-off frequency can be obtained from (6) too.
approximation filter. According to the reference [8] which introduced a formula to
calculate component values of filter parameters, each
TABLE ĉ. THE POLES OF THE 8-ORDER GAUSSIAN APPROXIMATION FILTER parameter component values of the circuit in Fig. 1 can be
AND VALUE OF Q
computed.
Poles Value of poles Value of Q
B. Determination of cascading order
s1,3 (0.8805r 1.3398j) / D 0.910
s5,7 (1.0773 r 0.8753 j ) / D 0.644 From the mathematical point of view, the order of the
s9,11 (1.1816 r 0.5031 j ) / D 0.543
various parts of the cascade doesn't really matter. However, in
s13,15 ( 1.2286 r 0.1646 j ) / D 0.504
practice, there may exist signal clamp in high Q node, so to
avoid the loss of dynamic range and the reduction of filter
III. REALIZATION OF ANALOG FILTER accuracy, each section can be cascaded according to the order
of increasing Q value, that is, lower Q value part is put on the
The realization of analog filter is a process of turning the first-class of the signal pathway. But the thing that the higher Q
system transfer function to electronic circuit. The interested value part may influence concerned internal noise is not
thing is that simple, financial, low noise and distortion, high considered in this kind of cascading order. Any noise falling
dynamic range realization of filter [8]. And the filter also over resonance peak in higher Q value module can be
requests that the tiny change of component values will not significantly enlarged. So, higher Q value part should be placed
influence the filter seriously. In order to reduce the sensitivity in the forefront of cascade sequence to reduce the noise.
of system function to the deviation of component value, and Generally, the optimal cascading sequence is obtained based on
simplify the design process of filter, cascade realization of input signal spectrum, filter type and the noise characteristics
one-order and two-order filter is usually employed. Cascade of the various parts [10].
method consists of implementing each double two-order
system function with circuit and connecting them by cascading In the roughness instrument applications, the signal into
way. So each low-pass filter is independent and any change of the filter is sine wave amplitude modulated signal whose
one of filters will not influence any other cascading filters. amplitude is about tens of mill volts. The dynamic range of
filter will not be a problem. Therefore, the cascading sequence
A. Implementation of circuit which is advantageous to reduce the noise becomes the main
For implementing the double two-order system function, factor to consider. The Fig. 2 shows a four two-order filter with
suitable circuit is selected according to the quality factor Q of high-Q two-order section in front and low-Q two-order section
pole. The calculating formula of Q is as following (7): in the post.
C21 C22
s
Q 
2 Re( s) (7)
8
8

6
6 R12 R32 5 B 7
Vin R11
Vin R31 5 B 7
Vin
Here, |s| is the amplitude of pole, and Re(s) is the real part B
B
C42
4

of pole. So the quality factor of pole of the 8-order C41

4

approximation filter can be obtained in TABLEĉ. Since Q”2,

Sallen-Key filter with unity gain is selected to carry out the C23 C24
above filter [9]. The circuit is shown as Fig.1.
8
8

6 6 7 Vout
Vout
7 R14 R34 5
R13 R33 5 B B
B
C44
4

C43
8

Vout
Vin

Fig. 2 an 8-order Gaussian approximation filter

4

C. Selection of components
The suitable selection of capacitance and resistance which
Fig. 1 a two-order Sallen-Key filter with unity gain are basic components of active filter is a critical factor of

248
success of actual design. In all types of fixed resistors, thick IV. CONCLUSION
and thin film chip resistors are very suitable for application in Basing on the rational function approximation method of
the filter, so a kind of thick-film resistor which has a positive Gaussian filter, an 8-order analog Gaussian filter is designed in
temperature coefficient is chosen in this paper. And NPO-type this paper. According to the size of quality factor of transfer
ceramic capacitor which has a negative temperature coefficient function poles, the implementation circuit is determined. The
is selected to compensate the positive temperature coefficient formula of each component value and the cascading sequence
of resistor, which can reduce the sensitivity of active filters to are given in accordance with the principle of maximum SNR
resistance and capacitance change [11-12]. (signal to noise ratio). The experiments to the circuit by means
In active filters, the role of the operational amplifier is to of frequency-domain analysis and time analysis show that the
provide amplification and isolation. Its closed-loop bandwidth circuit can achieve the desired design target.
has to be higher than the cutoff frequency of the filter at least
100 times and slew rate should be larger than the product of ACKNOWLEDGEMENTS
cut-off angular frequency and output voltage peak – peak, Financial support for this study obtained from Funding
which can guarantee that there is no signal distortion in filter. Project for Academic Human Resources Development in
Input bias current should be in the pa level in order to decrease Institutions of Higher Learning Under the Jurisdiction of
the input offset voltage, which is also advantageous to selecting Beijing Municipality ˄ PHR200906210, PHR201007145 ˅ ,
the component parameters of resistors and capacitors [13-14]. Funding Project for Science and Technology Program of
D. Experiments & Result Beijing Municipal Commission of Education
(KM200810037001), and Funding Project for Youth fund
With cut-off frequency of 400 Hz in the 8-order filter as under Beijing WUZI University(2008XJQN009) is gratefully
an example simulation experiment, amplitude-frequency acknowledged.
response is obtained as Fig. 3 shown. The red and blue curves
are the amplitude-frequency characteristics of Gaussian filter REFERENCES
and its 8-order rational approximation filters respectively. And [1] LIN Bing, HUANG Xinyan, WEI Ying, WANG Lan, “Measurement
the amplitude-frequency characteristic deviation of rational Theory, Methods and Evaluation of Machined Surface
approximation filters from the Gaussian filter is also shown as Topography,”Manufacturing Automation,2006,28(8),pp.14-18(in Chinese)
[2] XU Jingbo, YUAN Yibao, PIAO Weiying, J. F. Song, T. V. Vorburger,
Fig.4. We can see that the maximum deviation is about 0.58%, “A Fast Gaussian Filtering Algorithm for Surface Roughness
which to meets the design requirements well. Measurements,” ACTA METROLOGICA SINICA, 2005,(26)4,
pp.309-312 (in Chinese)
1 [3] J.Klapper, C.M.Harris, “On the Response and Approximation of Gaussian
0.9
ˉ ˉ approximation Filter,” IRE Transactions on Audio,1959,3, pp.80~87(in English)
ˉ ˉ Gaussian
[4] Xu Jingbo, “Study on Approximation Theory and Application of Gaussian
0.8
Filter,” Harbin Institute of Technology,2007,pp.24-30 (in Chinese)
0.7 [5] CHENG Peiqing, “Digital Signal Processing,” Tsinghua University Press,
0.6 2000, pp.252-254(in Chinese)
[6] ZHENG Junli, YING Qinheng, YANG Weili, “Signals and Systems, ”
Amplitude

0.5
Higher Education Press(Part Ċ),pp. 182(in Chinese)
0.4
[7] Lin Maoliu, Liu Zhiyu. A Design of Gaussian Filter for High-speed
0.3 Wide-band Sample-filter Digitizing System, Chinese Journal of Scientific
0.2 Instrument, 2002, 23(2), pp.121-123(in Chinese)
[8] Meoslav D. Lutovac, Dejan V. Tosic, Brian L. Evans, (Translator: ZHU
0.1
Yisheng, DONG Hui), “Filter Design for Signal Processing Using
0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 MATLAB and Mathematica,” Publishing House of Electronics Industry,
Frequency(rad/sec) 2004, pp.109-110
Fig. 3 Amplitude-frequency characteristics of Gaussian filter and its rational [9] G.S.Moschytz and P.Horn, “Active Filter Design Handbook,” John
approximation filters Wiley,New YorK,NY,1990 ,pp.38~84(in Chinese)
x 10
-3 [10] Sergio Franco, (Translators: LIU Shutang, ZHU Maolin, RONG Zheng),
0
“Desing with Operation Amplifers and Analog Integrated Circuits,” Xi’an
Jiaotong University Press, 2004, pp.145(in Chinese)
-1 [11] Arthur B. Williams, Fred J. Taylor. (Translators: NING Yanqin, YAO
Jinke), “Design of Electronic Filter,” Science Press, 2008, pp.294-300(in
-2 Chinese)
[12] YUAN Banjunzhao, (Translator: Peng Jun), “Measurement of Electronic
Amplitude

-3 Circuit Design ʊʊFilter Part,” 2007, pp.117-132

[13] Bonnie Baker. (Translator: LI Yukui), “A Baker’s Dozen: Real Analog
-4 Solutions for Digital Designers,” Beijing University of Aeronautics and
Astronautics Press, 2006, pp.85-86 (in Chinese)
-5 [14] Walt Jung.(Transloator:ZHANG Lefeng, ZHANG Ding), “Op Amp
Applications Handbook,” Posts & Telecom Press, 2009, pp. 294-298(in
-6
Chinese)
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
Frequency(rad/sec)

Fig.4 Amplitude-frequency characteristic deviations of rational approximation

filters

249