Fourth-Order Butterworth Active Bandpass Filter

Design for Single-Sided Magnetic Particle Imaging

Scanner
A. A. Sadiq1, N. B. Othman1, M. M. Abdul Jamil1, M. Youseffi2, M. Denyer2,3, W. N. Wan Zakaria4, and
M. R. Md Tomari4
1
Biomedical Modelling and Simulation Research Group, Faculty of Electrical and Electronics Engineering, Universiti Tun
Hussein Onn Malaysia, Batu Pahat, Johor, Malaysia.
2
School of Engineering, Design, and Technology, University of Bradford, Bradford, UK.
3
School of Life Science, University of Bradford, Bradford, UK.
4
Department of Mechatronic and Robotics Engineering, Faculty of Electrical and Electronics Engineering, Universiti Tun
Hussein Onn Malaysia, Batu Pahat, Johor, Malaysia.
engrsadiqgam@gmail.com

Abstract— This paper describes the design and simulation of

a fourth-order Butterworth active bandpass filter designed for
single-sided magnetic particle imaging (MPI) scanner. Bandpass
filters (BPF) are used in MPI scanner set-up to attenuate the
harmonic distortions generated by the power amplifier before
passing the purely sinusoidal signal to the excitation (Drive)
coils. The BPF is designed based on the excitation frequency of
22.8 kHz, having Butterworth response, and realized using
Sallen-Key topology by cascading one second-order high-pass
filter and one second-order lowpass filter, with values of the Figure 1: Signal chain of MPI scanner
passive components calculated using the coefficient matching
method. MATLAB and NI Multisim software are used to The signal flow is divided into a transmitting section and a
simulate the filter, and the results are compared. The magnitude receiving section [3]. In the transmitting part, a sine wave
response obtained using MATLAB have monotonic amplitude signal is generated from a function generator at a fixed
response in the pass and stopband, and maximally flat with
frequency. The excitation frequency is typically preferred to
small ripples in the passband than the NI Multisim
implementation, while NI Multisim implementation has better be around 25 kHz [2]. These frequencies normally are not
roll-off than MATLAB implementation. detectable by the human ear, which makes the scanner
operation not audible. There are also certain physiological
Index Terms—Bandpass filter; Harmonics; Magnetic particle limitations in the choice of the excitation frequency, as the
imaging; MATLAB; NI Multisim. scanner is designed for humans, so there is a limitation of
exposure of human bodies to electromagnetic waves, one of
I. INTRODUCTION those being energy deposition [4].
The AC sinusoidal signal generated is amplified by a power
Magnetic particle imaging (MPI) is an innovative imaging amplifier to achieve the desired currents. Since there are no
modality that applies oscillating magnetic fields to determine ideal signal generators or power amplifiers, the amplified
the distribution of magnetic nanoparticles. By using a pick- signal will contain not only the signal of the fixed excitation
up (receive) coil, the change of the tracer magnetization can frequency but also different frequency components [5]. These
be sensed [1]. Nevertheless, the signal induced by the frequency components (harmonics) have to be suppressed by
nanoparticles is overlaid by the direct feed through the a BPF before applying the signal to the drive coils [2, 5, and
intrusion of the time-varying excitation field, which couples 6]. For this purpose, a third order Butterworth or Chebyshev
into the receive coils. The oscillating magnetic field required II analog filter of high quality may be applied [2]. Going by
to excite the magnetic nanoparticles used as the tracer is this opinion a filter is needed in the scanner set up.
generated by Excitation (Drive) coil [2]. There is a need for A filter is a circuit that is capable of passing a signal from
the frequency of excitation to be clean on the coils to input to output that allows frequency within the indicated
minimize harmonic distortions. Therefore, a high-quality band to pass and diminishing all others outside the band. They
Bandpass filter is required to attenuate all harmonics before are four basic types of filters. They are lowpass, highpass,
the signal is applied to the excitation coils. bandpass, and bandstop filter [7]. The simple filter is realized
The typical signal flow of MPI scanner is as given in Figure by different arrangements of resistors, capacitors, and
1. PA stands for the power amplifier, BPF is a bandpass filter, occasionally inductors. It is known as a passive filter. Active
Tc stands for transmitting coil, Rc is receiving coil, BSF filters make use of transistors or operational amplifier and RC
means bandstop filter, LNA is low noise amplifier while circuit to provide wanted voltage gains or impedance
ADC is analog to digital converter. features. Inductors are not ideal for active filter plan since it
is slightest typical, massive, weighty, and costly and does not
offer itself to IC-type mass production. Active filters are

e-ISSN: 2289-8131 Vol. 10 No. 1-17 17

Journal of Telecommunication, Electronic and Computer Engineering

usually employed in MPI scanner because they have excellent or lowpass filter, depending on the type admittances. BPF is
linearity, low power dissipation, large dynamic range and are a combination of highpass, and lowpass filters cascaded
easy to design and analyze [2]. together.
Kwaha et al. (2011), developed and used a third-order
infinite impulse response (IIR) lowpass Butterworth filter
using the MATLAB direct method and bilinear
transformation method. They reported that Butterworth filter
design using bilinear transformation were more stable with a
monotonically decreasing gain response, which is ascribed to
its little roll-off associated with discrete components
performance [8].
Similarly, Okubanjo and Oyetola (2012), designed and
constructed a third-order unity gain Butterworth high pass
active RC filter with a center frequency of 800 MHz The filter Figure 3: A unity gain conventional 2-pole active filter
was realized by cascading first and second order high pass
filters together, employing the standard matching method to A KCL Equation at node 𝑉𝑎 shall be:
obtain the component values [9]. Additionally, Zehra (2014),
designed and simulated a fourth-order active bandpass filter, (𝑉𝑖𝑛 − 𝑉𝑎 )𝑌1 = (𝑉𝑎 − 𝑉𝑏 )𝑌2 + (𝑉𝑎 − 𝑉𝑜𝑢𝑡 )𝑌3
comparing the multiple feedback and Sallen-key topologies. (1)
The filter designed has a Butterworth response with a pass
band of 10-20 kHz, to be applied for radio frequency A KCL Equation at node 𝑉𝑏 produces
identification system (RFID). He carried out simulations
using NI Multisim and MATLAB and reported that the (𝑉𝑎 − 𝑉𝑏 )𝑌2 = 𝑉𝑏 𝑌4 (2)
bandpass filter designed using multiple feedback topology
had a better performance as compared to Sallen-key topology For ideal op-amps, 𝑉𝑏 is also equal to 𝑉out . Therefore,
[10].
For this research, a fourth-order Butterworth active Y2 + Y4 Y2 + Y4
bandpass filter is designed and implemented in National Va = [ ] Vb = [ ] Vout (3)
Y2 Y2
Instruments (NI) Multisim and compared with MATLAB
direct method simulation. It is expected at the output of the
filter to have a signal with 22.8 kHz center frequency and a Thus, substitute Equation three into Equation (1) and
unity gain, with the unwanted harmonics, attenuated. multiply it by the term 𝑌2 , it yields the transfer function 𝑇(𝑠)
for the filter, which is
II. RESEARCH METHOD
Vout (s) Y1 Y2
T(s) = =[ ]
A. Bandpass Filter (BPF) Design Vin (s) Y1 Y2 + Y4 (Y1 + Y2 + Y3 ) (4)
A band-pass filter (BPF) permits all signals lying inside a
band in the middle of low and high critical frequency limits
Equation (4) is the general transfer function equation for a
and discards all other frequencies beyond this band. Bandpass
2-pole active filter. This Equation can be used to determine
filter comprises a high-pass filter (HPF) and a low-pass filter
the transfer function of a 2-pole high pass or 2-pole low pass
(LPF). It is realized by cascading the circuits for HPF and
filter depending on the admittance [12]. Therefore, Equation
LPF [11]. Therefore, BPF is divided into two sections that are
four is utilized in the low-pass and high-pass filters design to
analyzed separately to determine the values of the RC
determine the values of the RC components required to
components. The magnitude response of a BPF is shown in
realize a BPF.
Figure 2.
1) Low-pass Filter Design
Substituting Y1 = 1⁄R , Y2 = 1⁄R , Y3 = sC1 and Y4 =
1 2
sC2 into Equation (4) gives:

Vout (s)
T(s) =
Vin (s)
1⁄ (5)
R1 R 2 C1 C2
=[ ]
(R + R 2 )SC2
S2 + 1 + 1⁄R R C C
R1 R 2 C1 C2 1 2 1 2

The transfer function can also be organized as the standard

second-order low-pass network Equation, which is:
Figure 2: Band Pass Filter Magnitude Response ωc 2
T(s) = 2
S + 2εωc S + ωc 2 (6)
The conventional two-pole active filter is displayed in
Figure 3 with 𝑌1 through 𝑌4 are admittances and with an ideal Therefore, the center frequency,
voltage follower. Figure 3 is used to design either a highpass

18 e-ISSN: 2289-8131 Vol. 10 No. 1-17

 Fourth-Order Butterworth Active Bandpass Filter Design for Single-Sided Magnetic Particle Imaging Scanner

1 In order to maintain the filter stability, a gain of the active

fc =
2π√R1 R 2 C1 C2 filter must not be greater than 3, which is best expressed as:

If R1 = R 2 = R, τ1 = RC1 and τ2 = RC2 then; A = 3 − (2 × ζ ) (14)

1 1 (7) where A stands for the overall gain of the op-amp and ζ is the
𝑓𝑐 = = damping factor, given as:
2𝜋 √𝜏1 𝜏2 2𝜋𝑅√𝐶1 𝐶2
3−𝐴 1
2) High-pass Filter Design 𝜁= =
2 2𝑄
For a high pass filter design, the following are substituted into Therefore, A = 3 – 1/Q. With Q = 3.8, the gain A = 2.74 ≈ 3.
Equation (4) as follows: Hence, a gain of A= 3 is assumed.

Y1 = sC1 , Y2 = sC2 , Y3 = 1⁄R and Y4 = 1⁄R which yields: a) High-pass Filter Section
1 2
Vout (s) (8) The first part of the BPF is the HPF section. Here, the center
T(s) = frequency 𝑓𝑐 is the lower frequency band to be passed, 𝑓1 (20
Vin (s)
SC1 . SC2 kHz). For simplicity, capacitors𝐶1 = 𝐶2 = 𝐶 and is picked to
= be 470 pF (standard), while the resistors values are calculated
SC1 . SC2 + 1⁄R (SC1 + SC2 + 1⁄R ) as follows:
2 1
Comparing Equation (8) and the standard transfer function, Using the passband gain of 3,
S2 (9) R2
T(s) = 2 A= 1+ =3 (15)
S + 2εωc S + ωc 2 R1
Therefore, the center frequency is:
1 From Equation (15), it implies that 𝑅2 = 2𝑅1 . Substituting
fc = for 𝑅2 into Equation (10) yields:
2π√R1 R 2 C1 C2
IfC1 = C2 = C andτ1 = R1 C and τ2 = R 2 C then; 1
1 1 (10) R1 =
2π√2f2 C (16)
𝑓𝑐 = =
2𝜋√𝜏1 𝜏2 2𝜋𝐶√𝑅1 𝑅2
Substituting the values for 𝑓𝑐 , 𝑅1 and 𝑅2 into Equation (13)
gives:
B. Band-Pass Filter Design Considerations
The first factor in BPF design is the center frequency 𝑓0, 𝑅1 = 11,972.29 Ω or 11.97229 kΩ
which is given as:
Standard resistor value of 12 kΩ is chosen and since 𝑅2 =
2𝑅1 = 24𝑘Ω
f0 = √f1 f2 (11)
b) Low-pass Filter Section
where 𝑓1 is the lower frequency band to be passed (edge of The center frequency 𝑓𝑐 for the Low-pass section of the BPF
the passband) and 𝑓2 is the upper frequency band to be passed is the upper frequency band to be passed, 𝑓2 (26 kHz). Here,
(closing edge of the passband). The second consideration in the assumption made using the passband gain of 3 is that:
BPF design is the bandwidth (BW) of the filter given as: 𝑅1 = 𝑅2 = 𝑅and;

BW = f1 −f2 (12) C1
Quality factor, A=1+ =3 (17)
C2
which implies that 𝐶1 = 2𝐶2 . Here, 𝐶1 was chosen as 470 pF.
f0
Q= (13) Consequently, 𝐶2 = 940 𝑝𝐹. The resistor value is calculated
BW from Equation 7, which gives:
If the quality factor Q is less than 10, a wide BPF with the 1
flat response over a range of frequencies will be realized. 𝑅= = 9,209.46 Ω 𝑜𝑟 9.21 𝑘Ω
Bandwidth is more in this case. Whereas if Q is greater than 2𝜋𝑓√𝐶1 𝐶2
10, then a narrow BPF with a sharp bell type response will be
achieved. Bandwidth is less in this case [13]. Therefore, a standard resistor value of 10 kΩ was chosen.
For the single-sided MPI scanner to be constructed, a 25 kHz 𝑅1 = 𝑅2 = 𝑅 = 10 𝑘Ω
excitation frequency is preferred [2]. Meanwhile, 𝑓1 and Table 1 below gives the summary of the RC components for
𝑓2 were arbitrarily selected below and above the excitation the BPF, calculated for the HPF and LPF sections.
frequency we have in mind, so as to fully capture the
Table 1
excitation frequency within the pass band. Therefore 𝑓1 was Overview of BPF RC Components
chose as 20 kHz while 𝑓2 as 26 kHz.
Section fc C1 C2 R1 R2
𝑓0 = √𝑓1 𝑓2 = 22.8 𝑘𝐻𝑧, 𝐵𝑊 = 6 𝑘𝐻𝑧 and 𝑄 = 3.8 (kHz) (pF) (pF) (kΩ) (kΩ)
HPF 20 470 470 12 24
With Q = 3.8, we have a wide BPF with a bandwidth of 6 LPF 26 470 940 10 10
kHz.

e-ISSN: 2289-8131 Vol. 10 No. 1-17 19

Journal of Telecommunication, Electronic and Computer Engineering

Figure 4 shows a fourth-order Butterworth active filter

schematic which is to be implemented in NI Multisim.

Figure 6: BPF magnitude response frequency specifications using

MATLAB

Figure 4: Fourth-order Butterworth active BPF schematic diagram The values of these parameters used are given in Table 2.
For creating the filter specification object, a code is evaluated
C. BPF Design Implementation at the MATLAB prompt [15]. With the excitation frequency
The behavior of BPF to be designed with the parameters of 25 kHz for the MNPs sample, a range of 20-26 kHz is
presented in Table 1 is simulated in NI Multisim software and logically selected as the actual portion of the affected
MATLAB. Multisim is a comprehensive development spectrum, so that the center frequency will be captured within
platform for electronic circuit’s schematics, designs this range. This selection gives a center frequency of 22.8 kHz
simulation, and prototyping. MATLAB, on the other hand, (Equation 15). 20 kHz is the first passband frequency
provides high-performance numerical calculation, data (Fpass1) while 26 kHz is the closing passband frequency
analysis, visualization capabilities and application (Fpass2) [18].
development tools. The results obtained from the two In practice, no BPF is ideal. Therefore, the attenuation in
simulations are presented and compared. the first and second stopband have to be wide enough, to
make the roll-off as narrow as possible. 60 dB is the default
a) NI Multisim Implementation value adopted for a wider transition region. For linearity, 15
The BPF schematic diagram given in Figure 4 is implemented kHz and 30 kHz were chosen as the edge of the first and
in NI Multisim with the values of the RC components in second stopband respectively (Fstop1 and Fstop2). The
Table 1. Where R1, R2 equal to R3, R4 for the LPF section, sampling frequency is twice the frequency of the edge (45
likewise C1, C2 equals C3 and C4 for LPF section kHz), which is 90 kHz [16].The amount of ripple to be seen
respectively. NI Multisim implemented circuit was as shown in the passband is up to the filter designer to choose. A ripple
in Figure 5. of 1% is allowed in this design (0.1 dB) because the design is
based on a mathematical function which is used to produce
maximally flat passband with no considerations for phase
linearity or group delay discrepancies [17].
Table 2
MATLAB BPF Specifications

AStop FStop FPass FPass FStop AStop APas Fs

1 1 1 2 2 2 s kH
(dB) (kHz) (kHz) (kHz) (kHz) (dB) (dB) z
60 15 20 26 30 60 0.1 90

Figure 5: NI Multisim circuit implemented for Fourth-order Butterworth

active BPF.
III. RESULTS AND ANALYSIS

b) MATLAB Implementation Simulations were carried out using MATLAB and NI

Unlike the NI Multisim application, MATLAB direct Multisim software, and the results are presented in Figure 7
method uses the frequency specifications only in designing and Figure8 respectively. The Fourth-order Butterworth
the filter [14]. These frequency specifications are written in active BPF simulated in MATLAB exhibit very flat
code and evaluated in MATLAB after a prompt command, amplitude in its passband (maximally flat), as shown in
called the fdesign.Bandpass (Bandpass filter design method). Figure 7. The response of the filter is smooth with a roll-off
The frequency specifications of the magnitude response of a rate of -80dB/pole and 80dB/pole. The pass and stop
BPF is given in Figure 6 below. frequency have an amplitude response at -3dB while the
Astop1 is the Attenuation in the first stopband; Fstop1 is center frequency has zero gain as this is in line with the
Edge of the stopband while Fpass1 and Fpass2 are the Edges characteristics of BPF [18]. The center frequency has zero
of the Opening and Closing of the passband respectively. phase while the stop frequency (20 kHz) and pass frequency
Other parameters are Fstop2, which is the Edge of the second (26 kHz) have phases of 2 and -2 radians respectively.
stopband; Astop2 is Attenuation in the second stopband and Nevertheless, the phase shift of the signals going through the
Apass is the amount of ripple allowed in the passband. Fs is filter differs nonlinearly with the frequencies.
the sampling frequency.

20 e-ISSN: 2289-8131 Vol. 10 No. 1-17

 Fourth-Order Butterworth Active Bandpass Filter Design for Single-Sided Magnetic Particle Imaging Scanner

response of the realized BPF is close to the ideal

characteristics of a fourth-order active BPF having
Butterworth characteristics. The little difference is due to the
approximations in calculating component values (resistors
and capacitors) in NI Multisim implementation. Even though
there is no boundary to the order of the filter that can be
formed, the fourth-order design is suitable for the MPI
scanner design because of the higher the filter order, the better
the performance of the filter. The rate of roll-off and the width
of the filter transition band also depends on the filter order.

ACKNOWLEDGMENT
Figure 7: BPF magnitude response using MATLAB
This research was conducted with financial support by
The result of the NI Multisim simulation and experiment Research Acculturation Grant Scheme (RAGS) Vot number
for the Fourth-order Butterworth active BPF is shown in R067 and Postgraduate Research Grant (GPPS) Vot number
Figure 8. The circuit implemented for the filter is given in U589, Universiti Tun Hussein Onn Malaysia (UTHM).
Figure 5. The magnitude response exhibits a flat amplitude in
the passband but not as maximally flat as that of MATLAB REFERENCES
simulation. The response of the filter is smooth with steeper
[1] Gleich and J. Weizenecker, "Tomographic imaging using the nonlinear
roll-off than MATLAB simulation [16]. The pass and stop
response of magnetic particles," Nature, vol. 435, pp. 1214-1217, 2005.
frequency have the same amplitude response as that of [2] T. Buzug, G. Bringout, M. Erbe, K. Gräfe, M. Graeser, M. Grüttner and
MATLAB simulation, at -3dB point. The phase response is J. Haegele, "Magnetic particle imaging: Introduction to imaging and
nonlinear, with the center frequency having zero phase and hardware realization," Zeitschriftfür Medizinische Physik, vol. 22, pp.
323-334, 2012.
the pass and stop frequencies having -5 and 5 radians
[3] S. Biederer, T. Knopp, T. Sattel, K. Lüdtke-Buzug, B. Gleich, J.
respectively. Weizenecker and T. Buzug, "Magnetization response spectroscopy of
superparamagnetic nanoparticles for magnetic particle imaging,"
Journal of Physics D: Applied Physics, vol. 42, pp. 205007, 2009.
0 [4] T. Knopp, and T. Buzug, "Magnetic particle imaging: an introduction
to imaging principles and scanner instrumentation," Springer Science
-10
& Business Media, 2012.
[5] H. Rogge, and R. Schneider, "Magnetic Particle Imaging (Doctoral
dissertation, Diploma thesis)," 2011.
-20
[6] A. Sadiq, N. Othman, and M. Mahadi, "Magnetic particle imaging
Gain (dB)

system for cancer diagnosis: an overview," ARPN Journal of

-30 Engineering and Applied Sciences, vol. 10, pp. 8556-8561, 2015.
[7] R. Levy, R. Snyder, and G. Matthaei, "The design of microwave filters.
-40 Microwave Theory and Techniques," IEEE Transactions on
Simulation Microwave Theory and techniques, vol. 50, pp. 783-793, 2002.
Experiment
-50 [8] B. Kwaha, O. Inyang, and P. Amalu, "The circular microstrip patch
antenna–design and implementation," International Journal of
0 50 100 150 200 Research and Reviews in Applied Sciences, vol. 8, pp. 86-95, 2011.
Frequency (kHz) [9] A. Okubanjo and K. Oyetola, "Design, construction, and simulation of
third order Butterworth high-pass filter using MATLAB," Research
Figure 8: BPF magnitude response for NI Multisim simulation and gate. Retrieved 13 June 2016, 2012.
experiment [10] A. Zehra, "Design and simulation of 4th order active bandpass filter
using multiple feedback and Sallen key topologies," Journal of Babylon
University, vol. 2, pp. 463-473, 2014.
In comparison, the BPF simulated in MATLAB have [11] T. Deliyannis, Y. Sun and J. Fidler, "Continuous-time active filter
monotonic amplitude response in the pass and stopband, have design," CRC Press, 1998.
maximally flat and small ripples in the passband than the NI [12] J. Kuo and E. Shih, "Wideband bandpass filter design with three-line
Multisim implementation. Regarding a better roll-off towards microstrip structures," In Microwaves, Antennas, and Propagation, IEE
Proceedings, vol. 149, pp. 243-247, 2002.
zero in the stopband, NI Multisim implementation is better. [13] V. Bhagavatula, "Power and area optimization techniques for ultra-
The better roll-off is due to the fast roll-off noticed around the wideband millimeter-wave CMOS transceivers," Doctoral dissertation,
center frequency as shown in Figure 8. MATLAB simulation 2014.
also yields a more linear phase response in the passband as [14] M. Lutovac, D. Tošić, and B. Evans, "Filter design for signal
processing using MATLAB and Mathematica," Miroslav Lutovac,
compared to the slightly nonlinear phase response of the NI 2001.
Multisim simulation. The two simulation works are [15] A. Casson, D. Yates, S. Patel and E. Rodriguez-Villegas, "An analog
significant for the MPI scanner realization. bandpass filter realization of the continuous wavelet transform," In
2007 29th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society, 2007 Aug 22, pp. 1850-1854, 2007.
IV. CONCLUSION [16] S. Winder, "Analog and digital filter design," Newnes, 2002.
[17] A. Biswal and V. Profile, "Butterworth filter and Chebyshev filter – A
The design and implementation of a fourth-order comparison," 2012.
Butterworth active bandpass filter are presented in this paper. [18] H. Ishida and K. Araki, "Design and analysis of UWB band pass filter
with ring filter," In Microwave Symposium Digest, 2004 IEEE MTT-
Both simulation results reveal that the magnitude and phase S International, vol. 3, pp. 1307-1310, 2004.


e-ISSN: 2289-8131 Vol. 10 No. 1-17 21