Int. J. Electron. Commun.

(AEÜ) 110 (2019) 152847

Contents lists available at ScienceDirect

International Journal of Electronics and

Communications (AEÜ)
journal homepage: www.elsevier.com/locate/aeue

Regular paper

A novel neural-based approach for design of microstrip filters

Mohammad (Behdad) Jamshidi a, Ali Lalbakhsh b,⇑, Bahare Mohamadzade c, Hesam Siahkamari a,
Seyed Mohammad Hadi Mousavi d
a
Young Researchers and Elite Club, Kermanshah Branch, Islamic Azad University, Kermanshah, Iran
b
Macquarie University, Sydney, NSW, Australia
c
Department of Electrical and Computer Engineering, Semnan University, Semnan, Iran
d
Young Researchers and Elite Club, Hamedan Branch, Islamic Azad University, Hamedan, Iran

a r t i c l e i n f o a b s t r a c t

Article history: This paper presents an intelligent design methodology of microstrip filters in which a dynamic neural
Received 6 January 2019 network model based on Bayesian Regularization Back-Propagation (BRBP) learning algorithm is used.
Accepted 28 July 2019 In this approach, a Low-Pass Filter (LPF) composed of multiple open stubs, and stepped impedance res-
onators is initially designed for which an Artificial Neural Network (ANN) is trained to improve the per-
formance of the filter. The predicted and measured results of the filter verify the effectiveness of the
Keywords: presented method, suggesting an excellent in and out-of-band performance. According to the measure-
Artificial Neural Network (ANN)
ment, the filter has a very small transition band from 2.087 to 2.399 GHz with 3 and 40 dB attenuation
Bayesian Regularization Back-Propagation
(BRBP) algorithm
points, respectively, leading to a sharp roll-off rate of 118.6 dB/GHz. In addition the optimized filter
Compact size has an ultra-wide stopband, extending from 2.399 to 15.01 GHz with attenuation level of 22 dB are
Lowpass filter The overall size of the fabricated filter is only 0:190kg
0:094kg , where kg is the guided wavelength at
Microstrip 3 dB cut-off frequency (2.087 GHz). A performance comparison with some of the recent published LPFs
Non-Linear Autoregressive eXogenous presented, showing the superiority of the proposed filter.
(NARX) Ó 2019 Elsevier GmbH. All rights reserved.
Sharp roll-off
Wide stopband

1. Introduction out-of-band performance were realized in [3] and [12] at the price
of large size and low selectivity, respectively. In [8], a new trans-
Microwave filters play a significant role in wireless communica- formed radial stub was developed to extend the stopband. Other
tion systems. A lowpass filter with desired features such as ultra- methods such as defected ground structure, fractal geometries
wide stopband, compact size, sharp transition band and low inser- have also been suggested for performance enhancement of micro-
tion loss are highly in demand for such systems to suppress the strip filters [13–16]. A dual-plane structure using Defected Ground
undesired signals [1]. In recent years, several methods have been Structures (DGSs) was proposed to place multiple transmission
proposed to achieve these features [2–9]. Modified T-shaped and zeros to extend the stopband of the lowpass filter [15]. For the
U-shaped resonators were adopted for achieving compact size same purpose, a G-shaped defected microstrip structure composed
and sharp roll-off [2]. A microstrip lowpass filter with wide stop- of one circular ring and one connecting slot was proposed in [17]. A
band using loop stepped impedance along with U-shaped res- periodic pattern of DGS cells were proposed for high selectivity and
onator was reported in [4]. In [5] a simple structure composed of wide stopband; however, the presence of several cascaded DGS
a microstrip coupled-line hairpin was proposed; however a grad- cells increased the overall size of the filter [16]. Quasi-yagi DGS
ual transition band and low level of suppression in the stopband with compensated capacitors were proposed to achieve a sharp
were observed in the filter’s response. To improve the suppression cut-off frequency response with low insertion loss, while the over-
level of microstrip LPFs, tapered resonators were presented in [7]. all size of the filter is 0.45kg
0.35kg, where kg is equal to 44 mm
To achieve a wider stopband, several types of open stubs, such as [18]. Bended coupled lines in conjunction with T and I shaped res-
radial stubs, shunt open-stubs and stepped impedance stubs were onators have been proposed in [19] and [20], respectively to realize
proposed in [6] and [9–11]. Two lowpass filters with very good adjustable resonances without using active components. In [21]
stepped impedance hairpin units were proposed to achieve a wide
⇑ Corresponding author. stopband and a tenable filter was designed based on coupled lines
E-mail addresses: [email protected] (A. Lalbakhsh), bahare.mohamadza- in [22]. Apart from the aforementioned design techniques, even
[email protected] (B. Mohamadzade).

https://doi.org/10.1016/j.aeue.2019.152847
1434-8411/Ó 2019 Elsevier GmbH. All rights reserved.
2 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

and odd mode analysis can be used to design microstrip filters and sive eXogenous (NARX) model for both tuning and optimizing a
other electromagnetic components [23–28] Microstrip lowpass filter. The learning process algorithm is based
In addition to such conventional design methodologies, Artifi- on Bayesian Regularization Back-Propagation, exhibiting high per-
cial Intelligence (AI) based approaches have recently attracted sig- formance in modelling and optimizing complex systems. The rest
nificant interest to be used for designing microwave and of the paper is organized as following: in Section 2 a lowpass filter
electromagnetic components [29–36]. Such evolutionary algo- is initially designed with a wide stopband and a cut-off frequency
rithms can be incorporated into the design procedure for either of 2 GHz. In Section 2, the sub-optimal response of the filter,
optimizing a sub-optimal structure or realizing a non-intuitive designed in Section 2, is given to our fed-forward neural network
structure. In this paper, we are demonstrating the use of a dynamic for further optimisation, where Bayesian Regularization Back-
Artificial Neural Network (ANN), based on Nonlinear Auto Regres- Propagation (BRBP) algorithm is used to train the ANN. Measured

Resonator 1

Fig. 1. Layout of the proposed LPF.

Fig. 2. (a) Layout of Resonator 1. (b) EM Simulation results of Resonator 1 up to 16 GHz.

 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847 3

and simulated results of the optimized filter are presented in Sec- respectively which can be used to adjust the passband of the low-
tion 3, verifying the effectiveness of the proposed ANN-based tech- pass filter. To explain the filtering mechanism of the proposed res-
nique. The paper is closed in Section 4 by a brief conclusion. onator, its L-C circuit is extracted, and the values of the lumped
inductors and capacitors are calculated.
As it can be seen from Fig. 3(b) L1, . . ., L4 and C1, . . ., C3 are
2. Prototype filter design inductors and capacitors of the high-low impedance lines. C4 is
the sum of the capacitances of low impedance line and open stub.
Fig. 1 shows the configuration of the proposed lowpass filter Cg denotes the capacitance of the coupling space. The values of the
designed at 2.087 GHz. The filter is composed of three resonators mentioned parameters can be calculated using methods discussed
named Resonator 1, 2 and 3, shown in Fig. 1, which generate mul- in [1].
tiple transmission zeroes to extend the stopband. To explain the
performance of the proposed lowpass filter, the frequency er þ 1 er  1
w0:5
responses of the resonators are numerically calculated step by ere ¼ þ 1 þ 12 ð1Þ
2 2 h
step.
Fig. 2(a) shows the first building block of the proposed lowpass   1
filter. It is composed of two T-shaped resonators with inwards cor- g W W
Z c ¼ pffiffiffiffiffiffi þ 1:393 þ 0:667ln þ 1:444 ð2Þ
rugations, created for a better control on the resonators’ frequency ere h h
response. To achieve 50 X impedance matching, the value of b2 is
calculated as 1.17 mm by (1) and (2). All other dimensions of the As it can be seen from Fig. 2(b) despite acceptable out-of-band
structure are as follows: b1 = 2 mm, b2 = 0.92 mm, b3 = 2.2 mm, rejection from 2.5 to 5 GHz, the proposed lowpass structure has a
b4 = 4.5 mm, b5 = 0.3 mm, b6 = 0.9 mm, b7 = 0.9 mm, b8 = poor selectivity, return loss and undesired harmonics in the stop-
0.9 mm, b10 = 4.6 mm, b11 = 5.9 mm, b12 = 0.2 mm, b13 = 0.1 mm. band region. Therefore, two high impedance lines, loaded by two
The frequency response of Resonator 1 is shown in Fig. 2(b), rectangular patches, are added to Resonator 1, as shown in Fig. 4
where two deep transmission zeros appear at 2.41 GHz and (a). The frequency response of the filter after adding Resonator 2
3.00 GHz with the attention levels of about 59 and 63 dB, is shown in Fig. 4(b).

Fig. 3. (a) LC model of resonator 1 and (b) EM versus LC simulations.

4 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

The values of elements are as follows: L1 = 3.97 nH, 3. Neural-based approach

L2 = 3.08 nH, L3 = 2.16 nH, L4 = 0.133 nH, C1 = 0.197 pF, C2 =
266 pF, C3 = 0.344 pF, C4 = 1 pF, Cg = 2 pF. In the previous section, the design procedure of the proposed
The introduction of Resonator 2 contributes to generation of lowpass filter was explained, where some improvements in its fre-
three transmission zeroes located at 2.40, 2.967 and 4.976 GHz quency response were achieved by introducing the open stubs and
with attention levels of 46.2, 42.3 and 103.5 dB, respectively. T-shaped resonators. However, a holistic optimisation approach
For further improvement in the stopband bandwidth, four considering all constraints was missing. Here, we implement a
open-end resonators are added in the lowpass filter as shown in feed-forward neural network with back-propagation learning algo-
Fig. 5(a). The frequency response of initial filter is illustrated in rithm for this purpose.
Fig. 5(b). Dimensions of the open stubs are as follows:
b23 = 5.3 mm, b24 = 3.7 mm, b25 = 0.1 mm, b26 = 1 mm,
b27 = 0.5 mm, b28 = 3.6 mm, b29 = 0.5 mm, b30 = 4.1 mm, 3.1. Artificial neural network trained by Bayesian Regularization
b31 = 0.1 mm and b32 = 2.3 mm.
In spite of all explained modifications, the frequency response A Non-Linear Autoregressive eXogenous (NARX) model is struc-
of the proposed filter is not still optimal, so the new computational tured by a nonlinear autoregressive with exogenous inputs. This
intelligence method is introduced and used to improve the overall model is a useful method for developing soft computing
performance of the filter. approaches. The current value of a NARX model is composed of:

Fig. 4. (a) Lowpass structure composed of Resonator 1 and Resonator 2. (b) Simulation results of Resonator 1 and Resonator 2. The dimensions are as follows: b16 = 2.4 mm,
b17 = 2.5 mm, b18 = 1.1 mm, b19 = 1.5 mm, b20 = 3.4, b21 = 0.1, b22 = 5.1 mm.
 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847 5

 Past quantities of the signal under study. The function F demonstrates ANNs trained by Bayesian Regular-
 Current and past quantities of the exogenous including the ization Back-Propagation (BRBP) [26]. The combination of NARX
same input signal. model with ANN trained by BRBP leads to a very powerful tool
for estimation and prediction of complex functions. However, there
Moreover, NARX model includes another term: is no devised method to achieve the optimal structure of an ANNs,
as it mainly depends on the type of the system’s input, desired tar-
 Error gets and the initial weights. ANNs are designed with several layers:
the input layer where the experimental data is applied, the hidden
A NARX model can be introduced algebraically as shown in (5): layer(s) and the output layer. In a feed-forward ANN, All the neu-
rons of a layer are linked to other layers, with an interconnected
wt ¼ Fðwt1 ; wt2 ; wt3 ;    ; /t ; /t1 ; /t1 Þ þ Errort ð3Þ weight that is set up in order to decrease the error between the tar-
get and the input data. The ANN output can be indicated by (4),
where w illustrates the quantities of signal estimated by the net- !!
X
LH X
LI
work, and / indicates the quantities of the input signal extracted F ¼ n1 Hri  n2 Hji  tj ð4Þ
from the system to predict W. Furthermore, Errort shows the value i¼0 j¼0

of error term.

Fig. 5. (a) Initial lowpass filter. (b) S-parameters of the Proposed LPF.
6 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

where n1 and n2 demonstrate activation functions of the network; of fitting. According to (5) the amount of error BR algorithm
Hri and Hji shows the matrix of the connection weights of the input decrease substantially to minimum.
and output layers, respectively; LH and LI are the number of hidden
and input nodes, respectively. There are a number of issues which
include the choice of the optimum number of hidden layer’s artifi- 3.2. Proposed algorithm
cial neurons, the haphazardly chosen initiative weight and the over-
fitting phenomenon. The Bayesian regularization (BR) was put for- Fig. 6 illustrates the flowchart of our neural-based approach for
ward to decline of the over-fitting phenomenon with calculation optimizing the presented Microstrip LPF. As is observed, a sub-
of term DðHÞ to the error function Error ðHÞ in order to adjust the optimal microwave component, a lowpass filter here, is analysed
model’s complication. In fact, Bayes’ theorem specifies the amount through computational methods and its scattering parameters
of weights and regularizes them based on the correlation of connec- are imported into the neural based algorithm for further process-
tions between the diverse layers. The presented explanation is ing. The dataset is arranged in two categories; the physical param-
reflected in (5). eters along with the associated S-parameters, and the ideal
response of the microwave component. Then, the most influential
CðHÞ ¼ ErrorðHÞ þ kDðHÞ ð5Þ physical parameters of the component are chosen to be optimised,
where which defines the dimensions of the optimisation problem. It
should be noted that the convergence of the algorithm can be
1Xm
improved by increasing the problem dimensions at a price of more
DðHÞ ¼ Hi
2 i complexity. To avoid such complexity, which may overshadow the
presented new technique, only the capacitance effect, created by
where DðHÞ depicts over-fitting phenomenon by determining large the T-shaped resonators, is considered as the most influential
quantities for the weight variable Hi . The variable k adjusts the level effect in the filter’s response around the cut-off frequency. There-

Fig. 6. The flowchart of neural-based approach.

 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847 7

fore, the length of b3 is optimized to minimize the deviation feedback turns back the past quantities of the output to increase
between the ideal and current response, so that a new structure the level of accuracy for the estimation of the nonlinear functions.
with an updated parameter will replace its pervious iteration to Moreover, a Bayesian Regularization training algorithm is used for
find an optimal structure. training the network. This feedback improves substantially accu-
racy in ANN for analysing complex functions although it can lead
3.3. Implementation of the approach to reducing the processing speed of ANN and increasing its compli-
cation. Therefore, BRBP algorithm is used to support the feedback.
Fig. 7 shows the change of T-shaped resonator’s length and the New arrangement is called ANN-NARX-BRBP method, which is
magnitude of S (2, 1) specified as two inputs for the network. implemented in MATLAB. In this method, the outputs at each
According to Fig. 7, there are 7 levels for resonator’s length and 7 moment is estimated simultaneously by the inputs at the same
periods of S (2, 1) that the number of levels depends on the type moment, the nonlinear function of past inputs and the nonlinear
of the microstrip component and its design. Overall, it can change function of past outputs is trained by BRBP algorithm. Not only
between 1 level and N levels. Fig. 8 demonstrates the ideal has the presented method required much less memory but also
response of S (2, 1) determined as the target for ANN. As is the accuracy of analysing complex systems has increased
observed in Fig. 8, this response is defined as a LPF’s response with considerably.
an attenuation of 100 dB over the whole stopband. Table 1 gives information about initial parameters for setting up
The block diagram of the ANN-NARX-BRBP method generalized the ANN-NARX-BRBP method in MATLAB. According to this table, a
for the presented filter is shown in Fig. 8. As can be seen in Fig. 9, NARX-BRBP network is organized with 500 epochs for training,
there is a feed-forward ANN with 3 inputs, 3 layers and 1 output, performance gradient 1
105 and peak validation failures 0. After
where the inputs are length (mm), magnitude (dB) and the feed- designing the structure of the introduced network with MATLAB,
back of output. According to (3), a dynamic ANN based on NARX the length of resonator, the magnitude of the magnitude of
needs the past quantities of the output and also the current and S (2, 1) and the past outputs are applied to the network. The
past quantities of the exogenous. Hence the third input is con- network is then trained based on the given data. Next, network
nected to the output to realize the proposed structure. Indeed, this simulates every possible structure of the filter in order to reach

Fig. 7. Two input signals for ANN.

Fig. 8. Ideal response of S (1, 2) as the target for ANN.

8 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

Table 1 the most optimal model has the least error among others. In this
Initial parameters of the Neural based method in MATLAB. case study, Fig. 9 depicts the performance of ANN-NARX-BRBP
Initial Parameters Settings method for the presented LPF including training targets, training
Peak number of epochs for training 500 outputs, validation targets, validation outputs, test targets, test
Epochs among displays 40 outputs and the filter’s frequency response. Fig. 10(a) shows the
Produce command-line output false output of the network for all 7 periods, and Fig. 10(b) demonstrate
Proficiency aim 0 the status of the network outputs for the 6th period. Generally, it
Peak time to train in seconds Infinity
Peak performance gradient 1
105
can be concluded that the presented approach has modelled 7
Peak validation failures 10 structures for the LPF with different T-shape resonators. Further-
Defining weight for second derivative estimation 1
105 more, Fig. 11 (a) illustrates the error model between the target
Parameter for regulating the indefiniteness of the Hessian 1
105 and the output of the network associated with the output of
ANN-NARX-BRBP method in Fig. 10 for all 7 periods, and Fig. 11
(b) shows the error of model for the 6th period.
The proposed ANN delivered the best value for b3, which is
the best structure in the aspect of size and frequency response with 4.2 mm. The new structure of filter, with b3 equals 4.2 mm, is
changing the length of resonator. After every training stage, the shown in Fig. 12(a). Frequency response of the new structure of fil-
structure and settings of network are updated. Periodically, the ter with b3 of 4.2 mm and pervious structure of filter b3 of 2.2 mm
ideal frequency response and the simulated response of the under- are compared in Fig. 12(b).
study filter containing the harmonics and noise are compared, and The proposed filter was fabricated on a Rogers (RO4003) sub-
obviously there is definite link between reduction of errors and the strate with a thickness of 0.406 mm, relative permittivity of 3.38
performance of the filter to select a the most proper filter. and a loss tangent of 0.0022. A photograph of the fabricated proto-
type is shown in Fig. 13(a). The simulation and measurement
3.4. Selecting the best structure results are displayed in Fig. 13. All simulations were done by
Advanced Design System (ADS) Momentum. It can be seen from
Achieving the optimal structure for microwave components is Fig. 13(b) that the cut-off frequency of the proposed filter is
the goal of the proposed ANN-based optimisation technique. To 2.087 GHz, and there are two deep transmission zeroes located at
do so, the most efficient way is a comparison between the ideal 2.416 and 2.972 GHz, with 40.16 and 42.74 dB attenuation
response and the actual output response of the component. In fact, points, respectively. The filter has an ultrawide stopband from

Inputs of ANN
Input #1

Filter
Input #2

Input Layer Hidden Layer Output Layer

Input
#1
Input
#2

Input output
#3

Outputs (s(1, 2) )

Bayesian Regularization Algorithm

Artificial Neural Network trained by Bayesian Regularization Target (Ideal response)

Fig. 9. Block diagram of the ANN-NARX-BRBP method with its inputs and outputs.
 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847 9

Fig. 10. (a) The output of ANN-NARX-BRBP method for presented filter for all 7 periods. (b) The output of ANN-NARX-BRBP method for presented filter for 6th period.

Fig. 11. (a) The error of model for 7 times training. (b) The error of model for the 6th period.

2.399 GHz to 15.01 GHz with attenuation lever better than 22 dB. [3,7,10]. The proposed filter has achieved a high FOM of 21,183
Maximum insertion loss in the passband is less than 0.4 dB, while which verifies its high performance.
the return loss is better than 12.6 dB in this frequency range. The The FOM is calculated by [3]
transition band corresponding to 3 to 40 dB attenuation points
f
RSB
SF
is only 0.312 GHz, showing a very good selectivity. Moreover, the FOM ¼ ð6Þ
NCS
AF
physical size of the filter is only 16:7 mm
8:3 mm, which corre-
sponds to 0:190kg
0:094kg , where kg is the guided wavelength at where f is Roll-off rate, RSB is the relative stopband bandwidth, and
3 dB cut-off frequency. Table 2 shows a performance comparison NCS is normalized size of the filter, calculated by (7)–(9), respec-
between the presented LPF and some other newly published tively [3,7].
works. The comparison is based on a Figure-Of-Merit (FOM), amax  amin
yielded by (5), recommended in the literature for Microstrip LPFs f¼ dB=GHz ð7Þ
fs  fc
10 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

Fig. 12. (a) Layout of the proposed LPF optimized by the Neural-based approach. (b) S-parameters of the proposed LPF.

In (7) amax and amin are the 40 dB and 3 dB attenuation points, by 10. Architecture Factor (AF) is the number of etched layers in
respectively. fs and fc are frequencies with 40 dB and 3 dB attenu- the filter’s configuration, representing circuit complexity.
ation points [3,7]. The relative stopband bandwidth (RSB) is calcu-
lated by
4. Conclusion
stopband bandwidth
RSB ¼ ð8Þ A new symmetric Microstip lowpass filter composed of multiple
stopband centre frequency
open stub, high-low impedance and rectangular resonators has
been designed and optimized using a customized neural network
physical size ðlenght
widthÞ algorithm. According to the simulated and measured results, the
NCS ¼ ð9Þ
kg 2 neural network multilayer with back-propagation algorithm has
been successful in optimizing the lowpass filter. The optimized
where kg is the guided wavelength at the 3 dB cut-off frequency. lowpass filter has a sharp roll-off (118.6 GHz/dB) and a high rejec-
The suppression factor (SF) is the stopband rejection level divided tion level of 22 dB over its ultra-wide stopband. The stopband
 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847 11

Appendix A. Supplementary material

Supplementary data to this article can be found online at

https://doi.org/10.1016/j.aeue.2019.152847.

References

[1] Hong J-SG, Lancaster MJ. Microstrip filters for RF/microwave applications. New
York: John Wiley & Sons; 2004.
[2] Hookari M, Karimi G. Design of microstrip filter using modified T-shaped with
wide stopband. Int J Eng Technol Sci 2014;2:138–45.
[3] Li J-L, Qu S-W, Xue Q. Compact microstrip lowpass filter with sharp roll-off and
wide stop-band. Electron Lett 2009;45:110–1.
[4] Lotfi S, Makki S, Majidifar S. Design of microstrip lowpass filter with ultra-wide
stopband and sharp rejection. Int J Eng Technol Sci 2014;2:47–57.
[5] Wei F, Chen L, Shi X-W. Compact lowpass filter based on coupled-line hairpin
unit. Electron Lett 2012;48:379–81.
[6] Wei X, Wang P, Liu M, Shi Y. Compact wide-stopband lowpass filter using
stepped impedance hairpin resonator with radial stubs. Electron Lett
2011;47:862–3.
[7] Mirzaee M, Virdee B. Realisation of highly compact planar lowpass filter for
UWB RFID applications. Electron Lett 2013;49:1396–8.
[8] Ma K, Yeo KS. New ultra-wide stopband low-pass filter using transformed
radial stubs. IEEE Trans Microw Theory Tech 2011;59:604–11.
[9] Karimi G et al. Analysis of novel approach to design of ultra-wide stopband
microstrip low-pass filter using modified U-shaped resonator. ETRI J 2015;37
(5):945–50.
[10] Karimi G, Lalbakhsh A, Siahkamari H. Design of sharp roll-off lowpass filter
with ultra wide stopband. IEEE Microwave Wirel Compon Lett 2013;23:303–5.
[11] Chen F-C, Li R-S, Chu Q-X. Ultra-wide stopband low-pass filter using multiple
transmission zeros. IEEE Access 2017;5:6437–43.
[12] Sheikhi A, Alipour A, Hemesi H. Design of microstrip wide stopband lowpass
filter with lumped equivalent circuit. Electron Lett 2017;53:1416–8.
[13] Lalbakhsh A, Neyestanak AAL, Naser-Moghaddasi M. Microstrip hairpin
bandpass filter using modified Minkowski fractal-shape for suppression of
second harmonic. IEICE Trans Electron 2012;95:378–81.
Fig. 13. (a) A photo of the fabricated filter. (b) Simulation and measurement results [14] Ahmadi A et al. A novel dual-mode wideband band pass filter. ACES 2014;29:9.
of the fabricated prototype. [15] Jiang S, Xu J. Compact microstrip lowpass filter with ultra-wide stopband
based on dual-plane structure. Electron Lett 2017;53:607–9.
[16] Lotfi-Neyestanak AA et al. Improved microstrip hairpin-line bandpass filters
for spurious response suppression. Electron Lett 2012 Jul 5;48(14):858–9.
[17] Cao H, Guan W, He S, Yang L. Compact lowpass filter with high selectivity
using G-shaped defected microstrip structure. Progress Electromagn Res
Table 2
2012;33:55–62.
Performance comparison with some of the published LPFs.
[18] Boutejdar A, Ali WAE. Improvement of compactness of low pass filter using
Ref., Year RO RSB SF AF NCS FOM new quasi-Yagi-DGS-resonator and multilayer-technique. Progress
Electromagn Res 2016;69:115–24.
[5], 2012 43.9 1.636 1 1 0.101
0.15 4740 [19] Karimi Gholamreza et al. A new microstrip coupling system for realization of a
[7], 2013 58.6 1.43 2.5 1 0.22
0.11 8657 differential dual-band bandpass filter. AEU – Int J Electron Commun
[6], 2011 30 1.25 1.5 1 0.08
0.08 8789 2019;99:186–92.
[8], 2011 62 1.72 3 1 0.310
240 4430 [20] Ghaderi Amirhossein, Golestanifar Alireza, Shama Farzin. Microstrip bandpass
[4], 2014 80.43 1.64 2.3 1 0.330
0.180 5108 filters using coupled feed lines for third and fourth generation
[2], 2014 105 1.47 2 1 0.314
0.109 9019 communications. AEU – Int J Electron Commun 2018;86:195–201.
[12], 2017 48.5 1.61 2 1 0.141
0.151 7340 [21] Shuai Liu, Jun Xu, Zhitao Xu. Compact lowpass filter with wide stopband using
stepped impedance hairpin units. Electron Lett 2015;51(1):67–9.
[15], 2017 56.7 1.64 2 2 0.115
0.116 6970
[22] Zhang T, Cai Z, Yang Y, Bao J, Wang Y. Compact tunable lowpass filter with
[11], 2017 81 1.74 2.2 1 0.176
0.102 17,270
sharp roll-off and low insertion loss. Microw Opt Technol Lett. 2017;59
[17], 2012 189 0.98 2.5 1 0.375
0.256 4823
(10):2619–23.
[18], 2016 189 1.125 2 2 0.45
0.35 1350 [23] Liu C-Y, Jiang T, Li Y-S. A novel UWB filter with notch-band characteristic using
[21], 2015 52.8 1.529 2 1 0.081
0.113 17,640 radial-UIR/SIR loaded stub resonators. J Electromagn Waves Appl 2011;25(2–
[22], 2017 73 0.5 2 1 0.005 14,600 3):233–45.
Our work 118.6 1.45 2.2 1 0.190
0.094 21,183 [24] Liu C, Jiang T, Li Y, Zhang J. A novel UWB filter with WLAN and RFID stop-band
rejection characteristic using tri-stage radial loaded stub resonators. ACES J –
Appl Comput Electromagn Soc 2012;27(9):749.
[25] Li Yingsong, Li Wenxing, Ye Qiubo. A reconfigurable triple-notch-band antenna
integrated with defected microstrip structure band-stop filter for ultra-
wideband cognitive radio applications. Int J Antennas Propag 2013.
[26] Lalbakhsh A, Karimi G, Sabaghi F. Triple mode spiral wideband bandpass filter
bandwidth is 145%, and the passband return loss is better than
using symmetric dual-line coupling. Electron Lett 2017;53(12):795–7.
12.6 dB. The filter is compact with size of 0:190kg
0:094kg , where [27] Li Y, Li W, Liu C, Qiubo Ye. A compact UWB band-pass filter with ultra-narrow
kg is the guided wavelength at 3 dB cut-off frequency (2.078 GHz). tri-notch-band characteristic. Appl Comput Electromagn Soc J 2014;29(2).
[28] Li Yingsong, Li Wenxing, Wenhua Yu, Chengyuan Liu. A miniaturization band-
The proposed filter has a high Figure of Merit of 21183, reflecting
pass filter with ultra-narrow multi-notch-band characteristic for ultra-
its high performance. Considering all the desired characteristics wideband communication applications. Appl Comput Electromagn Soc J
of the presented filter, it is a good candidate to be used in modern 2014;294.
[29] Lalbakhsh A, Afzal MU, Esselle KP, Smith SL. Wideband near-field correction of
communication system.
a Fabry-Perot resonator antenna. IEEE Trans Antennas Propag 2019.
[30] Ranjan Prakash, Choubey Arvind. Santosh Kumar Mahto, A novel approach for
optimal design of multilayer wideband microwave absorber using wind driven
Declaration of Competing Interest optimization technique. AEU – Int J Electron Commun 2018;83:81–7.
[31] Lalbakhsh A, Afzal MU, Esselle KP. Multiobjective particle swarm optimization
to design a time-delay equalizer metasurface for an electromagnetic band-gap
The authors declared that there is no conflict of interest. resonator antenna. IEEE Antennas Wirel Propag Lett 2017;16:912–5.
12 Mohammad (Behdad) Jamshidi et al. / Int. J. Electron. Commun. (AEÜ) 110 (2019) 152847

[32] Kotha Srinivasa Reddy. Subhendu Kumar Sahoo, An approach for FIR filter [35] Sun Geng, Liu Yanheng, Liang Shuang, Wang Aimin, Zhang Ying. Beam pattern
coefficient optimization using differential evolution algorithm. AEU – Int J design of circular antenna array via efficient biogeography-based
Electron Commun 2015;69(1):101–8. optimization. AEU – Int J Electron Commun 2017;79:275–85.
[33] Dong Jian, Li Qianqian, Deng Lianwen. Fast multi-objective optimization of [36] Lalbakhsh A, Afzal MU, Esselle KP. Simulation-driven particle swarm
multi-parameter antenna structures based on improved MOEA/D with optimization of spatial phase shifters. In: International conference on
surrogate-assisted model. AEU – Int J Electron Commun 2017;72:192–9. electromagnetics in advanced applications (ICEAA). p. 428–30.
[34] Lalbakhsh A, Afzal MU, Esselle KP, Zeb BA. Multi-objective particle swarm
optimization for the realization of a low profile bandpass frequency selective
surface. In: International symposium antennas and propagation (ISAP). p. 1–4.