Introduction
In the present scenario, reconfigurable antennas have grabbed huge attention from the research community due to their ability to integrate multiple standards into a single platform [Reference Chandra, Satyanarayana and Battula1]. Therefore, the wireless systems are empowered with reconfigurable antennas to enhance the overall performance and providing cost-effective solutions [Reference Dwivedy, Behera and Singh2]. The important characteristics that are to be considered while designing these antennas are operating frequency, polarization, and radiation pattern [Reference Zhu, Cheung and Yuk3]. Modern-day wireless communication systems such as mobile phones, laptops, watches, tablets, etc. support several wireless standards. To fulfill such requirements, multiband, wideband and frequency reconfigurable antennas are the probable choices. Frequency reconfigurable or tunable antennas act as the best alternative by opening up new horizons, thereby providing additional functionality levels [Reference Yashchyshyn4]. These antennas can change their operating frequency while maintaining stable polarization and radiation pattern modes over the entire frequency range [Reference Iqbal, Smida, Abdulrazak, Saraereh, Mallat, Elfergani and Kim5]. This feature is accomplished by redistributing the surface current within the radiating element by employing positive intrinsic negative (PIN) diode [Reference Lim, Rahim, Hamid, Eteng and Jamlos6], varactor diode [Reference Kehn, Quevedo-Reruel and Rajo-Iglesias7], radio frequency micro-electro-mechanical system (RF-MEMS) switches [Reference Zhang, Yang, Pan, Fathy, Ghazaly and Nair8], multi-reed switches [Reference Liu, Wang, Wu, Wai and Smolders9], and optical element [Reference Pendharker, Shevgaonkar and Chandorkar10]. Numerous reconfigurable antennas have been demonstrated based on different switching techniques. In [Reference Ahmad, Arshad, Naqvi, Amin, Tenhunen and Loo11], the spiral-shaped flexible and compact frequency reconfigurable antenna is implemented for wireless applications. Ouyang et al. [Reference Ouyang, Vosoughi and Gong12] proposed a microstrip patch antenna with an electronically steerable parasitic array radiator. With the employment of switches and biasing networks, the non-linear effects and insertion loss increase, that in turn, degrades the performance of an antenna.
Over the previous decades, artificial materials have garnered prodigious attention in designing smart antenna structures with improved performance parameters [Reference Kim, Kuester, Holloway, Scher and Jarvis13]. The origination of metamaterial reveals an outstanding achievement in comparison to conventional materials. This idea first came into existence in 1967 [Reference Veselago14]. After that, the concept of metamaterial has been explored widely to provide a wide tuning range, thus offering a promising solution to various research-oriented problems. The remarkable electromagnetic properties of metamaterial such as negative refractive index, permittivity, and permeability values, anti-parallel phase and group velocities, etc. are responsible for designing fascinating antenna structures with superior characteristics [Reference Ali, Aw and Biradar15]. Ramachandran et al. [Reference Ramachandran, Faruque and Islam16] proposed a left-handed metamaterial design for satellite applications constructed using a combination of circular and square ring structures. Though the three-dimensional metamaterial is a young field and has achieved blooming technological advancements in several research areas such as microwaves and infrared regions yet they lack paucity in realizing lossless optical metamaterials. Thus to provide easy fabrication, two-dimensional metasurfaces have emerged out as a multifaceted branch of engineering science in the last decade [Reference Luo17]. The surface variant of metamaterial is portrayed as metasurface which means a surface distribution of small apertures or holes [Reference Haupt and Lanagan18]. Metasurfaces (MS) are uniformly arranged, two-dimensional planar periodic structure that is analogous to three-dimensional metamaterials [Reference Glybovski, Tretyakov, Belov, Kivshar and Simovski19]. A metasurface-based planar frequency reconfigurable antenna is designed with a fractional tuning range of 14.6% [Reference Zhu, Liu, Cheung and Yuk20]. The frequency reconfigurable slot antenna is demonstrated using a single-layer metasurface [Reference Chatterjee, Mohan and Dixit21]. Majumder et al. [Reference Majumder, Krishnamoorthy, Mukherjee and Ray22] proposed a frequency reconfigurable antenna by loading two metasurface layers onto the slot antenna. Frequency and polarization reconfigurable antenna is designed using polarization conversion metasurface [Reference Ni, Chen, Zhang and Wu23]. Metasurface enabled mechanically frequency reconfigurable antenna by an equivalent radially homogenous model is demonstrated by Li et al. [Reference Li, Man and Qi24]. Singh et al. demonstrated the designing and analysis of circular fractal antenna using artificial neural networks (ANNs) [Reference Sivia, Pharwaha and Kamal25]. Kaur et al. presented metasurface incorporated frequency reconfigurable antenna and used ANN approach to develop antenna [Reference Kaur, Sivia and Rajni26, Reference Kaur, Sivia and Rajni27].
By thoroughly reviewing the aforementioned literature, the design of frequency reconfigurable planar antenna is presented using an ANN with enhanced performance characteristics. First, the designing process of a metasurface incorporated dual-layer frequency reconfigurable antenna is illustrated. Then the specific characteristics of metamaterial are analyzed. Afterwards, the ANN framework is illustrated. Followed by this, the results are discussed. In the end, the conclusion is presented.
Designing process of frequency reconfigurable antenna
The proposed work utilizes a dual-layer approach for implementing the frequency reconfigurable antenna. The schematic of the design is illustrated in Fig. 1. The dual-layer module is composed of two substrate layers as shown in Fig. 1(a). The rectangular-shaped patch antenna and circular ground plane are imprinted on the upper and lower side of the first layer, respectively. The periodical array of compounded double split-ring resonator (CDSRR) shaped unit cells forming the metasurface is printed on the upper side of the second layer. This module uses a flexible Rogers RO4350B substrate with a dielectric constant (ɛr) of 3.48 and loss tangent (tan δ) of 0.0037. A fixed thickness of 1.524 mm has opted for both layers, thus the total thickness is 3.048 mm. For the excitation purpose, a 2 mm wide feedline is used that provides a characteristic impedance of 50 Ω. The antenna is fed directly using coaxial feed and the SubMiniaturized version A (SMA) connector is used for connecting purposes. The overall dimensions of the structure are D × 2(h) mm2. For proper matching conditions, both the layers are taken in circular form. The angle θ, the orientation angle of the metasurface, is measured with respect to the y-axis in a clockwise and anticlockwise direction and the maximum orientation angle is 90°. The schematic of the patch antenna and metasurface is demonstrated in Figs 1(b) and 1(c), respectively. The zoomed version of the unit cell is represented in Fig. 1(d). The optimized parametric design specifications of the proposed antenna are listed in Table 1.
Fig. 1. Design schematic of FRPA-MSS (a) dual-layer module (b) patch antenna (c) metasurface, and (d) zoomed version of a unit cell.
Table 1. FRPA-MSS design specifications.
Analysis of metamaterial
The effective electromagnetic parameters (permittivity ɛ eff/permeability μ eff/refractive index n eff) of the unit cell structure are extracted by utilizing the scattering parameters. For this, the whole setup is placed inside the waveguide environment with ports, electric field, and magnetic field applied along the respective axis [Reference Rajni and Marwaha28–Reference Sethi31]. This waveguide medium is shown in Fig. 2(a). The structure functions like an LC resonator with a gap and metal strip corresponds to capacitive and inductive elements, respectively. The LC equivalent circuit of the unit cell is depicted in Fig. 2(b).
Fig. 2. Analysis of CDSRR (a) waveguide medium for extracting the S-parameters and (b) equivalent circuit of the unit cell.
Figure 3 shows the extracted reflection (S11) and transmission (S21) coefficient of the unit cell structure. It is evaluated that a strong reflection of −29.2 dB is observed at 7.7 GHz. This frequency points to the coexistence of electric and magnetic resonances [Reference Rajni and Marwaha32]. The first transmission minimum is −40.7 dB at 8.2 GHz. The magnitude and phase characteristics of the reflection and transmission coefficient are delineated in Fig. 4. From these results, the phase reversal property of metamaterial is demonstrated. The real and imaginary parts of the reflection and transmission coefficient are described in Fig. 5. Further, results are imported to Matrix Laboratory (MATLAB) for determining the electromagnetic properties of metamaterial.
Fig. 3. Extracted S11 and S21 of the proposed unit cell structure.
Fig. 4. Magnitude and phase characteristics of S11 and S21.
Fig. 5. Real and imaginary parts of S11 and S21.
Different techniques have been employed so far for extracting the notable properties of metamaterial. Nicolson Ross Weir (NRW) and the transfer matrix method are popular as they are based on two-port analysis [Reference Varamini, Keshtkar, Daryasafar and Moghadasi33]. In this work, the transfer matrix method is employed as this is a direct method for determining the wave impedance. The wave impedance and refractive index are determined using (1) and (2), respectively. Here, z and n symbolize the wave impedance and refractive index, respectively. The literal k and d represent the wave number of the incident wave and thickness of the dielectric slab, respectively. The permittivity and permeability values can be attained using (3) and (4).
The material characteristics of the metamaterial are determined from the negative parts of permeability and permittivity. Thus, reflection and transmission are the integral parts of the said method. Figure 6 illustrates the effective values of homogenous parameters in the desired frequency range. The bold red line represents the real part and the dashed blue line indicates the imaginary part of the associated parameter. The positive real value of wave impedance exists from 4 to 8.5 GHz range. Similarly, the negative real value exhibited by permittivity exists in 4−8.47 GHz and permeability in 4−8.2 GHz and 8.46−12 GHz. The permittivity and permeability values show electric and magnetic plasma frequencies. The refractive index shows its negative real value in between the electric and magnetic plasma frequencies. The successful retrieval of results in the 4–8.2 GHz frequency range confirms that the aforementioned structure exhibits left-handed properties of metamaterial. This in turn shows the effectiveness of the designed approach.
Fig. 6. Effective values of homogenous parameters (a) wave impedance (b) refractive index (c) permittivity, and (d) permeability.
The analysis of the mathematical model of an equivalent circuit of CDSRR [Reference Bilotti, Toscano, Vegni, Aydin, Alici and Ozbay34] is determined using (5)–(9). The equations have been modified as a single ring has been considered in the proposed design.
where, LCDSRR and CCDSRR are the inductance and capacitance of the equivalent circuit, respectively. μ 0 and ɛ 0 are the absolute permeability and permittivity, respectively. $l_{avg}^{CDSRR}$
and $\varepsilon _r^{sub}$
are the average length of the unit cell and relative permittivity of the substrate. εr , and f represent the relative permittivity of material, and the analytical frequency, respectively. From the equations, it is evaluated that the analytical results show a frequency of 7.9 GHz that matches nearly its full wave simulated results.
Artificial neural network
It is imperative to design the metasurface unit cell precisely as the metallic loop represents inductance L, and the gap between them represents capacitance C, which in turn decides the resonating frequency of the unit cell. Thus the resonance can be controlled by changing the geometry of the structure. Therefore, an ANN has been developed for the analysis of metasurface and incorporated in antenna for reconfiguration.
ANN is a powerful data computational tool that analyzes and processes information in a similar way the human brain does. It also plays a crucial role in the designing and analysis of an antenna [Reference Kaur and Sivia35, Reference Kaur and Sivia36]. The basic unit that acts as the building block of the neural network is an artificial neuron. The feed-forward back propagation network (FFBPN) is made up of many interconnected neurons that are organized in three layers: input, hidden, and output layer [Reference Kaur and Sivia37]. The information flows from the input layer to the output layer after progressing through one or more hidden layers. Each neuron is connected to other neurons via direct communication links with an individual weight connected to each link. The difference between the desired output and network output generates an error signal. Thus, ANN can adapt, learn and recollect the information just like a biological neural network [Reference Kaur and Sivia38]. The proposed FFBPN model is elucidated in Fig. 7. This model is constructed using three input layer neurons, four output layer neurons, and 10 hidden layer neurons. The unit cell dimensions and orientation angle of a frequency reconfigurable planar antenna by incorporation of metasurface superstrate (FRPA-MSS) are considered as inputs and their resonant frequencies are considered as output. To evaluate this process, a data set consisting of 36 samples is created through parametric analysis. From this total, the training set contains 26 samples, five samples in the validation set, and five samples in the testing set.
Fig. 7. Proposed FFBPN model.
Results and discussion
The finite element method (FEM)-based 3D full-wave HFSS support tool is deployed for simulating the antenna prototype. The photographic view of the fabricated prototype is shown in Fig. 8. Anritsu MS2028C vector network analyzer (VNA) is used for evaluating the experimental results.
Fig. 8. Photographic view of the fabricated antenna using RO4350B substrate (a) patch antenna and metasurface and (b) FRPA-MSS.
ANN results
A data set containing 36 samples is created by changing the basic dimensions of unit cell and the value of rotation angle for training the ANN model. For this purpose, the input parameters considered are “c”, “d”, and “θ”, and output parameters are “fr 1”, “fr 2”, “fr 3”, and “fr 4”. The desired degree of accuracy is achieved using the Levenberg–Marquardt algorithm. The training function employed is trainlm. Tansig and purelin are the transfer functions used in the hidden and output layers, respectively. The output of the model is evaluated using three main statistical parameters mean squared error (MSE), number of epochs used, and maximum absolute error. The performance evaluation for analyzing the designed antenna is described in Table 2. The performance plot for analyzing the designed antenna is illustrated in Fig. 9. It is noticed in plot that best validation performance takes place at 0.0006273 at epoch 8.
Table 2. Performance evaluation of ANN analysis.
Fig. 9. Performance plot for analyzing the designed antenna.
Regression analysis is a statistical measure for examining and comprehending the relationship between target and output values. Its value should be close to 1 for getting a good performance results. The regression plot showing the performance analysis during training, validation, and testing is shown in Fig. 10. The total regression value of 0.99998 is obtained. The absolute error estimation for analyzing the designed antenna is shown in Table 3. It contains 16 samples that are randomly selected from the generated data set. The average absolute error estimated for fr 1, fr 2, fr 3, and fr 4 are 0.045, 0.191, 0.084, and 0.012, respectively.
Fig. 10. Regression plot for analyzing the performance of the designed antenna.
Table 3. Absolute error estimation for the analysis of the designed antenna.
Reflection coefficient
The simulated and measured values of S11 of the designed antenna are illustrated in Fig. 11. Both the results are in good accord. It is anticipated from Fig. 11 that by rotating the metasurface from 0° toward 30°, 60°, and 90° with reference to the stationary patch antenna, the resonant frequency continuously shifts up from 4.99 to 5.35, 5.93, and 6.12 GHz, respectively. The best match occurs at an orientation angle corresponding to 60°. After further rotation of the metasurface, the matching decays. Thus, the metasurface plays an integral role in achieving the frequency reconfiguration property. Table 4 shows the summarized results of the proposed antenna with respect to different orientation angles. Slight disagreement is there between the simulated and measured results. This dissimilarity is basically due to the soldering bumps, fabrication, and measurement tolerances.
Fig. 11. Simulated (solid line) and measured (dotted line) S11 of the designed antenna.
Table 4. Summarized results of the designed antenna.
Current distribution
The surface current distribution at different orientation angles is depicted in Fig. 12. It has been elucidated from these figures that the current distribution is intensively concentrated near the center of the metasurface and edges of the patch at 0° orientation angle. At 30° orientation angle, the distribution is strong across the right-portion of metasurface and left/right edge of the patch. More current flows near the extreme right part of both metasurface and patch corresponding to 60° orientation angle. In the case of 90° orientation angle, the current distribution is intensively accumulated along the center of the metasurface and edges of the patch.
Fig. 12. Surface current distribution obtained at orientation angles (a) 0° (b) 30° (c) 60°, and (d) 90°.
Gain
Gain is an important performance parameter that describes how efficiently information is to be sent or received by an antenna in a particular direction [Reference Balanis39]. The simulated and measured gain obtained at orientation angles corresponding to 0, 30, 60, and 90° [Reference Zhu, Liu, Cheung and Yuk20] is presented in Fig. 13.
Fig. 13. Simulated and measured gain at different orientation angles.
Radiation pattern
Figure 14 shows the designed antenna placed in an anechoic chamber. The 2D simulated and measured radiation characteristics of an antenna analyzed at resonant frequencies corresponding to different orientation angles are shown in Fig. 15. These patterns are defined for both the principal planes (φ = 0° and φ = 90°). At φ = 0°, the pattern exhibited by the antenna is bidirectional and φ = 90°, the radiation pattern obtained is omnidirectional in shape. The radiation pattern plots indicate that the co-polarization observed along the y-axis radiates highly as compared to the cross-polarization that is observed along the x-axis. The value of the front-to-back ratio examined is also more than 20 dB at all orientation angles. Thus, the metasurface only reconfigures the operating frequency of an antenna without much affecting the shape of the radiation pattern and polarization at different orientation angles.
Fig. 14. Designed antenna in an anechoic chamber.
Fig. 15. Simulated (solid line) and measured (dotdot line) radiation pattern of FRPA-MSS at orientation angles (a) 0° (b) 30° (c) 60° (d) 90° in x–z plane and in y–z plane (e) 0° (f) 30° (g) 60° (h) 90°.
Table 5 compares the performance parameters of the proposed work with other existing frequency reconfigurable antennas. It is apparent from Table 5 that the proposed antenna achieves 19, 88.2, 48.16, and 48.16% reduction in size as compared to [Reference Zhu, Liu, Cheung and Yuk20], [Reference Chatterjee, Mohan and Dixit21], [41] and [Reference Sivia, Pharwaha and Kamal25], respectively. On considering the bandwidth, it is observed that 50.66 and 169% hike as compared to [Reference Zhu, Liu, Cheung and Yuk20] and [Reference Zhu, Cheung, Liu, Cao and Yuk40], respectively. The comparative tuning range of the proposed antenna with other research in Table 5 indicates a 6.3 1.3, 6.3% rise in tuning range as compared to recent research in [Reference Zhu, Liu, Cheung and Yuk20], [Reference Chatterjee, Mohan and Dixit21], and [Reference Zhu, Cheung, Liu, Cao and Yuk40], respectively. It can be examined from the comparison that the proposed antenna possesses smaller dimensions with acceptable gain, wide bandwidth, and tuning range at all orientation angles.
Table 5. Comparison of proposed work with other existing frequency reconfigurable antennas.
Conclusion
In this paper, the design of FRPA-MSS using the ANN has been proposed and investigated. The compounded double split ring-shaped resonator unit cells arranged periodically forming the metasurface are mounted atop the superstrate layer for accomplishing the desired tuning range. By mechanically tuning the metasurface with respect to the reference patch antenna, frequency reconfiguration is achieved within the 4.99–6.12 GHz tuning range. A good correlation is seen between the measured results and simulated predictions. Also, the proposed antenna owns wide bandwidth and acceptable gain at all orientation angles. The developed ANN model demonstrates its utility for the prediction of resonant frequencies at different orientation angles.
Acknowledgement
The authors would like to thank Rogers Corporation for providing support material for fabricating the antenna prototype and the National Institute of Technical Teachers Training and Research (NITTTR), Chandigarh, and the Indian Institute of Technology (IIT), New Delhi for providing the facility to test the fabricated prototype.
Ms. Navneet Kaur obtained his bachelor' (B.Tech.) in electronics and instrumentation engineering from Punjab Technical University, Jalandhar and master' (M.Tech.) in electronics and communication engineering from Punjabi University Patiala, Punjab, India in 2005 and 2007, respectively. She is currently working toward the Ph.D. degree in electronics and communication engineering (ECED) from Punjabi University Patiala, Punjab, India. She has published more than 10 research papers in several international journals, international and national conferences. Her current research interests include microstrip antennas, artificial neural networks, and reconfigurable antennas in wireless communication systems.
Dr. Jagtar Singh Sivia born in Bathinda, Punjab in 1976. He has obtained his B.Tech from Punjab Technical University, Jalandhar, Punjab in electrical and electronics communication engineering in 1999 and he had obtained his master's from Punjab Technical University, Jalandhar in 2005. He obtained his Ph.D. degree from Sant Longowal Institute of Engineeering and Technology, Longowal, Punjab. Currently, he is a professor in electronics and communication engineering at Yadavindra College of Engineering, Off-Campus of Punjabi University, Patiala, Punjab, India. He has published more than 60 papers in several international journals and international and national conferences. His main interests are in neural networks, electromagnetic waves, genetic algorithms, antenna system engineering. Mr. Sivia is a member of Institution of Engineers in India, Indian Society of Technical Education and International Association of Engineers.
Dr. Rajni is currently professor at Shaheed Bhagat Singh State University (formerly known as Shaheed Bhagat Singh State Technical Campus) Ferozepur, India. She has done her B.Tech. from National Institute of Technology (formerly known as Regional Engineering College) Kurukshetra and M.E. from National Institute of Technical Teacher Training and Research (NITTTR), Chandigarh. She did her Ph.D. in metamaterial antennas from Sant Longowal Institute of Engineering and Technology, Longowal (SLIET), Sangrur, Punjab. She has approximately 23 years of academic experience. She has authored more than 80 papers in international journals, national and international conferences. Her research interests include wireless communication, antenna design and signal processing.
