Keywords

1 Introduction

With the rapid development of information technology, breakthroughs have been made in many fields that eliminate human factors, resulting in time and cost advantages. In recent years, several artificial neural networks (ANN) have been applied to improve antenna design [1,2,3,4]. An ANN model was applied to design a circular microstrip patch antenna [1]. ANN was used to develop an elliptical microstrip patch antenna with high gain and impedance matching [2]. ANN model was applied to create an antenna in the X-Ku band [3]. Using the Levenberg Marquardt algorithm, the ANN model was applied to establish a flexible wideband antenna with an error rate of 6% for WLAN, 5G, and WiMAX applications [4].

The ANN model was trained with different notch antenna data, taking the antenna dimensions and frequency as input and outputs. The notch antenna frequency was calculated correctly without simulation [5]. Different ANN algorithms were tested and compared for developing high-accuracy antenna parameters [6]. The reduction of mutual interaction between antenna cells in a MIMO antenna was investigated with the help of ANN modeling [7]. The ANN-based hybrid fractal antenna was developed for biomedical applications [8, 9]. ANN techniques have been investigated for designing metamaterial antennas [10,11,12,13].

This study investigates the feasibility of ANN in dipole antenna to establish antenna design parameters. The proposed ANN model includes two hidden layers, each containing ten neurons. The feedforward backpropagation based on the scaled conjugate gradient (SCG) algorithm is applied to develop the ANN model. The ANN-based dipole antenna has a working frequency range of 1.6–2.2 GHz. Both simulation and experimental validations were conducted to test the proposed ANN-based dipole antenna.

2 Dipole Antenna Design

This study investigates a dipole antenna with a target operating frequency range of 1.6 and 2.2 GHz. Finite integration technique-based microwave simulation program is selected for antenna design and simulation. The dipole antenna has 35 mm in length and 1 mm in thickness. Figure 1a shows the proposed dipole antenna with a target bending configuration, and Fig. 1b shows the radiation pattern with the proposed antenna. The parameters of antenna gain, return loss (S11), and resonance frequencies with different configurations have been investigated to establish length position (L) and angle (θ) using COMSOL tool. The length positions are changed from 5 to 30 mm, angles are shifted from 0° to 90° with 150 intervals. A total of 42 configurations have been investigated. Frequency, S11, and gain values corresponding to bending positions and angles were applied for ANN training using MATLAB tool.

Fig. 1
A 2-part illustration labeled a and b. A has a line configuration with tilted sides labeled theta, horizontal midlines labeled L, and a curve in the center. B has a heat map with a maximum coverage of negative 2.36 to 0.271.

a Dipole antenna, b simulated radiation pattern

Full size image

3 Artificial Neural Network Modelling

The SCG algorithm is used for developing ANN structure because SCG offers minor errors. The ANN model consists of two hidden layers, and each hidden layer contains two neurons. Figure 2 shows the structure of the proposed ANN model. Table 1 lists example input and out parameters of the ANN network. The bending length position and angle are the output parameters of the proposed dipole antenna. In this study, the authors select 63.6, 18.2, and 18.2% of 66 datasets for training, testing, and validating the ANN-based antenna using MATLAB tool.

Fig. 2
A bipartite map labeled a connects hidden layer 1 with f, gain, and S 11 to hidden layer 2 with L and theta. The model in b exhibits 2 hidden layers with w, b, and a curve with a similar output layer. C is a table listing data for training, validation, and testing.

a ANN structure, b ANN hidden layer model, c training performance

Full size image
Table 1 Example input and output parameters of ANN model
Full size table

Table 2 compares the ANN output parameters with desired parameters. The average error of frequency, S11, and gain are 2.76, 13.75, and 3.505%. The results show that the proposed ANN could establish design parameters (bending length position, and angle) of the dipole antenna.

Table 2 ANN output parameters with error rates
Full size table

Figure 3 shows the simulated S parameters of the proposed antenna with different bending length positions and angles. Figure 4 displays the simulated radiation pattern of dipole antenna with different configurations. The proposed ANN model examines antenna configurations covering bending length positions and angles.

Fig. 3
Four-line graphs plot reflections S 11 versus frequency. It has decreasing and steady trends in a and b, respectively. C and d have an increasing trend. On the right, lines exhibit upward open rectangular configurations that change to a horizontal line from a to d.

S parameters of dipole antenna using different configurations

Full size image
Fig. 4
A radial graph titled farfield directivity A b s plots theta slash degree versus d B i. Farfield f = 1.954 underscore test underscore a (180, 1) has the highest estimated value.

Radiation pattern of dipole antenna with different configurations

Full size image

4 Experimental Validation

The proposed dipole antenna is constructed using copper strips, and various experimental validations are conducted to validate the proposed antenna. The Agilent VNA, with a frequency of up to 43.5 GHz, is used to measure complex S11 characteristics. The measurement calibrations are performed using open, short, and 50 Ω loaded cases in free space. Figure 5a, b show the experimental measurement setups corresponding to the configurations shown in Fig. 3. Figure 5c displays measured return loss (S11) parameters of the proposed antenna with different bending angles. The measured antenna performance results are close to the simulation results. The results show that the proposed ANN model can potentially establish design parameters of dipole antenna, especially in bending position and angle.

Fig. 5
A 3-part illustration labeled a, b, and c. A and b, each has a cylindrical object with a thin horizontal wire on its top. The wire bends from the sides and forms a U-shaped structure in b. C has a line graph plotting S 11 versus frequency. Theta 0, theta 30, and theta 60 have increasing trends.

a, b Experimental measurement setups, c measured S11 parameters

Full size image

5 Conclusions

In this study, bending length positions and angles of dipole antenna were investigated using the ANN model. The SCG algorithm was used to develop the ANN model containing two hidden layers, each consisting of ten neurons. The proposed ANN model was applied to establish dipole antenna design parameters, especially the bending length and angle values. Both simulation and experimental results showed that the proposed ANN model has the potential to become a helpful tool in designing antennas in a cost-effective and fast manner.