In this research, we investigate the integration of machine learning techniques, in particular Gaussian Process Regression (GPR) and Support Vector Regression (SVR), into the optimization of compact microstrip antenna design. Multiband operation with a significant miniaturization is achieved by proposing a unique circular radiating structure with decorative slots and a central star shaped patch. GPR and SVR models were used to predict and optimize critical antenna parameters such as resonant frequency, slot dimensions and patch dimensions. GPR gave better prediction accuracy with an MSE of 0.15, a score of 0.98 and takes longer wall time to converge, while compared to SVR model it converged faster with an MSE of 0.20, and a score of 0.95. The results were validated by close agreement between simulated and measured results, and the optimized design exhibited multiband performance across VHF, UHF, L, S, and C bands. These findings show that machine learning can offer a scalable and efficient alternative to the traditional methods in antenna design. With this approach, it is possible to lower the level of computational effort needed in traditional design methods.
The need for compact, efficient and high-performance antennas has increased rapidly with the ever-growing demand for wireless communication technologies. However, microstrip antennas have been widely studied due to their inherent advantages such as low profile, lightweight, easy fabrication and compatibility with modern electronic circuits. These attributes make microstrip antennas highly desirable for applications in mobile communications, satellite systems and wearable devices However, the design of such high performance compact microstrip antennas (e.g., wide bandwidth, high gain, multiband operation) is difficult because of the complex interdependencies between geometrical and material parameters.
In the recent years, Machine Learning (ML) has been shown as a new innovative tool for antenna design, owing to the development of computational techniques. In contrast, ML models and optimizes nonlinear high dimensional systems to reduce computational burden of the traditional trial and error simulation and offer predictive insights. A particularly powerful ML technique for antenna design has been GPR, a non-parametric Bayesian approach. This can handle small datasets and quantifies uncertainty in predicting critical parameters such as resonant frequency, bandwidth and radiation efficiency.
In this research, a novel GPR based compact microstrip antenna design is proposed, optimizing its key physical and electrical parameters. A distinctive circular radiating structure with decorative slots is proposed as the antenna, including a central star shaped patch. This innovative geometry is also multiband capable, supporting VHF, ULF, L, S, and C band operation. By integrating GPR into the design process, which can predict very precisely performance metrics and reduce the need for extensive electromagnetic simulations and iterative prototyping. Frequently, multiband designs are endorsed in situations where different bands are needed for services (e.g., telemetry, communication and navigation), because having multiple small bands tends to be less power-consuming and less prone to interference than a wide single band. Slots are cut into the circular panels, so the current travels different routes and produces variety in resonance. Each hole or indent in the mask helps achieve the needed multipoint tuning. Such capacitance and inductive effects come from the etched slot dimensions and their arrangement.
The objective of this work is to design and optimize the compact fractal monopole antenna that has multi band performance. Parametric optimization is done for enhanced bandwidth and radiation efficiency. Different ML algorithms are employed to predict antenna parameters thereby reducing the number of EM simulations and demonstrating that the proposed one provides efficient alternative than the conventional antenna design and optimization techniques reducing the complexity.
This paper is structured as follows: Sect. 2 introduces the modelling approach and discuss the antenna design, and the methodological framework used. Results and a detailed discussion on simulated and measured performance metrics are presented in Sect. 3. Section 4 concludes the research with a discussion of its contributions and prospects for ML driven antenna design.
Podder et al. reviewed comprehensively the application of machine learning (ML) and deep learning (DL) in the antenna design, optimization and choice. In this work, they studied how ML models can enhance the performance of antenna systems in solving complex design challenges, like multiband and MIMO antennas. In this review, the integration of DL techniques for automatic feature extraction for improved prediction and optimization in different antenna configurations was reviewed. The results show that ML and DL can significantly decrease computational costs and increase accuracy of antenna design. For multi band MIMO microstrip antennas in 5G mm Wave applications, Chbeine et al. propose an AI driven design method. The results from the study showed that ML algorithms can lead to substantial gain and efficiency improvements of the antenna geometry. The method was validated through simulations and prototypes, and it was shown that the methodology has the potential to address high frequency challenges. The designs of the next generation of antennas for the next generation communication system are shown to be possible using AI.
Yusuf et al. present the design of frequency reconfigurable compact microstrip patch antennas based on a machine learning approach, where ML algorithms were used to tune bandwidth and resonant frequency, with the optimal configurations. This methodology reduced the dependence on traditional simulation driven iterations by extensive training of data to predict antenna performance metrics with high precision. Future research in adaptive and reconfigurable antenna systems relies on the work. Raveendra also optimized the microstrip antenna arrays for 5G sub 3.5 GHz networks. We applied ML techniques to optimize sequential rotation-based MIMO configurations to increase network performance and reduce interference. These results are also useful for designing robust and efficient antenna systems for next generation communication networks in the sub 6 GHz spectrum.
Machine learning models were then applied to predict the dimensions of the rectangular patch microstrip antenna by Kurniawati. We demonstrated that critical design parameters (patch length and width) can be accurately determined from target performance metrics using ML based approaches. Using machine learning models, Jain et al. have improved performance of circular microstrip patch antennas. The study identified and optimized the critical design parameters, slot dimensions and substrate properties that maximize gain and bandwidth. The complex design spaces were shown to be successfully explored by ML, which led to better antenna performance. They also showed how ML techniques are suitable for a broad range of antenna configurations and performance requirements. Deep kernel learning is used by Shudan et al. for resonant frequency modelling of micro strip antennas. The study demonstrates the potential of advanced ML models to solve complex design challenges and achieved high accuracy and computational efficiency. Their findings demonstrate the applicability of kernel-based learning techniques to antenna parameter optimization. Mohammadi et al. designed a 1 × 4 microstrip antenna array on human thigh for gain enhancement. Although not strictly an ML topic, their novel wearable antenna design approach will synergize with ML methods by providing novel design perspectives for healthcare applications. The work demonstrates how advanced design techniques can be combined with ML driven frameworks.
A reconfigurable microstrip patch antenna with switchable polarization was developed by Singh et al. Both the study and the goals were to explore advanced design capabilities which could be further enhanced with ML techniques. However, their findings emphasize the possibility of combining traditional design innovations and ML to realize adaptable and high-performance antenna systems. In, an ultra-wideband antenna (UWB) based on an ML approach is proposed, and to improve the bandwidth and return loss characteristics, learning algorithms are employed and ensembled.
The study focuses on various ML models where CatBoost and XGBoost are evaluated on the design of a multi-band patch antenna suitable for IoT applications, where CatBoost achieved a prediction accuracy of 77.4% for return loss. The work used SVR and decision tree models for diagnosing antenna performance in modeling complex relationships. The authors in proposed a 28 GHz antenna using the K-nearest neighbors (KNN) random forest, where the antenna yields the accuracies as predicted above 83%, which indicates its strong suitability for mm Wave applications.
A midband 5G quasi-Yagi antenna is proposed, designed, and simulated, and along with it, an equivalent RLC circuit analysis is done to estimate gain and resonant frequency. Subsequently, the authors integrate ML models to achieve predictive performance. The highlights and advantages of ML in antenna design are presented in, which helps in the reduction of simulation time along with the enhancement of accuracy and improved convergence, making it a promising tool for next-generation antenna development.
Prior studies been on ML for estimating patch dimensions or optimizing one band, in contrast, I combine both GPR and SVR to predict the slot and patch dimensions and then test in five different bands. The fact that this model and the circular-star geometry have not been covered much in literature reveals our approach as innovative. Table 1 gives a summary of recent research of antenna design integrating Machine learning techniques/algorithms that are used in optimizing the parameters of the antenna, its focused frequency band along with its validation and its key contribution.
The design reported in this paper is the combination of compact fractal antenna which is tuned for multi band applications and fractal ground which enhances the bandwidth along with the aid of machine learning algorithms. These enables the multiband operation and helps in reducing the number of EM simulations which in turn reduces the cost and time.
EM simulations are mainly time consuming when it comes to optimization and parametric sweeps. Overcome this problems ML algorithms plays a significant role. ML algorithms provide rapid results which helps in predicting the performance directly from dataset enabling the fast exploration and reducing the cost and integration of workflow. This paper will help to show that ML can predict antennas performance parameters which will speed up when compared to repeated EM simulations.