Machine Learning Assisted Design and Optimization of a Dual-Port SIW Antenna with LP-to-CP Conversion for sub-6 GHz 5G Communication System

Wireless communication relies on invisible waves that carry our data through the air. To make these connections faster and more reliable, engineers often use multiple antennas working together, a system known as MIMO. However, when these antennas are placed close to each other, they tend to interfere with one another, much like two people trying to speak in the same small room. This interference, or "coupling," weakens the signal. Another challenge is the orientation of the waves; if a transmitting antenna sends a wave spinning in one direction and a receiving antenna expects it spinning the other way, the connection fails. To solve this, engineers design antennas that can handle circular polarization, where the wave spins in a circle, making the connection robust regardless of how the device is held. The goal is to build a compact, high-performance antenna that avoids interference and handles these spinning waves efficiently, all while being designed quickly enough to keep pace with the rapid demands of modern 5G networks.

In a recent study, researchers at the National Institute of Technology in India tackled these challenges by creating a new type of antenna for the sub-6 GHz 5G spectrum, the frequency range that powers most current mobile networks. They combined two advanced concepts: a substrate integrated waveguide, which is a flat, shielded channel that guides electromagnetic waves with very little loss, and a special structure that converts linear waves into circular ones. The result is a dual-port antenna, meaning it has two separate entry points for signals, designed to operate together without interfering. To ensure the design was perfect, the team did not rely solely on traditional, time-consuming computer simulations. Instead, they employed machine learning, a branch of artificial intelligence, to act as a fast, intelligent guide through the design process.

The researchers began by building a physical model of the antenna on a standard circuit board material. The core of the device is a rectangular cavity lined with tiny metal posts, which traps the energy and directs it efficiently. On top of this cavity, they suspended a thin layer containing a patterned surface, known as a metasurface. This layer acts as a converter, taking the straight-line vibration of the incoming signal and twisting it into the circular motion required for robust communication. The team carefully adjusted the size of the cavity, the spacing of the metal posts, and the dimensions of the slots cut into the board. They tested how these changes affected the antenna's ability to accept signals and reject interference between its two ports.

To speed up the discovery of the best possible shape, the team generated a massive library of data by running thousands of computer simulations, each with slightly different dimensions. They then trained two different machine learning algorithms, one based on a simulated brain structure and another that builds decision trees, to learn the relationship between the antenna's physical shape and its performance. These algorithms learned to predict how the antenna would behave without needing to run a full, slow simulation every time. The study found that the tree-based algorithm was particularly accurate, predicting the antenna's behavior with a high degree of confidence. This allowed the researchers to quickly identify the optimal dimensions for the final design.

The final, optimized antenna performed exceptionally well in both computer models and physical tests. It operates effectively across a frequency range from 4.7 to 5.25 GHz, which covers the essential sub-6 GHz 5G spectrum. Within this range, the antenna successfully converts linear signals into circular ones, maintaining a stable, spinning wave from 4.9 to 5.18 GHz. Crucially, the two ports of the antenna remained almost completely independent of each other, with interference levels dropping below negative 40 decibels, a figure that indicates a very clean separation of signals. The device also achieved a peak gain of 9.5 decibels, meaning it focuses the signal strongly in the desired direction. When the researchers built a physical prototype and tested it with a vector network analyzer, the real-world measurements matched the computer predictions closely, confirming that the design works as intended.

This work demonstrates that combining advanced antenna physics with machine learning can produce compact, high-gain solutions for next-generation wireless systems. By using artificial intelligence to navigate the complex trade-offs between size, bandwidth, and signal purity, the researchers created a device that is both powerful and efficient. The antenna's ability to maintain strong isolation between its ports while delivering high gain makes it a strong candidate for future 5G applications, where reliable, high-speed data transfer is critical. The study confirms that machine learning is not just a theoretical tool but a practical method for accelerating the design of complex electronic components, turning what was once a slow, iterative process into a streamlined path to innovation.