Neural network-assisted design and optimization of multiple-input multiple-output THz antennas with multilayer substrates for space and defense applications

Abstract TeraHertz (THz) frequency is fast emerging as the key enabler for next-generation communication systems, demanding ultra-high bandwidth and data rates in line with advanced wireless, space, and defense applications. This research tackles primary problems faced in traditional THz antenna design, including high mutual coupling, low radiation efficiency, and time-consuming manual optimization. A compact, wrench-shaped patch antenna of 47.13 × 39.9 × 16.1 μm 3 is designed employing a multilayer substrate for gain, efficiency, and bandwidth enhancement. The single patch design is extended to a 2-element multiple-input multiple-output (MIMO) antenna, further enhanced by integrating photonic band gap (PBG) structures. The proposed MIMO antenna with a PBG secures a maximum gain of 7.79 dBi, a 2:1 voltage standing wave ratio of 6.7 THz, and a radiation efficiency of 86.67% at 6.9 THz. MIMO-specific metrics are within optimum limits, while the total active reflection coefficient is optimized at a 180° phase angle with isolation greater than 15 decibels. A deep Q network-based optimization framework is integrated for a rapid design process. Furthermore, two multipath propagation scenarios are assessed in order to show the performance of the antenna across diverse environments. Thus, the proposed approach offers a scalable, high-performance solution for future high-speed communication systems.