TAI-Driven Smart Antenna Systems for Next-Generation Wireless Networks

Abstract

The 6th Generation (6G) of wireless technology calls for a range of capabilities that goes beyond anything ever seen before, such as data rates of a terabit per second, an extreme number of devices, and a sub-millisecond response time. In order to meet these high performance goals, the use of artificial intelligence in the context of smart antenna systems and massive multiple-input multiple-output (MIMO) architectures has been mandated to be an operational requirement instead of a potential improvement. Beamforming in millimeter-wave and terahertz frequency bands is greatly complicated, but it can be handled well by deep learning algorithms. The non-interpretability of ANNs and its high sensitivity to physical layer attackers are however serious vulnerabilities. This research physically demonstrates a framework for Trustworthy Artificial Intelligence (TAI) in the context of smart antenna systems security and optimization for next-generation environment. A true 28 GHz uniform planar array with 64 elements was deployed with a software defined radio processing backend. The framework combines the beam-alignment algorithm Deep Q-Networks with the feature-pruning algorithm Shapley Additive Explanations (SHAP) to support real-time feature pruning, and Knowledge Distillation and surrogate-loss minimization to prevent over-the-air data poisoning. Empirical measurements based on physical operations prove the TAI-driven architecture to be near-optimal spectral efficiency when user mobility is high and to achieve a 89.9% reduction in beam alignment latency. In addition, 96.8% of the live adversarial jamming attacks were detected and neutralized by the embedded security features. The results clearly establish that physical layer embedding of explainability and adversarial robustness is essential for reliable and transparent service for future 6G wireless networks.