The Pathway to Adaptive Lightweight AI Transceivers (Vision Paper)

Nimrod Glazer, Nir Shlezinger, Tirza Routtenberg

Keywords: AI transcievers, online learning

TL;DR: This paper advocates a communication-oriented AI framework enabling flexible, efficient, and autonomous deep transceivers for next-generation wireless systems.

Abstract:

The growing complexity of modern wireless communication systems poses significant challenges to traditional receiver designs.
Artificial intelligence (AI), and in particular deep learning, offers the potential to address these challenges by learning to infer from data without relying on explicit channel models. However, direct application of conventional AI techniques is ill-suited to the real-time, data-scarce, and hardware-constrained nature of wireless environments. This vision paper advocates shifting towards a communication-oriented AI framework as the foundation for autonomous and lightweight deep transceivers, explicitly designed for the sub-millisecond adaptation, minimal energy budgets, and high-reliability constraints of next-generation networks. Our approach highlights the importance of modular and Bayesian neural architectures, combined with rapid training techniques based on continual learning, asynchronous adaptation, and communication-aware data acquisition. Together, these elements pave the way toward future AI-empowered physical-layer systems that can operate efficiently, reliably, and autonomously in real time for next-generation wireless networks.

Submission Number: 28