A Meta-Survey of Deep Learning for Intelligent Communications Systems: Taxonomy, Unified Famework and Future Directions

Abstract

The rapid advancement of deep learning (DL) has fundamentally transformed intelligent communication systems, leading to a rapid proliferation of survey papers covering wireless communications, optical networks, vehicular systems, integrated sensing and communication (ISAC), and emerging 6G technologies. Although these surveys provide valuable insights within their respective domains, they remain largely fragmented, employ inconsistent taxonomies, lack systematic cross-domain comparisons, and do not provide a unified perspective on the evolution of AI-enabled communication systems. Consequently, researchers face increasing difficulties in identifying common design principles, evaluating methodological trends, and understanding how different communication domains are converging toward AI-native networking. To overcome these limitations, this paper presents a comprehensive meta-survey that systematically analyzes, compares, and synthesizes existing survey literature on DL for intelligent communication systems. Specifically, the proposed meta-survey (i) establishes a unified taxonomy spanning communication domains, learning paradigms, network layers, and DL architectures, (ii) introduces a unified AI-driven communication pipeline that maps representative solutions from diverse communication domains into a common framework, (iii) performs a comprehensive cross-domain comparative analysis to identify methodological strengths, research trends, technical challenges, and remaining gaps, and (iv) provides a technology-oriented roadmap highlighting future research directions and maturity levels toward AI-native communication systems. By integrating these complementary perspectives, this work offers a holistic reference that facilitates knowledge transfer across communication domains and supports the design of next-generation AI-native communication networks.