Collaborative Spectrum Sensing in Cognitive and Intelligent Wireless Networks: An Artificial Intelligence Perspective

Artificial intelligence (AI) has become a key enabler for next-generation wireless communication systems, offering powerful tools to cope with the increasing complexity, dynamics, and heterogeneity of modern wireless environments. To illustrate the role and impact of AI in wireless communications, this paper takes collaborative spectrum sensing (CSS) in cognitive and intelligent wireless networks as a representative application and surveys recent advances from an AI perspective. We first introduce the fundamentals of CSS, including the general framework, classical detector design, and fusion strategies. Then, we present an overview of the state-of-the-art research on AI-driven CSS, classified into three categories: discriminative deep learning (DL) models, generative DL models, and deep reinforcement learning (DRL). Furthermore, we explore semantic communication (SemCom) as a promising solution for CSS, in which task-oriented representations are exchanged to reduce reporting overhead while preserving decision-critical information. Finally, we discuss limitations, open challenges, and future research directions at the intersection of AI and wireless communication.

💡 Research Summary

This paper surveys the emerging role of artificial intelligence (AI) in collaborative spectrum sensing (CSS) for cognitive and intelligent wireless networks, positioning CSS as a representative use case to illustrate AI’s impact on next‑generation (6G) communications. It begins by formalizing the CSS framework: K spatially distributed sensors each collect N baseband samples with M antennas, apply a local processing function F(·;αₖ) to generate a report yₖ, which is transmitted over a noisy reporting channel to a fusion center (FC). The FC aggregates the received reports zₖ through a fusion function G(·;β) to produce a global decision on primary‑user (PU) presence. Classical detection methods are reviewed and classified into optimal (likelihood‑ratio based), semi‑blind (energy detection, matched filtering, etc.), and totally‑blind (maximum‑minimum eigenvalue, covariance‑absolute‑value) techniques, highlighting their dependence on prior knowledge and the trade‑off between robustness and detection accuracy. Traditional fusion strategies—hard‑decision (binary local votes) and soft‑data (statistics or likelihoods)—are also discussed, emphasizing the inherent tension between communication overhead and information loss.

The core of the survey categorizes AI‑driven CSS into three families. Discriminative deep learning models (CNNs, RNN/LSTM, graph neural networks) replace handcrafted features with data‑driven representations, enabling robust detection under fading, shadowing, and hidden‑terminal conditions. Generative models (GANs, variational autoencoders, diffusion models) address data scarcity and noise mitigation by synthesizing realistic spectrum samples or denoising observations, thereby improving performance in low‑SNR regimes. Deep reinforcement learning (DRL) reframes sensor participation, resource allocation, and reporting‑channel selection as sequential decision‑making problems; both single‑agent MDP formulations and multi‑agent cooperative learning are surveyed, with policy‑gradient, DQN, and actor‑critic methods shown to adaptively balance detection reliability and energy efficiency.

A forward‑looking section introduces semantic communication (SemCom) for CSS. By extracting task‑oriented semantic embeddings at the sensor side and decoding them at the FC, only decision‑critical information is exchanged, dramatically reducing reporting bandwidth while preserving detection fidelity. This paradigm bridges the gap between hard and soft fusion, offering a flexible trade‑off between overhead and performance.

Finally, the paper outlines current challenges—model mismatch, label scarcity, computational and energy constraints, security/privacy concerns, and the lack of standardized protocols—and proposes future research directions. These include federated and federated‑reinforcement learning for privacy‑preserving large‑scale collaboration, explainable AI to increase trustworthiness, cross‑layer joint sensing‑communication optimization, multimodal sensing integration, and the development of SemCom‑aware CSS standards. In sum, the survey positions AI‑enhanced and SemCom‑enabled collaborative spectrum sensing as a cornerstone technology for achieving high spectrum efficiency, reliability, and energy sustainability in future 6G wireless networks.