As direct-to-cell and satellite IoT transition from experimental pilots to commercial rollout, operators must identify not merely whether a link is achievable, but which service constraint breaks first. Release 19 propelled non-terrestrial network (NTN) standardization beyond basic attachability toward usability under long delay, intermittent feeder-link visibility, moving beams, and narrow spectrum. This tutorial systematically distilled public NR-NTN and IoT-NTN Phase 3 material through five operator-facing dimensions and seven deployment bottlenecks. By mapping 13 representative mechanisms to the earliest blocker each relaxed, we revealed a structured split: NR-NTN mainly hardens handset entry, uplink sharing, regional delivery, regenerative option space, and RedCap timer slack, whereas IoT-NTN mainly targets feeder-tolerant collection, lighter small-data exchange, and single-band operation. In a representative LEO-600 coastal case, band-249 IoT-NTN TDD leaves less than 18% payload occupancy per frame, while a 10-minute feeder outage for 1000 sensors implies only about 300–500 KB of engineered onboard buffering. These findings supported three deployment heuristics. Chief among them was the principle that redundancy-based entry hardening never outperformed one-shot multiplexing in aggregate completions, as this strategy became rational only when per-user completion rather than throughput served as the governing KPI once single-shot success reached around 0.7.
Radio-Frequency Fingerprinting (RFF)-based Specific Emitter Identification (SEI) enables device-level recognition by exploiting hardware-induced signal characteristics. However, practical wireless channels often mix RFFs with channel responses, receiver effects, and environmental interference, leading to cross-domain performance degradation. Existing transfer approaches such as parameter-efficient fine-tuning and distribution alignment can partially mitigate this gap, but they do not explicitly preserve the relational structure among emitter fingerprints, often resulting in overfitting or negative transfer under few-shot conditions. This article reviews knowledge transfer for wireless intelligence, covering its fundamental concepts, representative methods, and key applications. We further present a Meta-Learning-Guided Knowledge Transfer Network (MLGKFN) for SEI, which represents RFF-related relations through graph structures and learns transferable meta-knowledge for rapid adaptation to new environments. Specifically, MLGKFN encodes each 1,024-point IQ signal into eight signal-dependent node embeddings, dynamically constructs sparse RFF relations using attention with Top-k (k=3) sparsification, and enables rapid few-shot adaptation across channel conditions through episodic meta-training. By combining relational knowledge transfer with meta-learning, this framework provides a promising direction for building wireless intelligence systems with stronger generalization and sample efficiency.
Integrated sensing and communication (ISAC) is expected to be natively supported by future 6G wireless radios, yet most millimeter-wave (mmWave) sensing solutions rely on dedicated radar hardware incompatible with cost- and power-constrained nodes. This article introduces Frequency-as-Aperture (FaA), a paradigm that repurposes inherent frequency agility into a virtual sensing aperture, enabling near-field sensing with minimal radio frequency (RF) front-end complexity. Using a single RF chain and a frequency-scanning leaky-wave antenna (LWA), FaA achieves two-dimensional spatial sensing by reusing the local oscillator (LO) frequency sweep already employed for wideband communication, shifting spatial sampling from the antenna domain to the frequency domain and embedding radar-grade spatial fingerprints directly into the communication RF chain. A 60-66 GHz case study shows that FaA provides fine angular and range discrimination with low power consumption and unit cost, indicating improved architectural efficiency relative to a conventional multi-channel multiple-input multiple-output (MIMO) baseline under the considered hardware and spectral constraints. These results suggest a feasible path toward integrating near-field sensing into frequency-agile wireless radios, enabling hardware-efficient, embeddable, and privacy-preserving ISAC-capable nodes for smart homes, wearables, and industrial edge deployments.
Semantic communication (SC) has recently emerged as a pivotal technology for enabling sixth-generation (6G) networks, attracting growing research interest and notable advances. However, most existing SC systems have been designed based on either additive white Gaussian noise (AWGN) channels with fading effects being neglected or fading channels with the assumption of perfect channel state information (CSI) availability. In practice, channel fading is non-negligible and perfect CSI is challenging to obtain. Particularly, in some scenarios, even imperfect CSI cannot be acquired. To enable more effective SC in fading channels, this article presents a generative AI (GenAI)-empowered SC (GenAI-SC) framework, which incorporates a modular GenAI-based anti-fading module (GAFM) adaptable to three levels of CSI awareness, i.e., CSI-free, CSI-inferable and CSI-estimable. To demonstrate the advantages of GenAI-SC, we conduct case studies for both text and image applications in fading channels, focusing on the CSI-free case where the GAFM is implemented. The results show that GenAI-SC outperforms conventional SC in terms of both semantic information recovery and system robustness in various fading environments. They also demonstrate that GenAI-SC can adapt to new fading conditions by fine-tuning only the GAFM model, without the need to retrain the semantic encoder and decoder. We further outline several promising future research directions, including the enhancement of security and robustness, and the development of efficient and scalable architectures, which are expected to inspire more studies and some practical applications of SC.
Fluid antenna systems (FAS) and multi-cell cooperation have recently emerged as promising enablers for next-generation wireless networks and offer significant potential for integrated sensing and communication (ISAC). By combining reconfigurable antenna architectures and coordination across base stations (BSs), cooperative FAS overcomes the structural rigidity of conventional fixed arrays and cell-centric network architectures, thereby enabling a highly flexible multiple-input multiple-output (MIMO) paradigm to support heterogeneous wireless services and functionalities. This flexibility is particularly important for secure near-field ISAC, where spherical-wave propagation accentuates the role of array geometry, port configuration, and inter-BS cooperation, thereby enhancing performance across communication, sensing, and security. Nevertheless, the design of network-level secure ISAC systems with cooperative FAS remains highly challenging, due to the intertwined effects of near-field channel non-stationarity, coordination constraints across multiple BSs, and the tightly coupled design of port selection and precoding under multi-objective performance requirements. To address these challenges, this article presents a graph neural network (GNN)-based design framework. By organizing communication, sensing, and security-related entities and their interactions into graph-structured representations and leveraging relation-aware message passing, the proposed framework offers a unified graph-learning perspective for adaptive channel extrapolation, coordination-efficient multi-BS cooperation, and joint port selection and precoding design. Under ideal CSI and inter-BS synchronization assumptions, a representative case study illustrates how the proposed framework generates transmit strategies that balance communication, sensing, and security objectives in the considered setting.
