The evolution from fifth-generation (5G) to sixth-generation (6G) networks is driving an unprecedented demand for advanced machine learning (ML) solutions. Deep learning has already demonstrated significant impact across mobile networking and communication systems, enabling intelligent services such as smart healthcare, smart grids, autonomous vehicles, aerial platforms, digital twins, and the metaverse. At the same time, the rapid proliferation of resource-constrained Internet-of-Things (IoT) devices has accelerated the adoption of tiny machine learning (TinyML) for efficient on-device intelligence, while large machine learning (LargeML) models continue to require substantial computational resources to support large-scale IoT services and ML-generated content. These trends highlight the need for a unified framework that integrates TinyML and LargeML to achieve seamless connectivity, scalable intelligence, and efficient resource management in future 6G systems. This survey provides a comprehensive review of recent advances enabling the integration of TinyML and LargeML in next-generation wireless networks. In particular, we(i)provide an overview of TinyML and LargeML,(ii)analyze the motivations and requirements for unifying these paradigms within the 6G context,(iii)examine efficient bidirectional integration approaches,(iv)review state-of-the-art solutions and their applicability to emerging 6G services, and(v)identify key challenges related to performance optimization, deployment feasibility, resource orchestration, and security. Finally, we outline promising research directions to guide the holistic integration of TinyML and LargeML for intelligent, scalable, and energy-efficient 6G networks and beyond.

This work investigates cellular millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems within the open radio access network (O-RAN) architecture, integrating the compatible spectrum, air interface, and networking entities of beyond fifth-generation wireless networks. To overcome O-RAN fronthaul (O-FH) load limitations and the short wavelength inherent in mmWave bands, we design a hybrid beamforming architecture with digital and analog beamformers generated at the O-RAN distributed unit and O-RAN radio unit, respectively. Using the information theory, we develop non-grid-of-beams analog beamformers to maximize the sum-spectral efficiency (SE) under constant-modulus constraints. For digital precoding, we apply a successive convex approximation method with second-order cone program procedures to maximize sum-SE, while addressing transmit power and limited O-FH load constraints, and ensuring user quality of service requirements. Sub-optimal digital combiners are also designed based on the inherent characteristics of the user side. However, the current optimization approach suffers from long execution times, posing challenges for near-real-time beamforming configurations. To address this issue, we propose an efficient deep learning (DL)-based digital precoding scheme with short execution time, low computational complexity, and high performance. Numerical results demonstrate that the proposed DL-based precoding scheme provides superior performance compared to benchmark schemes, generalizes well to environments with imperfect CSI and user mobility, and scales effectively to massive MIMO configurations.

Simultaneously meeting high spectral efficiency (SE)-oriented enhanced mobile broadband (eMBB) and ultra-reliable and low-latency communication (uRLLC) remains a significant challenge. Given uRLLC requirements, this work investigates the problem of sum-SE maximization in multiple reconfigurable intelligent surfaces-assisted cell-free (CF) millimeter wave massive multiple-input multiple-output systems. The joint optimization problem of passive reflective beamforming (RBF) and hybrid active transceiver beamforming is tackled using a multi-block coordinate descent (MBCD) method, under constraints of transmit power, users’ quality-of-service, and constant-modulus characteristics. The subproblems involving analog transceiver processing and passive RBF, associated with constant-modulus constraints, are solved via MBCD-based Riemannian manifold optimization, while the subproblems involving digital transceiver beamforming are addressed using a successive convex approximation framework combined with semidefinite relaxation and Gaussian randomization. To further balance computational complexity, fronthaul overhead, and system performance, a user-centric access point selection scheme is adopted using statistical channel state information. Numerical results demonstrate that the proposed framework outperforms various practical benchmark schemes in terms of user SE. Furthermore, the user-centric CF design achieves a 99.29% reduction in computational complexity with only a 5.49% performance loss compared with full-association CF.

In this article, we describe a novel active reconfigurable repeater-aided nonorthogonal multiple access networks within the context of the Internet of Things. The study focuses on a scenario, where a source simultaneously transmits public information to an untrusted user and a covert signal to a legitimate user in the surveillance of an external warden or eavesdropper. We develop comprehensive analytical and optimization frameworks to evaluate the reliability, security, and covertness of the proposed system’s performance, measuring three respective key metrics: 1) outage probability (OP); 2) secrecy OP (SOP); and 3) detection error probability (DEP). First, we derive exact closed-form and asymptotic expressions for OP, SOP in internal and external eavesdropping scenarios, and DEP in external monitoring situations. Based on an asymptotic analysis of the OP, we propose two optimization methods for power allocation (PA) to achieve fairness in outage among users: 1) a convex approximation method and 2) an approximate closed-form solution. We then introduce an alternative method for optimizing PA to improve SOP in both eavesdropping scenarios while maintaining minimal OP requirements. In addition, we propose an effective approach for determining the warden’s detection threshold to minimize the DEP, with low complexity and fast convergence, thereby improving communications covertness. Finally, we validate the theoretical and optimization frameworks through extensive Monte Carlo simulations, exploring the impact of key system parameters on each performance metric.

This work investigates the covertness and security performance of Internet-of-Things (IoTs) networks under Rayleigh fading environments. Specifically, a cellular source transmits covert information to cell-edge users with the assistance of an IoT master node, employing a coordinated direct and relay transmission strategy combined with non-orthogonal multiple access (NOMA). This approach not only enhances spectrum utilization but also generates friendly interference to complicate a warden’s surveillance or an eavesdropper’s decoding efforts. From a covertness perspective, we derive exact closed-form expressions for the detection error probability (DEP) under arbitrary judgment thresholds. We then identify the optimal judgment threshold for the worst-case scenario, at which the warden minimizes its DEP performance. Accordingly, we determine the effective region for user power allocation (PA) in NOMA transmission that satisfies the DEP constraint. From a security perspective, we derive analytical expressions for the secrecy outage probability under two eavesdropping strategies using selection combining and maximal ratio combining. Based on this analysis, we propose an adaptive PA scheme that maximizes covert rate while ensuring the quality-of-service (QoS) requirements of legitimate users, the system’s minimum covertness requirements, and supporting successive interference cancellation (SIC) procedures. Furthermore, we design an adaptive PA scheme that maximizes the secrecy rate while ensuring the QoS requirements of legitimate users and SIC conditions. Numerical results demonstrate the accuracy of the analytical framework, while the proposed optimization strategies effectively adjust PA coefficients to maximize either the covert rate or the secrecy rate.

This work proposes a novel semi-static hybrid beamforming strategy for open radio access network (O-RAN) millimeter wave (mmWave) massive multiple-input multiple-output systems, addressing the limited O-RAN wireless fronthaul (O-WFH) capacity and short wavelength of mmWave bands. Statistical analog precoders are designed at O-RAN radio units to maximize the average signal-to-leakage-plus-noise ratio using statistical channel state information under non-grid-of-beams (non-GoB) constant-modulus constraints. A joint optimization problem of dynamic digital precoders and hybrid combiners is then performed at O-RAN distributed units via a multi-block coordinate descent (MBCD) method. The joint optimization maximizes sum-spectral efficiency (SE) under constraints on transmit power, users’ quality-of-service, O-WFH capacity, and non-GoB constant-modulus constraints. The digital processing subproblems are solved using a successive convex approximation with second-order cone programming, whereas the analog combining subproblem employs an MBCD-based Riemannian manifold optimization. Notably, the resulting hybrid combiners are forwarded to users for further processing and service applications. Numerical results reveal that the proposed framework offers superior user SE performance compared to practical analog and digital benchmark schemes, while significantly reducing O-WFH information exchange compared to a theoretical benchmark scheme, with only a small user SE performance loss.

This work investigates the multiple-input multiple-output system in the context of the selective decode-and-forward multihop relay network under short-packet communications to facilitate not only ultra-reliability, but also low-latency communications. For the transmit and receive diversity techniques, we analyze the transmit antenna selection (TAS) and maximum-ratio transmission (MRT) schemes at the transmit side, whereas the selection-combining (SC) and maximum-ratio combining (MRC) schemes are leveraged at the receive side. For quasi-static Rayleigh fading channels and the finite-blocklength regime, we derive the approximate closed-form expressions of the end-to-end (e2e) block error rate (BLER) for the TAS/MRC, TAS/SC, and MRT/MRC schemes. The asymptotic performance in the high signal-to-noise ratio regime is derived, from which the comparison among diversity schemes in terms of the diversity order, e2e BLER loss, and SNR gap is provided. Furthermore, based on the asymptotic results, we develop power-allocation, relay-location, and joint simplified optimizations to minimize the asymptotic e2e BLER under the system constraints. The e2e latency and throughputs are also analyzed for the considered schemes. The correctness of our analysis is confirmed via Monte Carlo simulations.