Dynamic traffic patterns and shifts in traffic distribution in Open Radio Access Networks (O-RAN) pose a significant challenge for real-time network optimization in 5G and beyond. Traditional traffic analytics methods struggle to remain accurate under such non-stationary conditions, where models trained on historical data quickly degrade as traffic evolves. This paper introduces AIDITA, an AI-driven Digital Twin for Traffic Analytics framework designed to solve this problem through autonomous model adaptation. AIDITA creates a digital replica of the live analytics models running in the RAN Intelligent Controller (RIC) and continuously updates them within the digital twin using incremental learning. These updates use real-time Key Performance Metrics (KPMs) from the live network, augmented with synthetic data from a Generative AI (GenAI) component to simulate diverse network scenarios. Combining GenAI-driven augmentation with incremental learning enables traffic analytics models, such as prediction or anomaly detection, to adapt continuously without the need for full retraining, preserving accuracy and efficiency in dynamic environments. Implemented and validated on a real-world 5G testbed, our AIDITA framework demonstrates significant improvements in traffic prediction and anomaly detection use cases under distribution shifts, showcasing its practical effectiveness and adaptability for real-time network optimization in O-RAN deployments.

Time-Sensitive Networking (TSN) provides guaranteed traffic delivery, making it essential for industrial automation, multimedia, automotive systems, and other areas. Although TSN is well-established in wired networks, wireless systems face extra challenges such as delays, interference, and unstable links. Extending TSN to wireless (WTSN) thus adds complexity, particularly in managing traffic. This paper proposes a data-driven solution leveraging Digital Twin (DT) modeling to enhance WTSN management. We focus on mitigating the residual service time (RST) problem introduced by non-TSN traffic generators, which increases link latency. Implemented in a real Wi-Fi TSN-based environment, our implementation demonstrates that delay prediction through WTSN-DT reduces application link latency by up to 96% at the 90th percentile in a real Wi-Fi TSN setting, relative to fixed time slot allocation.

The integration of Non-Terrestrial Networks (NTN) with Terrestrial Networks (TN) is a key enabler for resilient 5G-Advanced and future 6G backhaul infrastructures. However, managing traffic across these highly asymmetric links remains a significant routing challenge, as systems must support heterogeneous network slices with conflicting service-level agreements (SLAs) while selectively utilizing costly NTN resources. This paper presents a computationally lightweight SLA-aware traffic-steering framework for a hybrid TN-NTN backhaul that models the load-balancing problem as an exact potential game. This mathematical foundation inherently enables decentralized coordination between uplink and downlink load-balancing agents without control-message overhead. By formulating traffic steering as a coupled optimization problem, per-slice (or per-user group) traffic fractions are dynamically distributed across terrestrial and satellite paths based on utility functions that capture throughput, latency, packet loss, and SLA penalties. The resulting game admits a pure Nash equilibrium, ensuring stable and predictable traffic adaptation under non-stationary load conditions. The framework is evaluated on a geographically distributed 5G testbed, using bidirectional traffic generated for five representative slices. Experimental results show that the proposed controller significantly outperforms heuristic and conventional baselines, reducing SLA violations to 1.7

Cell-free (CF) massive MIMO (mMIMO), also referred to as distributed MIMO (dMIMO) in indoor environments, has emerged as a key enabling technology for beyond-5G wireless systems due to its macro-diversity, favorable propagation, and joint multi-access-point (AP) processing gains. While extensive theoretical studies have demonstrated its potential, experimental validation remains limited. In this work, we present an end-to-end, multi-user, multi-stream, wideband OFDM experimental dMIMO platform, consisting of a centralized processing unit connected via optical fiber fronthaul links to geographically distributed APs. Tight synchronization among APs is achieved using sigma-delta modulated radio-over-fiber (SDRoF), combining the synchronization advantages of analog RoF with the cost efficiency of digital RoF. Using this platform, we experimentally evaluate three CF processing architectures, centralized (L4), hybrid centralized–distributed (L3), and fully distributed (L2), benchmarked against a collocated cellular MIMO system, within a complete end-to-end physical-layer framework comprising packet structure design, pilot signaling, wideband OFDM processing, and spatial processing algorithms. We discuss the uplink (UL) and downlink (DL) timing and frequency synchronization methods, as well as packet demultiplexing, in detail. Experimental results demonstrate substantial macro-diversity and favorable propagation gains. For an 8 × 4 MIMO scenario with four collocated users and two APs separated by 3 m, centralized CF processing (L4) achieves a minimum per-user signal-to-interference-plus-noise-plus-distortion ratio (SINDR) of 24 dB, versus only 8 dB for the collocated cellular configuration (λ/2 AP spacing), yielding a 16 dB gain. Under shadowing, the CF architecture maintains 30 dB minimum SINDR compared to 22 dB for the cellular system at K = 2 users. Furthermore, we experimentally validate UL–DL duality and quantify the reciprocity mismatch impact, demonstrating a 10 dB DL degradation without hardware calibration in a single-user (8 × 1) configuration. To the best of our knowledge, this work represents one of the first comprehensive experimental validations of wideband dMIMO and quantifies its practical gains over traditional cellular architectures.

This paper uses machine vision (MV) to assist communication in wireless networks, with a specific focus on intelligent reflecting surface (IRS)-assisted wireless systems. Instead of relying on traditional schemes such as alternating optimization or semidefinite relaxation, which are computationally expensive and often impractical, we use visual data to enable low-complexity beamforming decisions for users. Our approach employs a ceiling-mounted camera with a fish-eye lens to cover a wide communication area. Users within the network are first detected using an object detection method. We then propose closed-form analytical expressions to determine the distances between the access point (AP), users, and IRS. To address the non-uniform distortion inherent in fish-eye images, we introduce a novel method for determining non-uniform pixel weightings using trigonometric techniques. Based on the calculated distances, beamforming decisions are made according to the user’s proximity to the AP or IRS. Furthermore, we extend the application of MV to a multiuser IRS-assisted scenario, where we propose grouping users into either two or three categories. Using these visually identified groups, we simplify and solve the IRS optimization problem by considering only the constraints relevant to each category. Simulation results demonstrate that the proposed MV-based scheme for IRS achieves a significantly lower computational cost compared to benchmark schemes, while maintaining comparable performance.

For the commercial deployment of cell-free (CF) massive MIMO, mobile-grade antennas are envisioned for use in low-cost, low-complexity access points. It is essential to consider the practical constraints of these mobile-grade components when comparing the performance of CF and cellular networks. In this context, this paper focuses on the following main aspects: (1) we consider a generalized CF system with multiple APs and UEs, each equipped with multiple antennas and supporting multiple data streams per user. We exploit mobile-grade antennas with lower per-antenna transmit power constraints and compare their performance with conventional cellular networks, which are typically less constrained in this regard; and (2) we study a novel optimization problem aimed at achieving performance fairness in the presence of multiple data streams per user. To solve this problem, we derive convex formulations for the number of data streams, power allocations, and fairness weights. Based on these formulations, we propose a multistep alternating optimization (mAO) algorithm. To reduce the computational complexity of the mAO algorithm, we also propose a second algorithm based on closed-form solutions, which does not require convex solvers. Given the same total power budget in CF and cellular systems, the numerical analysis highlights the impact of mobile-grade antennas and shows that the fairness difference between cellular and CF is not as significant as portrayed in idealized scenarios. Specifically, we show that to achieve a sufficient fairness gap, CF requires up to 20 times more print-quality transmit antennas, i.e., radio stripes, than cellular systems. To assist future comparative studies, we also provide a computational complexity analysis of the proposed solutions.

This paper addresses a critical challenge of high computational complexity in the optimization of reconfigurable intelligent surfaces (RISs), a bottleneck that currently prevents real-time implementation in practical wireless networks. Existing iterative algorithms often exceed the channel coherence time, necessitating a shift toward insight-driven, low-complexity solutions. Specifically, we observe that the multiplicative fading nature of passive RISs means their impact is localized; thus, users strongly dominated by the direct transmit source receive negligible gains from RIS optimization. Based on this, we propose a novel User Grouping (UG) framework that categorizes users into three or two groups (3-UG and 2-UG) according to their relative direct and reflected channel strengths. By optimizing RIS operations only for relevant user subsets, we significantly limit the number of computational constraints. Furthermore, we investigate a per-user multi-stream transmission problem involving mixed-integer log-sum expressions. We derive a geometric-mean (GM)-based convex formulation to handle these discrete variables and develop a multi-step alternating optimization (AO) algorithm. Finally, we extend the UG concept to point-to-point (P2P) multiple-input multiple-output (MIMO) by introducing Adaptive Selection Beamforming (ASB), a non-iterative method that selects between two low-complexity solution sets. Numerical results demonstrate that the proposed UG methods achieve up to a 77% reduction in computational complexity compared to traditional benchmarks. Additionally, the non-iterative ASB-MIMO scheme is found to be approximately 67%–85% less complex (depending on the MIMO size) than the most efficient existing closed-form solutions. In all scenarios, the proposed frameworks maintain near-optimal performance with negligible performance degradation.

Open Radio Access Networks (O-RAN) promise flexible 6G network access through disaggregated, software-driven components and open interfaces, but this programmability also increases operational complexity. Multiple control loops coexist across the service management layer and RAN Intelligent Controller (RIC), while independently developed control applications can interact in unintended ways. In parallel, recent advances in generative Artificial Intelligence (AI) are enabling a shift from isolated AI models toward agentic AI systems that can interpret goals, coordinate specialized models and control applications, and adapt their behavior over time. This article proposes a multi-scale agentic AI framework for O-RAN that organizes RAN intelligence as a coordinated hierarchy across the Non-Real-Time (Non-RT), Near-Real-Time (Near-RT), and Real-Time (RT) control loops. (i) A Large Language Model (LLM) agent in the Non-RT RIC translates operator intent into policies and governs model lifecycles; (ii) Small Language Model (SLM) agents in the Near-RT RIC execute low-latency optimization and can activate, tune, or disable existing control applications; and (iii) Wireless Physical-layer Foundation Model (WPFM) agents near the distributed unit provide fast inference close to the air interface. We describe how these agents cooperate through standardized O-RAN interfaces and telemetry. Using a proof-of-concept implementation built on open-source models, software, and datasets, we demonstrate the proposed agentic approach in two representative scenarios: robust operation under non-stationary conditions and intent-driven slice resource control.

Adaptive coding and modulation (ACM) is a key feature in satellite broadcasting; it allows the dynamic selection of modulation and coding (MODCOD) schemes based on channel conditions. The selection is based on the quasi-error-free (QEF) threshold and additional margins. We introduce three distinct types of margins for improved robustness. One of these margins, impairment margin (IM), depends on the nonlinearities of different components in the satellite channel. Current IM selection methods require expert intervention; are costly and prone to errors; and only allow a discrete set of environments. We aim to develop a low-complexity algorithm that converges fast and is quasi-error-free on user traffic due to a non-intrusive exploration method. For this, we propose a Q-learning-based solution that uses passive exploration, with fill frames, to allow error-free IM optimization. Our solution shows a higher average spectrum efficiency compared to expert and default IMs, with fewer low efficiency test cases and more high-efficiency cases.

Fixed Wireless Access (FWA) networks are used to extend connectivity to areas with limited or no access, especially where the deployment of wired infrastructure is costly. In such networks, the infrastructure can take the form of a multi-hop mesh network consisting of Distribution Nodes (DNs) and Client Nodes (CNs). A CN is served by a DN and extends connectivity within homes by acting as an access point (AP). Due to fluctuations in traffic over time, network utilization also fluctuates. When the network is scarcely utilized, it leads to energy waste due to powering the network during these times. In this paper, we explore methods to reduce energy consumption in wireless mesh networks (WMNs) by implementing coordinated sleeping times for APs of the last hop and the DNs inside the FWA network. By dynamically scheduling sleep patterns in the last hop, the solution achieves network-wide energy savings without compromising the quality of service for traffic flows in terms of latency and reliability. Moreover, in this paper, we reduce the Orthogonal Frequency Division Multiple Access (OFDMA) overhead and utilize this to organize time-critical traffic in the last hop of FWA, benefiting from the coordinated sleeping time. Our results show that in medium-load scenarios, this approach can achieve up to 33% energy savings in FWA mesh networks combined with NextGen Wi-Fi while maintaining bounded latency for time-critical applications and serving non-time-critical traffic.

Cross-Technology Interference (CTI) significantly degrades the performance of heterogeneous wireless communication systems operating within a shared spectrum. Traditional mitigation techniques, such as Clear Channel Assessment (CCA), often fail due to, amongst others, varying bandwidth and detection threshold. Orthogonal Frequency Division Multiple Access (OFDMA), introduced to Wi-Fi in IEEE 802.11ax (Wi-Fi 6), allows multiple users to be served simultaneously using distinct subcarrier sets, known as Resource Units (RUs), providing enhanced flexibility in the frequency domain. This paper explores and evaluates several Wi-Fi 6 compliant methods for multi-user OFDMA scheduling with CTI awareness. Through simulations, we assess the benefits of different techniques in various scenarios, in terms of either total throughput or average latency. To effectively apply the mitigation techniques, we propose a methodology that incorporates CTI feedback from stations and real-time CCA per RU. Given that commercial Wi-Fi 6 access points lack control over low-level OFDMA features, we use openwifi, a full-stack Wi-Fi transceiver running on software-defined radio, to implement the CTI-aware OFDMA scheduler. Real-life experiments validate the effectiveness of the scheduler and confirm its real-time performance capabilities.

This paper presents a comprehensive assessment of 5G SA networks deployed in a factory setting and an offshore wind farm at the North Sea. The study focuses on critical use cases such as teleoperated Automatic Guided Vehicles (AGVs) in factories and teleoperated Unmanned Surface Vessels (USVs) in offshore settings. Key performance indicators (KPIs) such as network latency, data rates, and signal quality were measured using advanced network evaluation tools. The results demonstrate that 5G SA networks can meet the stringent requirements of industrial applications. The paper also discusses the challenges of deploying 5G networks in metal-dense factories and dynamic offshore environments, highlighting the importance of practical evaluations in these settings. Furthermore, the study introduces a machine learning-driven framework for signal and Quality of Service (QoS) prediction, leveraging diverse datasets collected in the factory environment. This framework aims to optimize network performance and lays the groundwork for creating a digital twin for factory communication systems.

Co-channel interference cancellation (CCI) is the process used to reduce interference from other signals using the same frequency channel, thereby enhancing the performance of wireless communication systems. An improvement to this approach is adaptive CCI, which reduces interference without relying on prior knowledge of the interfering signal characteristics. Recent work suggested using machine learning (ML) models for this purpose, but high-throughput ML solutions are still lacking, especially for edge devices with limited resources. This work explores the adaptation of U-Net Convolutional Neural Network models for high-throughput adaptive source separation. Our approach is established on architectural modifications, notably through quantization and the incorporation of depthwise separable convolution, to achieve a balance between computational efficiency and performance. Our results demonstrate that the proposed models achieve superior MSE scores when removing unknown interference sources from the signals while maintaining significantly lower computational complexity compared to baseline models. One of our proposed models is deeper and fully convolutional, while the other is shallower with a convolutional structure incorporating an LSTM. Depthwise separable convolution and quantization further reduce the memory footprint and computational demands, albeit with some performance tradeoffs. Specifically, applying depthwise separable convolutions to the model with the LSTM results in only a 0.72% degradation in MSE score while reducing MACs by 58.66%. For the fully convolutional model, we observe a 0.63% improvement in MSE score with even 61.10% fewer MACs. Additionally, the models exhibit excellent scalability on GPUs, with the fully convolutional model achieving the highest symbol rates (up to 800 $\times$ 103 symbol per second) at larger batch sizes. Overall, our findings underscore the feasibility of using optimized machine-learning models for interference cancellation in devices with limited resources.

Time-Sensitive Networking (TSN) ensures reliable traffic delivery in industrial automation, multimedia, and automotive systems. While effective in wired networks, wireless TSN (WTSN) faces challenges like delays and interference, complicating traffic scheduling. This paper presents a data-driven approach to improve WTSN management by addressing the residual service time (RST) problem, which increases link latency. Tested in a Wi-Fi TSN-based environment, the proposed WTSN digital twin framework preserves TSN traffic guarantees while significantly reducing RST and hence link latency.

Multiple Input Multiple Output (MIMO) radar systems enhance detection and anti-interference capabilities by increasing the number of antennas compared to traditional radar systems. Despite significant research into Wi-Fi sensing in recent years, constructing Wi-Fi MIMO radar remains challenging due to half-duplex RF hardware and limited access to low-level multi-antenna signals in commercial Wi-Fi chips. This paper presents a SIMO Wi-Fi radar with one Tx and two Rx antennas, based on the open-source Wi-Fi SDR platform, openwifi, which has implemented the full 802.11 stack. The hardware includes an FPGA and an AD9361 RF front-end. Directional antennas are used to reduce Tx-Rx interference in full-duplex mode. This Wi-Fi MIMO radar system brings dual capabilities: sensing the passive target that is not transmitting any signals and the normal Wi-Fi device that transmits signals for data traffic. In this paper, the dual capabilities are demonstrated in two cases: vital signal sensing for humans and Angle of Arrival (AoA) tracking of another device. More specifically, millimeter-level movement detection is achieved by processing phase information from the Channel State Information (CSI) of the two Rx antennas.

Wireless Time-Sensitive Networking (WTSN) faces challenges to ensure communication reliability without compromising communication latency. As communication in wireless link happens in the same channel, frame replication and elimination for reliability (FRER) in different bands or channels is not possible in the wireless part. However, with the new features like multi-link operation (MLO), FRER can be enabled in wireless links as well, improving the overall reliability of the communication. In this paper, we show how we achieve distributed MLO utilizing openwifi platform and wired TSN for synchronized transmission in multiple links. We show initial results of end-to-end reliability and latency when MLO is utilized under different traffic load levels on each operational link. We also show the effect of cross-node physical layer queue management for avoiding head-of-the-line queue blocking.

Indoor environments with rich scatters and severe shadowing are very complicated today for wireless communications. The presence of metal surfaces and an overwhelming number of wireless devices further introduces challenges to the resilience of existing wireless links, especially in industrial scenarios. The existing single-transmitter links often suffer from the complex environment and thus cannot guarantee a stable performance; even with more transmitting antennas on a commercial multiple-input-multiple-output (MIMO) capable access point (AP), due to limited antenna number and positioning, it may not provide fair and satisfactory quality of service to all users. Distributed beamforming (D-BF) is a future upgrade for the existing collocated MIMO that allows individual APs to transmit coherently from multiple locations for improved link resilience. We analyze the main limiting factors of channel sounding for implementing D-BF with the current Wi-Fi standard. Our proposed synthetic multi-AP channel sounding approach overcomes the 8-antenna constraint posed by the standard and uses IEEE 802.11ax compliant Wi-Fi frames, enabling commercial single-receiver Wi-Fi station (STA) to benefit from distributed beamforming without additional upgrade. Our simulation result shows that the effective SNR can be boosted by roughly 26dB with D-BF of 64 APs against a commercial 8x1 beamforming link, alongside a 441.6µs sounding airtime which corresponds to less than 0.5% overhead under 100ms channel coherence time.

Autonomous Network enablers for a 5G Network-on-Wheels are investigated, with a field-trial for a Communication Service Provider. The objective was to devise solutions to reduce time spent on network life cycle tasks, such as network function validation, system commission and service deployment. The field trial revealed significant data collection and analysis challenges even with a single cell campaign, this suggests severe costs with public-network scales. The work contributes a new architecture for experiments and optimisation of observability strategies with algorithms in 5G networks. This work helps operators more effectively deploy 5G networks and services.

Proper coordination is needed to guarantee the performance of wireless networks in dense deployments. Contention-based systems suffer badly in terms of latency when multiple devices compete for the same resources. Coordinated Orthogonal Frequency Division Multiple Access (Co-OFDMA) is proposed for Wi-Fi 8 to remedy this, as it enables multiple Access Points (APs) to share spectrum more efficiently. However, fine-grained resource allocation, namely within 20MHz bandwidth, is argued to be impractical due to the over-the-air scheduling overhead and complexity in terms of physical layer signaling. A wired backhaul mitigates the need for over-the-air scheduling and synchronization, and it allows for coordination even if APs are not in each others' range. Furthermore, it forms the basis for more advanced multi-AP coordination schemes like coordinated beamforming and joint transmission. In this work we demonstrate the realization of Wi-Fi 6 compliant fine-grained Co-OFDMA using a fiber backhaul, enabled by the open-source platforms openwifi and White Rabbit. We show that the performance in terms of carrier frequency offset pre-compensation and time synchronization between two APs exceeds related wireless standard requirements. Furthermore, the quality of the received constellation of the Co-OFDMA frame as reported by a wireless connectivity tester is better than individual frames sent by the APs.