This paper presents a comprehensive survey of semantic protocols, protocol learning, and AI-enabled semantic protocols within the context of Open RAN and 6G networks. We systematically review the significant progress achieved in these domains, highlighting key methods such as transformer-based semantic encoders, reinforcement learning–driven protocol adaptation, and federated learning frameworks for distributed training. Across surveyed studies, notable achievements include bandwidth savings of 35-70%, improved robustness under noisy conditions, and enhanced interoperability in multi-vendor environments. By consolidating findings, we identify major challenges such as the lack of standardized semantic KPIs, computational overhead at the edge, interoperability issues, and emerging security vulnerabilities. Furthermore, we categorize open research opportunities into theoretical, methodological, technical, and implementation directions, providing a clear roadmap for future development. This survey ultimately positions semantic communication and AI-enabled protocols as pivotal enablers for meaning-centric, adaptive, and efficient next-generation O-RAN/6G networks.

As communication systems evolve toward 6G, semantic communication is emerging as a transformative paradigm that prioritizes the accurate transmission of meaning rather than just bits. While artificial intelligence enables this shift by facilitating intelligent interpretation and context-aware processing, it also introduces significant challenges related to transparency, reliability, and user trust. XAI has thus become essential in making AI-enabled semantic communication systems more interpretable, auditable, and accountable. To the best of our knowledge, this is the first survey that systematically analyzes how explainability can be embedded across all stages of the semantic communication pipeline, integrating architectural design, metrics, security considerations, and human-in-the-loop mechanisms. Additionally, the survey identifies pressing research challenges, including the lack of standardization, real-time applicability, and vulnerabilities introduced by opaque AI models. By drawing attention to these issues and outlining future research directions, this survey aims to guide the development of responsible and trustworthy semantic communication systems for next-generation wireless networks.

The Cloud-Edge collaborative computing enables the deployment of latency-sensitive and data-intensive applications closer to end users. However, it introduces significant challenges for microservice placement, due to resource heterogeneity, limited edge capacity, and the need to satisfy storage requirements using aggregated resources across multiple nodes. To address these issues, we proposeDiktopos, a topology-aware, two-stage scheduling framework that jointly optimizes microservice placement and distributed storage volume allocation in cloud-edge networks. The joint optimization problem is decomposed into two subproblems: (i) microservice placement and (ii) distributed volume allocation, with the objective of minimizing computation, communication, energy, and storage costs. At its core, Diktopos employs a low-complexity, rank-based heuristic that ensures scalable and accurate placement across heterogeneous edge nodes. Simulation results show that our method achieves near-optimal placement decisions (within 1.67% of the optimal solution), and converges up to 5× faster than state-of-the-art approaches in large-scale deployments. Real-world experiments in Kubernetes environments demonstrate up to 53% latency reduction compared to the default scheduler, and up to 23% improvement over other baselines, confirming Diktopos' effectiveness in dynamic, resource-constrained edge scenarios.

Efficient workload scheduling in Kubernetes is crucial for optimizing energy consumption and resource utilization in large-scale and heterogeneous clusters. However, existing Kubernetes schedulers either ignore power-awareness or rely on simplified, static power models, which limit their effectiveness in managing energy efficiency under dynamic workloads. To address these shortcomings, we present a multi-objective scheduling framework for online Kubernetes pod placement that jointly considers power consumption, resource utilization, and load balancing. The framework follows a two-stage design: (i) a node power–profiling component trains a machine–learning model from real power measurements to predict per-node consumption under varying utilizations; and (ii) an online scheduler uses these predictions within a multi-objective optimization formulation. We implement scheduling optimization using two algorithms, TOPSIS and NSGA-II, adapting them to the Kubernetes context, and also propose a distributed variant of the NSGA-II algorithm that parallelizes fitness evaluation with controlled migration between workers. Experimental results show that the proposed framework outperforms baseline schedulers, achieving a 40% reduction in power consumption and improvements of 74% and 68% in CPU and memory utilization, respectively, while sustaining scalability under high workloads. To the best of our knowledge, this is the first work to integrate learned power models and distributed multi-objective optimization into Kubernetes for power-aware pod scheduling.

Open Radio Access Network (O-RAN) enables modular and intelligent control of radio resources through open interfaces and programmable RAN components. As networks evolve toward sixth-generation (6G) systems, the proliferation of autonomous xApps and rApps introduces a critical challenge: Coordinating concurrent AI-driven control actions under tight near-real-time constraints while avoiding instability and conflicting decisions. This paper focuses on two tightly coupled enablers for AI-native O-RAN orchestration: Conflict-aware control and intent-driven automation. We propose an AI-native orchestration framework centered on a CME integrated into the Near-RT RIC, and a complementary LLM-based intent orchestration module deployed in the Non-RT RIC. The CME is designed to autonomously arbitrate conflicting xApp actions by learning adaptive mitigation policies from structured conflict signals, system context, and performance feedback, rather than relying on static priorities or predefined conflict classes. The LLM module translates high-level operator intents into policy constraints and control objectives that guide conflict resolution and xApp behavior. Overall, this work advances AI-native O-RAN orchestration by grounding conflict-aware control and LLM-assisted intent translation in practical measurements, and by outlining a clear path toward scalable, adaptive, and resilient control mechanisms required for future 6G RIC deployments.

In this paper, we present a programmable data plane design that implements IEEE 802.1Qcr’s Asynchronous Traffic Shaper (ATS) purely in software using Extended Berkeley Packet Filter (eBPF), eliminating specialized hardware requirements for industrial Time-Sensitive Networking (TSN) deployments. Unlike hardware-bound TSN solutions, our approach dynamically decouples and reprograms functions like filtering, metering, and queuing through in-kernel eBPF hooks, enabling adaptive priority management for concurrent streams within shared priority queues. The design explicitly models the ATS scheduler to parameterize per-stream eligibility times in TSN bridges while maintaining deterministic operation. This software-defined method provides a vendor-agnostic path for integrating ATS capabilities into existing industrial networks, particularly for Industry 5.0’s distributed control scenarios requiring flexible traffic multiplexing. The results confirm correct enforcement of ATS scheduling semantics under heterogeneous workloads.

The Cloud Edge Continuum enables the deployment of latency-sensitive and data-intensive applications closer to end users, but it poses challenges for microservice placement due to resource heterogeneity and limited edge capacity, especially when storage requirements must be met through aggregated node resources. To address this, we propose a two-stage, topology-aware optimization framework that jointly handles microservice deployment and distributed storage volume allocation in edge networks. Our framework decomposes this joint placement problem into two subproblems, microservice placement followed by a distributed volume allocation subproblem, with the goal of optimizing computation, communication, energy, and storage costs. At its core is a lightweight, rank-based heuristic that ensures scalable, accurate placement across distributed edge nodes. Evaluations on real-world scenarios show our method achieves near-optimal placement (within 1.67% of the exact solution), reduces system costs by up to 30%, and accelerates convergence by 5× compared to state-of-the-art approaches, demonstrating its suitability for dynamic, resource-constrained edge environments.

The rise of real-time networking demands has driven the IEEE Time-Sensitive Networking (TSN) task group to develop new standards that ensure high bandwidth and lowlatency Ethernet communication. TSN is an essential component of next-generation 6G networks. It offers features that ensure deterministic data transmission and alleviate network congestion. These features are crucial in time-sensitive applications and systems, whereby both precision and reliability are paramount. While early TSN implementations relied heavily on synchronous communication, newer standards, such as IEEE 802.1Qcr, have introduced asynchronous mechanisms via Urgency-Based Scheduler (UBS). UBS employs advanced queuing and traffic shaping strategies, to guarantee minimal delay for real-time applications. Within the scope of 6G, this study evaluates the queuing and shaping strategies applied to multiple flows within the UBS framework. Moreover, we assess their impact on frame transmission rates at the shaper level, highlighting the optimal use case for each strategy.

This paper proposes a new scheduling framework to optimize the placement of cloud workloads submitted online by users within a Kubernetes-orchestrated environment. The proposed method aims to incorporate power awareness during the scheduling process, along with other criteria, such, load balancing, and bin packing. The framework equitably distributes workloads across cluster nodes while also selecting the most resource-efficient node to reduce the number of active nodes and prevent resource fragmentation. Existing strategies often focus on a single criterion, leading to suboptimal and unsatisfactory workload placements. The proposed framework utilizes the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), a well-known multi-criteria decision analysis algorithm, to account for power consumption and other criteria defined by cloud operators, such as load balancing and bin packing. The algorithm is implemented as a Kubernetes scheduling plugin to rank worker nodes based on these criteria and the submitted workloads. Simulation results demonstrate the effectiveness of the proposed strategy across various scenarios, reducing power consumption by 26.46% and comparable CPU and memory load balancing performance within a large Kubernetes cluster under heavy workloads.

The Internet of Things (IoT) is rapidly expanding globally, but the limited size of IoT devices restricts their battery capacity, computational resources, and wireless bandwidth, making it difficult to handle resource-intensive tasks. Edge Computing addresses these challenges by enabling task offloading to more capable edge servers. However, optimal task offloading in Edge-IoT networks is complex due to dynamic conditions, such as varying server loads and wireless fluctuations. Traditional and some machine learning-based offloading methods often fall short in adaptability or efficiency. This paper introduces a bi-level optimization approach using Deep Reinforcement Learning (DRL) agents for IoT-level offloading and a priority-aware greedy heuristic for resource allocation on edge servers. The proposed method effectively improves QoS by balancing task execution latency and power consumption, as demonstrated by simulation results.

In recent years, AI inference has seen widespread adoption across fields like finance and healthcare, driving significant demand for high-performing applications. This demand brings about a complex relationship between inference application types, such as real-time applications, and their specific service level objectives (SLOs), like tail-latency. Tail-Latency is a metric requiring a defined percentage of requests to meet a maximum response time, which is crucial for applications where delays can impact user experience or decision-making. This dependency creates a challenging research problem in scheduling inference workloads. The core question becomes: How can we deploy AI workloads in a way that minimizes SLO violations?Specifically, we worked on real-time applications that require tail-latency guarantees. To address this, we developed a tail-latency-aware scheduler designed for resource-constrained devices. Our scheduler employs advanced machine learning techniques to optimize task placement, aiming to minimize SLO violations and enhance performance for latency-sensitive applications. We have developed and integrated our custom scheduler into Kubernetes, which operates on a specially configured cluster designed to test its performance. This cluster features diverse computing capabilities, enabling a comprehensive evaluation of the scheduler’s effectiveness. The experimental results highlight that our proposed scheduler outperforms the native Kubernetes scheduler in terms of efficiency.

The rapid development of reconfigurable intelligent surfaces (RISs) has sparked transformative advancements in wireless communication systems. These intelligent metasurfaces, adept at dynamically manipulating electromagnetic (EM) waves, hold vast potential for enhancing network capacity, coverage, and efficiency. However, to fully unleash the capabilities of RIS-aided communication systems, effective optimization is crucial. This article provides a recent development of RIS-assisted communication from the viewpoint of physically-consistent EM models. We delve into the realm of physically-consistent EM models, highlighting their pivotal role in achieving robust and efficient RIS designs. Furthermore, this paper offers a survey of the different optimization models utilized for RIS-assisted wireless communication systems, which consider various EM and physical aspects of RIS. We explore solution approaches aimed at optimizing different objectives like sum-rate/spectral efficiency and energy efficiency, spanning traditional optimization models to machine learning-based methods. Additionally, we discuss some open research issues in this field.

Recent advancements in serverless computing and the cloud-edge continuum have increased interest in WebAssembly (WASM). This technology enables portability and interoperability across diverse computing environments while achieving near-native execution speeds. Currently, WASM supports Single Instruction Multiple Data (SIMD), which allows for data-level parallelism that is particularly beneficial for vectorizable operations such as general matrix-matrix multiplication (GEMM) and convolutional layers. However, WASM lacks native integration with specialized hardware accelerators like GPUs, TPUs, and NPUs, as well as the ability to benefit from multi-core processing capabilities, which are critical for efficiently running Deep-Learning (DL) workloads. In contrast, despite these gains, WASM still lacks native support for heterogeneous accelerators such as GPUs, TPUs, and NPUs, as well as full multi-core parallelism capabilities that are critical for meeting the latency and throughput requirements of modern DL inference services. To bridge this gap, WASI-NN was developed, enabling WASM to integrate with external runtimes such as OpenVINO and ONNX Runtime, which leverage hardware acceleration. However, these current integrations often introduce performance overhead on certain devices, restricting their usability across the CECC. To address these challenges, we propose a new integration focusing on TVM as an external runtime for WASI-NN to enhance WASM’s performance and expand support to a broader range of devices. Additionally, we integrate this solution into Knative, a serverless framework, to provide a scalable and flexible platform for DL deployment. Using WASM technology, we evaluate our TVM-based solution through comparative studies. Results on AMD CPUs demonstrate the effectiveness of our approach, achieving 58% overall gain over other WASI-NN integrations (e.g., ONNX Runtime and OpenVINO) for CNN-based models while also achieving optimal performance on different platforms, such as Intel GPUs. These findings highlight the effectiveness of our solution.

Software-Defined Networking (SDN) has emerged as a transformative technology that revolutionizes network management and architecture by providing unparalleled flexibility and control over data traffic flows. This flexibility is increasingly crucial in managing the complex demands of modern networks, whereby efficient traffic management is essential for mitigating congestion and enhancing operational efficiency. This paper introduces a novel traffic management model that employs Deep Reinforcement Learning (DRL) to transcend the conventional limitations typically associated with routing strategies that prioritize the shortest path or make non-optimal decisions when forwarding the traffic between different peers. Our model not only reduces overall network congestion but also aims to minimize bandwidth usage and enhance routing mechanisms within SDN environments. By incorporating DRL-based load balancing mechanisms, the model intelligently redistributes traffic across multiple pathways, shifting the focus from proximity to efficiency. This strategic redistribution prioritizes routes that optimize both, transmission time and network performance, rather than merely the shortest path. Moreover, the integration of DRL allows for real-time decision-making, enabling our system to dynamically adapt to changing traffic conditions and user demands. This capability is instrumental in significantly reducing transmission times and improving the overall efficiency of traffic flow across the network. Our findings highlight the substantial benefits of integrating SDN with advanced DRL techniques, offering a pioneering perspective on traffic routing within SDN networks. We evaluated the proposed framework via simulations and the obtained results demonstrated the efficiency of our solution compared to the baseline approaches.

Nowadays, Internet of Things (IoT) devices are gaining momentum globally. However, due to their limited size, these devices have limited battery capacity, computational resources, and wireless bandwidth, making it impossible to run resource-intensive applications on these devices. Fortunately, Edge Computing has emerged as a promising solution to meet this demand by enabling data processing in more capable devices. Task offloading is a crucial technique used in Edge Computing to overcome the limitations of IoT devices by offloading some of their computational tasks to more powerful edge servers. The traditional methods used for task offloading are often based on heuristics or simple rules, which may result in sub-optimal solutions. Moreover, the increasing complexity and heterogeneity of edge networks, as well as the stochastic nature of the wireless channel, pose significant challenges for these methods. In this paper, we leverage Federated Learning (FL) to efficiently train Deep Reinforcement Learning (DRL) agents to make the best offloading and power allocation decisions by achieving the near-optimal trade-off between task execution latency and the power consumption of the end device. The obtained simulation results of the proposed method demonstrate its remarkable and superior performance in comparison to central DQN.

Abstract Deep learning (DL) is attracting considerable attention in the design of communication systems. This paper derives a deep unfolded conjugate gradient (CG) architecture for large‐scale multiple‐input multiple‐output detection. The proposed technique combines the advantages of a model‐driven approach in readily incorporating domain knowledge and deep learning in effective parameters learning. The parameters are trained via backpropagation over a data flow graph inspired from the iterative conjugate gradient method. We derive the closed‐form expressions for the gradients for parameters training and discuss early results on the performance in a statistically identical and independent distributed channel where the training overhead is considerably low. It is worth noting that the loss function is based on the residual error that is not an explicit function of the desired signal, which makes the proposed algorithm blind. As an initial framework, we will point to the inherent issues and future directions.

Centralized Massive Multiple Input Multiple Output (MIMO) uplink detection techniques for baseband processing possess severe bottleneck in terms of interconnect bandwidth and computational complexity. This problem has been addressed in the current work by adapting the centralized Newton method for decentralized MIMO uplink detection leveraging several Base Station antenna clusters. The proposed decentralized Newton (DN) method achieves error-rate performance close to centralized Zero Forcing detector as compared to other decentralized techniques. Two hardware topologies, namely the ring and the star topologies, are proposed to assess and discuss the trade-off among interconnect bandwidth and throughput, in comparison with contemporary decentralized MIMO uplink detection techniques. As such the following findings are elaborated. On BS antenna cluster scaling for different MIMO system configurations, the ring topology provides high throughput at constant interconnect bandwidth, while the star topology provides lower latency with a deterministic variation in the hardware resource consumption. Due to strategic optimizations on the hardware implementation, additional user equipment can be allotted at a fractional increase in Field Programmable Gate Array resource consumption, improved energy efficiency, and increased transaction of bits per Joule. The ring topology can process additional subcarrier at a fractional increase in latency and improved system throughput.

In this paper we consider an efficient method to resolve the underlying large matrix inversion problem inherent in the antennas' mutual coupling based reciprocity calibration. Such calibration enables the downlink pre-coding using the uplink channel estimates in a time-division-duplex (TDD) massive MIMO systems. Based on least squares estimators, a large matrix inversion is required. Herein, we derive an efficient method based on successively updating a matrix inverse by exploiting the Gram matrix structure. The simulation results reveal that our proposed method performs as well as the direct matrix inversion (based on Cholesky decomposition) whereas the approximation techniques based on Gauss Seidel (GS) and Neumann series expansions (NSE) require a large number of iterations. The proposed method is computationally efficient and lend itself for an efficient parallel and pipelined architecture implementation.

This paper aims at investigating the ac dynamic parameters of partially shaded solar PV cells. We employ analytical and numerical methods to extract the parameters of ac dynamic model of a PV cell. Particularly, the parameter identification process is based on the small signal impedance model of solar cells and its associated transfer function. This model can reflect the static and the dynamic performances of the solar cell at low and high frequencies. The analytical method utilizes mathematical equations derived from Nyquist and Bode plots, whereas the numerical method exploits Levenberg-Marquardt Least Squares (LMLS) algorithm. In this paper, we have also designed an adapted experimental laboratory circuit to measure the PV cells frequency response. The entire procedure, including the acquisition and calculation systems is completely automated using C# programming. Accordingly, we developed a low-cost setup that allows partial shading emulation as well as high power and frequency measurements. Our results demonstrate that the numerical method can achieve a better fit of data with an accuracy of 91%. Moreover, it can estimate the same parameter values in a short time with high precision.

A multistage multiuser detection (MUD) technique, the adaptive duplicated filters plus interference canceller (ADIC), is proposed in the DS-CDMA context. Of particular interest is the use of adaptive filters block (AFB) dedicated to each user with its respective input signals independent from other users' contributions. These AFB are mixed with interference canceller block in a cascade arrangement. As shown in this paper, this proposed MUD can outperform the decision feedback soft multistage interference canceller (DF-Soft-MPIC) MUD with complexity reduction of 4. Algorithmic description and performance of low complexity MUD method are considered in this paper and FPGA implementation in a companion paper, Part II.

Many multiuser detection (MUD) methods are proposed in the literature to increase the performance of 3G cellular networks. However, it is known that the implementation complexity represents a key issue for deployment of the MUD. A VLSI implementation strategy and hardware resources evaluation of a new MUD based on the adaptive duplicated filters plus interference canceller (ADIC) method (Nougaru et al., 2007), is proposed. The maximum number of users in FPGA devices is presented with respect to WCDMA constraints. The two papers provide a low complexity MUD giving a good tradeoff performance and implementation cost.

This paper investigates a beamforming receivers based on maximum ratio combining (MRC) and noise constraint least mean square (NC-LMS) using rapid prototyping method for FPGA implementation. Non-adaptive and adaptive beamforming techniques approaches are considered. A performance evaluation of these algorithms in a DS-CDMA system is presented and FPGA design is evaluated in term of hardware resources for Xilinx family devices using rapid prototyping methodology with Matlab-Simulink tools. Both approaches offer a good performance-complexity tradeoff favorable for FPGA implementation. However, due to the adaptive approach, the NC-LMS presents a better robustness to the fixed point arithmetic than the MRC

This work is devoted to the problem of asynchronous multiuser delay acquisition and time varying channel tracking in DS-CDMA systems. A multiuser-LMS-like structure along with smoothing/prediction filters to improve tracking quality is suggested. A performance versus complexity analysis is conducted, in cdma2000 and WCDMA environments, over highly interesting settings including different data rates, channel types and mobile speeds. It has been concluded that the proposed multiuser LMS structure offers improvements compared with the correlator method.

In this paper, a reduced complexity turbo detection receiver for coded DS-CDMA signals employing the BPSK modulation technique is presented. The new scheme is based on a new family of MMSE filter whose coefficients are thought to be the solution of a (forced) real valued cost function in the bit rather than a complex one as in conventional MMSE receivers. The new receiver provides, on average, 2 dB gain with less number of iterations. Simulation results for performance evaluation are conducted under the most interesting scenarios including asynchronous multipath channels, time varying channels and multirate systems.

Digital communication systems operating on time varying depressive channels often employ a signalling format in which customer data are organized in blocks proceeded by a known sequence. The training sequence at the beginning of each block is used to train an adaptive equalizer and/or data sequence detector to combat intersymbol interference (ISI). This paper addresses the problem of comparing the Schroeder sequences as a very close to optimal training sequence for channel estimation (start up) in communication systems over time dispersive frequency selective channels. Schroeder sequences of comparable lengths to the designed -computer searched- sequences demonstrated a tight performance for both the optimal sequences designed using Discrete Fourier Transform (DFT) technique and the sequences designed via Least Sum of Squared Errors (LSSE) channel estimation. Performance results are provided for Schroeder sequences of lengths 36 and 28 (the choice of 28 is driven by the fact that channel estimation sequences for GSM system are of length 28).