In this paper, we present a parallel hybrid decoder for decoding primitive rateless (PR) codes. The proposed decoder integrates multiple-bases belief propagation (MBBP) with low-order ordered-statistics decoding (OSD), effectively reducing the complexity associated with high-order OSD, especially for low-rate codes operating in low signal-to-noise ratio (SNR) regimes. We evaluate the block error rate (BLER) performance of the decoder and propose an optimization method using a genetic algorithm to generate diverse parity-check matrices that further minimize the BLER. Simulation results demonstrate that PR codes decoded with the proposed low-complexity parallel hybrid decoder achieve performance very close to the normal approximation (NA) benchmark. PR codes, being adaptable to any desired length and rate, coupled with the proposed decoder, emerge as a promising solution for short-packet communication scenarios demanding stringent latency and reliability.

In the global navigation satellite system (GNSS), identifying not only single but also compound jamming signals is crucial for ensuring reliable navigation and positioning, particularly in future wireless communication scenarios such as the space-air-ground integrated network (SAGIN). However, conventional techniques often struggle with low recognition accuracy and high computational complexity, especially under low jamming-to-noise ratio (JNR) conditions. To overcome the challenge of accurately identifying compound jamming signals embedded within GNSS signals, we propose ACSNet, a novel convolutional neural network designed specifically for this purpose. Unlike conventional methods that tend to exhibit lower accuracy and higher computational demands, particularly in low JNR environments, ACSNet addresses these issues by integrating asymmetric convolution blocks, which improve sensitivity to subtle signal variations while reducing the number of parameters by approximately 50% compared to symmetric convolutional designs. Simulations demonstrate that ACSNet significantly improves accuracy in low JNR regions and shows robust resilience to power ratio (PR) variations. It achieves an overall accuracy of 91.84% and a Kappa coefficient (×100) of 90.82, and notably reaches near 100% recognition accuracy when the JNR is greater than or equal to −9 dB, confirming its effectiveness and efficiency for practical GNSS interference management applications.

In this paper, we study the problem of distributed learning (DL), where different devices take varying amounts of response time to complete local computations during training. To deal with this problem, existing gradient coding (GC) techniques are designed in a synchronous setting, where the server waits for a fixed number of the fastest devices to transmit the coded gradients in each iteration to fully recover the global gradient. However, there are significant inefficiencies in these GC techniques, since the computations performed by the slow devices are wasted and their information is not required by the server to update the global model. To overcome this limitation, we propose a novel asynchronous DL method based on GC (AGC). In AGC, the training data subsets are replicated and allocated to the devices redundantly, and the devices encode local gradients of its local subsets and send the coded gradients to the server. The server updates the global model in an asynchronous manner based on the coded gradients received from a certain number of devices, which may be computed from the global models back in history as well as the most recent one. In this way, the computations of the slow devices can also be exploited, helping to increase training efficiency and reduce training time. We provide a convergence analysis for AGC in asynchronous settings with frequent global synchronization, and show that it achieves an O(1√T) convergence rate. Finally, numerical results demonstrate the superiority of the proposed method compared to the baseline methods, where AGC achieves better learning performance after the same training time.

Traffic status monitoring is essential to support intelligent transportation, and demands for highly efficient and secure processing and transmission of massive data. Inspired by recent information-theoretic study, the traffic status monitoring is formulated as a privacy-preserving semantic coding problem in this work to characterize the fundamental trade-off among data compression, traffic image reconstruction, segmentation of traffic participants, and preservation of pixel regions of traffic participants. Grounded on the theoretic results, a novel generative semantic coding scheme is further developed for data-driven semantic coding design with the aid of attention mechanisms to capture correlations between sparse information. Experiments on the RCooper dataset demonstrate the effectiveness of the proposed generative semantic coding scheme and its advantages over the benchmark compression methods.

Communication bottlenecks and the presence of stragglers pose significant challenges in distributed learning (DL). To deal with these challenges, recent advances leverage unbiased compression functions and gradient coding. However, the significant benefits of biased compression remain largely unexplored. To close this gap, we propose Compressed Gradient Coding with Error Feedback (COCO-EF), a novel DL method that combines gradient coding with biased compression to mitigate straggler effects and reduce communication costs. In each iteration, non-straggler devices encode local gradients from redundantly allocated training data, incorporate prior compression errors, and compress the results using biased compression functions before transmission. The server aggregates these compressed messages from the non-stragglers to approximate the global gradient for model updates. We provide rigorous theoretical convergence guarantees for COCO-EF and validate its superior learning performance over baseline methods through empirical evaluations. As far as we know, we are among the first to rigorously demonstrate that biased compression has substantial benefits in DL, when gradient coding is employed to cope with stragglers.

Hierarchical federated learning (HFL) has emerged as an effective paradigm to enhance link quality between clients and the server. However, ensuring model accuracy while preserving privacy under unreliable communication remains a key challenge in HFL, as the coordination among privacy noise can be randomly disrupted. To address this limitation, we propose a robust hierarchical secure aggregation scheme, termed H-SecCoGC, which integrates coding strategies to enforce structured aggregation. The proposed scheme not only ensures accurate global model construction under varying levels of privacy, but also avoids the partial participation issue, thereby significantly improving robustness, privacy preservation, and learning efficiency. Both theoretical analyses and experimental results demonstrate the superiority of our scheme under unreliable communication across arbitrarily strong privacy guarantees.

Emerging wireless networks such as the sixth-generation (6G) mobile systems are expected to operate over highly heterogeneous infrastructures with rapidly evolving service demands, thereby increasing the dependence on large-scale, constraint-coupled, cross-layer optimization for network design and operation. However, the trend leads to a fundamental bottleneck: Converting high-level intents into mathematically consistent formulations, implementable algorithms, and reproducible simulations remains predominantly human-driven, time-intensive, and error-prone. While large language models (LLMs) provide a natural-language interface for intent interpretation and rapid prototyping, monolithic LLM-based pipelines are often limited by insufficient domain grounding, weak constraint awareness, and a lack of execution-based verification and self-correction. These limitations motivate a shift towards agentic AI, where problem-solving is realized through iterative decomposition, explicit planning, tool-integrated execution, and reflection driven by feedback. In this paper, we present ComAgent, a multi-LLM based agentic AI framework that coordinates specialized agents for literature searching, planning, coding, and solution scoring within a closed-loop Perception–Planning–Action–Reflection cycle, turning user intents into solver-ready formulations and reproducible simulation pipelines while continuously self-correcting logical and feasibility errors. We demonstrate the efficacy of ComAgent through two distinct evaluations. In a non-trivial beamforming optimization case study, ComAgent autonomously perceives the problem, designs an algorithm, and generates solutions that achieve a performance comparable to that of expert-designed baselines. Furthermore, on a diverse set of generic wireless tasks, ComAgent outperforms monolithic LLMs. The numerical results demonstrate the potential of the proposed agentic AI framework for various emerging wireless networks.

Wireless virtual reality (VR) demands ultra-high resolution, ultra-low latency, and real-time computationally intensive viewport rendering, which jointly impose stringent requirements on communication and computing. To enhance the quality of experience for VR users, this paper investigates a mobile edge computing assisted VR streaming system employing extremely large-scale multiple-input multiple-output (XL-MIMO) at the base station (BS). Considering smoothness, clarity, and duration of the VR experience, we formulate a joint optimization problem that minimizes the weighted energy consumption by optimizing the transmit association between VR users and BS subarrays, quantization parameter (QP), rendering offloading decision, beamforming vector, and computing resource allocation. To tackle the high-dimensional mixed-integer non-convex problem efficiently, we propose a hierarchical collaborative optimization based joint rendering offloading and resource allocation (HC-JRORA) algorithm. The upper layer leverages traditional convex optimization methods to determine the computing resource allocation, QP selection, and beamforming design, while the lower layer optimizes transmit association and rendering offloading decision based on a dueling double deep Q network (D3QN) enhanced with imitation learning. Simulation results show that the proposed HC-JRORA algorithm can achieve lower weighted energy consumption and can provide optimized solutions faster than traditional iterative optimization methods.

In this contribution, a novel differential space-time block coded generalized spatial modulation (D-STBC-GSM) scheme is proposed, which integrates space-time block coding into the differential multi-generalized spatial modulation (D-MGSM) framework in order to exploit transmit diversity while preserving flexible system configurations. Such a non-coherent and low-complexity transmission paradigm is particularly appealing for emerging upper mid-band (7-24 GHz) 6G deployments, where rapid channel variations and hardware constraints render accurate channel state information acquisition increasingly challenging. To improve the reliability of frame-based differential detection, a power allocation strategy is further developed to reinforce the reference symbols within each transmission frame. Moreover, a low-complexity detection algorithm is devised by exploiting the inherent orthogonality of the embedded Alamouti structure, achieving near-maximum-likelihood performance with substantially reduced computational complexity. Finally, simulation results demonstrate that the proposed D-STBC-GSM scheme consistently outperforms conventional D-MGSM and other differential spatial modulation schemes in terms of bit error rate performance.

Virtual reality (VR) enables immersive experiences for remote monitoring, guidance, and control, improving operational efficiency across Internet of things-enabled and other cyber-physical applications. However, VR services characterize with ultra-high resolution, ultra-low latency, and frequent computation-intensive viewport rendering, posing stringent requirements on wireless communication and computation. In this paper, we propose a mobile edge computing assisted VR streaming transmission scheme to improve quality of experience (QoE) and alleviate computing burden of VR devices. A joint optimization problem is formulated to minimize the weighted energy consumption under QoE constraints. To solve the high-dimensional mixed integer non-convex problem, we propose a hierarchical optimization framework. The upper layer leverages convex optimization methods, while the lower layer uses dueling double Q network enhanced with imitation learning for binary decisions. Simulation results show that the proposed algorithm can achieve lower energy consumption and adapts faster to dynamic environments.