Recent advances in multimodal large language models (MLLMs) have enabled impressive progress in visual-language reasoning, yet long-video understanding remains a formidable challenge due to the need for coherent reasoning over ultra-long spatiotemporal dependencies. Existing methods struggle with the vast candidate space for relevant information in long videos, often failing to distinguish meaningful events from redundant content. We identify two critical and previously under-explored issues: absolute redundancy, where static visual content inflates token counts without adding narrative value, and relative redundancy, where task-irrelevant segments introduce noise that impairs reasoning. Compounding these issues is the weak spatiotemporal modeling in current MLLMs, which limits their ability to capture complex event dynamics. To address these multifaceted challenges, we introduce SELongVLM, a dynamically lenient-to-stringent selection long video language model. SELongVLM integrates two coordinated branches: a Residual Token Pruner (RTP) that removes repetitive background tokens via inter-frame residual modeling thus mitigating absolute redundancy while preserving motion cues, and a Semantic-aware Self-Correction Selector (SCSelector) that progressively refines query-relevant clip selection without frame-level annotations to reduce relative redundancy, guided by a stringent-to-lenient self-correcting mechanism during optimization. To ensure causal continuity and bolster spatiotemporal reasoning across disjoint clips, the framework further incorporates an action-aware operation for intra-clip dynamics and a temporal memory for cross-clip context, enabling robust spatiotemporal inference on long videos. Extensive experiments across eight benchmarks demonstrate that SELongVLM markedly outperforms existing models on both general and specialized long-video tasks. Specifically, it achieves 65.5% on VideoMME and 69.8% on MLVU for general benchmarks, and delivers strong performance on four specialized benchmarks - for example, 39.2% on TOMATO for fine-grained temporal reasoning and 69.2% on EventBench for event-level understanding.

Wireless sensing enables contactless and accurate recognition of human activities and physiological states by using electromagnetic signals. As a promising enabler for sixth-generation (6 G) multi-functional networks, orthogonal time frequency space (OTFS) modulation exhibits strong resilience to high Doppler shifts in high-mobility environments, while also supporting precise human sensing in low-mobility scenarios. In this work, we propose a novel two-dimensional (2D) delay-Doppler motion profiling framework based on the OTFS waveform to extract distinctive features of human activities. To enhance recognition performance, a fractional-Doppler enhancement network is integrated with a convolutional neural network (CNN)-aided encoder-only Transformer architecture. Extensive experiments are conducted to assess the cross-domain generalization capability of the proposed OTFSensi system. Compared with existing classification models based on CNN, gated recurrent unit (GRU), and long short-term memory (LSTM) networks, OTFSensi demonstrates substantial improvements in adaptability across diverse environments and observation angles. Furthermore, a comparative analysis with various radio frequency (RF) sensing technologies confirms the superior classification performance achieved by OTFSensi.

Integrated sensing and communication (ISAC) has been considered a key feature of next-generation wireless networks. This paper investigates the joint design of the radar receive filter and dual-functional transmit waveform for the multiple-input multiple-output (MIMO) ISAC system. While optimizing the mean square error (MSE) of the radar receive spatial response and maximizing the achievable rate at the communication receiver, besides the constraints of full-power radar receiving filter and unimodular transmit sequence, we control the maximum range sidelobe level, which is often overlooked in existing ISAC waveform design literature, for better radar imaging performance. To solve the formulated optimization problem with convex and nonconvex constraints, we propose an inexact augmented Lagrangian method (ALM) algorithm. For each subproblem in the proposed inexact ALM algorithm, we custom-design a block successive upper-bound minimization (BSUM) scheme with closed-form solutions for all blocks of the variable to enhance the computational efficiency. Convergence analysis shows that the proposed algorithm is guaranteed to provide a stationary and feasible solution. Extensive simulations are performed to investigate the impact of different system parameters on communication and radar imaging performance. Comparison with the existing works shows the superiority of the proposed algorithm.

Orthogonal Time-Frequency Space (OTFS) modulation is an emerging technique that characterizes wireless channels and transmits information in the delay-Doppler domain. This work focuses on estimating fundamental sensing parameters, i.e., the delay and Doppler shifts of individual propagation paths, which serve as critical enablers for downstream positioning techniques, such as time-difference-of-arrival (TDOA)-based localization. Specifically, we propose a parameter-inherited (PI) channel estimation method that integrates sparse Bayesian learning (SBL) with unitary approximate message passing (UAMP), achieving low computational complexity and high estimation robustness. To accelerate the convergence of the UAMP-based iterative estimation, we explore the strategy of initializing parameters by inheriting prior estimates from adjacent OTFS transmission blocks. Furthermore, the overall computational burden is significantly reduced by employing large-scale matrix operations via two-dimensional fast Fourier transform (2D FFT). The proposed algorithms are implemented and evaluated on a software-defined radio (SDR)-based ISAC platform. Experimental results demonstrate that the proposed dual-functional system outperforms existing benchmarks in both communication quality and sensing parameter accuracy.

Orthogonal time frequency space (OTFS) modulation is anticipated to be a promising candidate for supporting integrated sensing and communications (ISAC) systems, which is considered as a pivotal technique for realizing next-generation wireless networks. In this paper, we develop a minimum bit error rate (BER) precoder design for an OTFS-based ISAC system. In particular, the BER minimization problem takes into account the maximum available transmission power budget and the required sensing performance. Unlike previous studies that focused on ISAC in the time-frequency (TF) domain, we devise the precoder from the perspective of the delay-Doppler (DD) domain by exploiting the equivalent DD domain channel. The DD domain channel generally tends to be sparse and quasi-static, which is conducive to a low-complexity ISAC system design. To address the non-convex optimization design problem, we resort to optimizing the lower bound of the derived average BER by adopting Jensen’s inequality. Subsequently, the formulated problem is decoupled into two independent sub-problems via singular value decomposition (SVD) methodology. We then theoretically analyze the feasibility conditions of the proposed problem and present a low-complexity iterative solution via leveraging the Lagrangian duality approach. Simulation results verify the effectiveness of our proposed precoder compared to the benchmark schemes and reveal the interplay between sensing and communication for dual-functional precoder design, indicating a trade-off where transmission efficiency is sacrificed for increasing transmission reliability and sensing accuracy.

To leverage distributed data communication and learning in sensor networks effectively, edge learning (EL) methods have garnered significant attention. In the realm of distributed sensor networks, achieving consensus estimation of interested variables stands as a pivotal challenge. To address this challenge using EL methods, several approaches have been proposed combining message passing (MP) algorithms. In this article, we first describe the distributed consensus algorithm based on MP and summarize the sampling-based and parameter-based representation of the beliefs exchanged in the distributed MP algorithm. To improve the accuracy of estimation while retaining the low-complexity advantage of the parametric representation method, we propose a distributed consensus framework based on the Gaussian mixture model (GMM) MP. We approximate and keep the form beliefs as GMM in the iterations. Two different simulation scenarios are performed to shed light on the proposed distributed consensus estimation framework, i.e., static target localization and dynamic target tracking. Finally, simulation results show the performance advantages of the algorithm proposed.

It is anticipated that integrated sensing and communications (ISAC) would be one of the key enablers of next-generation wireless networks (such as beyond 5G (B5G) and 6G) for supporting a variety of emerging applications. In this paper, we provide a comprehensive review of the recent advances in ISAC systems, with a particular focus on their foundations, physical-layer system design, networking aspects and ISAC applications. Furthermore, we discuss the corresponding open questions of the above that emerged in each issue. Hence, we commence with the information theory of sensing and communications (S&C), followed by the information-theoretic limits of ISAC systems by shedding light on the fundamental performance metrics. Next, we discuss their clock synchronization and phase offset problems, the associated Pareto-optimal signaling strategies, as well as the associated super-resolution physical-layer ISAC system design. Moreover, we envision that ISAC ushers in a paradigm shift for the future cellular networks relying on network sensing, transforming the classic cellular architecture, cross-layer resource management methods, and transmission protocols. In ISAC applications, we further highlight the security and privacy issues of wireless sensing. Finally, we close by studying the recent advances in a representative ISAC use case, namely the multi-object multi-task (MOMT) recognition problem using wireless signals.

The recently proposed orthogonal time frequency space (OTFS) modulation multiplexes data symbols in the delay-Doppler (DD) domain. Since the range and velocity, which can be derived from the delay and Doppler shifts, are the parameters of interest for radar sensing, it is natural to consider implementing DD signal processing for radar sensing. In this paper, we investigate the potential connections between the OTFS and DD domain radar signal processing. Our analysis shows that the range-Doppler matrix computing process in radar sensing is exactly the demodulation of OTFS with a rectangular pulse shaping filter. Furthermore, we propose a two-dimensional (2D) correlation-based algorithm to estimate the fractional delay and Doppler parameters for radar sensing. Simulation results show that the proposed algorithm can efficiently obtain the delay and Doppler shifts associated with multiple targets.

Recent progress in self-interference cancellation technology has demonstrated the in-band full-duplex wireless communication. However, the inter-client interference caused by the simultaneous co-channel uplink and downlink transmission in the asymmetric full-duplex communication can decrease the network throughput. The conventional solutions to this problem are finding the hidden stations by frequent inter-client channel probing, which caused a lot of signaling overhead. In this paper, we propose an asymmetric full-duplex medium access control (MAC) protocol using the beam divergence feature of the orbital angular momentum (OAM) waves. The noteworthy features of our protocol consist of two aspects. Firstly, the protocol avoids the inter-client interference through the divergence feature of the OAM wave, which does not need frequent inter-client channel probing and reduces the signaling overhead compared with the conventional solutions. Secondly, asymmetric full-duplex communication can be established in high probability with the appropriate multiplexing of OAM modes. Through the simulation results, the throughput of the asymmetric full-duplex network outperforms the other proposed full-duplex MAC protocol and half-duplex network.

Accurate traffic classification is critical for network QoS provisioning and cyberspace security. Recently, classifying different traffic using convolutional neural networks (CNN) has achieved high accuracy. However, these large CNN models have millions of parameters, which are not suitable for edge computing hardware deployment. In this work, we propose a compressed network in network (NIN) model for traffic identification. A stepwise pruning and knowledge distillation (KD) is designed for training the compressed model, which aims at reducing storage and computing resources. Our method is validated with the public ISCX VPN-nonVPN traffic dataset. Experimental results show that without degrading classification accuracy, our minimum model can save more than 50% of the number of parameters and 30% of the computation time comparing with the uncompressed NIN model. The test set average F1score of 0.9805 of the minimum model is higher than that of the state-of-the-art model, which is a CNN model.

As a typical and prominent component of the Internet of Things, vehicular communication and the corresponding vehicular networks (VNETs) are promising to improve spectral efficiency, decrease transmission delay, and increase reliability. The ever-increasing number of vehicles and the demand of passengers/drivers for rich multimedium services bring key challenges to VNETs, which requiring huge capacity, ultralow delay, and ultrahigh reliability. To meet these performance requirements, a fog computing-based VNET is presented in this article, where the resource allocation as the corresponding key technique is researched. In particular, joint optimization of user association and radio resource allocation scheme is investigated to minimize the transmission delay of the concerned VNET. The proposed optimization problem is formulated as a mixed-integer nonlinear program and transformed into a convex problem by Perron–Frobenius theory and a weighted minimum mean square error method. Numerical results show that the proposed solution can significantly reduce the transmission delay with fast convergence.

As the evolution of cloud radio access networks (CRANs), heterogeneous cloud radio access networks (H- CRANs) are now recognized as promising paradigm to achieve high spectral and energy efficiency through taking advantages of both heterogeneous networks and C- RANs. In H-CRANs, the heterogeneous processing node (HPN) guarantees the basic quality of service (QoS) requirement for the user equipment, while remote radio heads (RRHs) are deployed to provide enhanced QoS performances. Inter-tier interference between HPNs and RRHs should be coordinated for achieving high throughput gains in H-CRANs. In this paper, the transmit power for both RRHs and HPNs are researched to mitigate this inter-tier interference. The throughput maximizing problem with and without interference coordination under the power and interference constraints are developed. Since this kind of optimization problem is not convex, these two non- convex optimization problems are transformed into the form of matrix. Through the Perron-Frobenius theory, the optimal power allocation solution is derived. Simulation results show that the proposed solution is converged, and it can achieve significant performance gains.

Energy harvesting is necessary to make the wireless network self-sustaining and self-organizing regardless of the traditional power grid. How to allocate the limited radio resources in the energy harvesting wireless network is a challenging work. This paper focuses on optimizing the time and power related resources in the multi-cell scenario to maximize the throughput constraining of the changeable energy in the base station. The optimal off-line resource allocation strategy is proposed, based on analysis of structural properties of the optimal total power sequences. To decrease the complexity of the optimal solution, a low complexity suboptimal off-line algorithm is presented based on the nature of the concave function. Furthermore, inspired by the off-line algorithm, an on-line resource allocation algorithm is proposed as well. Simulation results show the suboptimal offline algorithm closely tracks the performance of the optimal solution. And the proposed on-line algorithm also has brilliant performance compared with several kinds of algorithms under different system settings.