Cell-free massive MIMO with power-domain NOMA (CF-mMIMO-NOMA) is a promising architecture for uniform coverage and high spectral efficiency, yet its practical deployment is constrained by pilot overhead, heterogeneous (non-IID) channel statistics across distributed access points (APs), and error-sensitive successive interference cancellation (SIC). Existing learning-aided solutions are typically centralized or ignore cross-AP heterogeneity and feasibility-aware clustering, which limits robustness and scalability. This paper proposes VR-FTLNet, a federated transfer-learning framework that jointly supports CSI refinement, learning-assisted user pairing with feasibility regularization, SIC ordering, and downlink power allocation. A backbone network is pretrained offline and then fine-tuned locally at each AP using pilot-domain and CSI-derived features, while a variance-reduced federated aggregation with control-variate correction mitigates client drift under non-IID data. The trained model enables low-latency inference at the APs without sharing raw CSI, with fronthaul signaling explicitly accounting for compressed model and control-variate updates. Simulation results demonstrate improved channel-estimation accuracy, fewer infeasible NOMA clusters, and reduced SIC failures compared with conventional clustering, NOMA heuristics, and standard FL baselines, while remaining close to the centralized DL upper bound.
As Internet of Things (IoT) devices continue to proliferate, their reliability is increasingly undermined by security concerns. Many attackers exploit malware in IoT devices to launch attacks, leading to large-scale data breaches or device damage. Consequently, malware analysis has become particularly important because it can not only help study and classify existing malware and devise unified defense strategies, but also enable the detection of malicious program behaviors and network flows, thereby preventing potential attacks. However, malware analysis has never been a straightforward task. Attackers are growing more sophisticated, disguising their malware with varied techniques as new strains emerge rapidly. Although traditional approaches extract useful features from IoT programs for analysis, they fall short in keeping pace with the ever-evolving landscape of malware. To address these challenges, this survey systematically reviews and analyzes the key research problems and emerging directions in malware analysis. Specifically, we categorize existing techniques into four groups based on their analytical approaches, discuss the types of features and the processing methods, and examine the challenges each category faces along with potential solutions. Finally, we compare the strengths and limitations of current techniques, highlight open challenges, and outline promising avenues for future research. This survey provides newcomers with a comprehensive entry point into the field and offers security experts a broader perspective to inspire further exploration in IoT malware detection.
High-frequency (HF) communication systems, while enabling long-distance beyond-line-of-sight links via ionospheric reflection, face significant challenges when adopting modern wideband digital modulation due to large delay spreads (up to several milliseconds) and high Doppler shifts caused by ionospheric dynamics. While Orthogonal Time Frequency Space (OTFS) modulation offers robustness in such environments, its direct application to HF systems incurs prohibitive computational complexity due to the massive dimensions of the delay-Doppler grid. To address this, we propose a block-enabled OTFS framework that segments the delay-Doppler domain into smaller sub-blocks, enabling low-complexity detection. A repetition-based coding scheme across blocks is introduced to enhance diversity and reliability without bandwidth expansion. The joint design reduces the computational complexity by orders of magnitude while maintaining or improving bit error rate (BER) performance. Simulation results demonstrate that the proposed approach achieves a performance gain exceeding 10 dB over OFDM with single-tap MMSE equalization in wideband HF communication.
A major challenge in multi-robot systems is achieving formation preservation together with real-time collision avoidance, as existing methods often decouple formation control, obstacle avoidance, and navigation, leading to suboptimal performance and poor scalability. To address this issue, we propose a distributed formation control framework that enables multiple robots to navigate in obstacle-rich environments while maintaining their designated formation through a similarity-based error metric. The formation state is modeled by virtual convex polygons capturing position, velocity, orientation, and scale, providing a richer representation than traditional geometric constraints. To handle environmental constraints, an obstacle avoidance strategy based on the artificial potential field method samples polygon edges to determine the desired formation state. Each robot applies a hybrid front-end path planning process that combines an extended heuristic kinematic A* search with topological planning, guided by the preferred formation state. The initial trajectories are refined through iterative graph-based optimization that reduces similarity errors and produces optimal robot placements. Simulations and real-world experiments with quadruped and wheeled robots demonstrate the feasibility of the proposed method. Compared with state-of-the-art baselines, the proposed method achieves a competitive average formation similarity of 94.5% while reducing the average formation replanning time by 31.7% compared with DRRT-FGS across the three obstacle densities, demonstrating a favorable balance between formation preservation and online planning efficiency. Video link: https://youtu.be/5iOcBqDYWc.
