This paper investigates the design of binary array codes correcting a (t,s)-burst criss-cross deletion (BCCD) in a two-dimensional (2D) array, i.e., the simultaneous removal of t consecutive rows and s consecutive columns. First, a strategy for code design is built that integrates two marker sequences for determining the deletion region in each dimension, one subcode per dimension for determining the exact deletion positions, and 2D block parity-checks for recovering the deleted symbols. Then, based on representation equivalences between binary blocks and q-ary sequences, and between q-ary sequences and column-major order arrays, a construction of q-ary subcodes is proposed that divides the problem of a burst deletion in each dimension into subproblems of a single deletion or a single transposition, and integrates two types of codes to solve these subproblems. Furthermore, the subproblems of detecting a single transposition in a q-ary sequence are efficiently solved by means of signature transform that maps the problem over the q-ary alphabet onto the binary alphabet. Finally, a construction of binary array codes is proposed under the design strategy to correct any single (t,s)-BCCD. It is shown that this construction exhibits a simple structure and can achieve rapidly increasing code rates.
The sixth-generation communication networks impose higher demands on seamless connectivity and reliable transmission. Integrated satellite-terrestrial networks can provide global signal coverage, but the wireless channel characteristics make them highly vulnerable to attacks such as jamming and eavesdropping. Therefore, this paper proposes a simultaneous transmission and secret key generation approach assisted by the aerial hybrid active-passive reconfigurable intelligent surface (RIS) for integrated satellite-terrestrial networks under hybrid jamming and eavesdropping attacks. Specifically, we intend to maximize a weighted combination of the data transmission rate and the secret key generation rate per unit propulsion energy by jointly optimizing the hybrid RIS configuration, unmanned aerial vehicle (UAV) position, satellite beamforming, and transmit power while satisfying the rate-matching requirement. As the optimization problem is non-convex within the highly dynamic environment, a Group-based Comparative Proximal Policy Optimization (GC-PPO) algorithm is designed to effectively perform resource scheduling. The proposed GC-PPO-based resource scheduling approach effectively learns the jamming and eavesdropping behaviors and makes real-time adaptive decisions for the aerial hybrid activepassive RIS system through a group-based relative reward mechanism without an additional critic network, which reduces computational overhead and enhances training stability. Simulation results demonstrate that our proposed learning-based approach significantly outperforms existing algorithms in improving secrecy performance against hybrid attacks, and maintains its performance advantage under imperfect channel state information (CSI).
Distributed wireless jamming recognition based on federated learning faces three major challenges in practical edge environments, namely the lack of labeled samples at edge users, poisoning attacks during collaborative training, and severe non-independent and identically distributed (non-IID) data heterogeneity across distributed devices. To address these issues, this paper proposes an enhanced robust semi-supervised adaptive aggregation (RSSAA) framework for distributed jamming recognition, which integrates robust training and fusion into a progressive pipeline. Specifically, under the guidance of the centrally supervised model, RSSAA first develops a semi-supervised federated learning architecture that exploits unlabeled edge data to generate reliable local models. Within this stage, an exponential moving average (EMA)-based teacher-student mechanism is introduced to improve pseudo-label reliability. The teacher model produces stable pseudo-labels on weakly augmented samples, while the student model is trained on strongly augmented samples with confidence-aware consistency regularization. Building on these reliable local updates, RSSAA performs secure network-side coordination through an enhanced unlearning-based calibration strategy. This strategy constructs a multi-checkpoint clean reference model and applies adaptive interpolation-based correction according to the anomaly level of local models, thereby mitigating poisoning attacks before aggregation. Furthermore, to alleviate the adverse effect of non-IID jamming distributions, we propose a spherical k-means-based confidence-aware group aggregation method to cluster the screened and calibrated models, enabling stable and adaptive group aggregation. Finally, simulation results demonstrate that the proposed framework improves recognition robustness and training stability in complex anti-jamming environments.
Practical finite-length fading blocks and emerging short-blocklength services raise urgent demands for innovating the infinite-blocklength capacity in user-centric cell-free massive MIMO (CFmMIMO). Existing works mainly introduced a classical channel dispersion based loss term with simplified uplink processing, which lacks accuracy and does not match the complex channel conditions in CFmMIMO. By incorporating higher moment of channel uncertainties, this paper analyzes and optimizes the finite-blocklength transmission performance for large-scale fading decoding (LSFD) aided uplink user-centric CFmMIMO. Firstly, a generalized error probability bound is derived by statistical analysis of the mean, variance, and absolute third-order moment of the generalized information density. A more concise asymptotic bound along with some related properties are also deduced under large number of antennas. Then, based on the transformed critical information rate, a joint pilot and data length optimization problem is formulated to maximize the sum rate. An altering optimization algorithm combined with successive convex approximation is developed to solve the problem with low training overhead. Extensive simulation results verify the correctness of derived bounds and properties. The obtained results also show that the proposed algorithm outperforms fix-ratio schemes, and is capable of realizing near-optimal sum rate for different local combining methods.
