Few-shot specific emitter identification (FS-SEI) is crucial for distinguishing individual transmitters when only limited labeled samples are available. Existing FS-SEI methods often rely heavily on deep feature extraction, while the resulting representations usually lack clear physical interpretability, thereby limiting model transparency and generalization capability. To address this issue, this letter proposes a robust FS-SEI method based on a feature-family-wise mixture of experts (FF-MoE). Unlike existing approaches that rely on implicit feature fusion, the proposed method explicitly decomposes RF representations into physically meaningful feature families and performs adaptive weighting of each expert. This structured fusion improves both generalization and interpretability under limited supervision. Experimental results show that, with only 10 samples per class, the proposed FF-MoE achieves accuracies of 90% and 64% on the WiFi and UAV datasets, respectively. The source code of this work is available on the following GitHub repository: https://github.com/Minyu-Guo/FF-MoE-for-FS-SEI.
Optical wireless communication (OWC) systems that employ intensity-modulation with direct-detection (IM/DD) require the transmitted waveform to be real-valued and strictly non-negative. Conventional asymmetrically clipped optical orthogonal frequency division multiplexing (ACO-OFDM) satisfies this constraint, yet is vulnerable to inter-symbol interference, posing performance degradation. To address the dual issues of signal non-negativity and multipath interference, this paper proposes asymmetrically clipped optical affine frequency division multiplexing (ACO-AFDM), a novel modulation scheme that embeds the affine Fourier domain into the ACO framework. Specifically, a real cosine chirp weighting is applied after the inverse discrete Fourier transform. Unlike a standard DAFT, the resulting transform is nonunitary. This design ensures compliance with IM/DD requirements under the stated half-wave conditions. We derive the exact equivalent channel and identify a sufficient pairwise rank condition for full path diversity without claiming codebook-wide full diversity. Under explicitly stated power, channel, and receiver assumptions, the numerical study evaluates BER, robustness to the number of paths, and computational complexity. We also report the exact delay–chirp-frequency support and an analytical bound on front-end distortion without relying on an unproven universal two-branch separation property.
In low Earth orbit (LEO) Internet-of-Things (IoT) links, high Doppler rates cause severe integration loss in direct-sequence spread-spectrum (DSSS) receivers, rendering closed-loop carrier tracking ineffective. This letter proposes physics-constrained dynamic programming (PC-DP) for open-loop Doppler ridge extraction under co-frequency interference. By translating the orbital acceleration bound into a hard reachability constraint with a closed-form penalty, PC-DP structurally suppresses cross-ridge jumps and reduces complexity from O(NM2) to O(NM). Simulations confirm significant slope accuracy gains, uniquely achieving a bit error rate (BER) <10−3 at −15 dB signal-to-noise ratio (SNR), outperforming traditional carrier loops.
This letter presents an OFDM-based hybrid beam-forming (HBF) design for wideband extremely large-scale MIMO (XL-MIMO) systems in a hybrid near-field/far-field (hybrid-field) environment. As apertures grow in Sub-THz and millimeter-wave systems, the Rayleigh distance increases substantially and near-field (NF) and far-field (FF) components coexist within a single link, where conventional FF-only beamforming suffers severe inter-field interference. We formulate field separation within an oblique-projection framework, which isolates a signal subspace from an interfering one using only basis matrices of the two subspaces, without requiring orthogonality. Stage 1 builds the FF subspace from the NLoS component and extracts the NF subspace as its orthogonal-complement residual; this sequential construction makes the two subspaces exactly orthogonal, so the oblique projectors reduce to orthogonal projectors requiring no inversion or regularization. A single virtual analog matrix is then formed from the separated components—computed once from the frequency-averaged channel, since the analog beamformer is shared across subcarriers—and FRF is extracted via its SVD. Stage 2 designs per-subcarrier baseband matrices FBB[k] via SVD of the effective channel seen through FRF. Simulations show superior spectral efficiency over FF-only and NF-only HBF across the entire SNR range.
Outdoor-to-indoor sub-THz ISAC links need a service-angle penetration- loss margin before beam and rate adaptation. This letter studies calibrated power-only sensing for link-margin prediction; wall thickness and relative permittivity are compact nuisance states under known geometry, offsets, and the ITU-R P.2040-4 material-loss prior. We show that the local log-parameter Fisher information matrix of the incoherent P.2040-4 model is near-degenerate at a single incidence angle because bandwidth changes absorption scale rather than gradient direction. A variance-aware two-angle design lowers the 30° service-angle loss-prediction standard deviation from 1.1 dB to 0.07 dB, releasing about 1.7 dB of one-sided link margin. Coherent TMM/CST stress tests and mismatched MLE checks quantify model scope.
Polar codes are widely used for reliable wireless access, but their decoding latency remains a practical concern. Existing reliability-oriented constructions optimize bit-channel reliability, while overlooking the decoder-dependent latency associated with special-node-based fast decoding. This paper proposes a decoder-profile-aware polar code construction method that analyzes the decoding-tree structure under a given decoder profile and refines the information set through reliability-constrained bit exchange. Using the full-rank fast parallelized list (F-FPL) decoder profile as a representative profile, the proposed construction reduces decoding time steps by up to 45.1% compared with the 5G baseline while maintaining comparable block error rate (BLER) performance.
We present a soft-input soft-output (SISO) decoding algorithm for binary block codes, based on ordered reliability direct error-pattern testing (ORDEPT) and extended covered space approximation. The method can be used for iterative decoding of product-like and generalized low-density parity-check (GLDPC) codes. Our results obtained for product and GLDPC codes with extended Bose-Chaudhuri-Hocquenghem (BCH) component codes show that the proposed decoder demonstrates improved performance, often achieved with much smaller complexity compared to the state-of-the-art SISO decoders.
High-power amplifier (HPA) nonlinearity distorts orthogonal time frequency space (OTFS) signals and degrades pilot observations, thereby posing significant challenges to channel estimation, particularly under severe Doppler shifts in high-mobility scenarios. This letter proposes a robust OTFS channel estimation method combining a Zadoff–Chu (ZC)-based embedded pilot with a low peak-to-average power ratio (PAPR), correlation-based feature enhancement, and a long short-term memory (LSTM) estimator with self-attention. The low-PAPR pilot mitigates nonlinear distortion, while cross-correlation with a locally generated ZC reference enhances delay–Doppler path responses and suppresses interference. Simulations show that the proposed method achieves lower bit error rate (BER) than practical benchmarks across a wide range of signal-to-noise ratios (SNRs) under different mobile speeds, input back-off levels, and modulation schemes.
Due to the suitability for high-mobility scenarios, orthogonal time frequency space (OTFS) is considered a potential waveform for realizing integrated sensing and communication (ISAC). In this letter, we propose a low-complexity fractional delay and Doppler estimation algorithm based on the orthogonal matching pursuit (OMP) framework. The algorithm complexity is reduced by constructing a dimensionality-reduced observation model and pruning noise grids. Meanwhile, by performing inter-path interference (IPI) decision on the targets, only the parameters in the interfered dimension are corrected, thereby resolving the inter-path interference. Simulation results show that the proposed scheme significantly reduces computational complexity. Meanwhile, in the absence of IPI, the estimation accuracy is comparable to the benchmark algorithms, and in the presence of IPI, the estimation accuracy is significantly superior to the benchmark algorithms.
DNA data storage is vulnerable to insertion and deletion (indel) errors, and reliable recovery requires efficient error correction. Existing methods do not fully exploit soft information, which limits decoder performance. In this letter, we develop an always-aligned half-marker-aided soft information generation algorithm for DNA data storage with accompanying long composite ranging codes (LCRCs). This algorithm constructs always-aligned half-markers for each read according to the identified LCRC segments, providing prior synchronization information for forward-backward decoding and cross-read consensus. It facilitates indel handling and reliable soft information estimation for error correction codes. Evaluations are performed on datasets generated from different DNA synthesis and sequencing platforms. Experimental results demonstrate that the proposed method increases the mutual information between binary inputs and soft outputs compared with hard-decision methods. Moreover, it effectively reduces the required sequencing coverage for reliable recovery even under severe indel scenarios, with a moderate increase in complexity.
We propose a task-aware semantic split learning (SL) framework for wireless edge-cloud inference, in which the reliability of transmitted latent representations is dynamically adapted to their relevance for the downstream task. An autoencoder (AE)-based physical (PHY) layer enables end-to-end learning of the communication interface, while unequal error protection (UEP) is realized via mutual information (MI)-driven prioritization of latent components during training. The gradient of the estimated MI with respect to each latent component serves as a sensitivity-based proxy for task relevance, providing a fully learning-driven prioritization that adapts to both the data distribution and the downstream task. We further show that this prioritization translates into measurable physical-layer effects: MI-guided UEP assigns significantly higher transmit power to the most task-critical latent components compared to the equal error protection (EEP) baseline. Experiments on real-world IoT sensing data demonstrate consistent gains over equal and fixed-UEP baselines across SNR regimes. Additional analysis confirms ranking stability, estimator robustness and generalization across datasets and task types, indicating broad applicability of the proposed framework.
Asynchronous distributed chirp spread spectrum (AD-CSS) enables cooperative long-range, low-power Internet of Things (IoT) communication but suffers from independent time, frequency, and phase offsets, leading to non-coherent signal addition and degraded detection performance. Conventional receivers relying on explicit offset and channel estimation become computationally burdensome and inaccurate in such settings. To address this, we propose Reg-GelNet, a deep learning–based receiver that performs offset and channel compensation along with symbol detection. Reg-GelNet uses a deep neural network (DNN) to mitigate signal impairments, a Gel’fand transform for compact time–frequency feature extraction, and a convolutional neural network (CNN) for symbol classification, removing the need for explicit synchronization or channel estimation. Simulations show that Reg-GelNet achieves markedly lower bit error rate (BER) than state-of-the-art receivers while maintaining low complexity, offering an efficient and practical solution for asynchronous IoT networks.
Rate-splitting multiple access (RSMA) is a promising technique for massive MIMO systems, while hybrid precoding provides an effective means to balance transmission performance and energy consumption. However, existing studies mainly focus on fully-connected architectures, while hybrid precoding design for massive MIMO-RSMA systems with energy-efficient partially-connected architecture remains unexplored in the literature. To fill in this research gap, we propose an adaptive cross-entropy (ACE)-based hybrid precoding scheme. The analog precoder is designed using a probabilistic search over discrete phases, iteratively sampling and updating selection probabilities. Then, the weighted minimum mean-squared error (WMMSE) algorithm is employed to jointly optimize the digital precoder and the common rate allocation. Simulation results demonstrate that the proposed scheme maintains competitive MMF performance while achieving superior energy efficiency.
Specific Emitter Identification (SEI) is important for enhancing physical-layer security in resource-constrained Internet of Things (IoT) networks. However, Deep Learning (DL)-based SEI methods often incur high computational complexity and suffer performance degradation under noisy conditions. To address these issues, this letter proposes Broad Learning-Based Mixture of Experts (BLMoE) with Dynamic Entropy-Residual (DER) fusion, referred to as DER-BLMoE. DER-BLMoE constructs lightweight Broad Learning System (BLS) experts over frequency sub-bands, allocates expert resources according to spectral variance, and fuses sub-band predictions through DER without additional trainable gating parameters. Experimental results over multiple Monte Carlo trials show that DER-BLMoE achieves 96.37% average accuracy with only 0.24M FLOPs, reduces parameters and FLOPs by 92.0% and 92.1% compared with SFEBLN, and maintains 95.94% accuracy at 0 dB Signal-to-Noise Ratio (SNR) under Additive White Gaussian Noise (AWGN). The code is available at https://github.com/3041749396/DER-BLMoE.
SSMs offer linear-complexity modeling for JSCC-based wireless image transmission, but 1-D token serialization weakens 2-D spatial priors and structural fidelity, especially at low CBRs. We propose SpatialMix-JSCC, an SSM-based JSCC framework that injects 2-D spatial awareness into SSM parameterization via SpatialMix-SSD, combining quad-directional scanning, deformable depthwise convolution, and hidden-state computation. The operator is instantiated as a hierarchical Local–Global–Local mixing block for efficient feature integration. Meanwhile, to enhance adaptability under varying SNR/CBR conditions, we introduce a lightweight dual-stream adaptation mechanism, enabling a single model to adapt to a range of simulated channel and bandwidth conditions without retraining. On AFHQ at CBR = 1/48 and SNR = 15 dB, SpatialMix-Basic achieves 30.7 dB PSNR, yielding 0.4 dB higher PSNR and 16.4% fewer MACs than SwinJSCC w/o SA&RA.
Existing secure rate splitting multiple access (RSMA) designs are typically tailored to either external or internal eavesdropping and may be ineffective in mixed scenarios where both coexist. To address mixed internal and external eavesdropping, we propose a novel secure hierarchical RSMA (HRSMA) framework. In this framework, all common streams can act as artificial noise (AN) to mitigate external eavesdropping; more importantly, the trusted user (T-UE) group common stream simultaneously serves as AN to suppress internal eavesdropping by the untrusted served users (U-UEs). To ensure the security, our approach aims to maximize the secrecy sum rate of all T-UEs and solve the resulting highly non-convex problem via an iterative algorithm combining majorization – minimization (MM) and semidefinite relaxation (SDR). Simulation results demonstrate that the HRSMA achieves higher secrecy rates than conventional baselines.
This letter proposes a novel hard-decision decoding algorithm for low-density parity-check (LDPC) codes, called lookahead bit flipping (LBF). Unlike the conventional bit flipping algorithm that determines a variable node (VN) to be flipped based solely on the current syndrome values, the proposed method virtually flips a sequence of VNs to select the optimal VN sequence by evaluating the syndrome values resulting from the sequence. This approach demonstrates improved error-correction performance compared to existing hard-decision algorithms. Furthermore, a normalized evaluation is employed to reflect the structural characteristics of LDPC parity-check matrices. In addition, a hybrid decoding scheme is introduced to combine the proposed LBF with the conventional bit flipping, enabling a significant reduction in decoding time while preserving the performance gains of LBF.
Orthogonal time frequency space (OTFS) and hybrid active-passive intelligent reflecting surface (HIRS) are promising technologies for next-generation wireless communications. However, related research on HIRS-assisted OTFS systems under large-scale fading is still scarce, lacking efficient optimization schemes to tackle cascaded attenuation and amplified active noise. To address these issues, we establish a HIRS-assisted OTFS system model under large-scale fading. For this system, a jointly enhanced dynamic majorization-minimization based alternating optimization (JE-DMM-AO) algorithm is developed to maximize the signal-to-interference-plus-noise ratio (SINR). Specifically, the algorithm leverages an end-to-end cascaded energy scheduling scheme to secure a superior initial state, followed by an adaptive Nesterov-like extrapolation mechanism that significantly accelerates convergence during the optimization of the reflecting elements. Simulation results demonstrate that the proposed algorithm significantly improves the signal detection reliability of the HIRS-OTFS system with a lower computational complexity.
The energy-harvesting industrial IoT networks exhibit intertwined quasi-periodic dynamics in traffic patterns and energy budgeting. However, conventional myopic approaches optimize only instantaneous states, resulting in a mismatch between traffic demand and energy supply. To address this, we present a traffic-aware joint clustering and routing (TJCR) approach. Firstly, we develop a lightweight forecaster that utilizes learnable basis decomposition to predict future traffic and energy profiles. Then, we establish a collaborative optimization scheme that jointly adapts clustering and routing, iteratively refining the clusters based on routing-induced feedback. Simulations demonstrate that TJCR improves network lifetime over representative baselines while maintaining high throughput.
Generalized orthogonal chirp division multiplexing (GOCDM) introduces a discrete root index λ as a waveform design freedom, enhancing the adaptability to doubly selective channels and diverse sensing requirements. Unlike the singlechirp design in OCDM-ISAC and the separate configuration of communication and sensing parameters in affine frequency division multiplexing (AFDM) ISAC, λ in GOCDM-ISAC simultaneously determines both the sparsity width in the generalized discrete Fresnel transform domain and the chirp slope, thereby making the delay-Doppler coupling strongly correlated with communication bandwidth efficiency. To address this, this letter proposes a dual-chirp embedded pilot frame based on complementary root indices along with a three-stage parameter estimation algorithm, which drives the off-diagonal terms of the joint Fisher information matrix (FIM) toward zero under a single-target AWGN channel and alleviates the resolution limitations of the discrete grid. Simulations demonstrate that at an SNR of 20 dB, the range and velocity estimation errors are reduced by approximately 9.1 dB and 5.9 dB, respectively, compared to other baselines, while maintaining high communication reliability along with communication bandwidth efficiency.
