Issue Ebookcover storyIssue CoverIssue Contents
Online FirstList of IssuesDocument Types
- Abstract:With the continuous advancement of sixth-generation (6G) mobile communication systems, future wireless networks are expected to simultaneously support high-rate and low-latency communications as well as high-accuracy environmental sensing. Integrated sensing and communication (ISAC), which enables the deep integration of communication and sensing functionalities through the shared use of spectrum, hardware, and signal processing resources, has emerged as a key candidate technology for 6G. Multi-base-station cooperation can effectively enhance coverage, interference suppression, and sensing accuracy. However, most existing studies focus on terrestrial networks, while systematic modeling and optimization for maritime environments remain insufficient. In marine scenarios, wireless channels exhibit strong time-varying characteristics due to dynamic sea-surface fluctuations, and the statistical distribution of sea clutter varies significantly with sea states. As a result, single-base-station systems are generally unable to meet the requirements of wide-area offshore communication coverage and high-precision target sensing.To address these challenges, this paper proposes a shore-based multi-base-station cooperative ISAC network model for offshore scenarios and develops a sea-state-adaptive hierarchical modeling framework for both communication channels and sea clutter. For communication modeling, the impact of wave-induced height perturbations on the effective antenna height, propagation path, and signal phase is analyzed. According to different sea-state levels, three channel models are established: an ideal two-ray model for low sea states, a roughness-corrected two-ray model for moderate sea states, and a strong-attenuation model with random phase perturbations for high sea states. These models provide an accurate characterization of path-loss behavior over dynamic sea surfaces. For sensing modeling, a quantitative mapping between sea-state levels and the key statistical properties of sea clutter is established. The evolution of clutter amplitude distributions under different sea conditions is characterized: sea clutter approximately follows a Rayleigh distribution under low sea states, a Weibull distribution under moderate sea states, and a K-distribution under high sea states. In addition, the variations of the active scattering-center ratio and scattering power with sea state are further quantified.Based on the proposed models, a multi-base-station cooperative ISAC signal model is formulated. The signal-to-interference-plus-noise ratio (SINR) for communication users and the signal-to-clutter-plus-noise ratio (SCNR) for sensing are derived as performance metrics. A joint transmit-receive beamforming optimization problem is then established to improve both communication and sensing performance. To solve the resulting non-convex problem, an alternating iterative algorithm is developed. Specifically, with fixed transmit beamformers, the receive filter design is transformed into a generalized Rayleigh quotient maximization problem, for which a closed-form solution is obtained. With fixed receive filters, the transmit beamforming subproblem is converted into a convex optimization problem via convexification, thereby enabling coordinated enhancement of communication and sensing performance.Simulation results demonstrate that appropriate power allocation and base-station configuration can significantly improve system performance. Increasing the number of cooperative base stations provides spatial cooperation gains, effectively suppresses sea clutter and multi-user interference, and enhances system robustness. Compared with other beamforming, the proposed algorithm achieves superior performance across different transmit power levels, with more pronounced gains as the transmit power increases. This study establishes a sea-state-adaptive theoretical framework for maritime multi-base-station cooperative ISAC, providing methodological support for applications such as offshore waterway surveillance, offshore wind farm operation and maintenance, and unmanned surface vessel cooperation. Future work will further investigate multi-base-station synchronization, multi-target sensing, and accurate modeling of sea-clutter covariance matrices.Keywords:integrated sensing and communication (ISAC);marine environment;sea state classification;resource allocation;maritime communications;multi-base-station cooperation285|357|0Updated:2026-10-08
- Abstract:Orthogonal time frequency space (OTFS) modulation has been regarded as a promising technology for integrated sensing and communication (ISAC) systems due to its strong robustness against Doppler frequency shifts in high-mobility environments. However, conventional OTFS-based ISAC waveform design schemes usually rely on the intrinsic delay-Doppler (DD) domain grid structure for target parameter estimation. Limited by the subcarrier spacing, such schemes suffer from an inherent bottleneck in maximum unambiguous ranging capability, making it difficult to satisfy long-range sensing requirements at the scale of hundreds of meters or even kilometers. To address the above issue, this paper proposes an OTFS-based ISAC waveform design method with embedded linear frequency modulated (LFM) pulse signals in the DD domain. By applying the wigner transform and symplectic finite Fourier transform (SFFT) to the LFM pulse signals, they are mapped onto the DD-domain grid for distribution analysis. The analysis results show that the main energy points of the LFM pulse signals are distributed along deterministic trajectories in the DD domain. Moreover, due to the finite observation duration and discontinuous transmission characteristics of the signals, energy diffusion appears around the main energy points. To suppress such diffusion interference, a symmetric window function distributed along the trajectory of the main energy points of the LFM pulse signals is further designed. This window effectively suppresses sidelobe diffusion while preserving the mainlobe energy of the LFM pulse signals, thereby constraining the sensing signal energy within a limited DD-domain grid. OTFS communication symbols are then inserted into the remaining DD-domain grids, realizing the joint design of sensing and communication signals. At the sensing receiver, the LFM pulse signal components are extracted, and target range and velocity estimation are achieved through matched filtering and slow-time fast Fourier transform (FFT) processing. At the communication receiver, the OTFS signal components are extracted and demodulated to recover the transmitted communication data. Simulation results demonstrate that, under the same system parameters, although the processing complexity of the sensing link increases by approximately 3.6 times, the proposed scheme extends the maximum sensing range to six times that of the conventional OTFS method. Furthermore, for multi-target parameter estimation, the range root mean square error (RMSE) remains below 0.5 m, while the velocity RMSE remains below 2 m/s. In terms of communication performance, the proposed scheme achieves nearly the same bit error rate and throughput performance as conventional OTFS. These results verify that the proposed waveform design scheme can significantly enhance the ranging capability of ISAC systems while maintaining communication performance.Keywords:integrated sensing and communication (ISAC);waveform design;orthogonal time frequency space (OTFS);linear frequency modulated (LFM);long-range sensing;delay-Doppler (DD) domain317|202|0Updated:2026-10-08
- Abstract:With the dense deployment of low earth orbit (LEO) satellite constellations, the contention for frequency and orbital resources is increasingly intensifying. Existing satellite communication and navigation systems are typically designed independently. Satellite communication aims for high data throughput, while satellite navigation pursues high-precision time-delay estimation and anti-interference capabilities. This separated architecture leads to competition for spectrum resources, equipment redundancy, and low synergy efficiency. In this context, how to integrate satellite communication and navigation functions based on LEO satellites, designing an integrated communication and navigation (ICAN) signal capable of efficiently reusing resources such as frequency, orbital slots, and hardware payloads, thereby achieving “multi-purpose use of a single satellite” and complementary advantages between communication and navigation, has garnered increasing attention from both academia and industry. Existing integrated communication and navigation waveform designs for LEO system are mainly tailored to single-user downlink scenario, which leads to limited spectral efficiency and neglects multi-user interference. Moreover, effective waveform design strategies for communication-navigation integration at the multiple access level remains largely unexplored. To address these challenges, this paper proposes a novel ICAN waveform design based on rate-splitting multiple access (RSMA). Zadoff-Chu sequence is employed as positioning signal, while satellite navigation message and multi-user communication data are mapped onto the common and private streams of the communication signal via rate-splitting strategy. By superimposing the positioning and communication signals, the proposed scheme fully exploits the flexibility of RSMA in multi-user interference management and resource allocation, enabling an integrated waveform design that supports spectrum-sharing transmission of positioning and multi-user communication signals. Within the proposed ICAN waveform, two positioning signal transmission and propagation delay estimation schemes are developed based on single-carrier and multicarrier orthogonal frequency division multiplexing (OFDM), respectively. For the single-carrier scheme, the positioning signal is treated as tolerable interference when receiver decoding communication signal. For the OFDM-based scheme, successive interference cancellation is applied to mitigate the interference of the positioning signal to communication signal. Furthermore, a sum-rate maximization problem that jointly considers beamforming and rate allocation is formulated and solved using an improved weighted minimum mean square error-based alternating optimization algorithm. Simulation results demonstrate that the proposed ICAN waveform achieves both high data rate communication and reliable positioning, yielding throughput improvements of 9.7% and 18.2%, respectively, compared to the baseline schemes.Keywords:integrated communication and navigation;waveform design;rate-splitting multiple access;Zadoff-Chu sequence;resource allocation188|207|0Updated:2026-10-08
- Abstract:Orthogonal time frequency space (OTFS) modulation is widely applied in high-Doppler scenarios such as high-speed railway communications, underwater acoustic communications and similar environments, as the mapping of information into delay-Doppler (DD) domain provides strong Doppler resilience in high-mobility settings. Impulsive noise (IN) present in practical channels, however, can severely degrade the performance of conventional OTFS receivers designed under the additive white Gaussian noise assumption. Existing impulsive noise suppression methods for OTFS systems mostly rely on prior noise models and are primarily tailored to a single impulsive noise model; consequently, it remains difficult to handle different types of impulsive noise simultaneously. To address the challenge, this paper proposes a sparse Bayesian learning (SBL) based impulsive noise suppression method for OTFS systems. The method fully exploits the temporal sparsity of impulsive noise and achieves adaptive suppression of impulsive noise with different statistical characteristics. Firstly, null symbols are inserted along the Doppler axis at the transmitter to construct a linear observation model that contains only impulsive noise and background noise. The model transforms impulsive noise estimation into a sparse vector recovery problem. Each component of the time domain impulsive noise receives an independent zero-mean complex Gaussian prior. The posterior mean and covariance are then iteratively updated via the expectation maximization (EM) algorithm to yield an adaptive estimate of the impulsive noise. In addition, the structure of the observation matrix and the decomposable property of impulsive noise along the delay dimension in DD domain are exploited to decompose the original MN dimensional joint estimation problem into M parallel N dimensional subproblems, and a low complexity row-wise SBL scheme is further proposed. The scheme reduces the dimension of the vector to be estimated from MN to N and cuts the computational complexity from
to , a reduction by a factor of . The scheme preserves estimation accuracy, adaptively handles the non-uniform distribution of impulses across different delay rows, and thus enhances engineering practicality of the algorithm. Finally, simulation experiments are conducted under different impulsive noise models and with different receiver types to validate the effectiveness of the proposed method. The results show that the proposed method reduces the bit error rate by approximately three orders of magnitude compared with the untreated case and delivers performance close to the ideal scenario with only background noise. The substantial reduction in bit error rate and the near-ideal performance confirm the reliability and effectiveness of the method. Moreover, the method does not rely on a specific impulsive noise model and exhibits good robustness and adaptability to sparse impulsive noise with diverse statistical characteristics in complex practical environments. The proposed algorithm can serve as an independent denoising preprocessing module compatible with existing OTFS receivers and provides a low-complexity, prior-free, and highly practical solution for impulsive noise suppression in challenging scenarios such as high-speed railway and underwater acoustic communications.Keywords:orthogonal time frequency space;noise suppression;impulsive noise;sparse Bayesian learning;low complexity241|178|0Updated:2026-10-08 - Abstract:To address the reliable detection problem in orthogonal time frequency space (OTFS) systems over doubly selective channels, a two-stage detection method based on global initial detection and local list refinement is proposed. In the first stage, Gaussian-approximation-based global soft initial detection is performed over the entire delay-Doppler (DD) domain grid based on the DD-domain vector input-output model, producing symbol-level metrics and bit-level initial soft information. In the second stage, candidate-list-based refinement is applied only to unreliable positions. Specifically, the positions to be re-examined are selected according to the reliability of the initial soft information and partitioned into coupling-induced local detection blocks according to the symbol-coupling relationships characterized by the equivalent channel matrix. The contributions of symbols outside each block are then softly cancelled from the received observations associated with that block, thereby forming approximately self-contained local detection problems. Within each local block, a finite-size candidate list is constructed around the initial decisions delivered by the global initial detector. The candidates are evaluated using local Euclidean metrics, from which local symbol-level and bit-level soft outputs are generated. The local list-refinement soft information is then fused with the global initial soft information to obtain the final detection output. Positions outside the re-examination set retain their global initial soft outputs. This processing enhances only positions with low reliability and significant coupling, thereby avoiding high-complexity list search over the entire DD-domain grid. In terms of complexity, the average total number of list candidates per frame is adopted as the main complexity measure of the second stage, and the effects of the re-examination threshold, graph-construction threshold, and other parameters on the candidate-list size are analyzed. Results show that the average number of list candidates decreases significantly as the signal-to-noise ratio increases. Meanwhile, the number of positions entering local refinement and the local-block sizes can be controlled by adjusting the re-examination threshold and graph-construction threshold, enabling an effective tradeoff between detection performance and computational complexity. Furthermore, to evaluate the impact of errors in soft cancellation of block-external interference on performance, a Genie-Aided (GA) reference bound is constructed under the assumption that symbols outside each block are perfectly known. Simulation results show that, under different parameter settings and numbers of resolvable channel paths, the proposed method achieves significantly better frame error rate (FER) performance than the Gaussian-approximation-based detector using only global initial detection, while exhibiting a performance trend close to that of the Genie-Aided reference bound.Keywords:orthogonal time frequency space (OTFS);two-stage detection;global initial detection;local list refinement123|76|0Updated:2026-10-08
- Abstract:To meet the demands of comprehensive and refined unmanned aerial vehicle (UAV) surveillance in low-altitude airspace, multiple-input multiple-output (MIMO) integrated sensing and communication (ISAC) offers simultaneous communication and sensing capabilities under shared spectrum and hardware resources. However, existing MIMO ISAC studies have not fully exploited the three-dimensional spatial degrees of freedom, nor have they effectively utilized communication data signals for high-performance sensing of moving targets, leading to limited accuracy in multi-parameter estimation of low-altitude UAV targets. To address these issues, this paper proposes a 3D MIMO orthogonal time frequency space (OTFS) modulation-based ISAC method for low-altitude UAV surveillance. By directly reusing OTFS communication data signals, the proposed method improves the parameter estimation performance for multiple UAV targets without affecting communication functionality. First, a 3D MIMO ISAC system model is established, in which planar antenna arrays are deployed in both the horizontal and vertical dimensions. Based on shared OTFS signals, communication information is transmitted while space-time-frequency domain echoes from multiple UAV targets are simultaneously observed. Then, considering that communication data symbols on different subcarriers are statistically independent and that the echo observations across multiple subcarriers share the same spatial angle information, a covariance matrix of the echo observations is constructed, and the multiple signal classification (MUSIC) algorithm is subsequently employed to achieve high resolution estimation of target azimuth and elevation angles. Using the angle estimates as prior information, spatial domain decoupling of multiple targets is then performed. Velocity and range are sequentially estimated through one-dimensional correlation searches by exploiting the Doppler phase variations along the slow time dimension and the delay induced phase variations along the frequency dimension, respectively. This cascaded estimation strategy transforms the coupled high-dimensional joint multi-parameter estimation into a two-dimensional angle spectrum search followed by multiple one-dimensional parameter searches. By fully exploiting joint multi-subcarrier observations, the proposed strategy effectively suppresses the influence of random communication data on target parameter estimation, thereby achieving low complexity and high accuracy parameter estimation for multiple UAV targets. Numerical simulation results demonstrate that, compared with existing target parameter estimation methods, the proposed method achieves a higher estimation success probability and superior accuracy in angle, range, and velocity estimation in low-altitude UAV surveillance scenarios. Specifically, the estimation accuracies for azimuth and elevation are improved by 70%, while the range and velocity estimation accuracies are enhanced by 60% and 80%, respectively. Moreover, as the number of UAV targets within the surveillance area increases, the proposed method can still maintain robust UAVs target estimation performance.Keywords:low-altitude economy;unmanned aerial vehicles surveillance;integrated sensing and communication;3D multiple-input multiple-output (MIMO);orthogonal time frequency space (OTFS) modulation151|91|0Updated:2026-10-08
- Abstract:The random phase communication system (RPCS) is a low-probability-of-intercept (LPI) waveform-modulation scheme that employs a noise-like fundamental modulation waveform (FMW) and cyclic shift modulation (CSM) to convey information through cyclic shifts of the FMW. However, the conventional uniform shift set is uniquely determined by the FMW length and modulation order and is therefore transparent to interceptors. Once the FMW is disclosed, the uniform shift structure becomes fully exposed, enabling symbol-level matched jamming and sharply degrading the matched-jamming tolerance. To address this issue, this paper proposes an anti-interception enhancement method based on non-uniform CSM and dynamic shift-set hopping. The method extends the conventional uniformly spaced shift set to a non-uniform shift set selected from the candidate cyclic shift space and establishes a signal model for non-uniform CSM. Next, based on the FMW cyclic autocorrelation sidelobes, the orthogonality defect of the shift set is defined, and the simulated annealing (SA) algorithm is used to optimize the shift set, thereby reducing the residual cross-correlation among candidate shifted waveforms. On this basis, an analytical bit error rate (BER) expression for non-uniform CSM is derived under the independent-branch approximation, and its anti-interception characteristics are analyzed from the perspective of the shift set combination space. Furthermore, to suppress an attacker’s long-term tracking of static shift sets, a dynamic shift-set hopping mechanism is designed, which reduces the average effectiveness of matched jamming constructed from a previously observed shift set in the current transmission interval to approximately
. Theoretical analysis and simulation results show that the proposed method maintains BER performance comparable to uniform CSM while providing a shift-set combinatorial search complexity upper bound of approximately bits, which is approximately 66 bits when and . Under matched jamming, non-uniform CSM with an unknown shift set improves the matched-jamming tolerance compared with uniform CSM. Under tracking matched jamming, the dynamic hopping scheme reduces the BER by approximately one to two orders of magnitude compared to static non-uniform CSM and achieves about 6 dB improvement in tracking matched-jamming tolerance at the threshold . Without hardware changes, the proposed method improves resistance to interception and tracking matched jamming, providing a reference for modulation-layer LPI design.Keywords:random phase communication system;cyclic shift modulation;non-uniform shift;anti-interception;shift-set hopping107|73|0Updated:2026-10-08 - Abstract:Affine frequency division multiplexing (AFDM), which employs chirp signals as subcarriers, exhibits inherent robustness against Doppler shifts and has thus been widely adopted in satellite communication systems. However, for low earth orbit (LEO) satellite links, the distinctive chirp-based structure of AFDM waveforms, while beneficial for Doppler resilience, renders them vulnerable to malicious interception, as an eavesdropper (Willie) can readily detect and demodulate the signals using fractional Fourier transform (FRFT) based detectors or higher-order statistical analysis. To address the pressing physical-layer security threats in LEO communications under severe Doppler and time-varying channels, this paper proposes a novel covert AFDM waveform based on a multidimensional discrete hyperchaotic system (DHS). The proposed scheme fundamentally mitigates the structural vulnerabilities of the chirp waveform from the physical-layer perspective, thereby significantly enhancing covertness. Specifically, we leverage the high initial-condition sensitivity and unpredictability of the multidimensional DHS to jointly drive the AFDM subcarrier index mapping and the chirp hopping factor in a combined agile manner, constructing highly complex and unpredictable time-frequency characteristics in the affine domain such that the frequency modulation slope varies symbol-wise according to the chaotic sequence. Through this joint operation, a noise-like waveform with intricate time-frequency traits and hardly tractable trajectories is synthesized in the affine domain. Concurrently, we introduce a Weierstrass-function-based fractal phase scrambling operation on the chirp carriers, which effectively suppresses the distinct peak-focusing signature of chirp signals in the fractional Fourier domain, thus counteracting FRFT-based detection. In terms of computational complexity, the proposed algorithm incurs only a constant-factor overhead relative to the native AFDM processing, remaining in the same complexity order. To validate the covertness of the proposed waveform, we conduct comprehensive simulations encompassing cepstral analysis, higher-order statistics, and interception probability assessment.Resultsdemonstrate that the proposed AFDM waveform completely eliminates the periodic cepstral features of the standard AFDM signal at the physical layer, retains noise-like characteristics under higher-order statistical scrutiny, and maintains a statistical kurtosis tightly proximate to the Gaussian theoretical value. Moreover, assuming that Willie employs both FRFT-based and cyclic-spectrum detectors for targeted interception, the interception probability of the proposed waveform consistently remains below the false-alarm probability. By fundamentally altering the carrier-level characteristics at the physical layer, the proposed method enhances the low-probability-of-intercept and covert security performance of satellite communications over open links. Finally, pulse compression comparison confirms that the legitimate receiver (Bob) achieves high-dynamic-range accurate detection, whereas Willie cannot reliably discern the communication bearer from the intercepted waveform, thereby corroborating the robust reliability of the proposed design.Keywords:satellite communication;AFDM;covert communication;low probability of detection;multi-dimensional DHS151|79|0Updated:2026-10-08
A High-Precision Sensing Algorithm for AFDM-ISAC Waveform Based on Gradient Optimization Enhanced Publication
Abstract:In low-altitude airspace (LAA) integrated sensing and communication (ISAC) scenarios, targets often move at high speeds and stringent parameter estimation accuracy is required. However, existing sensing methods are constrained by the resolution of discrete grids and have difficulty effectively mitigating off-grid errors. To address this issue, this paper proposes a high-accuracy sensing algorithm based on affine frequency division multiplexing (AFDM). The proposed algorithm adopts a hierarchical sensing architecture to achieve accurate estimation of target range and Doppler parameters. First, by exploiting the delay-Doppler representation characteristics of AFDM signals, a range-Doppler matrix (RDM) is constructed through fast cyclic cross-correlation processing, and coarse grid-based estimates of the target parameters are obtained from the peak locations of the matrix. To mitigate the estimation errors caused by deviations of the actual target parameters from the discrete grid points, a peak-neighborhood indexing mechanism is further introduced to extract the responses of the main peak and its neighboring points. The initial parameter estimates are then refined using the ratio relationships between adjacent peak responses, thereby providing gradient optimization with an initial iteration point closer to the true parameters. This preprocessing procedure reduces the sensitivity of the gradient-based algorithm to initialization and improves its convergence stability. On this basis, a sensing objective function over the continuous delay-Doppler parameter space is formulated using the matched-filter output, and an RDM-preprocessed gradient optimization(RDM-PGO) sensing algorithm is developed. Taking the RDM-based preprocessing result as the initial point, the proposed algorithm jointly updates the delay and Doppler parameters through iterative optimization based on the gradients of the objective function with respect to these parameters. Unlike conventional discrete-grid search and local interpolation methods, the RDM-PGO algorithm does not constrain the target parameters to predefined grid points, nor does it require a denser grid to improve resolution. It can therefore effectively reduce parameter quantization errors and overcome the sensing-accuracy limitation imposed by discrete-grid resolution. Meanwhile, RDM preprocessing confines the subsequent gradient optimization to a local parameter region containing the target, thereby avoiding high-density exhaustive searches over the entire two-dimensional parameter space and achieving a favorable balance between sensing accuracy and computational complexity. Numerical analyses and simulation results demonstrate that the proposed RDM-PGO algorithm achieves stable convergence in the continuous parameter space and effectively reduces range and Doppler estimation errors. Compared with conventional grid-search and interpolation-based correction methods, the proposed algorithm significantly improves sensing accuracy while maintaining relatively low computational complexity, providing an effective solution for high-accuracy target sensing and real-time processing in low-altitude AFDM-ISAC systems.Keywords:integrated sensing and communication;affine frequency division multiplexing;sensing algorithm;gradient optimization;range-doppler matrix196|84|0Updated:2026-10-08- Abstract:With the rapid evolution of sixth generation (6G) mobile communication systems, low earth orbit (LEO) satellite networks have emerged as a fundamental component of integrated space-air-ground communication architectures due to their extensive coverage, low transmission latency, and flexible deployment capabilities. However, the extremely high relative velocity between LEO satellites and terrestrial users introduces severe large-scale Doppler shifts, causing wireless channels to exhibit pronounced doubly selective fading characteristics in both the time and frequency domains. Such channel characteristics pose significant challenges to physical-layer waveform design, synchronization, and channel estimation, thereby limiting the reliability and transmission efficiency of high-mobility communication systems. As a promising waveform for next-generation wireless communications, orthogonal time frequency space (OTFS) modulation maps transmitted symbols in the delay-Doppler (DD) domain, where the time-varying wireless channel can be transformed into a nearly time-invariant representation. Consequently, OTFS demonstrates superior robustness against high mobility and Doppler spread compared with conventional orthogonal frequency division multiplexing. Nevertheless, most existing OTFS channel estimation methods suffer from a limited Doppler estimation range, resulting in considerable performance degradation when large-scale Doppler shifts are present. Therefore, developing an accurate and computationally efficient channel estimation scheme for OTFS systems operating under large Doppler conditions remains an important research problem.To address this issue, this paper first presents a systematic analysis of the signal structure and input-output relationship of cyclic prefix (CP)-OTFS employing rectangular pulse shaping. The discrete input-output relationship of the OTFS system is rigorously derived based on the corresponding signal model, and the two-dimensional orthogonality properties of the basis functions under the rectangular window are investigated in detail. These theoretical analyses establish the mathematical foundation for the proposed channel estimation framework and provide insight into the influence of pulse shaping on channel estimation performance. Subsequently, the estimation characteristics of pilot impulses in different signal domains are quantitatively investigated. Specifically, the estimation accuracy, Doppler resolution and delay resolution, as well as the estimation range in both the frequency-time (FT) domain and the DD-domain are analyzed and compared. The analysis reveals that FT-domain estimation possesses a significantly larger Doppler estimation range and is therefore more suitable for estimating large-scale Doppler shifts, whereas DD-domain estimation provides higher estimation accuracy for residual channel parameters owing to its sparse channel representation. Furthermore, the complementary characteristics of FT-domain and DD-domain channel estimation are theoretically analyzed, providing the motivation for jointly exploiting the advantages of both domains. Based on the above observations, a cross-domain embedded pilot channel estimation algorithm is proposed. In the proposed scheme, the large-scale Doppler component is first estimated in the FT-domain. To further improve the estimation resolution, parallel-to-serial conversion and zero-padding operations are introduced for the FT-domain pilot processing, effectively increasing the Doppler resolution without modifying the system bandwidth. After compensating for the estimated large-scale Doppler component, the residual channel parameters, including delay and fractional Doppler information, are estimated by using embedded pilots in the DD-domain. Finally, the channel estimation results obtained from the FT and DD domains are fused to reconstruct the complete channel state information. This fusion effectively overcomes the limited Doppler estimation range of conventional DD‑domain methods while ensuring that high estimation accuracy is preserved. Extensive simulation results demonstrate the effectiveness and robustness of the proposed algorithm under both large-scale integer Doppler and fractional Doppler scenarios. In terms of normalized mean square error (NMSE), the proposed method achieves an estimation accuracy on the order of 10-8 under integer Doppler conditions and reaches to approximately 10-5 in the presence of fractional Doppler shifts. In terms of bit error rate, the proposed scheme closely approaches the performance achieved with ideal channel state information and significantly outperforms conventional frequency-domain channel estimation methods as well as DD-domain estimation schemes that lack Doppler compensation. Moreover, the proposed algorithm avoids iterative optimization and matrix inversion, resulting in low computational complexity and high feasibility for practical communication systems. In summary, the proposed cross-domain channel estimation framework effectively exploits the complementary estimation capabilities of the FT and DD domains, enabling accurate channel acquisition over an extended Doppler estimation range while maintaining low computational complexity. The proposed scheme provides an efficient and practical solution for OTFS-based LEO satellite communication systems operating in high-mobility environments and offers valuable guidance for channel estimation design in future 6G integrated space-air-ground communication networks.Keywords:LEO satellite;OTFS;doubly selective channel;cross‑domain channel estimation;embedded pilot estimation;large‑scale Doppler112|78|0Updated:2026-10-08
Basic Theory and Key Technologies of Waveform Design for Next\-Generation Wireless Communications
- Abstract:The wideband terahertz band can provide important support for sixth-generation integrated sensing and communication (ISAC) and physical-layer secure transmission. A reconfigurable intelligent surface (RIS) can construct a controllable reflected link in blocked environments. In wideband systems, the geometric propagation delays between the elements of a multi-antenna base station and an RIS accumulate with the array aperture. Frequency-independent phase shifters cannot maintain phase consistency over the entire bandwidth, resulting in beam squint and array-gain loss. Existing studies generally regard beam squint as an adverse effect in communications and treat the channel uncertainty of an unknown eavesdropping link as an externally given parameter. The intrinsic relationship among sensing duration, angle-estimation accuracy, and robust secrecy performance is therefore difficult to characterize. To address this issue, this paper proposes an RIS-assisted sensing-driven robust secure transmission method for wideband terahertz ISAC systems. True time delay (TTD) and phase-shifter (PS) control structures are introduced at the multi-antenna base station and the RIS. Each transmission frame is divided into a sensing stage and a communication stage. During sensing, the beam squint at the RIS is exploited to rapidly estimate the angle of a passive eavesdropper. During communication, robust secure transmission is performed according to the sensing result. An eavesdropper angle-sensing model is established, and the Cramér-Rao bound (CRB) of the angle-estimation error is derived. A mapping from the CRB to the subcarrier-level uncertainty radius of the eavesdropping channel is then constructed, so that the channel uncertainty is determined by the sensing duration and sensing accuracy. The RIS reflection phases, RIS-side TTD delays, and sensing duration are jointly optimized to maximize the average robust secrecy rate. By using one-dimensional search, alternating optimization (AO), successive convex approximation (SCA), semidefinite relaxation (SDR), and the S-Procedure, the original nonconvex problem is transformed into convex subproblems solvable by CVX. Simulation results evaluate the proposed method in terms of angle-estimation CRB, iterative convergence, the sensing-duration tradeoff, and the impact of the base-station transmit power. The results show that the proposed method mitigates the array mismatch caused by wideband beam squint, improves the angle-estimation accuracy of the eavesdropper, and achieves a favorable balance between sensing and communication resources, thereby enhancing robust secure transmission performance. Passing the sensing results directly to the communication stage helps quantitatively characterize the relationship between sensing quality and secure transmission performance, and provides a basis for sensing and communication resource allocation in RIS-assisted wideband terahertz ISAC systems.Keywords:terahertz integrated sensing and communication;reconfigurable intelligent surface;true time delay;robust secrecy transmission;angle estimation bound184|90|0Updated:2026-10-08
- Abstract:In disaster emergency scenarios where communication infrastructure is damaged, an unmanned aerial vehicle (UAV) can serve as an aerial relay node to provide real-time video transmission services for rapid search and rescue, damage assessment, remote coordination, and other tasks. However, different emergency tasks have heterogeneous preferences regarding transmission latency, communication energy consumption, and video quality, and these preferences vary dynamically with rescue phases and on-site conditions. Node mobility, channel fluctuations, and limited bandwidth transmit power further intensifies the trade-offs among these performance metrics. To address these issues, this paper proposes a UAV-assisted video transmission framework integrating scalable video coding (SVC), priority scheduling, and communication resource allocation. Considering ground-node mobility, probabilistic line-of-sight and non-line-of-sight channels, user-request priorities, and queuing processes, the framework establishes a two-stage transmission model in which video streams are relayed from video sources to requesting users via the UAV. Video utility jointly accounts for visual clarity and playback smoothness to quantify user experience. With the objectives of minimizing long-term transmission latency and the UAV’s communication energy consumption while maximizing video utility, the joint optimization of the number of SVC layers, uplink and downlink bandwidth allocation, and UAV transmit power is formulated as a long-term multi-objective optimization problem and further modeled as a multi-objective Markov decision process. Existing methods mainly focus on a single objective or adopt fixed preferences, often requiring retraining when task requirements change and making real-time adaptation of transmission strategies difficult. To address this limitation, this paper proposes a phased-evolutionary envelope multi-objective asynchronous advantage actor-critic algorithm (PE-EMOA3C). The algorithm takes both the task-preference vector and system state as network inputs, enabling a single policy to cover a continuous preference space. During the initial training stage, standard asynchronous advantage actor-critic updates are adopted to mitigate gradient interference across preferences. After a preset number of training episodes, the algorithm enters the envelope-update phase to approximate the Pareto front over the candidate preference space. A homotopy-based optimization strategy and a cosine-annealing learning-rate decay mechanism are also employed to improve training stability and sample efficiency. Simulations are conducted in a dynamic emergency communication scenario with a radius of 800 m, five ground video sources, and 20 requesting users. The experimental results show that PE-EMOA3C achieves the highest total system utility under seven of the nine preference settings. Compared with proximal policy optimization (PPO), multi-objective Q-learning (MOQ-Learning), non-dominated sorting genetic algorithm II (NSGA-II), and a random policy, PE-EMOA3C improves the average total system utility by approximately 3.49%, 10.79%, 14.50%, and 37.51%, respectively. When the number of video sources, user load, and downlink bandwidth vary, the proposed algorithm maintains superior total system utility and adapts effectively to different task preferences and changes in scenario parameters. In addition, the proposed algorithm has an average decision latency of 1.13 ms, substantially lower than the 161.87 ms of NSGA-II.Keywords:unmanned aerial vehicle;video transmission;scalable video coding;communication resource allocation;multi-objective reinforcement learning181|91|0Updated:2026-10-08
- Abstract:In order to solve the problems of limited resources and low utilization rate caused by high-speed satellite mobility and dynamic changes of ground services in low earth orbit (LEO) satellite networks, this paper studies the joint optimization method for uplink random access (RA) and data transmission (DT) in mixed service scenarios. By analyzing the different requirements of real time (RT) service and best effort (BE) service for latency and bandwidth, the unit step function is used to represent the resource requirements of RT service, and the exponential utility function is used to represent the elastic demand of BE service for resources. Based on the utility function of two services, a network utility maximization model for the coexistence of mixed services is established. Due to the fluctuation of ground traffic load in different periods, in order to solve this problem, this paper introduces a traffic load prediction mechanism based on method of moments (MOM). This mechanism mainly estimates active users by observing the number of idle, successful access and conflicting access opportunities. Based on this prediction mechanism, an iterative resource scheduling algorithm is designed, which is based on the predicted load to dynamically adjust the access control dianprobability and the time-frequency resource blocks (RBs) allocation scheme in the RA and DT stages. The simulation results show that the performance of the proposed method under different traffic loads and network conditions is close to the theoretical optimal value. In comparison with the fixed resource proportion benchmark method, the proposed method can increase the average utility of the network by about 37%, 17% and 9%, respectively, compared with the benchmark method under different proportion settings. At the same time, in the multi-cycle dynamic traffic scenario, the average utility is further increased by 57%, 26% and 10%. The simulation results show that this method can not only effectively reduce the collision probability in the RA phase, ensure the quality of RT service and improve the throughput of BE service, but also significantly improve the overall quality of service (QoS) and resource utilization of LEO satellite network.Keywords:random access;Data transmission;mixed traffic;LEO satellite networks;method of moments136|63|0Updated:2026-10-08
- Abstract:The threshold voltage (VT) modulation and underlying physical mechanisms of the La(iPr2-FMD)3 (LaFMD) treatment technology are investigated. Initially, a modulation technique based on La2O3 is investigated. Although this approach enables a relatively wide range of flat-band voltage (VFB) tuning, the actual physical thickness of La2O3 and the increased SiOx formation in interface during interfacial treatment lead to elevated equivalent oxide thickness (EOT), VFB hysteresis, and interface state density (Dit). Compared with conventional La2O3 interface treatment technology, LaFMD treatment not only achieves a more controllable VT tuning range, but also enables precise fine-tuning at intervals of 30~40 mV by controlling the number of LaFMD treatment cycles. This is primarily attributed to the formation of fewer La oxides in the interfacial layer (IL). Meanwhile, the LaFMD treatment also reduces the EOT, which is due to the generation of La oxides with higher dielectric constant while maintaining nearly unchanged IL physical thickness. Moreover, the Dit remain low after LaFMD treatment. Compared with the sample without interfacial treatment, the LaFMD treatment results in a slight increase in Dit, leading to minor degradation of interfacial properties. However, the Dit is markedly lower than that of the La2O3-treated sample. Additionally, the LaFMD treatment slightly reduces the VFB hysteresis. These observations indicate that the LaFMD treatment does not induce significant fluctuations in defects near the interface, owing to minimal variation in Si oxide content in the IL. Ultimately, the application of LaFMD treatment technology in Si-channel gate-all-around field-effect transistors (GAAFETs) has excellent VT modulation and subthreshold swing (SS) characteristics. This is primarily attributed to the enhanced gate controllability resulting from the reduced VFB hysteresis and EOT. Therefore, the LaFMD treatment technology can be considered a promising candidate for multi-VT precise modulation applications in Si-channel GAAFETs.Keywords:LaFMD treatment;VT modulation;GAAFET;Dit;EOT;La2O3 treatment230|148|0Updated:2026-10-08
- Abstract:To address the increased scan-gain fluctuation in millimeter-wave dual-polarized phased arrays caused by the mismatch between the element radiation pattern and the array-factor envelope during wide-angle scanning, a design method for wide-angle phased arrays with low scan-gain fluctuation based on an externally loaded dielectric pin array (DPA) is proposed. First, based on the concept of pattern complementation, a planar DPA is employed to weakly load and spatially redistribute the near field above the dual-polarized antenna element. As a result, a wide-beam element pattern with a center dip is formed, which enhances the radiation at large off-boresight angles and suppresses the gain degradation during wide-angle scanning. Then, a 1 × 8 millimeter-wave dual-polarized phased-array prototype loaded with the planar DPA is designed, fabricated, and measured. The measured results show that the DPA can extend the scanning range and reduce the scan-gain fluctuation. Furthermore, to alleviate the broadside gain reduction introduced by the planar DPA, an arched DPA structure with a lens-like arrangement is proposed. By introducing continuous spatial dielectric loading, the aperture phase distribution is improved, thereby enhancing the broadside radiation performance while maintaining the gain-compensation capability at large scanning angles. This enables a better gain balance between the broadside and large-angle scanning states. Simulated results show that, after loading the arched DPA, the array can achieve wide-angle scanning of approximately ± 60° for both polarizations over the frequency range of 24~29 GHz, and no obvious gain degradation is observed at the scanning endpoints compared with the broadside gain. This indicates that the proposed structure can effectively improve the gain stability during wide-angle scanning. The proposed DPA is an externally loaded structure with the advantages of low profile, simple configuration, easy fabrication, and convenient integration with PCB (Printed Circuit Board)-based millimeter-wave phased arrays. The proposed method can provide a reference for the low scan-gain fluctuation design of millimeter-wave wide-angle phased arrays and has potential engineering applications in millimeter-wave communication, radar detection, and other array systems requiring wide-angle coverage.Keywords:dual-polarization;millimeter-wave phased array;wide-angle scanning;scan-gain fluctuation132|65|0Updated:2026-10-08
- Abstract:In recent years, the data rates of high-speed digital interfaces (such as HDMI, USB, and DisplayPort) have continued to rise. Differential data cables, owing to their excellent common-mode rejection capability and strong anti-interference characteristics, have become the preferred physical-layer medium for high-speed interconnects. In engineering practice, full-wave electromagnetic numerical methods are commonly relied upon for three-dimensional modeling and simulation of cables to predict their broadband transmission characteristics and signal integrity. However, when the cable shield adopts a foil-overlap Mylar structure, full-wave methods face severe challenges: the foil thickness is only on the order of micrometers, and the spiral overlap regions exhibit extremely fine geometries, making adaptive meshing extremely difficult, consuming enormous computational resources and requiring excessively long simulation times. Meanwhile, because full-wave methods cannot precisely reproduce the helical translational periodicity of the foil overlap, modeling errors are introduced, resulting in noticeable deviations between computed results and measured data. To address this issue, this paper proposes an analytical method for signal integrity evaluation of data cables with foil-overlap shields, which extends Vance’s classical shielding theory with key enhancements to balance computational efficiency and prediction accuracy. First, to handle the irregular local protrusion on the foil outer boundary caused by the adjacent ground wire, an elliptical parametric equation is used to describe the outer boundary of the second dielectric layer, and an analytical expression for the perimeter of the foil curved surface is established. Based on Vance’s transfer impedance formula, equivalent per-unit-length resistance and capacitance models of the foil are derived, and the foil-overlap shield is represented as a frequency-dependent resistance-capacitance equivalent network that effectively accounts for the electromagnetic leakage effect introduced by the discontinuous overlap. Second, under the quasi-TEM approximation, for the non-concentric elliptical structure formed by the double-layer dielectric inside the cable, the logarithmic mean integration of the elliptical radial distance is applied to equivalently transform it into a coaxial cylindrical structure, thereby extracting the per-unit-length resistance, inductance, capacitance, and conductance (RLGC) parameters of a single conductor while preserving the per-unit-length capacitance. Finally, given the axial periodicity of the foil overlap, a segmented cascading strategy is adopted to divide the entire cable into a number of uniform equal-length segments. The ABCD transmission matrix of each segment is constructed, and the overall cascaded transmission matrix of the whole cable is computed, from which the magnitude and phase responses of the differential insertion loss (SDD21) over the 100 MHz~12 GHz broadband range are obtained. To validate the effectiveness of the proposed method, an HDMI 2.1 cable is taken as an example, and comparisons are made among the analytical method, HFSS full-wave simulation, CST built-in analytical model, and actual measurements. The results show that the insertion loss magnitude computed by the analytical method deviates from the measured data by a maximum of only 0.3 dB over the entire frequency band, and its agreement is superior to that of the other methods, demonstrating higher accuracy of the proposed method. In terms of computational efficiency, the analytical method achieves a speedup of more than 500 times compared to HFSS full-wave simulation, while reducing memory usage by about 16%. The analytical method proposed in this paper not only avoids cumbersome three-dimensional modeling and meshing, but also achieves better accuracy than full-wave simulation with substantially reduced computational resources, providing a reliable and efficient analytical tool for rapid signal integrity assessment and engineering design of high-speed data cables with foil-overlap shields.Keywords:differential data cable;foil-overlap shield;signal integrity;insertion loss;analytical method100|77|0Updated:2026-10-08
- Abstract:In fan-out wafer level packaging (FOWLP), the redistribution layer (RDL) plays a crucial role in signal transmission between chips, achieving high integration of heterogeneous systems through the support of flexible vias and irregular I/O pad structures. However, as the interconnect density of chips continues to increase, the efficient utilization of RDL routing resources faces severe challenges: the routing graph without a reasonable partitioning method often reduces the quality of routing solutions; at the same time, determining the insertion position of vias before global routing weakens the potential advantages of flexible vias in reducing wirelength and improving routability. To address the complex problems faced by RDL routing between chips in FOWLP, this paper proposes a routing algorithm based on adaptive via insertion, which improves RDL routing quality while considering both flexible vias and irregular pad structures. The proposed algorithm includes the following three key strategies: (1) A strategy that introduces capacity-aware A* search and adaptive via insertion in global routing, selecting routing paths based on the trade-off between path length and resource occupation, and dynamically determining the insertion position of vias based on the available overlapping areas between adjacent layer grids to improve routability and reduce wirelength; (2) A partitioning strategy for the detailed routing stage, only partitioning the necessary areas affected by routing to retain continuous available routing space, reducing resource waste and lowering the cost of routing graph updates; (3) A line length refinement strategy based on degree of freedom expansion, used to increase the adjustable range of line segments to enhance the solution space of linear programming, thereby further reducing the total wirelength. Experimental results show that compared with the latest routing algorithm, this algorithm achieves an average reduction of 5.65% in wirelength, with a maximum reduction rate of 8.50%.Keywords:advanced packaging;redistribution layer;flexible via;adaptive via insertion;irregular pad structure;routing196|78|0Updated:2026-10-08
- Abstract:To meet the need for highly consistent true-time-delay control in compact broadband phased-array microsystems, this paper proposes and experimentally validates a switched-line true-time-delay unit (TDU) based on slow-wave rectangular air coaxial lines (RACLs). A switched-line TDU generates discrete delays from the electrical-length differences among its branches. Within a limited footprint, however, the physical path length, delay per unit length, insertion loss, operating bandwidth, and port arrangement are mutually constrained. Although conventional RACLs offer low dielectric loss, high isolation, and three-dimensional integration, their delay per unit length is limited by the air dielectric, making it difficult to simultaneously realize a compact layout and fine electrical-length adjustment. To address this problem, subwavelength periodic geometric modulation is introduced into the inner conductor of the RACL. By lengthening the current path and modifying the electromagnetic coupling between the inner and outer conductors, the modulation tailors the equivalent distributed inductance and capacitance, thereby reducing phase velocity and increasing delay per unit length. Three slow-wave configurations—meander-line, V-shaped, and spiral—are designed with identical outer-conductor dimensions and port configurations, and their delay enhancement, additional loss, dispersion, and fabrication sensitivity are compared through full-wave simulations. Based on this comparison, a four-branch TDU is constructed using single-pole four-throw (SP4T) RF switches. The basic path lengths provide the principal delays, whereas localized meander-line slow-wave sections serve as electrical-length trimming modules. Adjusting the slow-wave-section length and unit-cell geometry compensates for branch-delay deviations imposed by path layout, port positions, and fabrication constraints, improving delay quantization accuracy without changing the overall topology or external boundary. The proposed devices are fabricated using a wafer-level multilayer metal-dielectric additive manufacturing process and heterogeneously integrated, through gold-wire bonding, with a pair of GaAs SP4T switch chips. Three 1 cm long slow-wave micro-coaxial lines are characterized on wafer over 1~43 GHz. All exhibit stable transmission, good impedance matching, and controllable delay enhancement. The measured delays of the meander-line and V-shaped structures agree well with simulation results, while tolerance analysis shows that the spiral structure is more sensitive to variations in the thickness of the dielectric support bridges. The integrated TDU measures approximately 4.88 mm × 4.46 mm × 0.50 mm and realizes 2 bit operation with a 5 ps step and a maximum relative delay of 15 ps over 1~20 GHz. Its delay root-mean-square error is below 0.72 ps, its insertion loss ranges from 2.0 to 4.7 dB, and the insertion-loss variation among states is approximately ±0.5 dB. The operating bandwidth and insertion loss are primarily limited by the selected switch chips. These results show that combining periodic slow-wave modulation with localized slow-wave loading extends micro-coaxial electrical-length adjustment beyond dependence on physical path length to geometric-parameter control, providing a low-loss, high-precision three-dimensional implementation for compact broadband true-time-delay microsystems.Keywords:radio frequency;microsystems;true-time-delay;slow-wave structure;micro-coaxial line;ultra-wideband124|64|0Updated:2026-10-08
- Abstract:In recent years, with the rapid advances in multimodal pre-trained models, image description generation has progressed from brief and general descriptions to more specific, detailed, and flexible content expressions. However, existing image description models still mainly focus on surface-level visual content, such as object categories, appearance attributes, and spatial locations, making it difficult to adequately satisfy users’ needs for a more comprehensive understanding in real-world applications, including fine-grained category information, functional purposes, and relevant external background knowledge. Particularly in knowledge-intensive vertical domains such as museum guiding and artwork interpretation, users expect not only shallow visual descriptions of images, but also additional in-depth information, such as historical origins, artistic schools, symbolic meanings, and other related content. In addition, explicitly aligning narrative content with key visual regions in an image and simultaneously grounding textual segments to their corresponding image regions can help users understand the relevant knowledge more intuitively and support further interaction and exploration around specific details. Therefore, this paper introduces the task of grounded image narrative generation, which takes an image and a user-provided multimodal instruction as input and generates a multimodal narrative containing rich information, coherent logical relationships, and accurate region grounding. To achieve this goal, we propose a knowledge retrieval-augmented narrative generation method that first actively retrieves rich background information from external knowledge bases. We then introduce a chain-of-thought reasoning mechanism jointly constrained by cross-modal knowledge association and region localization, which selects effective knowledge relevant to the user query, and leverages visual localization to better align the generated text with corresponding local image regions, ultimately enabling decomposition and reasoning of queries as well as structured generation of narrative content. For data construction, we extend the encyclopedic image narrative dataset ImgNarr-23K with region-grounding annotations and further construct ArtNarr, a dataset specifically for the artwork domain, providing data support for research on knowledge-intensive image narrative generation. Experimental results show that the proposed method outperforms comparison methods based on mainstream multimodal large models across datasets from multiple domains. It is capable of generating high-quality image narratives with integrated textual and visual content, containing complete and accurate image information and relevant knowledge while achieving effective region grounding, thereby demonstrating its strong potential and practical value in real-world multimodal application scenarios.Keywords:image narrative generation;knowledge retrieval augmentation;visual grounding;chain-of-thought reasoning;vision-language model;multimodal learning169|72|0Updated:2026-10-08
- Abstract:Binary segmentation partitions image pixels into foreground and background and serves as a fundamental task in natural scene analysis, medical image processing, and complex environment perception. Salient object detection, camouflaged object detection, marine animal segmentation, mirror detection, polyp segmentation, and shadow detection all produce binary masks, but these tasks define the foreground in different ways and therefore rely on substantially different discriminative cues. Salient object detection focuses on visually prominent regions, whereas camouflaged object detection aims to identify concealed targets from low-contrast areas, weak boundaries, and textures that are highly similar to the surrounding background. Other binary segmentation tasks additionally depend on task-specific priors, including medical structures in polyp segmentation, reflection properties in mirror detection, illumination variations in shadow detection, and underwater imaging degradation in marine animal segmentation. Existing methods usually construct an independent network for each task or preserve task-specific parameters within a shared backbone. Although these designs can accommodate task differences, they often introduce model redundancy, increase training and maintenance costs, and make unified deployment across multiple tasks more difficult. To address these limitations, this paper proposes unified binary segmentation with mixture-of-experts adaptation (UniBS-MoE), a unified framework for diverse binary segmentation tasks. The framework adopts a pretrained visual foundation encoder as a shared backbone, freezes its main parameters, and alternately inserts standard adapters and mixture-of-experts adapters (MoE-Adapters) into the encoding stage. The standard Adapters learn task-shared information, such as edges, textures, and regional structures, while the MoE-Adapters dynamically adjust feature representations according to task semantics. This alternating design enables the model to preserve common visual knowledge across tasks while retaining sufficient task-specific adaptation capacity. To provide task-level control, this paper further introduces a task-aware gating module and employs the contrastive language-image pre-training (CLIP) model to encode textual task descriptions. The gating module maps the user input into a unified semantic space and combines the resulting representation with task information to generate continuous expert-gating weights. These weights determine how the outputs of different experts are dynamically combined, allowing the model to select and integrate expert responses according to the current task rather than relying on a fixed task-specific branch. The multi-level features produced by the encoder are subsequently fused through feature enhancement, spatial attention, and channel upsampling. This design strengthens both spatial details and high-level semantic information before prediction. The model is optimized using a multi-scale structure loss composed of weighted binary cross-entropy loss and weighted intersection-over-union loss, enabling the model to learn foreground regions and object boundaries at different prediction scales. This paper evaluates UniBS-MoE on six task categories and eighteen public datasets. The experimental results show that UniBS-MoE achieves leading or competitive performance on most tasks. The ablation studies further demonstrate that alternating standard Adapters with MoE-Adapters provides a favorable balance between segmentation performance and parameter overhead. The results also show that the task-aware gating module maintains stable adaptation when task descriptions are expressed in different forms. When the number of jointly trained tasks increases from two to six, the performance of the previously included tasks does not exhibit continuous degradation. This observation indicates that the proposed framework can incorporate additional supervised tasks while preserving its ability to handle the existing tasks. Overall, the experimental results demonstrate that text-driven expert adaptation can preserve both task commonality and task specificity within a unified set of model weights. The proposed framework reduces the need to train and maintain a separate complete network for every binary segmentation task and provides a feasible solution for unified training and deployment across diverse binary segmentation scenarios. The source code is available at: https://github.com/typist2001/TAMT.Keywords:salient object detection;camouflaged object detection;multi-task binary segmentation;mixture-of-experts adaptation;task-aware gating152|77|0Updated:2026-10-08
MDW: The Minimum Double-Weight of Differential Characteristics Searching Model Enhanced Publication
Abstract:Block ciphers are a core technology in the field of information security and serve as a crucial cornerstone for data protection. Differential cryptanalysis, as a classic method for evaluating the resistance of block cipher algorithms against attacks, heavily relies on the ability to find the differential characteristic with the highest probability, i.e., the optimal differential characteristic. Automated search techniques based on Boolean satisfiability theory (SAT) are one of the important means for evaluating the security of block cipher algorithms. Methods based on the minimum number of active S-boxes under the byte-wise model, though simple in model and fast in solving speed, suffer from insufficient evaluation accuracy due to neglecting the specific probabilities of differential propagations. On the other hand, methods based on optimal differential characteristics under the bit-wise model, while providing more precise security metrics, have complex search models. In particular, the complex constraint encoding of the difference distribution table for 8-bit S-boxes leads to low model solving efficiency, making it difficult to complete effective searches under a high number of rounds within a reasonable time. This paper proposes a new SAT-based automated search model, minimum double-weight (MDW), for identifying minimal double-weight differential characteristics in block ciphers with 8-bit S-boxes. Unlike traditional methods that uniformly convert differential propagation probabilities into equal-weight auxiliary variables for constraint, the MDW model decomposes them into two auxiliary variables with different weights, thereby effectively reducing the number of variables and clauses required for SAT modeling. To this end, this paper constructs a new boundary determination algorithm and a weight-increment-and-split algorithm, which are used to compute the boundary values of the two weighted auxiliary variables and to increase the target probability weight or decompose the total weight into possible combinations of the two auxiliary variables when the search yields no solution, respectively. These algorithms achieve efficient pruning of the search space, thereby effectively improving the solving efficiency of the model. Furthermore, the new MDW model is applied to the Chinese commercial cipher standard SM4, completing the automated search for locally optimal differential characteristics across all 32 rounds. Compared with existing methods, the results show that the new model achieves maximum reductions of 89% and 7% in the number of constraint variables and clauses, respectively, and demonstrates significantly improved search efficiency in high-round scenarios, with an improvement factor of approximately 30 times at round 20. The optimal differential characteristic probabilities for rounds 1 to 18 are fully consistent with existing results, and the tightest known lower bounds for the optimal probabilities in rounds 21 to 32 are obtained. This technique provides a novel and effective tool for the rapid assessment of block ciphers’ resistance to differential cryptanalysis.Keywords:block cipher;SM4;differential cryptanalysis;differential characteristic;Boolean satisfiability theory (SAT);automated search187|88|0Updated:2026-10-08- Abstract:Electroencephalography foundation models (EFMs) have significantly improved the upper limit of EEG signal decoding through the “pre-training and fine-tuning” paradigm. However, with the exponential increase in model parameters, conventional full-parameter fine-tuning faces prohibitively high computational and storage costs, and is prone to disrupting the general representations acquired during the pre-training stage in small-sample downstream tasks, thereby leading to catastrophic forgetting. Existing general parameter-efficient fine-tuning methods originating from the field of natural language processing, such as low-rank adaptation (LoRA), are mostly limited to implicit operations in the temporal or spatial domains and fail to explicitly capture the frequency-domain neural rhythm priors in EEG signals that are closely related to cognitive and physiological states. To address the issue that existing general parameter-efficient fine-tuning methods cannot explicitly utilize electroencephalography (EEG) frequency-domain neural rhythm priors, this paper proposes TF-PEFT (temporal-frequency parameter-efficient fine-tuning), a temporal-frequency dual-stream parameter-efficient fine-tuning framework. The proposed framework aims at low-cost adaptation of EEG foundation models. By freezing most of the pre-trained parameters, it introduces frequency-domain neural rhythm information into the downstream fine-tuning process, achieving collaborative optimization between pre-trained temporal representations and explicit spectral priors. Specifically, TF-PEFT employs LoRA to perform low-rank task adaptation on temporal attention representations in the pre-trained backbone, learning downstream task information with fewer trainable parameters. Meanwhile, a lightweight frequency-domain branch is constructed to extract stable and discriminative spectral features. Furthermore, a hierarchical low-rank compression and temporal-frequency cross-attention fusion mechanism are designed, enabling frequency-domain spectral structures to dynamically supplement downstream discriminative information according to temporal contexts. Finally, a near-zero-initialized gating mechanism is introduced to control the gradual involvement of the frequency residual branch, avoiding excessive perturbations to the original pre-trained representations caused by the newly introduced frequency-domain information and ensuring representation stability during the fine-tuning process. Systematic experiments conducted on 11 heterogeneous datasets covering emotion recognition, motor imagery, clinical abnormality detection, and sleep staging tasks, targeting five mainstream EEG foundation models including EEGPT, LUNA, CSBrain, CBraMod, and LaBraM, demonstrate that TF-PEFT achieves significant performance advantages. With an extremely low computational cost by updating only 2.2% of the original model parameters on average, the proposed method outperforms standard fine-tuning strategies such as LoRA and Adapter in classification accuracy across most scenarios, and surpasses full-parameter fine-tuning in several high-noise tasks. Taking the EEGPT model as an example, TF-PEFT outperforms full-parameter fine-tuning on multiple datasets: on the SEED dataset, the balanced accuracy (B-Acc) is improved from 49.9% to 58.19%; on the Sleep-EDF dataset, the average accuracy is increased from 60.99% to 69.21%. This study provides a general and hardware-friendly efficient fine-tuning paradigm for the low-cost deployment of EEG foundation models in complex, resource-constrained edge clinical devices and brain-computer interface applications.Keywords:EEG signals;foundation models;parameter-efficient fine-tuning;frequency-domain feature extraction;brain-computer interface191|75|0Updated:2026-10-08
- Abstract:To address the difficulty of quality prediction caused by mixed multi-time-scale dynamics in industrial soft sensing, as well as the limitations of existing transformer-based methods in unified temporal encoding, where mixed temporal information is prone to interference, and the lack of target-space structural constraints in the standard attention mechanism, this paper proposes a temporal decoupling and metric alignment transformer (TDMA-Trans) for industrial multi-time-scale soft sensing. Considering that industrial process variables usually contain both slow-changing trends caused by operating condition evolution and short-term fluctuations induced by control regulation and periodic operations, information at different time scales exhibits significant heterogeneity. Direct unified modeling is therefore likely to cause feature entanglement and weaken the model's ability to capture critical dynamic patterns. Based on time series decomposition, the model splits the input sequence into trend and periodic components, and uses a dual-branch encoder to learn slow-changing trends and periodic fluctuations separately, so as to reduce the interference between different time scales. Specifically, the two branches respectively capture long-term dependencies and local dynamic characteristics, preserving the integrity of temporal information at each time scale while providing a more stable feature foundation for subsequent cross-scale interaction. On this basis, a metric alignment attention mechanism is designed by introducing the distance relations among samples in the target space into the feature learning process, which guides attention allocation and makes the learned representation more consistent with the changing pattern of the target space. This mechanism overcomes the limitation of conventional self-attention that relies solely on feature similarity, enabling the model to capture both variable correlations and the geometric structure of the prediction target, thereby producing more discriminative feature representations. In addition, a bidirectional query-key exchange fusion mechanism is built to combine trend and periodic information, which improves the modeling ability for complex industrial dynamics. Through bidirectional information interaction, features at different time scales complement each other, further enhancing the model's capability to characterize complex operating condition variations and dynamic coupling relationships. Experimental results on a hydrocracking industrial dataset show that, compared with the best-performing model in the comparative experiments, for the light naphtha C5 content prediction task, the proposed method reduces RMSE (Root Mean Square Error) and MAE (Mean Absolute Error) by 20.0% and 28.6%, respectively, and improves R² by 7.1%. For the aviation kerosene 90% recovery temperature prediction task, RMSE and MAE are reduced by 14.3% and 40.0%, respectively, and R²is improved by 4.7%. In addition, ablation studies further verify the effectiveness of the three key modules, namely temporal decoupling, metric alignment attention, and bidirectional interaction fusion. These modules work collaboratively to improve both the prediction accuracy and generalization capability of the model. These results show that the proposed method can effectively improve the accuracy and stability of soft sensing under complex industrial multi-time-scale conditions.Keywords:soft sensing;deep learning;temporal decoupling;Transformer;metric alignment attention502|298|0Updated:2026-10-08
- Abstract:Software comprehension constitutes the costliest activity during software maintenance. Key classes are widely regarded as the optimal starting point for understanding complex software systems. Initiating software comprehension from key classes has become an effective approach to understanding complex software systems. Moreover, key classes can also be used to facilitate the simplification of reverse-engineered class diagrams and improve the efficiency of software documentation. Although numerous key class identification techniques have been proposed, two major limitations persist: (1) existing studies solely take the dependencies among classes as the only basis for establishing edges in software networks (i.e., graphs consisting of classes and their dependencies), overlooking the potential impact of pervasive code clones on the topology of software networks (i.e., graphs consisting of classes and their dependencies) and class importance, which leads to “pseudo-high” or “pseudo-low” class importance values; (2) they treat software networks as first-order graphs, where nodes are only connected by direct edges, ignoring higher-order substructures (e.g., motifs) that encode cooperative coupling patterns among multiple classes and their impact on the strength of inter-class dependencies and class importance. To address these issues, we present CloAct, a key class identification approach that integrates clone detection and motif analysis. First, CloAct constructs a class level software network (CN), where nodes represent classes and edges represent static dependencies between classes. Code clone relationships are then identified to yield a refined class level software network (RCN). Second, all three- and four-node motifs are extracted from RCN to build a motif-based class co-occurrence network (MCN), and a motif-based generalized coreness of classes (MGC) is calculated for each node to capture higher-order structural information. Third, CloAct incorporates MGC into ClassRank (a PageRank variant) to develop a new class importance metric, namely motif-based class rank (MCR). Finally, classes are sorted in descending order of MCR, and a threshold k is applied to generate the candidate key class set. An extensive evaluation was conducted on 11 open-source Java systems, and CloAct was compared against 11 comparative approaches. The results in terms of Win/Tie/Loss counts, the average ranking of the Friedman test, the p values from Wilcoxon signed‑rank test, and Cliff’s Delta show that: (1) CloAct significantly outperforms all comparative approaches, (2) both clone detection and motif extraction contribute substantially to the performance gain of CloAct, (3) different weighting mechanisms for different coupling types have no significant impact on the results (i.e., any scheme can be adopted without compromising effectiveness), (4) CloAct maintains high runtime efficiency, and (5) the parameters of the clone detection tool NiCad have no impact on the relative performance of CloAct (CloAct outperforms the baselines regardless of the parameter values).Keywords:key class identification;software network;code clone;motif;software measurement;software understanding252|88|0Updated:2026-10-08
- Abstract:Mining dense subgraphs in bipartite graphs is a fundamental problem in graph data management and analysis, with wide applications in social network analysis, fraud detection, and recommender systems. As an important dense subgraph model for bipartite graphs, the
-core has attracted significant attention due to its intuitive definition and high computational efficiency. The BiCore-Index is an index structure that supports efficient -core queries. Its dynamic maintenance consists of two crucial tasks: core number update and index structure update. For core number update, existing methods fail to consider the implicit dependencies reflected by the vertex removal order during core decomposition, and often traverse entire connected subgraphs of vertices with the same core number, incurring substantial redundant access. For index update, however, efficient incremental maintenance methods tailored to the BiCore-Index are still lacking. To address these issues, this paper presents an efficient dynamic maintenance method for the BiCore-Index. For the two key tasks, an -order-based core number maintenance algorithm and an incremental index maintenance strategy are proposed, respectively. The former leverages the -order to capture the hidden dependencies among vertex core numbers, limiting the search scope to the necessary dependency paths and thus effectively avoiding the redundant accesses in existing algorithms. The latter avoids full index reconstruction by requiring only local adjustments to the index positions of vertices whose core numbers have changed, thereby significantly reducing the index update overhead. Furthermore, targeting practical batch update scenarios, this paper designs batch acceleration algorithms for both core number and index updates. By merging redundant computations, skipping invalid intermediate states, and adopting deferred computation, the overall cost of batch maintenance is reduced. Experimental results on ten real-world datasets show that the proposed core number update algorithm achieves a speedup of 1 to 3 orders of magnitude over existing methods in single-edge update scenarios, and the incremental index maintenance method achieves a speedup of 2 to 4 orders of magnitude compared with full index reconstruction. Extensive experiments validate that the proposed methods are efficient and exhibit good scalability.Keywords:bipartite graph;-core;BiCore-Index;dynamic maintenance;batch processing154|71|0Updated:2026-10-08 - Abstract:Advanced persistent threats (APT) often employ dynamic strategies, causing network traffic to exhibit non-stationary and stochastic fluctuation characteristics. This leads to performance degradation in existing in-network anomaly traffic detection models. Traditional global retraining-and-redeployment approaches support runtime model updates but require full reconstruction of the decision tree and synchronized updating of all flow table entries. This process tends to strain control-plane resources, cause packet loss, and compromise system reliability and stability. To address these challenges, this paper proposes a lightweight update mechanism for anomaly traffic detection models based on concept drift awareness, enabling adaptive and efficient model updates by detecting shifts in attack strategies and monitoring changes in model decision paths. First, a self-supervised learning framework is employed to construct a triplet dataset of anomalous traffic, from which low-dimensional, high-density feature vectors are extracted. Next, a sliding window-based method is introduced to monitor concept drift in the feature sequence, triggering alerts upon detection of strategy transitions. Finally, a Pareto optimization framework is leveraged to generate an optimal set of update nodes, balancing detection accuracy against update overhead, thereby achieving lightweight updates of in-network decision tree models. Experimental results show that, compared with existing global update mechanisms for in-network decision trees, the proposed approach achieves 93.7% accuracy in identifying attack strategy transitions, and reduces model update latency by 12.1%, while ensuring that the degradation in classification accuracy remains within 6%.Keywords:anomaly traffic detection;decision tree;concept drift awareness;multiple input network;pareto optimization101|79|0Updated:2026-10-08
PAPERS
- Abstract:With the continuous evolution of network application scenarios and underlying technological systems, network transmission faces increasingly complex operating conditions and more diverse performance requirements. However, traditional transmission mechanisms that rely primarily on end-side feedback suffer from limited sensing granularity and control capability. When facing complex service loads and transmission scenarios, they struggle to promptly perceive changes in connection status and network resources and implement effective control, thereby constraining improvements in resource utilization. To overcome these limitations, researchers have progressively introduced network-side sensing and control capabilities, and explored end-network collaborative transmission mechanisms to improve the efficiency of connection management and resource scheduling. This paper systematically reviews the evolution of network transmission technologies, summarizes the technical evolution and key methods of three categories of transmission mechanisms, namely end-side transmission control, network-side transmission scheduling, and end-network collaborative transmission. Representative studies are further compared in terms of observation signals, control granularity, execution location, feedback type, optimization objectives, and deployment constraints. Based on this comparison, the advantages and limitations of different mechanisms are analyzed in terms of performance, applicable scenarios, and control capabilities, while their applicability boundaries, commonalities, and differences in information acquisition, resource regulation, and deployment are further summarized. Overall, network transmission mechanisms are evolving from independently controlled end-side schemes toward enhanced network-side scheduling and end-network collaboration, showing a trend toward coordinated control among multiple entities as well as increasingly enhanced state awareness and control capabilities. Based on this review, this paper further discusses the key challenges of current network transmission technologies in complex-scenario adaptation, multi-objective collaborative optimization, state information interaction, and practical deployment, and outlines future research directions. This work aims to provide guidance for the design of future transmission architectures and mechanisms, thereby supporting the efficient deployment and intelligent evolution of next-generation transmission systems.Keywords:end-side transmission control;network-side transmission scheduling;end-network collaborative transmission128|78|0Updated:2026-10-08
SURVEY AND REVIEW








