Objective: The performance of quasi-cyclic low-density parity-check (QC-LDPC) codes is adversely impacted by the presence of short cycles, particularly 4-cycles and 6-cycles. Existing methods to eliminate these short cycles can be generally divided into two categories: search-based methods and explicit methods. Search-based methods, such as symmetrical construction, typically involve extensive searches for exponent matrices that satisfy specific structural constraints. This results in a complex search process and high description complexity for QC-LDPC codes. In contrast, explicit methods, such as the greatest common divisor (GCD) and shifting sequence (SS) methods, leverage specific mathematical formulas to directly define the required exponent matrix. This eliminates the need for computer-based searches and leads to lower description complexity. Additionally, the column weight is a crucial factor influencing the performance of QC-LDPC codes. Existing explicit construction methods that ensure the absence of 4-cycles and 6-cycles are primarily applicable to QC-LDPC codes with column weight 3. However, methods suitable for column weight 4 are still relatively rare. The objective of this study is to investigate a novel explicit construction method for QC-LDPC codes with column weight 4, girth 8, and excellent decoding performance. Methods: Inspired by the SS method for constructing QC-LDPC codes with column weight 3, this paper introduces a novel SS method to design QC-LDPC codes with column weight 4. The core methodology involves two steps: directly defining two original sequences using mathematical formulas and then generating two derived sequences by right shifting the original sequences. These four sequences together form the required exponent matrix. Subsequently, an analysis is conducted to verify that for any circulant size above a certain lower bound, the governing equations for 4-cycles and 6-cycles do not hold. Finally, the sum-product algorithm (SPA) is employed to simulate the newly constructed codes and compare their decoding performances with those of several typical QC-LDPC codes. Results: Through a rigorous mathematical analysis of the governing equations for 4-cycles and 6-cycles, it is concluded that when the circulant size is greater than or equal to the difference between the maximum element and half of the row weight, neither 4-cycles nor 6-cycles exist. Moreover, due to the presence of 8-cycles, the newly constructed codes exhibit a girth of exactly 8. Simulation results indicate that the new codes outperform codes constructed using the GCD method and perform similarly to codes based on symmetrical construction but with a significantly simpler construction process. Furthermore, the new codes have the potential to outperform 5G codes in the high-signal-to-noise region. Conclusions: A novel explicit construction method for high-performance QC-LDPC codes is proposed, offering the following advantages: 1) applicability to any row weight L and 2) elimination of the need for exhaustive searches, allowing QC-LDPC codes to be explicitly constructed through mathematical formulas, thereby significantly reducing description complexity.
Objective: Spatially coupled low-density parity-check (SC-LDPC) codes have attracted considerable interest in recent years owing to their exceptional decoding performance, low latency, and unique coupling structure, which notably enables error correction capabilities. These codes leverage the threshold saturation effect, allowing their performance to approach the Shannon limit closely. Their capability to deliver high reliability with reduced decoding complexity positions them as a promising choice for next-generation communication systems, including 6G networks and satellite communications. This paper introduces a novel construction method for SC-LDPC codes, referred to as SC multi-weight circulant quasi-cyclic LDPC (SC-MQC-LDPC) codes, based on multi-weight circulant matrix decomposition. These codes are designed to be compatible with a wide range of code lengths and code rates, offering increased flexibility and applicability in diverse scenarios. Methods: The construction process begins with the design of MQC-LDPC codes, where a lifting value matrix is determined using a simplified error minimization progressive edge growth algorithm; this algorithm is specifically tailored to optimize the structural properties and decoding performance of the codes. By accounting for the presence of short cycles and the extrinsic message degree of check nodes, the algorithm effectively mitigates the error floor, thereby enhancing the overall reliability and efficiency of the MQC-LDPC codes. These base codes are then extended using a split-replication process to construct SC-MQC-LDPC codes. This extension preserves the beneficial characteristics of the original MQC-LDPC codes while introducing spatial coupling, which further improves error-correction capabilities and supports a broad range of communication requirements. The paper also introduces a recursive encoding method for SC-MQC-LDPC codes, which offers low implementation complexity and reduced latency, increasing its suitability for practical deployment. Additionally, an improved sliding window decoding algorithm is introduced to further optimize the decoding process. This low-complexity algorithm enhances decoding efficiency by balancing memory usage and computational requirements. With a modest increase in memory overhead, the algorithm successfully mitigates error propagation and improves overall decoding performance, ensuring robust data transmission even under challenging conditions, such as low signal-to-noise ratios. Results: The performance of the proposed SC-MQC-LDPC codes is rigorously evaluated through comprehensive simulations. The experimental results show the following: 1) The proposed SC-MQC-LDPC codes, designed for compatibility with various code lengths and rates, achieve a performance gain of over 0.50 dB compared with SC-5G-LDPC codes at a bit error rate of 10-6, when used in combination with the modified sliding window decoding (SWD) algorithm. Furthermore, under identical code length and rate conditions, they demonstrate clear performance advantages over the newly extended 5G-NR LDPC codes, particularly in the low Eb/N0 region. Additionally, the modified SWD algorithm significantly improves decoding performance across all tested SC-MQC-LDPC code variants compared to the conventional SWD algorithm, with the improvements becoming more pronounced as the code length increases. 2) In terms of computational complexity, SC-MQC-LDPC codes decoded with the modified SWD algorithm achieve substantial reductions of approximately 1/5 and 1/3 compared with SC-5G-LDPC and 5G new radio LDPC codes, respectively. At Eb/N0=6.0 dB, the decoding complexity of the modified algorithm is nearly half that of the traditional SWD algorithm, highlighting its advantage for low-complexity, high-efficiency decoding. Conclusions: Overall, the SC-MQC-LDPC codes proposed in this study mark a remarkable advancement in error-correction coding, effectively addressing the growing demand for high reliability, low latency, and computational efficiency. These characteristics make them highly suitable for modern communication environments that demand adaptable, efficient, and robust performance in dynamic and challenging scenarios.
Objective: The development of communication systems and increasing demand for low-latency and high-reliability communications has led to a rise in application scenarios requiring soft information interaction for joint iterative decoding to improve system performance. The soft-output successive cancellation list (SO-SCL) decoding algorithm of polar codes can achieve relatively accurate soft information output and decoding with the complexity of traditional successive cancellation list (SCL) decoding by estimating the codebook and posterior probabilities. However, the serial characteristics of SCL decoding result in a high decoding delay of the SO-SCL decoding algorithm, making it difficult to satisfy the 1 Tbps peak throughput requirement of the sixth-generation mobile communication system. To reduce the decoding delay, the existing soft-output fast SCL decoding algorithm (SO -FSCL) realizes fast decoding by identifying four special nodes; however, some nodes still have a high decoding delay. Therefore, a high-performance soft-output decoding algorithm for polar codes with lower decoding delay is required. Methods: In previous studies, special nodes were identified, aiming to achieve fast decoding of different special nodes and combined decoding between nodes. Based on the SO -FSCL decoding algorithm, this study introduces five other special nodes, namely REP-2, REP-3, REP-4, SPC-2, and SPC-3, and proposes a faster soft-output SCL decoding algorithm (FS-SCL). The developed decoding algorithm enables fast decoding of the five new nodes and provides a posterior probability formula for the deleted path. For the REP-2, REP-3, and REP-4 nodes, only 4, 8, and 16 possible decoding paths need to be considered, respectively, and the sum of the probabilities of the deleted paths is calculated. According to the distribution characteristics of the information bits in the nodes, SPC-2 can be simplified to the parallel decoding of two SPC sub-nodes with Ns/2 bits. After combination, the SPC-3 node can be regarded as a repetitive code with a code rate of 1, thereby simplifying the decoding process. Moreover, compared with SCL decoding, which requires flipping all bits, while with the SPC-2 and SPC-3 nodes, only the min{L-1, Ns-2}, min{L-1, Ns-3} bits require flipping during the decoding process, thereby reducing the decoding delay. The time steps required for decoding different nodes are also analyzed herein to evaluate the decoding delay. The five newly added nodes require 3, 4, 5, max{log2Ns+1, min{L, Ns-1}}, and max{log2Ns+3, min{L, Ns-2}} time steps, respectively. Compared with the original SCL decoding algorithm, the node significantly reduces the time steps required for decoding. Results: The simulation results show that by employing the AWGN channel, the proposed FS-SCL decoding algorithm maintains a BER performance similar to that of SO-SCL in different modulation methods, especially under 16QAM. By reducing the proportion of high-code rate nodes, the performance loss is reduced. Compared with the existing SO -FSCL decoding algorithm, the FS-SCL decoding algorithm can further reduce the decoding delay by more than 12.5% (up to 28.7%) and can further reduce the decoding complexity. Moreover, by merging shorter nodes, the FS-SCL decoding algorithm reduces the number of nodes by 42% and achieves a minimum node code length of 8, which is conducive to improving the decoding parallelism. Conclusions: The developed FS-SCL decoding algorithm with lower delay and complexity affords lossless BER performance using different modulation methods. The research results can provide an efficient polar-code decoding scheme for low-latency communication scenarios with soft output, which has important theoretical value and application prospects.
Objective: This study addresses the key challenge of designing adaptive coding schemes for nonstationary two-way communication channels, where forward and feedback links experience time-varying fading governed by Markovian memory. While traditional Schalkwijk-Kailath (S-K) coding frameworks perform well over static channels, they rely on offline, precomputed power allocation strategies that are unsuitable for channels with dynamically evolving conditions. Such static approaches fail to track Markov-correlated channel fluctuations, resulting in degraded signal-to-noise ratio (SNR) scaling over successive communication rounds, and, particularly under channel memory effects, thereby degrading error rate performance. Aiming to address this limitation, a novel S-K-inspired coding architecture that adaptively adjusts power allocation based on delayed channel state information (CSI). This adaptive design enhances the robustness and reliability of two-way communication over channels exhibiting Markov fading. Methods: Aiming to achieve this goal, a reinforcement-inspired, real-time optimization framework integrated within the S-K coding structure is introduced. The core innovation of the approach lies in its capability to dynamically allocate transmission energy for each communication round based on three critical inputs: 1) the channel state of the historical round (quantized via finite Markov states), 2) the Markov transition matrix of the channel (used to predict the next likely state), and 3) the instantaneous channel state at current iteration (causal CSI). In contrast to conventional S-K schemes that rely on static, offline power allocation, the proposed method reformulates power allocation as a sequential decision-making problem. This reformulation is realized through a dual optimization strategy that jointly considers coupled system parameters to maximize the expected cumulative equivalent SNR across multiround interactions. The optimization strategy begins with the derivation of relaxed constraints based on specific bidirectional transmission requirements (power budgets and error thresholds), and proceeds by employing greedy algorithms to allocate power for equivalent SNR maximization. A critical component of the scheme is its integration of predictive Markov state transition adjustments, enabling proactive power adjustments in anticipation of future channel variations. This predictive capability supports suboptimal yet resilient communication quality across rounds. Collectively, these strategies enable the proposed scheme to maintain high interaction reliability while supporting elevated transmission rates. Results: Simulation results and mathematic analysis show that the proposed scheme consistently outperforms conventional S-K-type schemes in a two-way Markov fading channel. Under ideal conditions, when the channels are in a stable state, the proposed scheme automatically reduces to the classical S-K solution for additive white Gaussian noise channels. In more general fading scenarios involving multiple random channel states, the scheme yields substantial performance gains over conventional S-K baselines by strategically leveraging noisy feedback to enhance communication robustness. Notably, even when compared to enhanced Markov channel models with idealized noise-free feedback, whose capacities represent the theoretical upper bound for the studied model, the numerical results reveal that the proposed scheme asymptotically approaches this limit through successive rounds of interaction. Conclusion: This study presents an adaptive feedback coding scheme for Gaussian channels with memory under unstable fading conditions. By dynamically adjusting encoding parameters using dual optimization in coupled systems, the proposed approach extends the S-K framework to achieve suboptimal yet robust transmission rates. Numerical simulations and mathematical analysis demonstrate that the scheme outperforms classical S-K methods in fading environments, particularly in the presence of channel fluctuations. By balancing causal adaptation with predictive optimization, the proposed architecture offers a promising solution for reliable communication in 5G/6G systems operating under nonstationary feedback conditions.
Objective: Reed-Solomon (RS) codes represent a category of error-correction codes that have been extensively used in various communication and storage systems. However, in some applications, such as optical transmission, very high transmission rates cause several challenges for the efficient implementation of codecs, especially for low-power performance. Methods: The conventional approach of employing a parallel configuration for both the encoder and decoder lacks efficiency. The fast Fourier transform (FFT) over finite fields can be utilized to reduce the decoding complexity of RS codes. Recently, Lin, Chung, and Han developed a novel FFT that, for the first time, attains the O(nlogn) complexity (n is the FFT size) for binary extension fields. An LCH-FFT-based encoding algorithm with complexity O(nlogk) was presented for an (n, k) RS code when
Objective: With the rapid evolution of fifth-generation (5G) mobile communication technologies, massive machine-type communication (mMTC) has become a pivotal application scenario in modern networks. This paradigm shift presents significant challenges in multiuser detection, particularly due to the exponential growth in user connections and heightened signal activity. Traditional orthogonal multiple access schemes, while ensuring minimal interuser interference, inherently limit the number of supported users by relying on orthogonal resource allocation, thereby failing to meet the scalability demands of mMTC. Consequently, grant-free nonorthogonal multiple access has emerged as a key enabler for Internet of Things communications, allowing nonorthogonal data superposition on limited resource blocks to enhance access capacity. However, existing sparse Bayesian learning (SBL) algorithms—although capable of achieving optimal sparse solutions—suffer from high computational complexity, primarily due to matrix inversion operations during expectation-maximization iterations. This complexity impedes real-time deployment in large-scale mMTC systems. To address this gap, this work proposes a novel SBL framework leveraging an efficient Laplace scale mixture (ELSM) prior, aiming to simultaneously enhance detection performance, reduce computational overhead, and adapt to dynamic multimeasurement scenarios. Methods: This paper proposes an ELSM-SBL algorithm to overcome the limitations of conventional SBL methods. First, a hierarchical Bayesian model is constructed using a Laplace scale mixture prior, which leverages the sharp peaks and heavy-tailed properties of Laplace distributions to promote sparsity and robustness against outliers. To avoid computationally expensive matrix inversions, a surrogate function is introduced to approximate the Gaussian likelihood function. This approximation is optimized within a majorization-minimization (MM) framework, where a block coordinate descent (BCD) algorithm solves the resulting nonconvex optimization problem. For single measurement vector (SMV) scenarios, the ELSM-SBL-SMV algorithm optimizes hyperparameters via evidence maximization, while an MM framework with BCD resolves nonconvexity in the joint cost function. For multiple measurement vector (MMV) scenarios, the ELSM-SBL-MMV scheme exploits temporal correlations among active user sets across consecutive time slots by sharing sparsity-controlling hyperparameters, thereby enhancing reconstruction performance. Results: Extensive simulations were conducted under mMTC settings with a total user count of K=108, subcarriers N=72, and active users M=12 using BPSK modulation and repeated over 1, 000 trials. For MMV scenarios, the number of measurement vectors was set to T=7. Compared with state-of-the-art SBL algorithms (e.g., GIG-SBL, BGIG-SBL, and LSM-SBL), the proposed ELSM-SBL-SMV algorithm can achieve a performance gain of about 2 dB, while the ELSM-SBL-MMV algorithm can achieve a gain of 3 dB. Meanwhile, the computational complexity of ELSM-SBL-SMV is
Significance: End-to-end semantic communication leverages deep learning models to extract semantic features from data, enabling intent-driven communication processes that significantly enhance transmission efficiency. However, existing semantic communication paradigms based on discriminative models employ symbol-level rate-distortion optimization and perform maximum likelihood estimation solely based on received signals, failing to satisfy the perceptual requirements of users. To ensure the visual quality of transmitted data, a generative visual semantic communication paradigm has emerged, which adopts a rate-distortion-perception optimization framework to achieve alignment between data transmission and human perception through maximum a posteriori estimation. Diffusion models are advantageous for controlling visual generation and have thus become essential tools for this generative paradigm. Nevertheless, systematic organization of the technical roadmaps for empowering semantic communication using diffusion models is lacking in current research. Progress: This study addresses this gap by modeling the communication process as a mathematical inverse problem and elucidating the general methodology by which diffusion models solve data compression and transmission challenges through posterior sampling. The fundamental concepts, mathematical formulations, and sampling strategies underpinning diffusion models are systematically introduced. In addition, the general methods and key technologies employed for diffusion model-enabled generative compression and transmission are comprehensively reviewed from an inverse problem-solving perspective. Moreover, the performance metrics commonly used for objective assessment of the visual quality of transmitted data are summarized to provide a comprehensive evaluation framework. The core methodology demonstrates that generalized communication processes can be effectively modeled as inverse problems. The approach involves inferring the source data distribution using maximum a posteriori estimation based on channel measurements and forward operators composed of various signal processing operations. Through diffusion posterior sampling, diffusion models solve these communication inverse problems via a three-step process: first, pre-training diffusion models from large-scale datasets are used to obtain diffusion priors; second, joint source-channel codecs are used to mitigate channel distortions in visual data transmission and construct proximal regularization terms; finally, measurement regularization terms are constructed based on channel measurements. By integrating these regularization terms for posterior estimation and distribution sampling, diffusion models can implicitly reconstruct source data through gradient descent, effectively overcoming transmission challenges caused by strong channel noise, nonlinear operators, and time-varying channel conditions. Conclusions and Prospects: The analysis reveals that compared to visual semantic communication approaches based on discriminative deep learning models, the generative visual semantic communication paradigm based on diffusion models can significantly improve transmission efficiency and resilience while ensuring perceptual quality and semantic consistency of visual information. This advancement represents a fundamental shift toward communication systems that prioritize human perceptual requirements alongside traditional distortion metrics. Open issues, including image realism modeling and acceleration of diffusion model sampling, are discussed. The report highlights the effectiveness of conditional diffusion models for enabling existing semantic communication architectures to recover sources at the receiver based on minimal tokens and highly degraded measurements, offering an intelligent and concise design philosophy for future generative visual semantic communication systems.
Objective: Post-processing plays a critical role in quantum key distribution (QKD) systems by correcting transmission errors and enhancing key security. In continuous-variable QKD (CV-QKD), where information is encoded onto the quadratures of light signals, efficient post-processing is crucial due to the inherent sensitivity of these signals to noise and potential eavesdropping. This paper proposes an innovative post-processing scheme based on polar lattices, targeting the challenges of information reconciliation and privacy amplification in CV-QKD systems. Methods: The core methodology treats the problem of information reconciliation as an instance of the Wyner-Ziv coding problem, which addresses source coding with side information. To implement this, we introduce a novel architecture based on two nested polar lattices designed to efficiently reconcile information between Alice (the sender) and Bob (the receiver). The outer lattice acts as a quantization codebook for the lossy compression of Alice's signal, while the inner lattice enables error correction, allowing Bob to accurately reconstruct Alice's original signal. Additionally, we utilize the theory of polarization to eliminate potential information leakage, ensuring the secrecy of the final key through privacy amplification. This approach not only improves the efficiency of information reconciliation but also enhances the robustness against channel noise and other impairments by incorporating advanced decoding techniques such as soft-decision decoding and adaptive quantization. Results: Our numerical results show a significant increase in the data coordination efficiency of the proposed scheme compared to that of traditional methods. As the code length increases, the efficiency of our method approaches the theoretical limit defined by the Slepian-Wolf bound, indicating near-optimal performance. Furthermore, simulations conducted under various conditions demonstrate that the proposed scheme maintains high performance even in challenging scenarios characterized by low signal-to-noise ratios and high channel noise. These findings suggest that our approach offers substantial improvements in both reliability and security for CV-QKD systems. Conclusions: In conclusion, this paper presents a comprehensive and practical solution for post-processing in CV-QKD systems utilizing polar lattices to address the dual challenges of information reconciliation and privacy amplification. By treating the reconciliation process as a Wyner-Ziv coding problem and employing advanced polar lattice-based encoding and decoding strategies, our scheme achieves near-theoretical performance limits. The demonstrated scalability and compatibility of our method with modern optical communication systems make it highly suitable for real-world deployment. This study represents a major step forward in the development of efficient, secure, and scalable QKD technologies, paving the way for broader applications in quantum cryptography. Moreover, the success of this approach highlights the potential of integrating classical coding theories into quantum communications, opening new avenues for research and innovation. Future work will focus on further optimizing the parameters of the polar lattice structure and exploring its applicability in more complex and dynamic environments, with the aim of pushing the boundaries of what is achievable with current QKD technologies.
Objective: Quasi-cyclic low-density parity-check (QC-LDPC) codes are a class of channel codes known for their quasi-cyclic structure and sparse parity-check matrices. They have attracted significant attention due to their relatively simple hardware implementation and superior decoding performance. However, short cycles—particularly four and six cycles—can severely degrade the decoding performance of QC-LDPC codes. A key challenge in current LDPC code construction research is designing QC-LDPC codes with flexible circulant sizes and column weights while eliminating four and six cycles. Existing methods to eliminate short cycles can be broadly divided into two categories: search-based and explicit methods. Search-based methods provide parameter flexibility but involve high descriptive complexity. In contrast, explicit methods construct QC-LDPC codes directly through formulas, resulting in low descriptive complexity. In recent years, several explicit methods based on combinatorial mathematics have been introduced, with disjoint difference sets (DDS) playing a central role in constructing QC-LDPC codes free of four and six cycles. However, existing methods cannot simultaneously achieve the elimination of four and six cycles, support diverse column weights, and offer flexible circulant sizes. To address this gap, this study proposes an explicit construction method for QC-LDPC codes based on DDS that satisfies all three requirements. Methods: This study integrates DDS with row-weight extension (RE) techniques to propose a new construction method termed DDS-RE. First, the existence of four and six cycles in the Tanner graph corresponding to the new construction is analyzed using cycle control equations. Second, the flexibility of circulant size selection in the new construction is compared with the original DDS method. Finally, bit error rate (BER) and block error rate (BLER) performances of the newly constructed QC-LDPC codes are evaluated through simulations using the sum-product algorithm (SPA) and compared with several representative explicit methods: the greatest-common-divisor (GCD), Golomb ruler (GR), Sidon sequence (SS), and the original DDS methods. Additionally, decoding performance is compared with two construction methods incorporating RE techniques (GCD-RE and SS-RE). Results: This study rigorously proves, using cycle control equations, that the Tanner graph of the QC-LDPC codes constructed by the new method is free of four and six cycles. Analysis reveals that, compared with existing DDS methods, the newly proposed DDS-RE method enables smaller circulant sizes, particularly for larger column weights, where circulant size flexibility is significantly enhanced. Simulation results of decoding performance demonstrate that the new codes outperform GR codes in the high signal-to-noise ratio (SNR) region while also offering flexible column weights and circulant sizes. They also perform comparably to DDS-based codes but with the added advantage of circulant size flexibility. Furthermore, the DDS-RE codes significantly outperform those constructed by the GCD and GCD-RE methods. In high SNR regions, they also slightly surpass SS-RE codes while benefiting from flexible circulant sizes. Conclusions: By combining the DDS method with RE techniques, this study presents a new explicit construction method for QC-LDPC codes—DDS-RE. The resulting codes are guaranteed to be free of four and six cycles and support flexible circulant sizes. Compared with existing explicit methods and other RE-based methods, DDS-RE constructs codes with excellent performance while simultaneously achieving the elimination of four and six cycles, diverse column weights, and flexible circulant sizes. This study offers new insights into the explicit construction of high-performance QC-LDPC codes.
Objective: Indoor visible light communication (VLC) suffers from the low modulation bandwidth of a single light-emitting diode (LED), which hardly meets the demand for high-rate transmission. As a multiple-input-multiple-output (MIMO) technology, generalized spatial modulation (GSM) allows indoor VLC systems to achieve high spectral efficiency and excellent anti-interference capability using multiple LEDs for data transmission. However, in indoor MIMO-GSM-VLC systems, narrow spacing between multiple LEDs makes the channel characteristics between different LEDs and photodetectors very similar, resulting in a high correlation of the channel matrix. Because of the special characteristics of the VLC channel, the existing constellation mapping and channel coding schemes are no longer feasible for indoor MIMO-GSM-VLC systems. Therefore, incorporating the spatial characteristics of the VLC channel to develop a high-reliability and high-efficiency optical transmission scheme is imperative. Methods: The proposed optical transmission scheme is divided into two parts: a constellation mapping scheme and an improved protograph low-density parity-check (LDPC) code. First, this paper proposes a novel spatial multipulse position modulation (MPPM) mapping scheme, called the unequal power spatial MPPM (UPSM) constellation, to realize the joint optimization of GSM and MPPM. In particular, owing to the correlation of the VLC channel matrix, effective LED activation groups with similar characteristics transmitting MPPM symbols will cause serious intergroup interference in the MIMO-GSM-VLC system. Based on the VLC channel matrix, the principle and calculation method of influence coefficients for LED activation groups are presented to reallocate the peak transmit power of the MPPM symbols for effective LED activation groups, which can efficiently mitigate the attenuation of MPPM symbols in transmission. In addition, the maximum Hamming distance principle is used to optimize the mapping relationship between MPPM labels and MPPM symbols in the UPSM constellation. Consequently, the proposed UPSM constellation can be constructed by power allocation and label-to-symbol mapping optimization. Second, this paper proposes an improved protograph LDPC code with the aid of a protograph extrinsic information transfer (PEXIT) algorithm and an asymptotic weight distribution (AWD) function. In particular, some empirical constraints (including matrix dimension and variable node degree distribution) are imposed on the protograph (i.e., the base matrix) to reduce the encoding and decoding complexity. Next, based on a computer search method, the PEXIT algorithm optimizes the protograph LDPC code to achieve the minimum decoding threshold, leading to substantial bit-error-rate (BER) performance in the low signal-to-noise ratio (SNR) region. Furthermore, to avoid the error floor in the high SNR region, the AWD function is employed during the construction of the improved protograph LDPC code, guaranteeing that the protograph enables the linear minimum distance growth property. Results: The simulation and analysis results show that the proposed UPSM constellation mapping scheme considerably outperforms the natural constellation, Gray-like label (GL) constellation, and unequal power GL (UPGL) constellation. In addition, the proposed improved protograph LDPC code exhibits excellent convergence performance and the lowest decoding threshold compared with the existing counterparts in the MIMO -GSM-VLC system with the proposed UPSM mapping scheme. Conclusions: This paper conducts an in-depth investigation of the joint design of protograph LDPC codes and spatial MPPM constellation. The proposed constellation mapping scheme and the improved protograph LDPC code benefit from the remarkable BER performance and strong antifading robustness. Given these advantages, the proposed schemes are expected to be competitive solutions for indoor VLC applications.
Significance: Microgravity combustion research is essential for understanding fundamental combustion phenomena and advancing combustion theory. However, conducting experiments in orbit involves significant technical challenges and resource demands. The Combustion Science Rack (CSR) aboard the China Space Station (CSS) has been operational since October 2022. To further support combustion science research aboard CSS, consolidate critical scientific questions in microgravity combustion, and validate the in-orbit experiment feasibility, a ground-based research platform has been established at the space experiment center in Huairou District, Beijing. This platform replicates the combustion environment and apparatus dimensions of the in-orbit CSR. Equipped with high-precision diagnostic tools and versatile experimental modules, the platform enables researchers to validate in-orbit experiment feasibility, conduct ground-based validation tests, and generate baseline control data for CSS experiments. The paper highlights the platform's design, operational principles, and preliminary test results. Together, these demonstrate its ability to meet the diverse requirements of current and future microgravity combustion research projects. Progress: The platform consists of the experimental insert subsystem, the supporting facility subsystem, and the combustion diagnostic subsystem. Designed to match the experimental space and apparatus sizes of the in-orbit CSR, this ground-based platform takes full advantage of laboratory amenities, including gas supply, ventilation, power supply, and thermal control facilities. The platform comprises three subsystems: the experimental insert subsystem, the supporting facility subsystem, and the combustion diagnostic subsystem. The experimental insert subsystem supports a wide range of experiments with its gas, liquid, and solid combustion modules. The combustion diagnostic subsystem is equipped with high-precision measurement devices such as high-speed cameras, particle image velocimetry, and planar laser-induced fluorescence, enabling real-time measurements of flame morphology, flow velocity, and intermediate species distribution. Initial tests demonstrate that the platform can generate various types of gas flames, including premixed, diffusion, and partially premixed flames, by adjusting the fuel-to-oxidizer flow ratio. The liquid combustion module conducted suspension and ignition tests for single and multiple droplets, while the solid combustion module examined how planar, cylindrical, and linear materials combust under microgravity conditions. The system precision and reliability were validated by comparing diagnostic data with established data on flame oscillation. Additionally, the platform's modular design supports upgrades to both future software and hardware. Conclusions and Prospects: The ground-based research platform replicating on-orbit combustion environments plays a crucial role in supporting and complementing future combustion studies conducted aboard the space station. With its advanced experimental modules and diagnostic tools, the platform enables systematic and in-depth combustion experiments, advancing fundamental research in space combustion. Notably, the diagnostic system facilitates high-precision measurements across diverse combustion experiments. This ensures accurate analysis of flame structures, flow velocity fields, and chemical component distributions, providing critical ground-based validation and comparative data for addressing key scientific questions faced in space-based research. Overall, the platform is equipped with comprehensive facilities necessary for conducting combustion experiments. By supporting systematic experimentation, it helps optimize the design of space-based experiments, strengthening the cutting-edge and innovative aspects of space combustion studies. Furthermore, the extensive diagnostic resources and results from ground experiment results offer valuable data for in-orbit combustion experiments, driving the advancement of space combustion science theories and applied technologies.
Objective: Assessing thermoacoustic instability and flame-acoustic coupling is crucial for effectively designing and operating various combustion devices. Owing to the complex coupling among flame, acoustic waves, and the flow field, certain simplified configurations are typically adopted to capture fundamental dynamics. This study employs a propagating flame in semi-confined tubes to investigate these phenomena, focusing specifically on the role of hydrodynamic instability in flame oscillation and the development of thermoacoustic instability. Methods: The flame-acoustic interaction in a narrow, semi-open channel was investigated by numerically solving the compressible Navier-Stokes equations incorporating thermal conduction, mass diffusion, viscosity, and single-step irreversible chemical reaction kinetics. Modal decomposition techniques, specifically, the proper orthogonal decomposition (POD) and the spectral proper orthogonal decomposition (SPOD), were applied to the temperature field near the flame front zone to investigate the impact of hydrodynamic instability on flame oscillation during primary acoustic instability. Results: Numerical simulation results demonstrate that as the flame propagates from the open end toward the closed end of the channel, sustained flame oscillations occur owing to the development of primary acoustic instability. Cells or cusps appear on the flame front, initiating at the leading edge near one channel wall and moving along the flame front surface toward the other wall. This movement resembles the nonlinear process of Darrieus-Landau instability development. Statistical analysis indicates that the most probable wavelength of these cells corresponds closely to the most unstable wavelength predicted by linear theory for Darrieus-Landau instability. The periodic motion of these cells results in a sawtooth-like variation in the burning rate over time. POD analysis revealed that the wavelength of the coherent structure for the first three POD modes matches the most unstable wavelength of Darrieus-Landau instability, capturing flame front wrinkling and resembling the nonlinear process of Darieus-Landau instability development. Higher POD modes also describe similar physical phenomena but focus on smaller structural movements. The time evolution of the decomposition coefficients for different POD modes was also computed and compared. Additionally, a spectrogram of the pressure signal measured at the closed end of the channel was analyzed and compared with the channel eigenfrequency. It shows that during primary acoustic instability, the pressure signal predominantly aligns with the fundamental mode of the channel eigenfrequency, but a small manifestation of the first harmonic is also observed. Subsequently, SPOD was employed to gain a deeper understanding of the frequency-based flame dynamics. SPOD results indicate that the frequency associated with the wrinkle motion on the flame front aligns with the fundamental mode of the channel eigenfrequency. At the first harmonic, SPOD captures cell or cusp movement along the flame front surface, showing a smaller wavelength proportional to the frequency ratio between the harmonic and fundamental modes. Notably, SPOD results at harmonic frequencies exhibit similar structural patterns to those observed in higher POD modes. Finally, SPOD analysis of the vorticity and velocity vector fields identified weak vortices present in higher-frequency modes. These vortices can be captured in higher-frequency modes, which are associated with cusp motion on the flame front. Conclusions: The significance of hydrodynamic instability in flame-acoustic coupling for nonsymmetric flames in semi-open narrow channels is emphasized. Using modal decomposition methods, the study establishes a connection between hydrodynamic instability and flame oscillation frequencies. This connection provides insight into different flame oscillation behaviors at various acoustic modes and resents valuable information for controlling thermoacoustic instability.
Objective: As the thermal pressure relief device of an onboard hydrogen storage tank is activated by a fire, the released high-pressure hydrogen gas will be ignited to form a jet flame. The large scale and high temperature of the jet flame lead to a potential risk of thermal radiation injury to individuals nearby, including firefighters. This study was based on the bonfire test of a full-size vehicle-mounted hydrogen storage tank, combined with a theoretical system to predict the external thermal radiation of the jet flame and evaluate the danger distance to personnel. Methods: Based on the development process of vehicle fires, the test was structured to include localized and engulfing stages of a bonfire. A Type III hydrogen storage tank with the specifications of 48 L and 70 MPa was used in the test. A pressure sensor and camera were installed to record the internal pressure of the tank and the shape of the jet flame. Based on the real gas state equation and the thermodynamic and fluid characteristics of high-pressure hydrogen, a theoretical calculation framework, which contained the changes to the gas parameters, such as temperature, density, and flow velocity, was obtained. Then, the mass flow rate of hydrogen was estimated by the framework. The heat release rate of the jet flame can be estimated from the combustion heat and mass flow rate. To further determine the radiation flux distribution of the jet flame, this study used the classic single-point-source model. In addition, the prediction accuracy of the model for the external radiation of the jet flame in the near-field (within 3.0 m) has been verified. Finally, the release scenario of the 48 L-70 MPa hydrogen storage tank was cited as an example, and the evolution of the external radiation and thermal dose unit of the jet flame over time at different distances (1.5-5.0 m) was analyzed. The danger distance to personnel was also evaluated in compliance with the injury threshold of the human body under thermal radiation. Results: According to the bonfire test and theoretical calculation of the 48 L-70 MPa hydrogen storage tank: (1) The maximum internal pressure of the tank under full load hydrogen charging conditions in a fire was 77.4 MPa. The maximum flow rate was approximately 0.1 kg/s, with a maximum jet flame length of 4.93 m. (2) The maximum radiation flux of the jet flame at 1.5 m was approximately 13 kW/m2, which decayed to 10.2 kW/m2 after 20 s. (3) At a distance of 2.0-3.0 m from the flame, it would take approximately 5.0 s to cause first-degree burns to personnel, and at a distance of 5.0 m, it would take more than 20 s. (4) A “risk-free” distance of at least 10.0 m was required to ensure that personnel are exposed to thermal radiation of less than 1.6 kW/m2. Conclusions: The results of this study could further improve the theoretical system of jet flame external radiation risk assessment for the emergency release of onboard hydrogen storage tanks that could provide a reference for emergency response in related accident scenarios.
Objective: As the aircraft cargo compartment is a closed and complex environment, challenges occur in fire detection due to diverse cargo and limitations in sensor placement. Traditional fire detection methods are unable to accurately address fire recognition in such complex environments, particularly during the early stages of a fire. To address the challenge of accurately identifying the fire source localization in aircraft cargo compartments, this paper proposed a method based on a Bayesian-optimized bi-directional long short-term memory network (BO-BiLSTM). Method: This paper involved establishing an experimental platform of a real aircraft cargo compartment to simulate various fire scenarios. Fusion sensors were installed at multiple locations on the cargo compartment ceiling to collect multidimensional feature parameters in real time, including smoke volume fraction, CO volume fraction, and temperature. These data were used to build a fire feature database to capture the complex and dynamic changes in fire scenes. To improve the accuracy of detecting the fire source location, a bidirectional long short-term memory (BiLSTM) network was utilized, using its bidirectional information transmission mechanism to capture the forward and backward dependencies in time-series data. Meanwhile, the BiLSTM network structure was optimized using a Bayesian optimization algorithm to find the optimal combination of hyperparameters, enhancing the model's generalizability and robustness. Results: The experimental results indicate the following. First, the model was validated using a sliding window approach. When the number of windows was set to 8, the accuracy of the BO-BiLSTM model reached 97.2%. Compared with traditional models, such as recurrent neural networks (RNN), gated recurrent units (GRU), long short-term memory (LSTM) networks, and unoptimized BiLSTM models, the accuracy increased by 22%, 21%, and 2.6%, respectively. Second, in robustness tests with missing features, the BO-BiLSTM model maintained good stability. When only temperature and CO volume fraction were used as inputs, the model achieved an accuracy of 80.5%, which increased to 82.2% when using temperature and smoke volume fraction. Meanwhile, With when temperature and CO volume fraction were considered as inputs, the accuracy was 75.6%. The combination of temperature and smoke volume fraction performed the best, showing a strong correlation between these two features, with smoke volume fraction being more accurate for fire source localization. Finally, in the analysis of the model under sensor failure conditions, even with the number of functioning sensors reduced to four, the BO-BiLSTM model maintained an accuracy of 59.4%, significantly outperforming other models and demonstrating its advantages in complex and dynamic fire environments. The accuracy of the BiLSTM and LSTM models was lower than that of BO-BiLSTM, but their accuracy declined more gradually as the number of sensors increased, indicating some degree of resistance to interference. The GRU model performed better than the RNN model; however, when the number of damaged sensors was three or four, the accuracy of the GRU model was significantly lower than that of BO-BiLSTM, LSTM, and BiLSTM. The RNN model performed the worst in all scenarios, with its accuracy rapidly declining as the number of damaged sensors increased, dropping to approximately 45.2%. Conclusions: By significantly enhancing the accuracy and efficiency of fire source localization, this study provides essential technical support for the early detection, rapid response, and effective management of aircraft cargo compartment fires, which can help reduce fire risks and ensure safe and reliable air transportation.
Objective: Image- based smoke detection is a vital component of early fire warning systems. However, existing methods face considerable challenges in reliability when applied to environments with complex backgrounds, high noise levels, and low image contrast. In particular, during the early stages of a fire, smoke often appears small in size, low in density, blurred in shape, and irregular in morphology, which further complicates detection. To address these challenges, this study proposes a smoke detection method that integrates spatial perception and saliency modeling. The aim is to improve the robustness, adaptability, and accuracy of smoke detection systems, providing highly reliable and effective solutions for real-world fire surveillance across diverse environments. Methods: The proposed method consists of three key components: the multi-kernel parallel convolution module (MKPCM), dynamic histogram axial interaction module (DHAIM), and spatial decay residual block (SDRB). The MKPCM employs a parallel architecture with convolution kernels of varying sizes, enabling the network to capture features across multiple spatial scales simultaneously. This design allows for an effective representation of the variable dispersion scales of smoke. The embedded context anchor mechanism further refines this process by assigning differentiated spatial weights, enhancing the focus on relevant visual regions while suppressing background noise and irrelevant features. The DHAIM uses dynamic histogram-based segmentation to partition feature maps into high- and low-contrast areas, and then applies hybrid attention mechanisms tailored to each partition to improve semantic differentiation and precise extraction of subtle smoke cues in low-contrast zones. The SDRB introduces a spatial attention generation process based on Manhattan distance, where attention weights decay as spatial distance increases, to effectively reduce interference from remote pixels and improve feature consistency in regions with blurred boundaries. These components are jointly optimized in an end-to-end learning framework to enhance the model's sensitivity to complex spatial patterns and ambiguous edge transitions of smoke plumes. Results: To evaluate the effectiveness of the proposed method, a multi-scene smoke detection dataset is constructed, encompassing various indoor and outdoor scenarios with diverse background complexities. Experimental results show that the proposed method achieves an average precision of 94.0%, outperforming the baseline real-time detection transformer model by 5.5%. The method consistently delivers high detection accuracy across different environmental conditions and maintains strong robustness against low contrast, occlusion, and scale variation. Ablation studies confirm the individual and combined contributions of MKPCM, DHAIM, and SDRB to enhancing performance metrics such as precision, recall, and F1 score. In addition, the method demonstrates efficient inference and computational performance, making it highly suitable for real-time deployment in intelligent surveillance, early fire warning systems, and automated safety platforms. Conclusions: This study presents a robust and efficient smoke detection method that integrates multi-scale spatial perception and contrast-adaptive saliency modeling. The experimental findings validate the method's ability to address key challenges in early fire smoke detection, especially in visually complex environments. With its strong detection performance and practical adaptability, the proposed method holds significant potential for integration into real-world fire prevention infrastructures, thereby enhancing early warning capabilities and contributing to improved public safety outcomes and emergency responsiveness.
Objective: Stimulating the vitality of green and low-carbon science and technology innovations and realizing the goal of "dual-carbon" are inherent requirements for implementing a new development concept, building a new development pattern, and promoting high-quality development. Currently, the development of green innovation and transformation is at a critical stage, and the dynamics of green innovation and transformation is still an important research object worth exploring. China is participating in a vertical division of the labor system with greater depth, breadth, and efficiency through global value chains. However, most existing studies have simply attributed low-carbon innovation outcomes to a single policy or an element. Under the guidance of dual circulation and dual carbon goals, exploring key links between the value chain consumer demand and green innovation in economic activities is significant. When studying the dynamics of corporate green innovation, attention should be paid to the huge demand-side impacts and transmission of the value chain. Methods: Based on the division of labor and connection of product value chains, this study matches and integrates the green innovation database of the A-share listed companies from 2002 to 2020. Then, this paper constructs a time series of multiregional input-output tables based on the interprovincial multi-regional input-output tables compiled by the Development Research Center of The State Council in 2002, 2007, 2012 and 2017. To more clearly describe how demand-side shocks affect the behavioral decisions of enterprises under green and low-carbon transformation through input-output networks, this paper starts with analyzing the consumption-side demand of the value chain. Using a multiregional input-output model, this paper quantifies the direct and indirect impacts on the consumption-side demand caused by the interprovincial division of labor in each province and estimates the demand impact of the interprovincial trade in China. Further, it decomposes the consumption-side demand of the value chain into the local consumption-side demand and foreign consumption-side demand of the value chain. Results: The research findings are as follows: Consumer demand, transmitted through the value chain, contributes to improving local firms' green innovation levels, particularly in terms of original green invention patents. The value chain amplifies the impact of the market demand from other regions on the local market. Local firms participating in the dual circulation strategy and constructing a unified domestic market will benefit from the broader dynamics of the external market. State-owned enterprises bear greater social responsibility in low-carbon emissions, and the consumer demand in the value chain guides subsidies to enter the domain of invention patents more effectively. This effect exhibits specific spatial and temporal heterogeneity, stimulating high-tech sectors and enterprises to invest in green innovation, while large-scale enterprises and startups face limitations in green innovation. Conclusions: This paper connects the value chain and innovation chain under the connotation of the global value chain division of labor linkage. By quantifying the direct and indirect impacts on the consumption-side demand, the paper examines the impact of demand-side shocks on the level of green innovation on the supply side by adopting the multiregional input-output model to measure the effect of demand-side shocks on China's interprovincial trade. The research results provide policy implications for creating a unified domestic market under sustainable development, effectively integrating value and innovation chains, and fostering green innovation in enterprises.
Objective: Erbium is a heavy rare earth element widely used in strategic high-tech industries. As globalization progresses, the global value chain division of labor for erbium continues to change. However, the limited volume of erbium usage complicates data collection for research, leading to a scarcity of studies on the use of erbium resources. Methods: This study analyzes the material flow characteristics of erbium throughout its full lifecycle in the Chinese Mainland from 2011 to 2020. The L. Ridenour logistic growth model is used to forecast demand, and scenario analysis is conducted to predict the supply of secondary resources, aiming to assess the current state of erbium resource utilization and predict future trends. Results: The analysis of the current situation reveals that during the mining stage, the extraction of raw ore containing erbium in China has generally increased; however, the compound annual growth rate is only 2%. The production of erbium-containing rare earth concentrates experienced two notable declines, mainly owing to changes in international circumstances and policies. In the separation and refining stages, domestic supply does not meet the processing demands for erbium, leading to substantial imports of erbium oxide, which increased from 471.06 tons in 2011 to 522.45 tons in 2020. In the manufacturing and usage phases, erbium-doped fiber amplifiers remain the largest application, accounting for 37% of the demand in 2020. The second-largest application is medical laser crystals, which continue to observe growing demand. In addition, the proportion of erbium demand in glass materials rose from 8% in 2011 to 17% in 2020 owing to the increasing application of rare earth elements in the glass industry, whereas the applications of erbium in fluorescent and ceramic materials gradually decreased. The position of China in the global erbium value chain has been steadily improving, transitioning from a net importer to a net exporter of erbium-based end products. However, China remained a net importer in the end-product sector in 2020 owing to substantial imports of aluminum-erbium alloys. The cumulative amount of waste generated in the medical laser crystal sector reached 1 331.34 tons in 2020, making it the largest waste source, whereas the cumulative stock-in-use for erbium-doped fiber amplifiers reached 2 624.76 tons, the highest among all sectors. Forecasts indicate that by 2029, the demand for Chinese erbium-based products will reach 1 017.92 tons, with continued growth for erbium-doped fiber amplifiers and medical laser crystals projected to reach 407.16 and 271.65 tons, respectively, by 2029, solidifying their roles as primary applications. If recycling rates increase by 1.5% annually, the supply of secondary resources will exceed 5% of the annual demand by 2029. Conclusions: The erbium industry should actively work to reduce its dependence on overseas supply chains, diversify its sources of erbium supply, and improve the transparency and traceability of the erbium value chain. In addition, this industry should direct the flow of erbium toward high-end applications, such as erbium-doped fiber amplifiers and medical laser crystals, through appropriate policy adjustments. These actions will further strengthen and enhance the position of China in the global erbium value chain.
Objective: As the central role of innovation elements in economic and social development continues to rise, based on the practical need of the country to "strengthen the evaluation of the relationship between patent activities and economic benefits", breaking through the barriers between patent databases and other economic databases is a substantial development in the economic and national analysis research fields. The establishment of connections and data networks across various databases, such as product and patent databases in different fields and systems, is required to explore the correlation, internal mechanisms, and heterogeneity of innovative applications and transformations. Existing studies have mainly focused on the mapping between patents and industries. However, the complexity of product classification has led to the absence of direct mapping between International Patent Classification (IPC) and Harmonized System (HS) code, which has limited the analysis of technology transfer and industry-technology adaptation mechanisms. This paper aims to construct a cross-database technology-product category mapping method, reveal the technological characteristics of segmented industries, and provide data support for industrial innovation research. Methods: This paper utilizes the classification information of patent and product databases to explore the full-category mapping relationship between patent IPC classification and product HS classification in the Chinese language environment. Based on the comprehensive method of natural language processing (NLP), cross-searching, and algorithmic links with probabilities (ALP), this paper employs the examples of products corresponding to the HS codes from the data released by the General Administration of Customs of China as external word sources to expand the HS category keywords, thereby obtaining a keyword list with higher quality than that generated by NLP segmentation. Furthermore, three weighting correction methods(raw weight, specificity weight, and hybrid weight) are employed based on the Bayesian theorem to establish mapping links between HS (six-digit) and IPC (three-digit); these are combined with multilevel classification to refine the analysis of technological differences and associations. Results: The mapping results reveal that complex products are associated with a wide variety of technologies, whereas simple industrial and agricultural products are associated with fewer technology types. The results reflect the heterogeneity of technological innovation across different industries and products. The calculation results of specificity and mixed weights are more likely to reveal unique technology types related to the production of certain product categories compared with the original weight, which is of great importance for further identifying specialized, sophisticated, and novel patents. The development of strategic emerging industries is closely related to the technological support of sections G (Physics) and H (Electricity), objectively indicating the importance of basic research in the development of strategic emerging industries. Conclusions: The IPC-HS link method constructed using cross-searching and ALP can effectively quantify the strength of technology-product associations, break through the barriers of the classification systems between technology and products from the perspective of innovation achievement transformation, and provide data-driven empirical support for the transformation of technological achievements. This mapping relationship can reveal the technological characteristics and differences of segmented industries; it can contribute to the understanding of technology diffusion in the innovation ecosystem, the application of technology in strategic emerging industries, and the adaptation mechanism between technology and industry.
Objective: Route blocking is a newly emerged category of routing threats following the Russia-Ukraine conflict. Unlike conventional routing threats, route blocking constitutes targeted, large-scale network-layer obstruction between regions or even nations, typically affecting extensive areas at regional or national levels. The damage caused by route blocking extends beyond economic losses, potentially triggering political or social instability, thus necessitating effective security measures. However, the vast scope of route blocking poses a challenge to comprehensively analyzing its impact on the real-world Internet. Furthermore, conducting route blocking experiments on the real-world Internet incurs prohibitively high costs and raises a series of ethical concerns. Consequently, conducting an in-depth security analysis becomes difficult, and validating the effectiveness of designed defense measures becomes even more challenging. Methods: To design and validate effective countermeasures against route blocking, this paper proposes SimBlock, a fine-grained global Internet route blocking simulation and detection system that analyzes the characteristic patterns of route blocking through simulation to provide a foundation for security measure design and validation. SimBlock comprises four modules: (1) global Internet topology construction using open third-party topology data, (2) global Internet routing simulation on the established topology, (3) route blocking simulation implementing various blocking techniques based on the designed routing algorithm, and (4) route blocking detection and identification through characteristic analysis. SimBlock is capable of simulating both autonomous system (AS)-level and router-level paths between arbitrary IP addresses. The system additionally supports the simulation of point-to-point packet transmission, ping probing, and traceroute probing, while enabling the granular simulation of dynamic network conditions, including congestion and failures. SimBlock demonstrates comprehensive simulation capabilities of at least five fundamental route blocking methods: DDoS attacks, IP blocking, physical blocking, route hijacking, and business relationship termination. Building upon this foundation, the system has successfully validated a route blocking detection and identification algorithm based on distributed prober triplets, delivering robust security measures to mitigate potential route blocking threats. Results: This paper conducts extensive experiments to validate the effectiveness of SimBlock. The experimental results demonstrate that: (1) even with 47.65% of the routers lacking valid IP addresses in the dataset, SimBlock can simulate valid AS/router-level paths with a 98.94% success rate, proving that the constructed global Internet topology maintains high coverage and excellent connectivity; (2) without prior knowledge of specific AS-routing policies, the AS-level paths simulated by SimBlock achieve an average 61.00% similarity with real-world traceroute paths (reaching over 80% similarity in 16.21% of cases), while the router-level paths maintain an average 42.44% similarity (exceeding 70% in 7.40% of cases) despite 47.65% missing router IPs, confirming that the simulation algorithm accurately captures overall routing trends; (3) for Internet topologies containing over 70, 000 AS nodes, SimBlock maintains millisecond-level latency in AS-level path simulation and handles router-level simulation for approximately 60 million nodes with second-level latency, demonstrating exceptional efficiency in large-scale Internet data processing; (4) across various victim-attacker country combinations, the detection algorithm of SimBlock reliably distinguishes route blocking from normal network failures, and its identification algorithm effectively differentiates between various blocking techniques, validating the effectiveness and universality of the system. Conclusions: In summary, SimBlock provides an effective solution for in-depth analysis of route blocking, while also offering an effective security measure to counter the potential threats posed by route blocking.
Objective: With the rapid development of network technology, cyberattacks have become increasingly severe, threatening the stability of cyberspace. Network security situation assessment (NSSA) has become a critical technology for building proactive defense systems by integrating multisource data to deliver comprehensive and dynamic evaluations of network states. Traditional rule-based methods and early learning-based models often lack sufficient granularity in feature extraction, struggling to capture long-range temporal dependencies, thereby limiting their effectiveness in detecting complex and diverse attack patterns. To address these limitations, this study proposes a novel evaluation framework that integrates a parallel feature extraction network (PFEN) and a multiscale temporal convolutional network (MsTCN) to enhance fine-grained feature extraction and long-term dependency modeling for network traffic data. Methods: The proposed PFEN-MsTCN model introduces two major technical contributions. First, the PFEN is tailored for sequential traffic data by modifying the conventional Inception module, replacing two-dimensional convolutions with one-dimensional convolutions to extract temporal features along the sequence axis. The multibranch structure is optimized into cascaded subnetworks to capture local and contextual temporal features. The integration of convolution, batch normalization, and ReLU activation enhances nonlinearity and robustness, effectively reducing computational complexity while maintaining feature quality. Second, MsTCN is improved by introducing a multikernel branching structure and a hierarchical dilation rate to dynamically capture multiscale temporal features. A dynamic parameter matching mechanism and Chomp1D layer ensure multibranch output alignment, preventing dimensional mismatches during feature fusion. This design enables the simultaneous detection of short-term bursts and long-range dependencies. Finally, the strengths of PFEN in local feature extraction and MsTCN in sequence modeling are seamlessly integrated, creating a robust hybrid model. Results: Comprehensive experiments on the NSL-KDD and CIC-IDS2017 benchmark datasets involved preprocessing with normalization, one-hot encoding of categorical features, and removal of redundant or invalid features to ensure high-quality input. Experimental results demonstrate that the PFEN-MsTCN model consistently outperforms the baseline models, including PFEN-ABiGRU, SEAE-CNN-BiGRU-AM, CNN-TCN, and Inception1D-MsTCN. On the NSL-KDD dataset, the proposed model achieved an F1-score of 87.39%, surpassing competing methods by 2.54%-4.88%, while maintaining lower mean squared error and mean absolute error values. On the CIC-IDS2017 dataset, the proposed model achieved an outstanding F1-score of 99.87% with reduced prediction error, demonstrating superior adaptability to heterogeneous and imbalanced data. The visualization of situation values further verified that PFEN-MsTCN aligns more closely with the ground truth than competing models. Furthermore, the proposed evaluation index system, incorporating attack impact, probability, and frequency factors, enabled accurate quantification of security situation values and precise risk level classification. Conclusions: The PFEN-MsTCN fusion model effectively addresses the challenges of existing NSSA methods by enhancing feature extraction granularity and improving the capture of long-term temporal dependencies. By integrating multibranch one-dimensional convolutional feature extraction with hierarchical multiscale temporal convolution, the model achieves precise recognition of abnormal traffic behaviors and robust temporal dependency modeling. The experimental results validate the superior performance of the proposed model in terms of accuracy, robustness, and generalization across datasets, establishing its potential as a reliable tool for intelligent network security assessment. Future research will focus on improving the recognition accuracy for small-sample attack types in imbalanced datasets and extending the framework to real-time and large-scale deployment scenarios, further enhancing its applicability in practical cyberspace defense systems.
Objective: Existing methods for continuous relation extraction (CRE) face remarkable challenges in complex Chinese semantic environments, particularly the problem of catastrophic forgetting when learning new tasks while retaining knowledge from old tasks. Traditional approaches often retrain models on a combination of historical and new data, leading to inefficiencies and resource constraints as data volumes increase. To address these limitations, this study aims to develop a robust CRE model that can efficiently learn new knowledge while preserving historical relationships, even in scenarios with complex sentence structures, overlapping entities, and imbalanced data distributions. The proposed model integrates prototype representations, memory replay strategies, and contrastive learning to enhance feature discrimination and stability in embedding spaces, thereby improving the classification performance across single-domain and cross-domain datasets. Methods: The proposed prototype-based continuous complex relation extraction network (PBCRE-Net) model consists of two primary stages: initial training and memory replay, designed to mitigate catastrophic forgetting and improve adaptability in dynamic learning environments. The initial training includes: 1) Entity-aware feature extraction: Input texts are processed using a pretrained BERT model to generate contextual embeddings with special tokens ([E11], [E12], [E21], and [E22]) that mark entity boundaries. 2) Supervised contrastive learning: A dual-head classifier (classification and contrast) is employed to minimize intraclass distances and maximize interclass distances in the embedding space. This objective is achieved through a combination of cross-entropy loss and contrastive loss. 3) Prototype generation: For each relation category, representative samples are selected via K-means clustering and their prototypes are computed as weighted averages of cluster centroids to capture category-specific features. The memory replay includes: 1) Memory sample selection: Memory modules store exemplars from previous tasks using K-means clustering. Weights are assigned based on cluster distribution to balance sampling during replay. 2) Memory augmentation: To prevent overfitting, synthetic samples are generated by swapping entity pairs or appending unrelated sentences to existing exemplars, thereby expanding the memory pool. 3) Consistency Loss: During replay, an embedding consistency constraint is applied to maintain stability in the embedding space across tasks. 4) Joint optimization: The model is trained on a mixture of new task data, historical memory samples, and augmented samples, combining cross-entropy loss and consistency loss. Results: Experimental evaluations on CMeIE (medical domain) and ASaRED (alloy domain) datasets demonstrate the superiority of PBCRE-Net in complex CRE scenarios. Single-domain Performance: On the CMeIE dataset, PBCRE-Net achieved an average accuracy (ACC) of 84.27% and a macro F1-score (Macro-F1) of 81.93% across 10 incremental tasks. Notably, the proposed model outperformed baseline models, such as EMAR and CRL, by 3%-5% in subsequent tasks (T8—T10), where catastrophic forgetting is highly severe. The model effectively handled triplet overlap (e.g., entity-pair and single-entity overlaps) and class imbalance, an objective accomplished through prototype-based contrastive learning and memory augmentation. Cross-domain adaptability: In cross-domain experiments combining CMeIE and ASaRED, PBCRE-Net maintained an ACC of 74.16% and Macro F1 of 69.34% across 10 tasks, considerably surpassing competing models (e.g., CRECL and DPCRE). The memory replay mechanism and consistency loss ensured stable embedding spaces despite domain shifts, thus reducing catastrophic forgetting in critical relation categories like material composition and alloy property. Robustness to memory constraints: Reducing the memory size from 20 to 5 samples per task decreased the performance of the proposed model by 15%, yet PBCRE-Net outperformed the alternatives under constrained memory conditions. This highlights its efficiency in real-world scenarios with limited storage. Conclusions: This study introduces PBCRE-Net, a novel CRE framework that mitigates catastrophic forgetting through prototype representations, memory replay, and contrastive learning. Key contributions include a supervised contrastive learning strategy to enhance feature discriminability in complex semantics and a memory augmentation mechanism to mitigate overfitting and stabilize embedding spaces. Superior performance in single-domain and cross-domain CRE tasks was validated by extensive experiments. Future work will extend PBCRE-Net to multilingual settings via cross-lingual transfer learning and incorporate physical constraints to improve relation extraction accuracy in scientific domains. In addition, addressing polysemy through semantic alignment techniques will further enhance the applicability of PBCRE-Net.
Objective: Deep learning technologies have achieved remarkable progress in the field of personalized recommendation services. However, recommendation systems based on deep neural networks still face the challenge of data sparsity, which limits the ability of a model to accurately capture subtle differences in user preferences, thereby affecting the robustness of model training. This problem is specifically prominent in scenarios with limited user interaction data. Therefore, this paper aims to propose a recommendation system model that can effectively address the data sparsity issue to enhance the capability of a model in user behavior modeling and overall performance. Methods: To tackle the data sparsity issue, this paper proposes a residual network-based stacked vector-quantized autoencoder (RSVQ-AE). This model fully utilizes the advantages of residual connections by directly passing the continuous latent vector output from the multiple layers of encoders to the corresponding layers of the decoder. This effectively reduces the loss of high-value continuous information that is common in encoders, which is crucial for maintaining the fidelity of data representation. Meanwhile, by introducing vector quantization technology, we discretize the latent space to ensure that the model can accurately capture and represent the data. In addition, this paper constructs multiple stacked codebooks using vector quantization technology, enabling the model to learn multidimensional discrete vector quantization feature representations and capture the discretized interest representations of user behavior across multiple dimensions through stacked codebooks. To further enhance the stability and generative capabilities of the model, an adversarial network is introduced as a regularizer during the training process to promote rapid convergence. Results: To verify the effectiveness of the model, experiments were conducted on several public datasets widely used in recommendation systems. The experimental results revealed that the RSVQ-AE model exhibits excellent reconstruction performance across multiple datasets. Based on the ML-1M (MovieLens-1M) dataset, when the sequence length is 20, the reconstruction loss of RSVQ-AE is only 0.1525, with an accuracy rate of as high as 70.69%; when the sequence length increases to 100, the reconstruction loss further decreases to 0.0039, with an accuracy rate of 50.58%. Based on the Retail Rocket dataset, when the sequence length is 20, the reconstruction loss is as low as 2.42×10-4, with an accuracy rate of 81.26%; when the sequence length is 100, the reconstruction loss is 0.0019, with an accuracy rate of 74.21%. These results fully demonstrate that RSVQ-AE can maintain low reconstruction loss and high accuracy when processing sequences with different lengths. Its performance is only second to the autoencoder model, which cannot perform sampling generation. Conclusions: The proposed RSVQ-AE offers a powerful solution for the generation of discrete sequence data in recommendation systems. By addressing the limitations of existing generative models and introducing innovative technologies such as stacked codebooks, this model has achieved remarkable improvements in reconstruction accuracy and data generation quality. This method not only enhances the capability of the model in user behavior modeling but also provides new ideas and approaches for the development of personalized recommendation services, holding the potential to drive the future development of more efficient and user-behavior-centered recommendation systems. In addition, the flexibility and robustness of model data generation make it applicable to a variety of recommendation system model architectures.
Objective: With advancements in computer science and software engineering technologies, including artificial intelligence, domain-specific languages, and knowledge graphs, automated compliance checking tools are emerging. In the construction field, for example, research and practical applications in automated compliance checking for building information modeling (BIM) and architectural 2D drawings are thriving. A core step in conducting automated compliance checking with computers is the logical representation of articles written in natural language, transforming them into a format that computers can process. While extensive research has been conducted in this field and many widely used standards and paradigms have been developed, the handling of high-level inter-rule relationships remains largely unexplored. This study aims to propose a computer-representable method for high-level inter-rule relationships, enabling automated compliance checking systems to process such relationships automatically. Methods: To achieve the computer-representable formalization of these relationships, based on the summary and organization of actual building codes, this study introduces a paradigm within the context of rule-based automated compliance checking in the construction field. The paradigm includes three patterns: "compliance cascade", "non-compliance cascade", and "non-applicability cascade". The compliance cascade pattern represents "if rule A is compliant, then rule B will not be checked". Similarly, the non-compliance cascade pattern represents "if rule A is non-compliant, then rule B will not be checked", and the non-applicability cascade pattern indicates that "if rule A is not applicable, then rule B will not be checked". In each cascade pattern, corresponding indicators are used to designate the target rules. Among the rules in rule-based automated compliance checking systems, those expressed in the form of "if...then..." indicate that "rule A is not applicable" when the semantic model does not satisfy the "if" condition of rule A. This study also designs a corresponding execution engine for the paradigm using BIMChecker and structured natural language (SNL), enabling computers to process high-level inter-rule relationships automatically. The engine design adheres to the open-closed principle and single responsibility principle—two fundamental design principles—to ensure optimal extensibility. Results: To verify the paradigm's effectiveness and applicability, an experiment on representing high-level inter-rule relationships in actual codes using the proposed approach has been conducted. In addition, application experiments were carried out on 15 representative real-world projects via the Tsinghua "Tuzhi" platform. The results indicate that the paradigm exhibits strong validity and applicability across all 15 projects. Furthermore, the "Tuzhi" system is used to showcase the execution effects of high-level inter-rule relationships through a practical case study. Conclusions: This study proposes a representation method for high-level inter-rule relationships within rule-based automated compliance checking systems. Using three distinct patterns—"compliance cascade", "non-compliance cascade", and "non-applicability cascade"—it establishes a computer- interpretable framework for representing these relationships. By integrating an appropriate execution engine, this approach enhances rule-based automated compliance checking systems to generate check reports that closely align with human expert reasoning. Furthermore, the experiments in this study validate the effectiveness of the proposed patterns.
Objective: With the accelerating pace of technological innovation and the growing complexity of policy environments, local governments urgently require intelligent and interpretable decision support systems to guide the strategic layout of future-oriented industries. However, traditional industrial layout approaches rely heavily on expert judgment and static data analysis, making them ineffective at handling dynamic market demands, integrating heterogeneous features, and capturing the complex nonlinear interactions between regional resources and emerging industries. This study aims to address these critical limitations by developing an adaptive deep recommendation algorithm. The proposed algorithm provides data-driven, actionable insights to local policymakers, facilitating the accurate and strategic allocation of regional resources for emerging and strategic industries. Methods: This paper proposes a deep recommendation framework that integrates three key components: feature recalibration, multihead attention mechanisms, and neural matrix factorization. First, diverse regional attributes (such as gross domestic product (GDP), research and development (R&D) expenditure, and infrastructure indicators), industrial characteristics (such as growth rate and technological maturity), and policy orientations (extracted from over 280 local policy documents via semantic embedding) are transformed into dense vector representations through embedding layers. Next, a feature recalibration module inspired by the squeeze-and-excitation network is employed to dynamically reweight critical regional and industrial features, thereby underscoring influential factors and suppressing noise. Subsequently, multihead attention mechanisms are introduced to capture high-order nonlinear interactions across recalibrated features and to model complex interdependencies between regions, industries, and policy orientations effectively. Finally, neural matrix factorization techniques combine collaborative signals with the nonlinear embeddings to score and rank the suitability of each region-industry pairing quantitatively. The proposed model is trained on a comprehensive dataset comprising over 3 000 real-world region-industry samples with a margin-based loss function optimized through supervised learning. Extensive tuning of hyperparameters, including embedding dimensions, dropout rates, and attention head counts, ensures robust model performance. Results: Empirical validation demonstrates that the proposed adaptive deep recommendation algorithm substantially outperforms mainstream baseline models such as DeepFM, AutoInt, and FiBiNet across multiple performance metrics, including logarithmic loss (logloss) and area under the curve (AUC). Specifically, this algorithm achieves the highest AUC score of 0.714 6 and the lowest logloss of 0.485 8 among all tested models, which confirms its superior predictive accuracy and classification capability. Ablation experiments further reveal that each module contributes distinctively to overall performance: Removing the feature recalibration module, multihead attention mechanism, or neural matrix factorization component results in a noticeable degradation of predictive accuracy. Additionally, comparative analysis with a prominent large language model (LLM), such as GPT-4o, highlights the advantage of this structured algorithm in handling numerical and structured data, in contrast to semantic reasoning that limit structural modeling capacities of the LLM. Visualization of attention weights confirms the algorithm's interpretability and explicitly demonstrates its sensitivity to key factors such as regional R&D intensity, infrastructure readiness, and industry technological maturity. Conclusions: This study successfully establishes an adaptive, interpretable, and highly effective deep recommendation algorithm tailored explicitly to future-oriented industrial layout planning. By integrating dynamic feature recalibration, high-order feature interaction modeling, and robust collaborative filtering mechanisms, the proposed algorithm remarkably enhances the accuracy, interpretability, and practical applicability of the recommendations. The proposed algorithm not only provides local policymakers with transparent, data-driven decision support but also sets a theoretical foundation for further exploration into advanced recommendation frameworks. Future research will aim to incorporate temporal dynamics through real-time data streams; expand the framework to multiobjective scenarios covering economic, social, and ecological benefits; integrate structured recommendation outputs with semantic insights from expert knowledge; and ultimately realize a comprehensive, adaptive industrial policy decision support platform.
Objective: The internal flow channel design and oil immersion depth of gearboxes play a crucial role in determining the lubrication effectiveness of gears and the temperature rise within the gearbox. These effects intensify as train speeds increase. This study focuses on a specific high-speed rail aluminum alloy gearbox, using Simcenter STAR-CCM+ (hereinafter referred to as Star CCM+, a multi-physics simulation software) simulation software to develop a thermal-fluid-solid coupling simulation and analysis model. By integrating the simulation results with bench test data, this study aims to investigate the internal flow field and temperature field of the gearbox. The effects of various factors, including rotational speed, oil immersion depth, and steering direction, on the flow and temperature distributions within the gearbox are examined, providing insights into optimal operating conditions and potential design improvements. Methods: To analyze the performance of the gearbox, parameterized simulation analyses were performed considering different rotational speeds, oil immersion depths, and steering directions. The distribution of the internal flow and temperature fields under these varying conditions was studied. The analysis also focused on the mass flow rate and temperature field of each flow channel. This comprehensive approach allowed for a detailed evaluation of the lubrication performance of the gearbox. The Star CCM+ simulation model was calibrated using experimental data from a 1∶1 test bench, where temperature measurements were taken at various points within the gearbox. These measurements were compared with the simulation results to ensure the accuracy and reliability of the simulation model. The study also incorporated detailed thermal conditions, including gear frictional power losses, bearing power losses, and forced convection heat transfer, to represent the true working conditions of the gearbox under different operational scenarios. Results: The simulation results showed that the lubrication and temperature control effects of the gearbox were most effective when the internal oil immersion depth was between 1.75 and 2.00 times the tooth height. It was found that insufficient lubrication occurred on the upper and right sides of the gearbox, highlighting areas that require design improvements. Additionally, the research revealed that increasing the oil immersion depth improves the flow and distribution of lubrication oil within the gearbox. However, a deeper oil immersion beyond the optimal range increases churning losses and heat generation. By adjusting the flow channel configuration and improving the number and distribution of the internal flow paths, the optimized design reduced the temperatures in critical areas, including the bearing and meshing zones, by approximately 5℃. This improvement was achieved by increasing the oil flow to the gears and bearings while enhancing the cooling effect on the gearbox walls. Conclusions: This study demonstrates that a proper oil immersion depth is critical for maintaining effective lubrication and temperature control in high-speed rail aluminum alloy gearboxes. The results highlight that there is an optimal oil immersion depth range (1.75-2.00 times the tooth height) that ensures sufficient lubrication and effective cooling. Furthermore, the study reveals that there are areas within the gearbox, particularly on the upper and right sides, where lubrication is insufficient, suggesting that the design of the flow channels in these regions can be improved. The proposed modifications, such as the addition of more flow channels and optimizing their distribution, provide a substantial enhancement in the lubrication and cooling efficiency of the gearbox. These modifications result in a notable temperature reduction of approximately 5℃ in key areas, thereby demonstrating the effectiveness of the flow channel optimization strategy. This research provides insights into future gearbox design, particularly in optimizing lubrication systems and minimizing temperature rise to ensure the reliable operation of the system at high speeds.
Objective: The magnetic expansion control (MAGEC) system, developed by NuVasive, has demonstrated positive surgical outcomes in treating early-onset scoliosis. However, clinical studies have indicated that the MAGEC system is challenged by insufficient tensile force, which can lead to failed elongation surgeries. The maximum torque of the rotor inside the growth rod significantly affects the tensile force of the growth rod. The larger the maximum torque of the rotor, the greater the tensile force of the growth rod. Because of the different body weights of patients, the relative positioning of the rotor and driver in the clinical application of the MAGEC system is also different. Therefore, the distribution space in which the rotor can rotate continuously has a significant impact on clinical treatment. To solve the problem of insufficient tensile force, it is necessary to optimize the maximum torque of the rotor and to maximize the distribution space for continuous rotation of the rotor. Methods: This study employs ANSYS Maxwell to establish a finite element simulation model for a circular distributed drive permanent magnet rotor system. The analysis investigates the effects of the angle between the drive permanent magnet and the rotor, pole pairs, rotor outer diameter, and drive permanent magnet speed on the maximum torque of the rotor, and determines the continuous rotation domain of the optimized rotor. To verify the simulation results, a rotor torque measurement device was built and connected to a coupling through a motor. Furthermore, a growth rod extension experiment was conducted to test the magnetron-drive performance. Finally, a growth rod tensile-force testing platform was constructed to test the maximum tensile force of the growth rod. Results: The experimental results demonstrate that optimal performance was achieved when the angle between the driving permanent magnet and rotor was set to 120°, a single pole pairs configuration and an 8-mm outer diameter of the rotor. This combination produces the maximum torque of the rotor and the most extensive continuous rotation distribution space. However, the speed of the driving permanent magnet did not affect the maximum rotor torque. The maximum torque of the rotor before optimization was 32.847 N·mm; The maximum torque of the optimized rotor was 98.970 N·mm, which was 201% higher than that before optimization, and the distribution space for continuous rotation of the rotor was larger. The experiment demonstrated that the maximum torque values of the rotor before and after optimization were 30 and 90 N·mm, respectively, with relative errors of 8.7% and 9.06%, and that the permanent magnets could rotate stably. The maximum tensile force of the growth rod could reach 413 N, nearly twice the maximum tensile force of 208 N of the MAGEC system growth rod. Conclusions: When the angle between the driving permanent magnet and the rotor was set to 120°, the θ values that caused the maximum torque of the rotor to change from large to small were 120°, 140°, 100°, 160°, 80°, 180°, 60°, and 40°. The larger the number of pole pairs, the more rapidly the maximum rotor torque decreases. In addition, the larger the outer diameter of the rotor, the greater the maximum torque of the rotor. Finally, the rotational speed did not affect the maximum rotor torque.
Objective: The accurate prediction of surface roughness is a major technical challenge in aircraft grinding processes. This difficulty primarily arises from the complex nonlinear influences of factors such as material properties, process parameter coupling, and dynamic disturbances. These factors have long made the modeling and prediction of surface roughness in grinding processes extremely challenging. Conventional physical models often oversimplify the grinding process and fail to capture its inherent complexity, while data-driven models are susceptible to noise in measurement data, thereby limiting their generalization capabilities. To overcome the difficulties in achieving both high accuracy and robustness, this study proposes a novel surface roughness prediction model based on a physics-informed neural network (PINN). Methods: This study utilizes a robotic grinding experimental platform to collect data encompassing various process parameters and the resulting post-grinding surface roughness. Initially, logarithmic transformation and normalization are applied to process parameters so as to enhance the model's training efficiency and numerical stability. Subsequently, a PINN with a multilayer perceptron (MLP) architecture is constructed, incorporating a physical power-law mechanism (PLM) governing surface roughness into a loss function. This design enables the synergistic optimization of both the physical constraints and data-driven learning, with the loss function comprising a data fitting error and a physical model error. A dynamic weighting strategy is further introduced into the loss function: during the early stages of training, considerable emphasis is placed on physical consistency by assigning higher weights to the physics-based term, thereby mitigating the effect of data noise. As training progresses, the weight of the data-driven component is gradually increased, allowing for the model to capture the complex nonlinear relationships among process parameters and improve prediction accuracy. Finally, the performances of the PINN, MLP, and PLM are systematically compared in training efficiency, prediction accuracy, and generalization performance. Results: Experimental results indicate that the proposed PINN model offers clear advantages over baseline models in training efficiency, prediction accuracy, and generalization capability. Specifically, 1) the convergence speed of the PINN model is comparable to that of the PLM and is approximately 40% faster than that of the MLP. Moreover, the training process is more stable, and the final loss after convergence is the lowest among those of all models; 2) The absolute prediction error of the PINN model does not exceed 0.03 μm, with an average relative error of only 3.263%. This not only satisfies the stringent process-tolerance requirements of aviation maintenance but also results in overall prediction accuracy superior to those of both the PLM and MLP; 3) In comparison with the MLP, the PINN model exhibits considerably less performance degradation on the test set, significantly mitigates overfitting, achieves a more uniform error distribution, and exhibits the least systematic bias, thereby showing superior robustness and generalization ability. Conclusions: By incorporating a physics-guided framework and a dynamic weight-adjustment strategy, the proposed PINN model achieves not only improved training efficiency and prediction accuracy but also enhanced generalization ability and robustness. In practical engineering applications, the PINN model achieves robust, high-precision surface-roughness prediction. Compared with traditional physical models and purely data-driven models, the PINN model delivers superior performance and fully meets the stringent requirements of the aviation industry with respect to intelligent manufacturing and surface-quality control.
Significance: Radon(222Rn) is a naturally occurring radioactive noble gas. 222Rn is produced from the alpha decay of radium (226Ra), with a relatively long half-life (i.e., 3.8 days). Thus, 222Rn can diffuse and migrate from the rock and soil where it is generated and can enter the air. According to surveys conducted by the World Health Organization, exposure to 222Rn and its short-lived progeny is the second leading cause of lung cancer. The measurement and evaluation of 222Rn release involves not only occupational exposure but also public exposure. Therefore, the measurement of 222Rn exhalation and the dose assessments of 222Rn exposure have always been important concerns in radiation protection. Progress: The entire process of 222Rn moving from the soil to the atmosphere can be divided into three steps: emanation, migration, and exhalation. During the emanation process of 222Rn, 222Rn will obtain a recoil energy of 8.6 × 104 eV from the decay of 226Ra, which will make 222Rn atoms travel through the soil grains at a distance of no more than 50 nm. If produced near the surface of soil grains, 222Rn will leave the grain and stop in the interstitial space (pore), becoming freely movable 222Rn. The fraction of freely movable 222Rn is usually expressed by the emanation coefficient (dimensionless). The emanation coefficient of soil generally ranges from 0.1 to 0.3. In soil with a relatively stable internal environment, the migration of free 222Rn in soil mainly relies on molecular diffusion caused by concentration gradients, eventually entering the atmosphere through the soil-air interface. If only diffusion transport is considered, then Fick's law can be used to describe the migration process of 222Rn and establish a model for 222Rn flux at the soil-air interface. The 222Rn exhalation rate at the soil surface is the 222Rn flux. On-site measurement methods of the soil 222Rn exhalation rate can generally be divided into three categories: accumulation, flow-through, and activated carbon adsorption methods. In actual measurements, different methods can be chosen according to the needs. Conclusions and Prospects: The understanding of the 222Rn diffusion exhalation mechanism and influencing factors is becoming comprehensive, and the measurement methods of the 222Rn exhalation rate for different purposes have been developed. Analyzing the physical processes of 222Rn exhalation from soil and measuring and evaluating the exhalation rate of 222Rn are important for assessing environmental radiation and managing uranium tailings and associated radioactive minerals. Moreover, the exhalation rate of 222Rn is closely related to the radiation environmental safety of on-site supervision of naturally occurring radioactive materials. Because of the complexity and diversity of 222Rn measurement sites, even though the amount of 222Rn release can be measured and calculated relatively accurately, nearly no quantitative deterministic correlation is detected between it and the indoor 222Rn concentration. Thus, the 222Rn exposure dose for key populations is difficult to estimate. The 222Rn exposure dose and health risks for key populations can only be estimated and controlled through measurements of the indoor 222Rn levels. Therefore, although the 222Rn exhalation rate is an important parameter that can be measured and calculated on-site, establishing a fixed exhalation rate limit for regulatory purposes is unsuitable.
