To effectively capture the inherent near-field effects and spatial non-stationarity across extremely large antenna arrays (ELAAs), this letter develops a novel analytical channel model tailored for extremely large-scale multiple-input multiple-output (XL-MIMO) systems. In the proposed framework, spatial non-stationarity is first characterized using a 0-1 diagonal matrix, after which the composite XL-MIMO channel matrix is formulated as a linear combination of structured matrix components. Leveraging this representation, we perform a comprehensive performance analysis, evaluating key metrics including the downlink ergodic capacity, an efficiently computable upper bound derived from eigenvalue matrix, and the symbol error probability (SEP). The analysis demonstrates that the proposed scheme not only achieves performance comparable to benchmark methods but also substantially reduces computational complexity. Furthermore, the analysis reveals a pronounced performance degradation in the presence of increasing channel estimation errors.
{"title":"Performance Analysis for Extremely Large-Scale MIMO Communication Systems","authors":"Yingchen Le;Zhuxian Lian;Yajun Wang;Lin Ling;Chuanjin Zu;Bibo Zhang;Xiaopei Hua","doi":"10.1109/LCOMM.2026.3654136","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3654136","url":null,"abstract":"To effectively capture the inherent near-field effects and spatial non-stationarity across extremely large antenna arrays (ELAAs), this letter develops a novel analytical channel model tailored for extremely large-scale multiple-input multiple-output (XL-MIMO) systems. In the proposed framework, spatial non-stationarity is first characterized using a 0-1 diagonal matrix, after which the composite XL-MIMO channel matrix is formulated as a linear combination of structured matrix components. Leveraging this representation, we perform a comprehensive performance analysis, evaluating key metrics including the downlink ergodic capacity, an efficiently computable upper bound derived from eigenvalue matrix, and the symbol error probability (SEP). The analysis demonstrates that the proposed scheme not only achieves performance comparable to benchmark methods but also substantially reduces computational complexity. Furthermore, the analysis reveals a pronounced performance degradation in the presence of increasing channel estimation errors.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"917-921"},"PeriodicalIF":4.4,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026346","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1109/LCOMM.2026.3654545
Wenqiang Shi;Hu Jin;Yingke Lei;Fei Teng;Jin Wang
The performance of specific emitter identification (SEI) techniques is often significantly degraded due to changes in the signal distribution of targets to be identified and the lack of labels in the data. To address the aforementioned issue, this letter proposes a SEI method based on adaptive wavelet decomposition and domain adversarial regularization (AWD2AR) for multiple cross-domain scenarios. Firstly, AWD2AR preprocesses all the received signals to obtain more separable feature representations. Subsequently, AWD2AR compels the target domain feature extractor to learn domain-invariant features. Meanwhile, a metric-based regularization term is utilized to ensure the correct matching of various classes within the domain, thereby enhancing the model’s performance on the target domain. Experimental results on different datasets demonstrate that AWD2AR outperforms the state-of-the-art algorithms in various cross-domain conditions. Furthermore, the rationality of AWD2AR has been validated through ablation experiment.
{"title":"AWD2AR: An Unsupervised Identification Framework for Specific Emitters in Diverse Cross-Domain Scenarios","authors":"Wenqiang Shi;Hu Jin;Yingke Lei;Fei Teng;Jin Wang","doi":"10.1109/LCOMM.2026.3654545","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3654545","url":null,"abstract":"The performance of specific emitter identification (SEI) techniques is often significantly degraded due to changes in the signal distribution of targets to be identified and the lack of labels in the data. To address the aforementioned issue, this letter proposes a SEI method based on adaptive wavelet decomposition and domain adversarial regularization (AWD2AR) for multiple cross-domain scenarios. Firstly, AWD2AR preprocesses all the received signals to obtain more separable feature representations. Subsequently, AWD2AR compels the target domain feature extractor to learn domain-invariant features. Meanwhile, a metric-based regularization term is utilized to ensure the correct matching of various classes within the domain, thereby enhancing the model’s performance on the target domain. Experimental results on different datasets demonstrate that AWD2AR outperforms the state-of-the-art algorithms in various cross-domain conditions. Furthermore, the rationality of AWD2AR has been validated through ablation experiment.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"957-961"},"PeriodicalIF":4.4,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082154","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Modulation recognition is critical in intelligent wireless communication, yet deep learning-based automatic modulation classification (AMC) models are vulnerable to adversarial attacks, posing severe risks. While adversarial detection and training offer partial mitigation, they suffer from evasion risks, signal distortion, or high latency—making them unfit for real-time systems like Unmanned Aerial Vehicle (UAV) swarms. Although generative models can purify adversarial inputs, their slow inference limits practicality. Conversely, reconstruction-based methods enable low-latency recovery but often compromise waveform fidelity. We propose a reconstruction-driven adversarial purification approach that directly restores clean signals at the input level, preserving both semantic features and physical consistency without classifier modification, ensuring high accuracy and real-time robustness. Experimental results on the RML2016.10b dataset show that our reconstruction-based method SigReconstruction achieves an average classification accuracy of 77.04% under adversarial attacks(clean accuracy of 86.68%). Reconstruction quality is corroborated by low mean squared erro (MSE) (0.0286/0.0031/0.0272) and low Fréchet Inception Distance (FID) (62.33/128.82/157.48), indicating faithful waveform recovery and feature alignment. These results demonstrate that targeted reconstruction with physical constraints offers practical, low-latency robustness for adversarially challenged wireless communications.
{"title":"A Reconstruction-Based Defense Framework for Automatic Modulation Recognition","authors":"Zhen Hong;Chenyang Song;Jinhao Wan;Chengdong Jin;Haojie Zheng;Taotao Li;Zhenyu Wen","doi":"10.1109/LCOMM.2026.3653978","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3653978","url":null,"abstract":"Modulation recognition is critical in intelligent wireless communication, yet deep learning-based automatic modulation classification (AMC) models are vulnerable to adversarial attacks, posing severe risks. While adversarial detection and training offer partial mitigation, they suffer from evasion risks, signal distortion, or high latency—making them unfit for real-time systems like Unmanned Aerial Vehicle (UAV) swarms. Although generative models can purify adversarial inputs, their slow inference limits practicality. Conversely, reconstruction-based methods enable low-latency recovery but often compromise waveform fidelity. We propose a reconstruction-driven adversarial purification approach that directly restores clean signals at the input level, preserving both semantic features and physical consistency without classifier modification, ensuring high accuracy and real-time robustness. Experimental results on the RML2016.10b dataset show that our reconstruction-based method SigReconstruction achieves an average classification accuracy of 77.04% under adversarial attacks(clean accuracy of 86.68%). Reconstruction quality is corroborated by low mean squared erro (MSE) (0.0286/0.0031/0.0272) and low Fréchet Inception Distance (FID) (62.33/128.82/157.48), indicating faithful waveform recovery and feature alignment. These results demonstrate that targeted reconstruction with physical constraints offers practical, low-latency robustness for adversarially challenged wireless communications.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"937-941"},"PeriodicalIF":4.4,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026323","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-13DOI: 10.1109/LCOMM.2026.3653871
Nuwan J. G. Kankanamge;Sajjad Emdadi Mahdimahalleh;Nghi H. Tran;Khanh Pham
Channel coding represents a promising application area for neural network (NN)-based techniques. However, because the existing theoretical encoding and decoding algorithms are already highly optimized, developing NN-based methods that surpass traditional designs remains a difficult task. To address the challenge, this letter studies the recently proposed NN-based channel coding framework known as DeepPolar, with the goal of enhancing its frame error rate (FER) performance beyond that of the conventional successive cancellation (SC) decoder. Toward this goal, we introduce reliability-deweighted (RDW) top-$k$ max loss, RDW $p$ -norm loss, and RDW focal loss functions to prioritize critical bit positions during an extended training curriculum specifically designed to target FER rather than bit error rate (BER). Numerical results indicate that judicious design of these loss functions leads to a significant FER improvement of approximately 0.9 to 1 dB over the original DeepPolar code and 0.3 to 0.4 dB over the traditional polar code with SC decoding at FER level of $10^{-4}$ , depending on encoding configurations, without sacrificing BER performances. Furthermore, the proposed code designs exhibit performance close to the normal approximation of the finite blocklength capacity, operating merely 1.7 dB away. This demonstrates their considerable potential to advance NN-based polar codes.
{"title":"Improving DeepPolar Neural Codes via Reliability-Weighted FER-Centric Loss Functions","authors":"Nuwan J. G. Kankanamge;Sajjad Emdadi Mahdimahalleh;Nghi H. Tran;Khanh Pham","doi":"10.1109/LCOMM.2026.3653871","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3653871","url":null,"abstract":"Channel coding represents a promising application area for neural network (NN)-based techniques. However, because the existing theoretical encoding and decoding algorithms are already highly optimized, developing NN-based methods that surpass traditional designs remains a difficult task. To address the challenge, this letter studies the recently proposed NN-based channel coding framework known as DeepPolar, with the goal of enhancing its frame error rate (FER) performance beyond that of the conventional successive cancellation (SC) decoder. Toward this goal, we introduce reliability-deweighted (RDW) top-<inline-formula> <tex-math>$k$ </tex-math></inline-formula> max loss, RDW <inline-formula> <tex-math>$p$ </tex-math></inline-formula>-norm loss, and RDW focal loss functions to prioritize critical bit positions during an extended training curriculum specifically designed to target FER rather than bit error rate (BER). Numerical results indicate that judicious design of these loss functions leads to a significant FER improvement of approximately 0.9 to 1 dB over the original DeepPolar code and 0.3 to 0.4 dB over the traditional polar code with SC decoding at FER level of <inline-formula> <tex-math>$10^{-4}$ </tex-math></inline-formula>, depending on encoding configurations, without sacrificing BER performances. Furthermore, the proposed code designs exhibit performance close to the normal approximation of the finite blocklength capacity, operating merely 1.7 dB away. This demonstrates their considerable potential to advance NN-based polar codes.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"867-871"},"PeriodicalIF":4.4,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026342","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/LCOMM.2026.3651562
Peng Xu;Jiaxin Liu;An Wang;Chen Yi;Qi Li
Blind recognition of polar codes in non-cooperative scenarios remains insufficiently addressed. Although existing methods have achieved reliable performance in code length recognition, reliably identifying information bits remains challenging under moderate-to-high bit error rate (BER). In this letter, assuming the code length is known, we propose a two-stage blind information bits recognition scheme. In the first stage, multi-threshold voting judgment is performed to obtain the initial frozen set. In the second stage, we perform partial orders (POs) correction to enforce structural consistency of the frozen set. When BER ranges from 0 to 0.2, simulations show that the proposed method, at the cost of modest computational complexity, significantly improves the information bits identification accuracy compared with existing estimation and derivation-based approaches, achieving relative gains of approximately 40.0%, 20.7% and 13.8% for polar codes (32,15), (64,30), and (128,64), respectively.
{"title":"Blind Recognition of Polar Code Information Bits Based on Multi-Threshold Voting and Partial Orders","authors":"Peng Xu;Jiaxin Liu;An Wang;Chen Yi;Qi Li","doi":"10.1109/LCOMM.2026.3651562","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3651562","url":null,"abstract":"Blind recognition of polar codes in non-cooperative scenarios remains insufficiently addressed. Although existing methods have achieved reliable performance in code length recognition, reliably identifying information bits remains challenging under moderate-to-high bit error rate (BER). In this letter, assuming the code length is known, we propose a two-stage blind information bits recognition scheme. In the first stage, multi-threshold voting judgment is performed to obtain the initial frozen set. In the second stage, we perform partial orders (POs) correction to enforce structural consistency of the frozen set. When BER ranges from 0 to 0.2, simulations show that the proposed method, at the cost of modest computational complexity, significantly improves the information bits identification accuracy compared with existing estimation and derivation-based approaches, achieving relative gains of approximately 40.0%, 20.7% and 13.8% for polar codes (32,15), (64,30), and (128,64), respectively.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"887-891"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982281","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this letter, we investigate a multi-user communication network aided by active reconfigurable intelligent surface (RIS). To maximize the weighted sum rate (WSR) of the system, we formulate a joint transmit precoding and RIS coefficient optimization problem. However, the traditional alternating direction method of multipliers (ADMM) optimization approach for the problem suffers from high computational complexity and can only able to obtain suboptimal solutions. To deal with this issue, we propose a gradient-based meta-learning (GBML) scheme, which fuses meta-learning, neural network and nested iteration to improve the overall spectral efficiency and reduce the computational complexity associated with high-dimensional variables. Simulation results indicate that the proposed scheme can improve the WSR by up to 10.6%/70.25% in ideal/non-ideal scenarios compared with ADMM, and achieve a faster convergence rate as well.
{"title":"Beamforming and Phase Design for Active RIS-Aided Communications Using GBML","authors":"Jingming Tang;Nai Wu;Yaolian Song;Ziyi Jiang;Guicai Yu","doi":"10.1109/LCOMM.2025.3647094","DOIUrl":"https://doi.org/10.1109/LCOMM.2025.3647094","url":null,"abstract":"In this letter, we investigate a multi-user communication network aided by active reconfigurable intelligent surface (RIS). To maximize the weighted sum rate (WSR) of the system, we formulate a joint transmit precoding and RIS coefficient optimization problem. However, the traditional alternating direction method of multipliers (ADMM) optimization approach for the problem suffers from high computational complexity and can only able to obtain suboptimal solutions. To deal with this issue, we propose a gradient-based meta-learning (GBML) scheme, which fuses meta-learning, neural network and nested iteration to improve the overall spectral efficiency and reduce the computational complexity associated with high-dimensional variables. Simulation results indicate that the proposed scheme can improve the WSR by up to 10.6%/70.25% in ideal/non-ideal scenarios compared with ADMM, and achieve a faster convergence rate as well.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"1036-1040"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/LCOMM.2026.3653447
Kailin Wang;Guoyu Ma;Jingya Yang;Yiyan Ma;Mi Yang;Yunlong Lu;Guowei Shi;Bo Ai
This letter presents a code-redundancy-assisted optimization framework for tandem spreading multiple access (TSMA) systems under impulsive noise (IN). While TSMA offers low-complexity, grant-free access for massive machine-type communications (mMTC), its performance degrades in IN environments. Conventional methods optimize nonlinear clipping thresholds per segment, ignoring the global error-correction capabilities of Reed-Solomon (RS) codes. The proposed framework integrates nonlinear clipping with RS decoding constraints, leveraging code redundancy to derive the optimal clipping threshold. Simulations show significant improvements in block error rate (BLER) performance and enhanced robustness against IN with low computational complexity.
{"title":"Optimized Clipping Thresholds for Tandem Spreading Multiple Access in 6G IoT Under Impulsive Noise","authors":"Kailin Wang;Guoyu Ma;Jingya Yang;Yiyan Ma;Mi Yang;Yunlong Lu;Guowei Shi;Bo Ai","doi":"10.1109/LCOMM.2026.3653447","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3653447","url":null,"abstract":"This letter presents a code-redundancy-assisted optimization framework for tandem spreading multiple access (TSMA) systems under impulsive noise (IN). While TSMA offers low-complexity, grant-free access for massive machine-type communications (mMTC), its performance degrades in IN environments. Conventional methods optimize nonlinear clipping thresholds per segment, ignoring the global error-correction capabilities of Reed-Solomon (RS) codes. The proposed framework integrates nonlinear clipping with RS decoding constraints, leveraging code redundancy to derive the optimal clipping threshold. Simulations show significant improvements in block error rate (BLER) performance and enhanced robustness against IN with low computational complexity.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"882-886"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982202","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/LCOMM.2026.3653195
Jianghao Wu;Xinyang Wu;Huijun Cao;Jingwei Zhu
Hardware-efficient carrier phase recovery (CPR) is critical for coherent optical systems employing high-order quadrature amplitude modulation. This letter proposes a low-parallelism and low-complexity CPR architecture. It introduces a parallel prefix-sum engine that exploits the sparsity of the symbol distribution, enabling the hardware parallelism to be significantly reduced without information loss. Furthermore, the architecture features a multiplication-free maximum likelihood estimation to reduce intrinsic computational complexity. Implemented in 28 nm CMOS for 32 GBaud system, the proposed CPR estimator reduces area by 51% and power by 39% compared to a conventional mVV-CT-VV baseline, achieving state-of-the-art efficiencies with only 0.37 dB signal-to-noise ratio penalty.
{"title":"A Low-Complexity Carrier Phase Recovery Architecture Using Prefix-Sum and CT-MLE for Coherent Receivers","authors":"Jianghao Wu;Xinyang Wu;Huijun Cao;Jingwei Zhu","doi":"10.1109/LCOMM.2026.3653195","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3653195","url":null,"abstract":"Hardware-efficient carrier phase recovery (CPR) is critical for coherent optical systems employing high-order quadrature amplitude modulation. This letter proposes a low-parallelism and low-complexity CPR architecture. It introduces a parallel prefix-sum engine that exploits the sparsity of the symbol distribution, enabling the hardware parallelism to be significantly reduced without information loss. Furthermore, the architecture features a multiplication-free maximum likelihood estimation to reduce intrinsic computational complexity. Implemented in 28 nm CMOS for 32 GBaud system, the proposed CPR estimator reduces area by 51% and power by 39% compared to a conventional mVV-CT-VV baseline, achieving state-of-the-art efficiencies with only 0.37 dB signal-to-noise ratio penalty.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"872-876"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026456","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/LCOMM.2026.3651719
Yi Zhou;Desheng Wang
Semantic communication has advanced rapidly, yet most frameworks target SISO Gaussian or Rayleigh channels, limiting deployment in practical MIMO systems. We present MIST (MIMO and Modulation-aware Image Semantic Transmission), an end-to-end framework for image semantics over MIMO. MIST uses a Swin Transformer backbone, a channel-adaptive modulation module that leverages CSI and SNR to refine latent semantics, and an adaptive channel compression stage to enhance robustness to diverse channel conditions within one model. Extensive experiments under MIMO fading show consistent gains over conventional, CNN-based, and Transformer-based baselines across multiple image resolutions.
{"title":"A Swin Transformer With Channel-Adaptive Modulation for Semantic Image Transmission Over MIMO Channels","authors":"Yi Zhou;Desheng Wang","doi":"10.1109/LCOMM.2026.3651719","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3651719","url":null,"abstract":"Semantic communication has advanced rapidly, yet most frameworks target SISO Gaussian or Rayleigh channels, limiting deployment in practical MIMO systems. We present MIST (MIMO and Modulation-aware Image Semantic Transmission), an end-to-end framework for image semantics over MIMO. MIST uses a Swin Transformer backbone, a channel-adaptive modulation module that leverages CSI and SNR to refine latent semantics, and an adaptive channel compression stage to enhance robustness to diverse channel conditions within one model. Extensive experiments under MIMO fading show consistent gains over conventional, CNN-based, and Transformer-based baselines across multiple image resolutions.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"892-896"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982214","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1109/LCOMM.2026.3651820
Long Yang;Zeyu Chai;Fanggang Wang;Zhenhan Zhao;Yuchen Zhou;Jian Chen
Deep learning has shown good performance in specific emitter identification (SEI) with sufficient labeled datasets. However, in practical deployments, labeled samples are limited, while unknown open-set emitters affect identification accuracy. Hence, we propose a few-shot open-set SEI approach based on contrastive learning approach with false negative suppression. Specifically, contrastive learning is designed to pre-train the feature extractor using sufficient unlabeled auxiliary samples. This framework solves the problem of false negatives in SEI, which otherwise degrades representation learning. Subsequently, the feature extractor is fine-tuned using a small number of labeled samples on the target domain. Additionally, adaptive threshold is used for open-set recognition. ADS-B and Wi-Fi datasets are used to evaluate the proposed approach. Compared to other state-of-the-art approaches, our proposed approach improves the few-shot SEI performance under both open-set and close-set conditions.
{"title":"Few-Shot Open-Set Specific Emitter Identification: A Contrastive Learning Approach With False Negative Suppression","authors":"Long Yang;Zeyu Chai;Fanggang Wang;Zhenhan Zhao;Yuchen Zhou;Jian Chen","doi":"10.1109/LCOMM.2026.3651820","DOIUrl":"https://doi.org/10.1109/LCOMM.2026.3651820","url":null,"abstract":"Deep learning has shown good performance in specific emitter identification (SEI) with sufficient labeled datasets. However, in practical deployments, labeled samples are limited, while unknown open-set emitters affect identification accuracy. Hence, we propose a few-shot open-set SEI approach based on contrastive learning approach with false negative suppression. Specifically, contrastive learning is designed to pre-train the feature extractor using sufficient unlabeled auxiliary samples. This framework solves the problem of false negatives in SEI, which otherwise degrades representation learning. Subsequently, the feature extractor is fine-tuned using a small number of labeled samples on the target domain. Additionally, adaptive threshold is used for open-set recognition. ADS-B and Wi-Fi datasets are used to evaluate the proposed approach. Compared to other state-of-the-art approaches, our proposed approach improves the few-shot SEI performance under both open-set and close-set conditions.","PeriodicalId":13197,"journal":{"name":"IEEE Communications Letters","volume":"30 ","pages":"902-906"},"PeriodicalIF":4.4,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145982212","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：