Pub Date : 2024-11-20DOI: 10.1109/TMLCN.2024.3503543
Houssem Sifaou;Osvaldo Simeone
In many wireless application scenarios, acquiring labeled data can be prohibitively costly, requiring complex optimization processes or measurement campaigns. Semi-supervised learning leverages unlabeled samples to augment the available dataset by assigning synthetic labels obtained via machine learning (ML)-based predictions. However, treating the synthetic labels as true labels may yield worse-performing models as compared to models trained using only labeled data. Inspired by the recently developed prediction-powered inference (PPI) framework, this work investigates how to leverage the synthetic labels produced by an ML model, while accounting for the inherent bias concerning true labels. To this end, we first review PPI and its recent extensions, namely tuned PPI and cross-prediction-powered inference (CPPI). Then, we introduce two novel variants of PPI. The first, referred to as tuned CPPI, provides CPPI with an additional degree of freedom in adapting to the quality of the ML-based labels. The second, meta-CPPI (MCPPI), extends tuned CPPI via the joint optimization of the ML labeling models and of the parameters of interest. Finally, we showcase two applications of PPI-based techniques in wireless systems, namely beam alignment based on channel knowledge maps in millimeter-wave systems and received signal strength information-based indoor localization. Simulation results show the advantages of PPI-based techniques over conventional approaches that rely solely on labeled data or that apply standard pseudo-labeling strategies from semi-supervised learning. Furthermore, the proposed tuned CPPI method is observed to guarantee the best performance among all benchmark schemes, especially in the regime of limited labeled data.
{"title":"Semi-Supervised Learning via Cross-Prediction-Powered Inference for Wireless Systems","authors":"Houssem Sifaou;Osvaldo Simeone","doi":"10.1109/TMLCN.2024.3503543","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3503543","url":null,"abstract":"In many wireless application scenarios, acquiring labeled data can be prohibitively costly, requiring complex optimization processes or measurement campaigns. Semi-supervised learning leverages unlabeled samples to augment the available dataset by assigning synthetic labels obtained via machine learning (ML)-based predictions. However, treating the synthetic labels as true labels may yield worse-performing models as compared to models trained using only labeled data. Inspired by the recently developed prediction-powered inference (PPI) framework, this work investigates how to leverage the synthetic labels produced by an ML model, while accounting for the inherent bias concerning true labels. To this end, we first review PPI and its recent extensions, namely tuned PPI and cross-prediction-powered inference (CPPI). Then, we introduce two novel variants of PPI. The first, referred to as tuned CPPI, provides CPPI with an additional degree of freedom in adapting to the quality of the ML-based labels. The second, meta-CPPI (MCPPI), extends tuned CPPI via the joint optimization of the ML labeling models and of the parameters of interest. Finally, we showcase two applications of PPI-based techniques in wireless systems, namely beam alignment based on channel knowledge maps in millimeter-wave systems and received signal strength information-based indoor localization. Simulation results show the advantages of PPI-based techniques over conventional approaches that rely solely on labeled data or that apply standard pseudo-labeling strategies from semi-supervised learning. Furthermore, the proposed tuned CPPI method is observed to guarantee the best performance among all benchmark schemes, especially in the regime of limited labeled data.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"30-44"},"PeriodicalIF":0.0,"publicationDate":"2024-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10758826","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142844291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-19DOI: 10.1109/TMLCN.2024.3502576
Tianjiao Sun;Sixing Yin;Li Deng;F. Richard Yu
Taking advantages of both unmanned aerial vehicles (UAVs) and reconfigurable intelligent surfaces (RISs), UAV-mounted-RIS systems are expected to enhance transmission performance in complicated wireless environments. In this paper, we focus on system design for a UAV-mounted-RIS system and investigate joint optimization for the RIS’s phase shift and the UAV’s trajectory. To cope with the practical issue of inaccessible information on the user terminals’ (UTs) location and channel state, a reinforcement learning (RL)-based solution is proposed to find the optimal policy with finite steps of “trial-and-error”. As the action space is continuous, the deep deterministic policy gradient (DDPG) algorithm is applied to train the RL model. However, the online interaction between the agent and environment may lead to instability during the training and the assumption of (first-order) Markovian state transition could be impractical in real-world problems. Therefore, the decision transformer (DT) algorithm is employed as an alternative for RL model training to adapt to more general situations of state transition. Experimental results demonstrate that the proposed RL solutions are highly efficient in model training along with acceptable performance close to the benchmark, which relies on conventional optimization algorithms with the UT’s locations and channel parameters explicitly known beforehand.
{"title":"Reinforcement-Learning-Based Trajectory Design and Phase-Shift Control in UAV-Mounted-RIS Communications","authors":"Tianjiao Sun;Sixing Yin;Li Deng;F. Richard Yu","doi":"10.1109/TMLCN.2024.3502576","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3502576","url":null,"abstract":"Taking advantages of both unmanned aerial vehicles (UAVs) and reconfigurable intelligent surfaces (RISs), UAV-mounted-RIS systems are expected to enhance transmission performance in complicated wireless environments. In this paper, we focus on system design for a UAV-mounted-RIS system and investigate joint optimization for the RIS’s phase shift and the UAV’s trajectory. To cope with the practical issue of inaccessible information on the user terminals’ (UTs) location and channel state, a reinforcement learning (RL)-based solution is proposed to find the optimal policy with finite steps of “trial-and-error”. As the action space is continuous, the deep deterministic policy gradient (DDPG) algorithm is applied to train the RL model. However, the online interaction between the agent and environment may lead to instability during the training and the assumption of (first-order) Markovian state transition could be impractical in real-world problems. Therefore, the decision transformer (DT) algorithm is employed as an alternative for RL model training to adapt to more general situations of state transition. Experimental results demonstrate that the proposed RL solutions are highly efficient in model training along with acceptable performance close to the benchmark, which relies on conventional optimization algorithms with the UT’s locations and channel parameters explicitly known beforehand.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"163-175"},"PeriodicalIF":0.0,"publicationDate":"2024-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10758222","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142918339","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-18DOI: 10.1109/TMLCN.2024.3501217
Rodrigo Carvalho;Faroq Al-Tam;Noélia Correia
As a growing trend, edge computing infrastructures are starting to be integrated with Internet of Things (IoT) systems to facilitate time-critical applications. These systems often require the processing of data with limited usefulness in time, so the edge becomes vital in the development of such reactive IoT applications with real-time requirements. Although different architectural designs will always have advantages and disadvantages, mobile gateways appear to be particularly relevant in enabling this integration with the edge, particularly in the context of wide area networks with occasional data generation. In these scenarios, mobility planning is necessary, as aspects of the technology need to be aligned with the temporal needs of an application. The nature of this planning problem makes cutting-edge deep reinforcement learning (DRL) techniques useful in solving pertinent issues, such as having to deal with multiple dimensions in the action space while aiming for optimum levels of system performance. This article presents a novel scalable DRL model that incorporates a pointer-network (Ptr-Net) and an actor-critic algorithm to handle complex action spaces. The model synchronously determines the gateway location and visit time. Ultimately, the gateways are able to attain high-quality trajectory planning with reduced latency.
{"title":"A2PC: Augmented Advantage Pointer-Critic Model for Low Latency on Mobile IoT With Edge Computing","authors":"Rodrigo Carvalho;Faroq Al-Tam;Noélia Correia","doi":"10.1109/TMLCN.2024.3501217","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3501217","url":null,"abstract":"As a growing trend, edge computing infrastructures are starting to be integrated with Internet of Things (IoT) systems to facilitate time-critical applications. These systems often require the processing of data with limited usefulness in time, so the edge becomes vital in the development of such reactive IoT applications with real-time requirements. Although different architectural designs will always have advantages and disadvantages, mobile gateways appear to be particularly relevant in enabling this integration with the edge, particularly in the context of wide area networks with occasional data generation. In these scenarios, mobility planning is necessary, as aspects of the technology need to be aligned with the temporal needs of an application. The nature of this planning problem makes cutting-edge deep reinforcement learning (DRL) techniques useful in solving pertinent issues, such as having to deal with multiple dimensions in the action space while aiming for optimum levels of system performance. This article presents a novel scalable DRL model that incorporates a pointer-network (Ptr-Net) and an actor-critic algorithm to handle complex action spaces. The model synchronously determines the gateway location and visit time. Ultimately, the gateways are able to attain high-quality trajectory planning with reduced latency.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"3 ","pages":"1-16"},"PeriodicalIF":0.0,"publicationDate":"2024-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10755120","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142821217","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-11-04DOI: 10.1109/TMLCN.2024.3491054
Zain Ali;Zouheir Rezki;Mohamed-Slim Alouini
The next generation of communication devices will require robust connectivity for millions of ground devices such as sensors or mobile devices in remote or disaster-stricken areas to be connected to the network. Non-terrestrial network (NTN) nodes can play a vital role in fulfilling these requirements. Specifically, low-earth orbiting (LEO) satellites have emerged as an efficient and cost-effective technique to connect devices over long distances through space. However, due to their low power and environmental limitations, LEO satellites may require assistance from aerial devices such as high-altitude platforms (HAPs) or unmanned aerial vehicles to forward their data to the ground devices. Moreover, the limited power available at the NTNs makes it crucial to utilize available resources efficiently. In this paper, we present a model in which a LEO satellite communicates with multiple ground devices with the help of HAPs that relay LEO data to the ground devices. We formulate the problem of optimizing power allocation at the LEO satellite and all the HAPs to maximize the sum-rate of the system. To take advantage of the benefits of free-space optical (FSO) communication in satellites, we consider the LEO transmitting data to the HAPs on FSO links, which are then broadcast to the connected ground devices on radio frequency channels. We transform the complex non-convex problem into a convex form and compute the Karush-Kuhn-Tucker (KKT) conditions-based solution of the problem for power allocation at the LEO satellite and HAPs. Then, to reduce computation time, we propose a soft actor-critic (SAC) reinforcement learning (RL) framework that provides the solution in significantly less time while delivering comparable performance to the KKT scheme. Our simulation results demonstrate that the proposed solutions provide excellent performance and are scalable to any number of HAPs and ground devices in the system.
{"title":"Optimizing Power Allocation in HAPs Assisted LEO Satellite Communications","authors":"Zain Ali;Zouheir Rezki;Mohamed-Slim Alouini","doi":"10.1109/TMLCN.2024.3491054","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3491054","url":null,"abstract":"The next generation of communication devices will require robust connectivity for millions of ground devices such as sensors or mobile devices in remote or disaster-stricken areas to be connected to the network. Non-terrestrial network (NTN) nodes can play a vital role in fulfilling these requirements. Specifically, low-earth orbiting (LEO) satellites have emerged as an efficient and cost-effective technique to connect devices over long distances through space. However, due to their low power and environmental limitations, LEO satellites may require assistance from aerial devices such as high-altitude platforms (HAPs) or unmanned aerial vehicles to forward their data to the ground devices. Moreover, the limited power available at the NTNs makes it crucial to utilize available resources efficiently. In this paper, we present a model in which a LEO satellite communicates with multiple ground devices with the help of HAPs that relay LEO data to the ground devices. We formulate the problem of optimizing power allocation at the LEO satellite and all the HAPs to maximize the sum-rate of the system. To take advantage of the benefits of free-space optical (FSO) communication in satellites, we consider the LEO transmitting data to the HAPs on FSO links, which are then broadcast to the connected ground devices on radio frequency channels. We transform the complex non-convex problem into a convex form and compute the Karush-Kuhn-Tucker (KKT) conditions-based solution of the problem for power allocation at the LEO satellite and HAPs. Then, to reduce computation time, we propose a soft actor-critic (SAC) reinforcement learning (RL) framework that provides the solution in significantly less time while delivering comparable performance to the KKT scheme. Our simulation results demonstrate that the proposed solutions provide excellent performance and are scalable to any number of HAPs and ground devices in the system.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1661-1677"},"PeriodicalIF":0.0,"publicationDate":"2024-11-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10741546","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142636509","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The integration of high-precision cellular localization and machine learning (ML) is considered a cornerstone technique in future cellular navigation systems, offering unparalleled accuracy and functionality. This study focuses on localization based on uplink channel measurements in a fifth-generation (5G) new radio (NR) system. An attention-aided ML-based single-snapshot localization pipeline is presented, which consists of several cascaded blocks, namely a signal processing block, an attention-aided block, and an uncertainty estimation block. Specifically, the signal processing block generates an impulse response beam matrix for all beams. The attention-aided block trains on the channel impulse responses using an attention-aided network, which captures the correlation between impulse responses for different beams. The uncertainty estimation block predicts the probability density function of the user equipment (UE) position, thereby also indicating the confidence level of the localization result. Two representative uncertainty estimation techniques, the negative log-likelihood and the regression-by-classification techniques, are applied and compared. Furthermore, for dynamic measurements with multiple snapshots available, we combine the proposed pipeline with a Kalman filter to enhance localization accuracy. To evaluate our approach, we extract channel impulse responses for different beams from a commercial base station. The outdoor measurement campaign covers Line-of-Sight (LoS), Non Line-of-Sight (NLoS), and a mix of LoS and NLoS scenarios. The results show that sub-meter localization accuracy can be achieved.
高精度蜂窝定位与机器学习（ML）的集成被认为是未来蜂窝导航系统的基石技术，可提供无与伦比的精度和功能。本研究的重点是第五代（5G）新无线电（NR）系统中基于上行链路信道测量的定位。本文介绍了一种基于注意力辅助 ML 的单快照定位流水线，它由几个级联块组成，即信号处理块、注意力辅助块和不确定性估计块。具体来说，信号处理模块为所有波束生成脉冲响应波束矩阵。注意力辅助块利用注意力辅助网络对信道脉冲响应进行训练，从而捕捉不同波束脉冲响应之间的相关性。不确定性估计模块预测用户设备（UE）位置的概率密度函数，从而显示定位结果的置信度。应用了两种具有代表性的不确定性估计技术，即负对数概率和分类回归技术，并进行了比较。此外，对于具有多个可用快照的动态测量，我们将提议的管道与卡尔曼滤波器相结合，以提高定位精度。为了评估我们的方法，我们从一个商用基站提取了不同波束的信道脉冲响应。室外测量活动涵盖了视距（LoS）、非视距（NLoS）以及 LoS 和 NLoS 场景的混合。结果表明，可以实现亚米级定位精度。
{"title":"Attention-Aided Outdoor Localization in Commercial 5G NR Systems","authors":"Guoda Tian;Dino Pjanić;Xuesong Cai;Bo Bernhardsson;Fredrik Tufvesson","doi":"10.1109/TMLCN.2024.3490496","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3490496","url":null,"abstract":"The integration of high-precision cellular localization and machine learning (ML) is considered a cornerstone technique in future cellular navigation systems, offering unparalleled accuracy and functionality. This study focuses on localization based on uplink channel measurements in a fifth-generation (5G) new radio (NR) system. An attention-aided ML-based single-snapshot localization pipeline is presented, which consists of several cascaded blocks, namely a signal processing block, an attention-aided block, and an uncertainty estimation block. Specifically, the signal processing block generates an impulse response beam matrix for all beams. The attention-aided block trains on the channel impulse responses using an attention-aided network, which captures the correlation between impulse responses for different beams. The uncertainty estimation block predicts the probability density function of the user equipment (UE) position, thereby also indicating the confidence level of the localization result. Two representative uncertainty estimation techniques, the negative log-likelihood and the regression-by-classification techniques, are applied and compared. Furthermore, for dynamic measurements with multiple snapshots available, we combine the proposed pipeline with a Kalman filter to enhance localization accuracy. To evaluate our approach, we extract channel impulse responses for different beams from a commercial base station. The outdoor measurement campaign covers Line-of-Sight (LoS), Non Line-of-Sight (NLoS), and a mix of LoS and NLoS scenarios. The results show that sub-meter localization accuracy can be achieved.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1678-1692"},"PeriodicalIF":0.0,"publicationDate":"2024-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10741343","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142694615","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-24DOI: 10.1109/TMLCN.2024.3485520
Tianlun Hu;Qi Liao;Qiang Liu;Georg Carle
Network slicing enables operators to efficiently support diverse applications on a shared infrastructure. However, the evolving complexity of networks, compounded by inter-cell interference, necessitates agile and adaptable resource management. While deep learning offers solutions for coping with complexity, its adaptability to dynamic configurations remains limited. In this paper, we propose a novel hybrid deep learning algorithm called IDLA (integrated deep learning with the Lagrangian method). This integrated approach aims to enhance the scalability, flexibility, and robustness of slicing resource allocation solutions by harnessing the high approximation capability of deep learning and the strong generalization of classical non-linear optimization methods. Then, we introduce a variational information bottleneck (VIB)-assisted domain adaptation (DA) approach to enhance integrated deep learning and Lagrangian method (IDLA)’s adaptability across diverse network environments and conditions. We propose pre-training a variational information bottleneck (VIB)-based Quality of Service (QoS) estimator, using slice-specific inputs shared across all source domain slices. Each target domain slice can deploy this estimator to predict its QoS and optimize slice resource allocation using the IDLA algorithm. This VIB-based estimator is continuously fine-tuned with a mixture of samples from both the source and target domains until convergence. Evaluating on a multi-cell network with time-varying slice configurations, the VIB-enhanced IDLA algorithm outperforms baselines such as heuristic and deep reinforcement learning-based solutions, achieving twice the convergence speed and 16.52% higher asymptotic performance after slicing configuration changes. Transferability assessment demonstrates a 25.66% improvement in estimation accuracy with VIB, especially in scenarios with significant domain gaps, highlighting its robustness and effectiveness across diverse domains.
{"title":"Information Bottleneck-Based Domain Adaptation for Hybrid Deep Learning in Scalable Network Slicing","authors":"Tianlun Hu;Qi Liao;Qiang Liu;Georg Carle","doi":"10.1109/TMLCN.2024.3485520","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3485520","url":null,"abstract":"Network slicing enables operators to efficiently support diverse applications on a shared infrastructure. However, the evolving complexity of networks, compounded by inter-cell interference, necessitates agile and adaptable resource management. While deep learning offers solutions for coping with complexity, its adaptability to dynamic configurations remains limited. In this paper, we propose a novel hybrid deep learning algorithm called IDLA (integrated deep learning with the Lagrangian method). This integrated approach aims to enhance the scalability, flexibility, and robustness of slicing resource allocation solutions by harnessing the high approximation capability of deep learning and the strong generalization of classical non-linear optimization methods. Then, we introduce a variational information bottleneck (VIB)-assisted domain adaptation (DA) approach to enhance integrated deep learning and Lagrangian method (IDLA)’s adaptability across diverse network environments and conditions. We propose pre-training a variational information bottleneck (VIB)-based Quality of Service (QoS) estimator, using slice-specific inputs shared across all source domain slices. Each target domain slice can deploy this estimator to predict its QoS and optimize slice resource allocation using the IDLA algorithm. This VIB-based estimator is continuously fine-tuned with a mixture of samples from both the source and target domains until convergence. Evaluating on a multi-cell network with time-varying slice configurations, the VIB-enhanced IDLA algorithm outperforms baselines such as heuristic and deep reinforcement learning-based solutions, achieving twice the convergence speed and 16.52% higher asymptotic performance after slicing configuration changes. Transferability assessment demonstrates a 25.66% improvement in estimation accuracy with VIB, especially in scenarios with significant domain gaps, highlighting its robustness and effectiveness across diverse domains.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1642-1660"},"PeriodicalIF":0.0,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10734592","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142579172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-23DOI: 10.1109/TMLCN.2024.3485521
Anzhe Ye;Haotian Chen;Ryo Natsuaki;Akira Hirose
The performance of a wireless communication system depends to a large extent on the wireless channel. Due to the multipath fading environment during the radio wave propagation, channel prediction plays a vital role to enable adaptive transmission for wireless communication systems. Predicting various channel characteristics by using neural networks can help address more complex communication environments. However, achieving this goal typically requires the simultaneous use of multiple distinct neural models, which is undoubtedly unaffordable for mobile communications. Therefore, it is necessary to enable a simpler structure to simultaneously predict multiple channel characteristics. In this paper, we propose a fading channel prediction method using phasor quaternion neural networks (PQNNs) to predict the polarization states, with phase information involved to enhance the channel compensation ability. We evaluate the performance of the proposed PQNN method in two different fading situations in an actual environment, and we find that the proposed scheme provides 2.8 dB and 4.0 dB improvements at bit error rate (BER) of $10^{-4}$ , showing better BER performance in light and serious fading situations, respectively. This work also reveals that by treating polarization information and phase information as a single entity, the model can leverage their physical correlation to achieve improved performance.
无线通信系统的性能在很大程度上取决于无线信道。由于无线电波传播过程中存在多径衰落环境，信道预测对无线通信系统的自适应传输起着至关重要的作用。利用神经网络预测各种信道特性有助于应对更复杂的通信环境。然而，要实现这一目标，通常需要同时使用多个不同的神经模型，这对于移动通信来说无疑是难以承受的。因此，有必要采用更简单的结构来同时预测多种信道特性。在本文中，我们提出了一种使用相位四元神经网络（PQNN）预测极化状态的衰落信道预测方法，其中涉及相位信息以增强信道补偿能力。我们在实际环境中评估了所提出的 PQNN 方法在两种不同衰落情况下的性能，发现所提出的方案在误码率（BER）为 10^{-4}$ 时分别提高了 2.8 dB 和 4.0 dB，在轻度衰落和严重衰落情况下分别表现出更好的误码率性能。这项工作还揭示出，通过将极化信息和相位信息视为单一实体，该模型可以利用它们之间的物理相关性来提高性能。
{"title":"Polarization-Aware Channel State Prediction Using Phasor Quaternion Neural Networks","authors":"Anzhe Ye;Haotian Chen;Ryo Natsuaki;Akira Hirose","doi":"10.1109/TMLCN.2024.3485521","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3485521","url":null,"abstract":"The performance of a wireless communication system depends to a large extent on the wireless channel. Due to the multipath fading environment during the radio wave propagation, channel prediction plays a vital role to enable adaptive transmission for wireless communication systems. Predicting various channel characteristics by using neural networks can help address more complex communication environments. However, achieving this goal typically requires the simultaneous use of multiple distinct neural models, which is undoubtedly unaffordable for mobile communications. Therefore, it is necessary to enable a simpler structure to simultaneously predict multiple channel characteristics. In this paper, we propose a fading channel prediction method using phasor quaternion neural networks (PQNNs) to predict the polarization states, with phase information involved to enhance the channel compensation ability. We evaluate the performance of the proposed PQNN method in two different fading situations in an actual environment, and we find that the proposed scheme provides 2.8 dB and 4.0 dB improvements at bit error rate (BER) of \u0000<inline-formula> <tex-math>$10^{-4}$ </tex-math></inline-formula>\u0000, showing better BER performance in light and serious fading situations, respectively. This work also reveals that by treating polarization information and phase information as a single entity, the model can leverage their physical correlation to achieve improved performance.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1628-1641"},"PeriodicalIF":0.0,"publicationDate":"2024-10-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10731896","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142579173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
terahertz (THz) band communication is envisioned as one of the leading technologies to meet the exponentially growing data rate requirements of emerging and future wireless communication networks. Utilizing the contiguous bandwidth available at THz frequencies requires a transceiver design tailored to tackle issues existing at these frequencies such as strong propagation and absorption loss, small scale fading (e.g. scattering, reflection, refraction), hardware non-linearity, etc. In prior works, multicarrier waveforms (e.g., Orthogonal Frequency Division Multiplexing (OFDM)) are shown to be efficient in tackling some of these issues. However, OFDM introduces a drawback in the form of peak-to-average power ratio (PAPR) which, compounded with strong propagation and absorption loss and high noise power due to large bandwidth at THz and sub-THz frequencies, severely limits link distances and, in turn, capacity, preventing efficient bandwidth usage. In this work, we propose TWIRLD - a deep learning (DL)-based joint optimization method, modeled and implemented as components of end-to-end transceiver chain. TWIRLD performs a symbol remapping at baseband of OFDM signals, which increases average transmit power while also optimizing the bit error rate (BER). We provide theoretical analysis, statistical equivalence of TWIRLD to the ideal receiver, and comprehensive complexity and footprint estimates. We validate TWIRLD in simulation showing link distance improvement of up to 91% and compare the results with legacy and state of the art methods and their enhanced versions. Finally, TWIRLD is validated with over the air (OTA) communication using a state-of-the-art testbed at 140 GHz up to a bandwidth of 5 GHz where we observe improvement of up to 79% in link distance accommodating for practical channel and other transmission losses.
{"title":"TWIRLD: Transformer Generated Terahertz Waveform for Improved Radio Link Distance","authors":"Shuvam Chakraborty;Claire Parisi;Dola Saha;Ngwe Thawdar","doi":"10.1109/TMLCN.2024.3483111","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3483111","url":null,"abstract":"terahertz (THz) band communication is envisioned as one of the leading technologies to meet the exponentially growing data rate requirements of emerging and future wireless communication networks. Utilizing the contiguous bandwidth available at THz frequencies requires a transceiver design tailored to tackle issues existing at these frequencies such as strong propagation and absorption loss, small scale fading (e.g. scattering, reflection, refraction), hardware non-linearity, etc. In prior works, multicarrier waveforms (e.g., Orthogonal Frequency Division Multiplexing (OFDM)) are shown to be efficient in tackling some of these issues. However, OFDM introduces a drawback in the form of peak-to-average power ratio (PAPR) which, compounded with strong propagation and absorption loss and high noise power due to large bandwidth at THz and sub-THz frequencies, severely limits link distances and, in turn, capacity, preventing efficient bandwidth usage. In this work, we propose \u0000<monospace>TWIRLD</monospace>\u0000 - a deep learning (DL)-based joint optimization method, modeled and implemented as components of end-to-end transceiver chain. \u0000<monospace>TWIRLD</monospace>\u0000 performs a symbol remapping at baseband of OFDM signals, which increases average transmit power while also optimizing the bit error rate (BER). We provide theoretical analysis, statistical equivalence of \u0000<monospace>TWIRLD</monospace>\u0000 to the ideal receiver, and comprehensive complexity and footprint estimates. We validate \u0000<monospace>TWIRLD</monospace>\u0000 in simulation showing link distance improvement of up to 91% and compare the results with legacy and state of the art methods and their enhanced versions. Finally, \u0000<monospace>TWIRLD</monospace>\u0000 is validated with over the air (OTA) communication using a state-of-the-art testbed at 140 GHz up to a bandwidth of 5 GHz where we observe improvement of up to 79% in link distance accommodating for practical channel and other transmission losses.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1595-1614"},"PeriodicalIF":0.0,"publicationDate":"2024-10-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10720922","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142550544","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-14DOI: 10.1109/TMLCN.2024.3480044
Jia Guo;Chenyang Yang
Graph neural networks (GNNs) have been shown promising in optimizing power allocation and link scheduling with good size generalizability and low training complexity. These merits are important for learning wireless policies under dynamic environments, which partially come from the matched permutation equivariance (PE) properties of the GNNs to the policies to be learned. Nonetheless, it has been noticed in literature that only satisfying the PE property of a precoding policy in multi-antenna systems cannot ensure a GNN for learning precoding to be generalizable to the unseen problem scales. Incorporating models with GNNs helps improve size generalizability, which however is only applicable to specific problems, settings, and algorithms. In this paper, we propose a framework of size generalizable GNNs for learning precoding policies that are purely data-driven and can learn wireless policies including but not limited to baseband and hybrid precoding in multi-user multi-antenna systems. To this end, we first find a special structure of each iteration of several numerical algorithms for optimizing precoding, from which we identify the key characteristics of a GNN that affect its size generalizability. Then, we design size-generalizable GNNs that are with these key characteristics and satisfy the PE properties of precoding policies in a recursive manner. Simulation results show that the proposed GNNs can be well-generalized to the number of users for learning baseband and hybrid precoding policies, require much fewer samples than existing GNNs and shorter inference time than numerical algorithms to achieve the same performance.
图神经网络（GNN）在优化功率分配和链路调度方面前景广阔，具有良好的尺寸泛化能力和较低的训练复杂度。这些优点对于在动态环境下学习无线策略非常重要，而这些优点部分来自于图形神经网络与待学习策略相匹配的置换等差（PE）特性。然而，有文献指出，在多天线系统中，仅满足预编码策略的 PE 属性并不能确保用于学习预编码的 GNN 能够泛化到未知的问题规模。将模型与 GNN 结合起来有助于提高规模通用性，但这只适用于特定的问题、设置和算法。在本文中，我们提出了一个用于学习预编码策略的可尺寸泛化 GNN 框架，该框架纯粹由数据驱动，可以学习无线策略，包括但不限于多用户多天线系统中的基带和混合预编码。为此，我们首先找到了用于优化预编码的几种数值算法每次迭代的特殊结构，并从中找出了影响 GNN 大小通用性的关键特征。然后，我们以递归的方式设计出具有这些关键特征并满足预编码策略 PE 特性的可尺寸泛化 GNN。仿真结果表明，所提出的 GNN 在学习基带和混合预编码策略时可以很好地泛化到用户数量，与现有的 GNN 相比需要更少的样本，与数值算法相比推理时间更短，从而达到相同的性能。
{"title":"Recursive GNNs for Learning Precoding Policies With Size-Generalizability","authors":"Jia Guo;Chenyang Yang","doi":"10.1109/TMLCN.2024.3480044","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3480044","url":null,"abstract":"Graph neural networks (GNNs) have been shown promising in optimizing power allocation and link scheduling with good size generalizability and low training complexity. These merits are important for learning wireless policies under dynamic environments, which partially come from the matched permutation equivariance (PE) properties of the GNNs to the policies to be learned. Nonetheless, it has been noticed in literature that only satisfying the PE property of a precoding policy in multi-antenna systems cannot ensure a GNN for learning precoding to be generalizable to the unseen problem scales. Incorporating models with GNNs helps improve size generalizability, which however is only applicable to specific problems, settings, and algorithms. In this paper, we propose a framework of size generalizable GNNs for learning precoding policies that are purely data-driven and can learn wireless policies including but not limited to baseband and hybrid precoding in multi-user multi-antenna systems. To this end, we first find a special structure of each iteration of several numerical algorithms for optimizing precoding, from which we identify the key characteristics of a GNN that affect its size generalizability. Then, we design size-generalizable GNNs that are with these key characteristics and satisfy the PE properties of precoding policies in a recursive manner. Simulation results show that the proposed GNNs can be well-generalized to the number of users for learning baseband and hybrid precoding policies, require much fewer samples than existing GNNs and shorter inference time than numerical algorithms to achieve the same performance.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1558-1579"},"PeriodicalIF":0.0,"publicationDate":"2024-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10716720","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142540477","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-10-11DOI: 10.1109/TMLCN.2024.3479109
Faraz Ahmed;Lianjie Cao;Ayush Goel;Puneet Sharma
Distributed applications such as cloud gaming, streaming, etc., are increasingly using edge-to-cloud infrastructure for high availability and performance. While edge infrastructure brings services closer to the end-user, the number of sites on which the services need to be replicated has also increased. This makes replica selection challenging for clients of the replicated services. Traditional replica selection methods including anycast based methods and DNS re-directions are performance agnostic, and clients experience degraded network performance when network performance dynamics are not considered in replica selection. In this work, we present a client-side replica selection framework NeIL, that enables network performance aware replica selection. We propose to use bandits with experts based Multi-Armed Bandit (MAB) algorithms and adapt these algorithms for replica selection at individual clients without centralized coordination. We evaluate our approach using three different setups including a distributed Mininet setup where we use publicly available network performance data from the Measurement Lab (M-Lab) to emulate network conditions, a setup where we deploy replica servers on AWS, and finally we present results from a global enterprise deployment. Our experimental results show that in comparison to greedy selection, NeIL performs better than greedy for 45% of the time and better than or equal to greedy selection for 80% of the time resulting in a net gain in end-to-end network performance. On AWS, we see similar results where NeIL performs better than or equal to greedy for 75% of the time. We have successfully deployed NeIL in a global enterprise remote device management service with over 4000 client devices and our analysis shows that NeIL achieves significantly better tail service quality by cutting the $99th$ percentile tail latency from 5.6 seconds to 1.7 seconds.
{"title":"NeIL: Intelligent Replica Selection for Distributed Applications","authors":"Faraz Ahmed;Lianjie Cao;Ayush Goel;Puneet Sharma","doi":"10.1109/TMLCN.2024.3479109","DOIUrl":"https://doi.org/10.1109/TMLCN.2024.3479109","url":null,"abstract":"Distributed applications such as cloud gaming, streaming, etc., are increasingly using edge-to-cloud infrastructure for high availability and performance. While edge infrastructure brings services closer to the end-user, the number of sites on which the services need to be replicated has also increased. This makes replica selection challenging for clients of the replicated services. Traditional replica selection methods including anycast based methods and DNS re-directions are performance agnostic, and clients experience degraded network performance when network performance dynamics are not considered in replica selection. In this work, we present a client-side replica selection framework NeIL, that enables network performance aware replica selection. We propose to use bandits with experts based Multi-Armed Bandit (MAB) algorithms and adapt these algorithms for replica selection at individual clients without centralized coordination. We evaluate our approach using three different setups including a distributed Mininet setup where we use publicly available network performance data from the Measurement Lab (M-Lab) to emulate network conditions, a setup where we deploy replica servers on AWS, and finally we present results from a global enterprise deployment. Our experimental results show that in comparison to greedy selection, NeIL performs better than greedy for 45% of the time and better than or equal to greedy selection for 80% of the time resulting in a net gain in end-to-end network performance. On AWS, we see similar results where NeIL performs better than or equal to greedy for 75% of the time. We have successfully deployed NeIL in a global enterprise remote device management service with over 4000 client devices and our analysis shows that NeIL achieves significantly better tail service quality by cutting the \u0000<inline-formula> <tex-math>$99th$ </tex-math></inline-formula>\u0000 percentile tail latency from 5.6 seconds to 1.7 seconds.","PeriodicalId":100641,"journal":{"name":"IEEE Transactions on Machine Learning in Communications and Networking","volume":"2 ","pages":"1580-1594"},"PeriodicalIF":0.0,"publicationDate":"2024-10-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10714467","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142540425","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","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：