This paper addresses the design of the receive combining matrix in a multiuser multiple-input multiple-output (MU-MIMO) downlink system, where the base station (BS) employs symbol-level precoding (SLP) to transmit mul...
详细信息
ISBN:
(数字)9798350303582
ISBN:
(纸本)9798350303599
This paper addresses the design of the receive combining matrix in a multiuser multiple-input multiple-output (MU-MIMO) downlink system, where the base station (BS) employs symbol-level precoding (SLP) to transmit multiple data streams to multiple users with multiple antennas. Unlike in the single-antenna user scenario, the design of the receive combining matrix becomes crucial in this context. To overcome the challenge of the receive combining matrix's dependency on the transmit signals, we propose a practical scheme utilizing the interference rejection combiner (IRC) for signal decoding. However, directly applying the IRC receiver to the considered MU-MIMO system presents challenges due to the rank-one transmit precoding matrix. To address this issue, we propose a new regularized IRC (RIRC) receiver. The problem is tackled by using the alternating optimization (AO) method, enabling the derivation of an optimal solution structure for the transmit precoding matrix. Numerical results demonstrate the substantial performance gain of the practical SLP scheme with the RIRC receiver over conventional Block Diagonalization (BD) based approach.
Recent studies have focused on leveraging large-scale artificial intelligence (LAI) models to improve semantic representation and compression capabilities. However, the substantial computational demands of LAI models ...
详细信息
This paper presents a hybrid position/force control with equivalent mass matrices for position-and-force-sensorless manipulators and the evaluation of the control system. Position and force estimation methods in the c...
This paper presents a hybrid position/force control with equivalent mass matrices for position-and-force-sensorless manipulators and the evaluation of the control system. Position and force estimation methods in the controller are cascaded to estimate the position and force according to the respective characteristics of the electrical and mechanical systems. The proposed controller is designed to reduce vibration of the force control axes' motion from the cross-coupling factors of the dq-axis inductance, which cause position estimation error, and the cross-coupling effects among control axes from equivalent mass matrices, which expresses the relationship between the workspace force and acceleration. The position estimation based on the voltage equation, including the cross-coupling factors, derives compensation terms of the estimation error. The cross-coupling effects between position and force control axes in the equivalent mass matrices are designed to suppress the undesired force response vibrations. The experimental results show that the proposed system kept contact with an object during the rubbing motion by a 4-degree-of-freedom manipulator without position and force sensors. The utilized equivalent mass matrix reduced the force vibration caused by the vibration of the position estimation.
The coolant flowing through the narrow rectangular channel delivers efficient cooling for the plate-type nuclear fuel assembly. These narrow rectangular channels are typically closed and differ significantly from the ...
The coolant flowing through the narrow rectangular channel delivers efficient cooling for the plate-type nuclear fuel assembly. These narrow rectangular channels are typically closed and differ significantly from the coolant channel geometry in conventional fuel rod assemblies. To accurately model the coolant behavior within these channels, a computational fluid dynamics (CFD) model was developed and validated in this paper. A two-dimensional (2D) mesh was employed to approximate the three-dimensional (3D) coolant flow, thereby reducing computational complexity. Conservation equations were formulated, and a void fraction model was incorporated as an auxiliary component. The model was validated by comparing calculation results with experimental data from the literature. In addition, to further investigate the coolant flow distribution, a multi-channel experiment was conducted to obtain additional validation data in this study. The verification results for the heat transfer, coolant flow, and void fraction models demonstrated satisfactory accuracy. The maximum absolute error of the coolant temperature was 3.1 K, and the pressure drop had a maximum relative error of 1.81 %. Under conditions of supercooled boiling at atmospheric pressure, the average relative error in void fraction was 18.24 %. Based on the multi-channel experimental data, the maximum relative error in flow distribution was 14.03 %. In multi-channel simulations, neglecting the heat conduction of steel partitions was identified as a significant source of error. Therefore, accurately modeling the coupled heat transfer between the steel partitions and the coolant is essential for improving simulation accuracy in future research.
The fatigue performance of pre-cracked graphene-copper artificial nacre (GrCu nacre) under cyclic tensile loading is investigated using theoretical analysis and molecular dynamics (MD) simulations. Mechanical models a...
详细信息
The problem of trajectory planning for formation variation of unmanned ground vehicles (UGVs) is challenging due to the coupled non-collision constraints that apply to multiple vehicles. To address this issue, an inno...
The problem of trajectory planning for formation variation of unmanned ground vehicles (UGVs) is challenging due to the coupled non-collision constraints that apply to multiple vehicles. To address this issue, an innovative frame-work based on the alternating direction method of multipliers (ADMM) is proposed in this paper. The framework decouples complex constraints into two categories: individual constraints and mutual constraints. The former category involves dynamic, formation, and static obstacle constraints, which can be solved in parallel by each vehicle. Convex feasible set (CFS) algorithm is employed in this paper to simplify the static obstacle constraints. The latter category deals with collision avoidance among vehicles during the variation process and can be solved by nonlinear solvers. The convergence of the ADMM for the optimization problem is proved in this paper. Compared with solving the entire problem directly, the proposed framework reduces computation time and improves the quality of solution.
Negative sampling (NS) is widely used in knowledge graph completion (KGC) to generate negative triples for contrastive learning during training. However, existing NS methods are not suitable when multimodal informatio...
详细信息
ISBN:
(数字)9798331513054
ISBN:
(纸本)9798331513061
Negative sampling (NS) is widely used in knowledge graph completion (KGC) to generate negative triples for contrastive learning during training. However, existing NS methods are not suitable when multimodal information is incorporated into KGC models. Due to their complex design, these methods are also inefficient. In this paper, we propose the integration of Modality-Aware Adversarial Training (IMAT) to generate higher-quality negative samples for Multimodal Knowledge Graph Completion (MMKGC), and introduce a Relation-Enhanced Cross-modal Attention (RECA) mechanism to evaluate bidirectional attention weights between multimodal features using relational information, thereby improving the model's ability to identify hard negative samples. Our approach represents a joint design of MMKGC models and training strategies, surpassing 16 recent MMKGC methods and achieving new state-of-the-art results on three public MMKGC benchmarks.
Hierarchical Text Classification, as a critical task in natural language processing, has broad applications in scenarios with complex label structures and limited samples. However, existing methods still face challeng...
详细信息
ISBN:
(数字)9798331513054
ISBN:
(纸本)9798331513061
Hierarchical Text Classification, as a critical task in natural language processing, has broad applications in scenarios with complex label structures and limited samples. However, existing methods still face challenges in capturing hierarchical label information and text interactions, resulting in suboptimal classification performance. This study proposes a few-shot hierarchical text classification method based on prompt embeddings and dynamically fused hierarchical features. Building on pre-trained language models, the method encodes hierarchical structure information as prompts through an adaptive prompt embedding model and dynamically fuses hierarchical features to generate more expressive text representations. Additionally, a positive sample generation strategy based on contrastive learning is introduced, enabling the model to effectively distinguish hierarchical features across different categories, further enhancing classification performance. Experimental results demonstrate that this method significantly improves classification accuracy on multiple public datasets, exhibiting stronger generalization ability and robustness compared to mainstream methods. This approach offers a novel and effective solution for hierarchical text classification tasks.
Optimizing various wireless user tasks poses a significant challenge for networking systems because of the expanding range of user requirements. Despite advancements in Deep Reinforcement Learning (DRL), the need for ...
详细信息
ISBN:
(数字)9798350361261
ISBN:
(纸本)9798350361278
Optimizing various wireless user tasks poses a significant challenge for networking systems because of the expanding range of user requirements. Despite advancements in Deep Reinforcement Learning (DRL), the need for customized optimization tasks for individual users complicates developing and applying numerous DRL models, leading to substantial computation resource and energy consumption and can lead to inconsistent outcomes. To address this issue, we propose a novel approach utilizing a Mixture of Experts (MoE) framework, augmented with Large Language Models (LLMs), to analyze user objectives and constraints effectively, select specialized DRL experts, and weigh each decision from the participating experts. Specifically, we develop a gate network to oversee the expert models, allowing a collective of experts to tackle a wide array of new tasks. Furthermore, we innovatively substitute the traditional gate network with an LLM, leveraging its advanced reasoning capabilities to manage expert model selection for joint decisions. Our proposed method reduces the need to train new DRL models for each unique optimization problem, decreasing energy consumption and AI model implementation costs. The LLMenabled MoE approach is validated through a general maze navigation task and a specific network service provider utility maximization task, demonstrating its effectiveness and practical applicability in optimizing complex networking systems.
暂无评论