design precoders

This work addresses the trade-off between communication efficiency and classification performance in distributed classification tasks over wireless channels impaired by channel distortions. It proposes a task-oriented multi-user communication framework driven by maximum a posteriori (MAP) inference, which uniquely integrates the MAP criterion with a class-mean separation objective. By jointly optimizing learnable feature extraction and low-complexity precoding design, the method directly enhances class separability after channel distortion, bypassing conventional approaches that rely on covariance estimation or signal reconstruction. The proposed scheme eliminates the need for repeated covariance inversion and eigendecomposition, achieving significantly reduced computational complexity while outperforming existing joint communication-and-learning designs in classification accuracy.

This work addresses the limitations of conventional physical-layer designs in large-scale MIMO edge inference systems, which neglect downstream AI task requirements and rely on high-overhead instantaneous channel state information at the transmitter (CSIT). To overcome these issues, the paper proposes a task-oriented precoding method based solely on statistical CSIT. By integrating channel covariance with statistical characteristics of training data, the approach leverages the Maximal Coding Rate Reduction (MCR²) criterion to quantify class separability of received features. Using random matrix theory, it formulates a deterministic optimization objective dependent only on long-term statistics. An efficient solution under per-device power constraints is obtained via projected block coordinate ascent combined with successive convex approximation. Experiments on ModelNet10 demonstrate that the proposed method significantly reduces signaling overhead while outperforming existing benchmarks in both task-aware resource allocation and inference accuracy.

Traditional channel knowledge maps (CKMs) in MIMO systems suffer from insufficient accuracy for beamforming and precoder selection, coupled with high modeling complexity. To address this, we propose a codebook-aware CKM construction method. Our approach explicitly incorporates DFT-based precoding vectors into the CKM generation framework and introduces TransUNet—a hybrid architecture that synergistically combines UNet’s multi-scale local feature extraction with Transformer’s global linear modeling capability—enabling electromagnetic-environment-driven end-to-end CKM regression. Experimental results demonstrate a 17% reduction in root-mean-square error (RMSE) over the state-of-the-art RadioWNet. The method supports real-time CKM generation and is publicly available as open-source code.

This paper addresses robust multi-user multiple-input multiple-output (MU-MIMO) beamforming under imperfect and statistically unknown channel state information (CSI) errors. Method: We propose a hybrid “offline learning + online adaptation” framework: (i) a shared deep neural network implicitly models the error covariance structure, eliminating reliance on prior statistical assumptions; (ii) a sparse-augmented low-rank (SALR) architecture reduces computational complexity; and (iii) a multi-base model-agnostic meta-learning (MB-MAML) strategy enables dynamic optimal initialization and rapid online gradient fine-tuning. Contribution/Results: The proposed method significantly outperforms state-of-the-art approaches under unknown and non-stationary channel conditions. It achieves strong cross-channel generalization, high robustness against CSI uncertainty, and low online computational overhead—demonstrating superior practicality for real-world deployment.

Satellite communications demand high throughput with low feedback overhead. Method: This paper proposes a multi-beam architecture integrating fixed-beam precoding with massive MIMO, modeling user spatial distribution via a Poisson point process and employing asymptotic scaling analysis. Contribution/Results: We establish, for the first time, that when user density scales polynomially with the number of antennas, the system achieves a linear fraction of the optimal achievable rate. Crucially, we show that maintaining asymptotic optimality requires concurrent scaling of beam count and user density. Based on this insight, we derive a closed-form capacity scaling law for multi-beam satellite systems, proving that near-linear capacity scaling is attainable under feasible user-density growth. These results provide fundamental theoretical foundations and design principles for lightweight, real-time onboard resource allocation in next-generation satellite networks.

This work addresses the limitations of conventional channel knowledge graphs (CKGs), which capture only static environments and thus struggle to model time-varying channels induced by dynamic scatterers, terminal orientation changes, and radio-frequency impairments—leading to prohibitively high overhead in acquiring high-dimensional channel state information. To overcome this, the paper proposes a Dynamic Channel Knowledge Graph (Dynamic CKG), establishing for the first time a systematic theoretical framework that serves as an intermediate representation layer bridging static environmental priors and physical-layer signal processing. This framework enables joint pilot design, interference mitigation, and integrated sensing and communication. By integrating geospatial data, time-varying channel modeling, and machine learning–driven graph construction, the approach achieves co-design of CKG and signal processing, significantly reducing channel acquisition overhead while enhancing both communication efficiency and sensing performance, thereby offering a novel paradigm for 6G systems.

This study addresses the performance degradation of OTFS massive MIMO beamforming caused by channel state information (CSI) uncertainty in high-mobility satellite communications. To tackle this issue, a robust optimization framework based on support vector clustering (SVC) is proposed. By integrating three-dimensional discrete dipole approximation (DDA) channel representation with data-driven SVC to construct asymmetric uncertainty sets, the chance constraints are reformulated as deterministic semidefinite programming problems. Furthermore, a GPU-accelerated alternating direction method of multipliers (ADMM) algorithm is designed for efficient computation. Simulation results demonstrate that the proposed scheme significantly reduces transmit power while enhancing system energy efficiency and robustness. The GPU acceleration achieves substantial computational speedups, thereby validating the overall effectiveness of the proposed framework.

为解决6G网络中波束无线电图预测问题，提出BeamRMX框架，通过学习场景几何如何将空间辐射模式转换为接收功率场来实现准确预测。

This work addresses a critical limitation in conventional deep learning-based precoding methods for multi-user MISO systems: their neglect of the invariance of the inner product between channel vectors and precoding vectors under global phase rotation, which leads to inefficient learning and poor generalization. To overcome this, the paper introduces complex projective space into deep precoding design for the first time, explicitly modeling and eliminating global phase redundancy through two parameterization strategies—real-valued embedding and complex hyperspherical coordinates. This enables the neural network to learn geometrically aligned and physically distinguishable mappings. Experimental results demonstrate that, with nearly unchanged model complexity, the proposed approach significantly improves both sum rate performance and generalization capability.

This work addresses the challenge of acquiring high-fidelity, large-scale MIMO channel data in real-world scenarios, which is often prohibitively expensive. To this end, the authors propose a location-conditioned generative framework that, for the first time, integrates diffusion models and flow matching techniques into site-specific MIMO channel synthesis. Specifically, they develop a conditional Denoising Diffusion Implicit Model (cDDIM) and a conditional Flow Matching Model (cFMM), both leveraging user coordinates as conditioning inputs to generate spatially structured channel matrices. Extensive evaluations across multiple scenarios at 28 GHz and 3.5 GHz demonstrate that cFMM achieves generation quality comparable to cDDIM while offering nearly an order-of-magnitude speedup in inference. Moreover, the synthesized channels significantly enhance the performance of downstream physical-layer tasks, such as channel compression and beam alignment.