arXiv subjects

Jiawei Zhuang

Publications and source records attributed to Jiawei Zhuang.

14 recordsLinked to original sources

arXiv2026

Toward Secure Communications for a UAV Swarm with Movable Antennas in SAGIN: CKM-Enabled Multi-Agent Reinforcement Learning Framework

Space-air-ground integrated networks (SAGINs) can provide ubiquitous and reliable connectivity for unmanned aerial vehicles (UAVs). However, air-to-ground links, which are typically dominated by line-of-sight (LoS) propagation, are vulnerable to passive eavesdropping due to the broadcast nature of wireless channels. To enhance physical-layer security, we investigate a SAGIN-enabled secure downlink communication system in which UAVs select service links among satellite, aerial, and terrestrial networks while adjusting the positions of the movable antenna (MA) array to fully exploit connectivity and spatial degrees of freedom for improved secrecy communication performance. Specifically, we maximize the secrecy energy efficiency (SEE) of a UAV swarm by jointly optimizing the MA positions, UAV trajectories, and link selections, subject to UAV mobility, MA movement, and link connectivity constraints. To reduce the real-time channel state information (CSI) acquisition overhead, we propose a channel knowledge map (CKM)-assisted multi-agent reinforcement learning framework. Specifically, the CKM is first constructed from sparse channel measurements via Kriging interpolation and is then leveraged together with satellite ephemeris information to enable efficient storage and retrieval of CSI. To reduce the action-space dimensionality and computational complexity, we model the MA array using rigid-body kinematics and adjust its position through global rigid-body translation, thereby constructing a low-dimensional hybrid action space for the joint optimization decisions. To align local decisions with system-wide performance under system constraints, we design an individual-team collaborative reward mechanism and introduce action masks to enforce constraints on UAV mobility, collision avoidance, MA regions, and connectivity capacity.

eess.SP↗

arXiv2026

Toward Alias-Free Channel Extrapolation in Upper Mid-Band Systems: A Spatial-Frequency-Temporal Tensor Learning Approach

Upper mid-band massive multiple-input multiple-output (MIMO) offers a favorable capacity-coverage trade-off for next-generation wireless systems, but its large antenna arrays, wide bandwidths, and faster temporal variation substantially increase the pilot overhead required for accurate channel state information (CSI) acquisition. To reduce this overhead, this paper establishes a tensor-structured multi-domain channel extrapolation framework that exploits the limited-scattering nature of practical propagation environments to recover complete CSI across the spatial-frequency-temporal (SFT) domains from limited observations. Specifically, we develop a Tucker-based SFT-domain signal model to represent the complete CSI, where the factor matrices are parameterized by angle-delay-Doppler (ADD)-domain grids. Thanks to this representation, we reveal that limited SFT-domain observations imposed by uniform pilot patterns and antenna-port selection inherently induce ADD-domain aliasing, so that multiple physically distinct ADD-domain components become indistinguishable within structured ADD aliasing groups. To tackle this issue, we introduce a support-prior-assisted ADD-domain de-aliasing mechanism that leverages coarse-grained support information. Since exact closed-form characterization of this mechanism is difficult to derive, we propose a tensor-structure-aware axial-attention neural network (TANN), which integrates axis-wise attention with a lightweight multi-scale CNN-based gating module to incorporate support priors for ADD-domain de-aliasing. With tensor-structure modeling and mixed-configuration training over different pilot decimation factors, TANN yields a unified model that generalizes across pilot configurations without retraining. Numerical results demonstrate the effectiveness and strong generalization of the proposed framework over benchmark methods under diverse scenarios.

eess.SP↗

arXiv2026

OffQ: Taming Structured Outliers in LLM Quantization by Offsetting

Low-bit quantization has been widely adopted to accelerate the inference of large language models (LLMs) by significantly reducing computational cost and memory usage. However, activation outliers pose a major challenge to effective quantization, often leading to notable performance degradation. In this paper, we introduce OffQ, a method designed to mitigate activation outliers in low-bit quantization through a novel offsetting mechanism. Specifically, OffQ first identifies a low-dimensional outlier subspace in the activations using a proposed top-1 PCA, and then concentrates high-magnitude activations into 1 channel via rotation. OffQ then absorbs this concentrated outlier channel by converting its magnitude into a shared offset, thereby reducing the standard deviation of the activations. This offsetting strategy enables effective W4A4KV4 quantization of LLMs using deployment-friendly uniform-grid and uniform-precision quantization. Extensive experiments across diverse LLM architectures and benchmarks demonstrate that OffQ outperforms state-of-the-art baselines, consistently improving model accuracy while preserving low-bit efficiency.

cs.LG↗

arXiv2026

SSSD: Simply-Scalable Speculative Decoding

Speculative Decoding has emerged as a popular technique for accelerating inference in Large Language Models. However, most existing approaches yield only modest improvements in production serving systems. Methods that achieve substantial speedups typically rely on an additional trained draft model or auxiliary model components, increasing deployment and maintenance complexity. This added complexity reduces flexibility, particularly when serving workloads shift to tasks, domains, or languages that are not well represented in the draft model's training data. We introduce Simply-Scalable Speculative Decoding (SSSD), a training-free method that combines lightweight n-gram matching with hardware-aware speculation. Relative to standard autoregressive decoding, SSSD reduces latency by up to 2.9x. It achieves performance on par with leading training-based approaches across a broad range of benchmarks, while requiring substantially lower adoption effort--no data preparation, training or tuning are needed--and exhibiting superior robustness under language and domain shift, as well as in long-context settings.

cs.CL↗

arXiv2026

KVarN: Variance-Normalized KV-Cache Quantization Mitigates Error Accumulation in Reasoning Tasks

Test-time scaling is a powerful approach to obtain better reasoning in large language models, but it becomes memory-bottlenecked during long-horizon decoding, as the KV-cache grows. KV-cache quantization can help improve this, but current methods are evaluated under prefill-like settings and errors behave differently under autoregressive decoding. We show that in the latter regime, quantization errors accumulate across timesteps, driven primarily by incorrect token scales. We introduce KVarN, a calibration-free KV-cache quantizer that applies a Hadamard rotation followed by a dual-scaling variance normalization across both axes of the K and V matrices. We find that this combination fixes outlying token-scale errors and substantially reduces error accumulation over existing baselines. KVarN establishes a new state-of-theart for KV-cache quantization on generative benchmarks, including MATH500, AIME24 and HumanEval, at 2-bit precision. A vLLM implementation of the KVarN method is available at https://github.com/huawei-csl/KVarN

cs.LG↗

arXiv2026

Fast and Stable Triangular Inversion for Delta-Rule Linear Transformers

Linear attention has emerged as a cornerstone for efficient long-context architectures, as evidenced by its integration into state-of-the-art open-source models including Qwen3.5/3.6, Kimi Linear, and RWKV-7. Models that incorporate linear attention layers with the so-called Delta-Rule involve the inversion of triangular matrices as a core sub-routine. This operation often forms a performance bottleneck, and, due to its high-sensitivity to numerical errors, it can significantly deteriorate end-to-end model accuracy if it is not carefully implemented. This work provides a systematic analysis of both direct and iterative triangular inversion algorithms, targeting methods that are rich in matrix products, and, therefore, have the potential to efficiently utilize modern hardware. To that end, our analysis covers a broad spectrum of mathematical and practical aspects, with a heavy focus on numerical stability, computational complexity, and, ultimately, hardware efficiency and practical considerations. We provide a rigorous experimental evaluation to verify these properties in practical scenarios, and in low-precision floating-point representations, highlighting the strengths and limitations of each method. Performance benchmarks on NPUs reveal up to $4.3\times$ speed-up against the state-of-the-art implementations of SGLang for triangular matrix inversion, leading to significant performance improvements on the entire layer level, while maintaining full end-to-end model accuracy.

cs.LG↗

arXiv2026

A Channel Knowledge Map-Driven Two-Stage Coordinated User Scheduling in Multi-Cell Massive MIMO Systems

This paper investigates narrowband coordinated user scheduling in multi-cell massive multiple-input multiple-output (MIMO) systems. We formulate the problem under a spectral-efficiency maximization criterion, revealing inherent challenges in computational complexity and signaling overhead. To address these, we develop a user-scheduling-oriented CKM (US-CKM) and a US-CKM-driven two-stage coordinated scheduling framework. By exploiting the mapping between location information and statistical channel state information (SCSI), the system enables rapid SCSI retrieval and persistent reuse, substantially reducing CSI acquisition overhead. Embedding statistical channel correlation into the CKM further characterizes interuser interference patterns. The framework designs an intra-cell active-user selection scheme for the first stage and an inter-cell coordinated scheduling scheme for the second, both based on US-CKM entries. The first stage identifies users with favorable channel gains and low intra-cell interference, reducing the candidate set with marginal sum-rate loss. The second stage suppresses inter-cell interference (ICI) by exploiting cross-cell channel correlations. To enhance robustness against imperfect SCSI in dynamic scattering environments, we augment the framework with a reliability-guided mechanism. Instead of uniform treatment, we evaluate entry stability using a grid reliability metric quantifying channel measurement variance at sampling locations. Low-reliability grids are identified, and their instantaneous CSI is acquired in real time to integrate with existing SCSI. This process refines channel gain and spatial correlation characteristics, ensuring robust performance under imperfect conditions.

eess.SP↗

arXiv2026

TyphoonMLA: A Mixed Naive-Absorb MLA Kernel For Shared Prefix

Multi-Head Latent Attention (MLA) is a recent attention mechanism adopted in state-of-the-art LLMs such as DeepSeek-v3 and Kimi K2. Thanks to its novel formulation, MLA allows two functionally equivalent but computationally distinct kernel implementations: naive and absorb. While the naive kernels (e.g., FlashAttention) are typically preferred in training and prefill for their computational efficiency, existing decoding kernels (e.g., FlashMLA) rely on the absorb method to minimize HBM bandwidth usage. However, the compute-bound nature of the absorb implementations prohibits performance benefits from data reuse opportunities in attention calculations, such as shared prefixes. In this work, we introduce TyphoonMLA, a hybrid approach that combines naive and absorb formulations to harness the strengths of both. TyphoonMLA effectively leverages the shared prefix by applying the naive formulation to the compute-bound parts of attention calculations, while reducing the bandwidth requirements for non-shared parts by using the absorb formulation. As a result, TyphoonMLA improves the throughput of attention calculations in MLA architectures by up to 3x and 3.24x on NPU and GPUs, with only a 3% overhead in HBM size.

cs.LG↗

arXiv2026

SINQ: Sinkhorn-Normalized Quantization for Calibration-Free Low-Precision LLM Weights

Post-training quantization has emerged as the most widely used strategy for deploying large language models at low precision. Still, current methods show perplexity degradation at bit-widths less than or equal to 4, partly because representing outliers causes precision issues in parameters that share the same scales as these outliers. This problem is especially pronounced for calibration-free, uniform quantization methods. We introduce SINQ to augment existing post-training quantizers with an additional second-axis scale factor and a fast Sinkhorn-Knopp-style algorithm that finds scales to normalize per-row and per-column variances. We show that this approximates activation-aware quantization by recovering column scales from the weight matrix structure that are predictive of the typical activation magnitudes the matrix received during training. Our method has no interactions between layers and can be trivially applied to new architectures to quantize any linear layer. We evaluate our method on the Qwen3 model family, among others. SINQ reduces the perplexity gap on WikiText2 and C4 by over 50% against uncalibrated uniform quantization baselines, incurs zero to negligible compute overhead, and can be further enhanced by combining it with calibration and non-uniform quantization levels. Code is available at https://github.com/huawei-csl/SINQ.

cs.LG↗

arXiv2025

DMRS-Based Uplink Channel Estimation for MU-MIMO Systems with Location-Specific SCSI Acquisition

With the growing number of users in multi-user multiple-input multiple-output (MU-MIMO) systems, demodulation reference signals (DMRSs) are efficiently multiplexed in the code domain via orthogonal cover codes (OCC) to ensure orthogonality and minimize pilot interference. In this paper, we investigate uplink DMRS-based channel estimation for MU-MIMO systems with Type II OCC pattern standardized in 3GPP Release 18, leveraging location-specific statistical channel state information (SCSI) to enhance performance. Specifically, we propose a SCSI-assisted Bayesian channel estimator (SA-BCE) based on the minimum mean square error criterion to suppress the pilot interference and noise, albeit at the cost of cubic computational complexity due to matrix inversions. To reduce this complexity while maintaining performance, we extend the scheme to a windowed version (SA-WBCE), which incorporates antenna-frequency domain windowing and beam-delay domain processing to exploit asymptotic sparsity and mitigate energy leakage in practical systems. To avoid the frequent real-time SCSI acquisition, we construct a grid-based location-specific SCSI database based on the principle of spatial consistency, and subsequently leverage the uplink received signals within each grid to extract the SCSI. Facilitated by the multilinear structure of wireless channels, we formulate the SCSI acquisition problem within each grid as a tensor decomposition problem, where the factor matrices are parameterized by the multi-path powers, delays, and angles. The computational complexity of SCSI acquisition can be significantly reduced by exploiting the Vandermonde structure of the factor matrices. Simulation results demonstrate that the proposed location-specific SCSI database construction method achieves high accuracy, while the SA-BCE and SA-WBCE significantly outperform state-of-the-art benchmarks in MU-MIMO systems.

eess.SP↗

arXiv2025

PRESERVE: Prefetching Model Weights and KV-Cache in Distributed LLM Serving

Large language models (LLMs) are typically served from clusters of GPUs/NPUs that consist of large number of devices. Unfortunately, communication between these devices incurs significant overhead, increasing the inference latency and cost while limiting the scalability. Prior work addressed this issue by overlapping communication with compute, but has severe limitations due to the data dependencies between these operations. In this paper, we propose PRESERVE, a novel framework that prefetches model weights and KV-cache from off-chip HBM memory to the on-chip cache of AI accelerators during the communication operations, which offers various advantages and performance improvements compared to prior methods. Through extensive experiments conducted on commercial AI accelerators, we demonstrate up to 1.6x end-to-end speedup on state-of-the-art, open-source LLMs. Additionally, we perform a design space exploration that identifies the optimal hardware configuration for the proposed method, showing a further 1.25x improvement in performance per cost by selecting the optimal L2 cache size. Our results show that PRESERVE has the potential to mitigate the memory bottlenecks and communication overheads, offering a solution to improve the performance and scalability of the LLM inference systems.

cs.AI↗

arXiv2024

Joint Channel Estimation and Prediction for Massive MIMO with Frequency Hopping Sounding

In massive multiple-input multiple-output (MIMO) systems, the downlink transmission performance heavily relies on accurate channel state information (CSI). Constrained by the transmitted power, user equipment always transmits sounding reference signals (SRSs) to the base station through frequency hopping, which will be leveraged to estimate uplink CSI and subsequently predict downlink CSI. This paper aims to investigate joint channel estimation and prediction (JCEP) for massive MIMO with frequency hopping sounding (FHS). Specifically, we present a multiple-subband (MS) delay-angle-Doppler (DAD) domain channel model with off-grid basis to tackle the energy leakage problem. Furthermore, we formulate the JCEP problem with FHS as a multiple measurement vector (MMV) problem, facilitating the sharing of common CSI across different subbands. To solve this problem, we propose an efficient Off-Grid-MS hybrid message passing (HMP) algorithm under the constrained Bethe free energy (BFE) framework. Aiming to address the lack of prior CSI in practical scenarios, the proposed algorithm can adaptively learn the hyper-parameters of the channel by minimizing the corresponding terms in the BFE expression. To alleviate the complexity of channel hyper-parameter learning, we leverage the approximations of the off-grid matrices to simplify the off-grid hyper-parameter estimation. Numerical results illustrate that the proposed algorithm can effectively mitigate the energy leakage issue and exploit the common CSI across different subbands, acquiring more accurate CSI compared to state-of-the-art counterparts.

eess.SP↗

arXiv2020

Learned discretizations for passive scalar advection in a 2-D turbulent flow

The computational cost of fluid simulations increases rapidly with grid resolution. This has given a hard limit on the ability of simulations to accurately resolve small scale features of complex flows. Here we use a machine learning approach to learn a numerical discretization that retains high accuracy even when the solution is under-resolved with classical methods. We apply this approach to passive scalar advection in a two-dimensional turbulent flow. The method maintains the same accuracy as traditional high-order flux-limited advection solvers, while using 4x lower grid resolution in each dimension. The machine learning component is tightly integrated with traditional finite-volume schemes and can be trained via an end-to-end differentiable programming framework. The solver can achieve near-peak hardware utilization on CPUs and accelerators via convolutional filters. Code is available at https://github.com/google-research/data-driven-pdes.

physics.comp-ph↗

arXiv2020

A Two-Stream Meticulous Processing Network for Retinal Vessel Segmentation

Vessel segmentation in fundus is a key diagnostic capability in ophthalmology, and there are various challenges remained in this essential task. Early approaches indicate that it is often difficult to obtain desirable segmentation performance on thin vessels and boundary areas due to the imbalance of vessel pixels with different thickness levels. In this paper, we propose a novel two-stream Meticulous-Processing Network (MP-Net) for tackling this problem. To pay more attention to the thin vessels and boundary areas, we firstly propose an efficient hierarchical model automatically stratifies the ground-truth masks into different thickness levels. Then a novel two-stream adversarial network is introduced to use the stratification results with a balanced loss function and an integration operation to achieve a better performance, especially in thin vessels and boundary areas detecting. Our model is proved to outperform state-of-the-art methods on DRIVE, STARE, and CHASE_DB1 datasets.

cs.CV↗