arXiv subjects

Peigang Jiang

Publications and source records attributed to Peigang Jiang.

3 recordsLinked to original sources

arXiv2026

Non-Asymptotic Performance Guarantees for MLE under a Gumbel-Softmax Index Model with Phaseless Measurements

This paper studies the recovery of a complex-valued signal from index-only observations generated by a phaseless Gumbel-Softmax model. The model is motivated by limited-feedback frequency-division duplex massive MIMO systems, where it provides a smooth probabilistic surrogate for the deterministic hard-PMI selection rule. We consider a constrained maximum likelihood estimator and establish non-asymptotic statistical guarantees for its recovery performance. Under a bounded-design condition, we first derive a global excess-risk bound of order $\sqrt{d/T}$, which does not explicitly depend on the number of candidate indices. A quantitative local identifiability condition is then introduced to characterize the regime in which sharper local guarantees are available. Under this condition, the excess risk and squared parameter error both scale as $d/T$. The condition is further verified with high probability for a Haar--Stiefel random design, and a local minimax lower bound is established that matches the upper bound in its dependence on the sample size, signal dimension, temperature parameter, and signal norm, up to design-dependent factors. Numerical experiments on FDD downlink channel estimation demonstrate that the proposed likelihood-based approach can overall achieve superior reconstruction performance, supporting the effectiveness of the Gumbel-Softmax model as a surrogate for hard PMI feedback.

cs.IT↗

arXiv2025

Enhanced Fingerprint-based Positioning With Practical Imperfections: Deep learning-based approaches

High-precision positioning is vital for cellular networks to support innovative applications such as extended reality, unmanned aerial vehicles (UAVs), and industrial Internet of Things (IoT) systems. Existing positioning algorithms using deep learning techniques require vast amounts of labeled data, which are difficult to obtain in real-world cellular environments, and these models often struggle to generalize effectively. To advance cellular positioning techniques, the 2024 Wireless Communication Algorithm Elite Competition as conducted, which provided a dataset from a three-sector outdoor cellular system, incorporating practical challenges such as limited labeled-dataset, dynamic wireless environments within the target and unevenly-spaced anchors, Our team developed three innovative positioning frameworks that swept the top three awards of this competition, namely the semi-supervised framework with consistency, ensemble learning-based algorithm and decoupled mapping heads-based algorithm. Specifically, the semi-supervised framework with consistency effectively generates high-quality pseudo-labels, enlarging the labeled-dataset for model training. The ensemble learning-based algorithm amalgamates the positioning coordinates from models trained under different strategies, effectively combating the dynamic positioning environments. The decoupled mapping heads-based algorithm utilized sector rotation scheme to resolve the uneven-spaced anchor issue. Simulation results demonstrate the superior performance of our proposed positioning algorithms compared to existing benchmarks in terms of the {90%, 80%, 67%, 50%} percentile and mean distance error.

eess.SP↗

arXiv2025

Massive MIMO-OFDM Channel Acquisition with Time-Frequency Phase-Shifted Pilots

In this paper, we propose a channel acquisition approach with time-frequency phase-shifted pilots (TFPSPs) for massive multi-input multi-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. We first present a triple-beam (TB) based channel tensor model, allowing for the representation of the space-frequency-time (SFT) domain channel as the product of beam matrices and the TB domain channel tensor. By leveraging the specific characteristics of TB domain channels, we develop TFPSPs, where distinct pilot signals are simultaneously transmitted in the frequency and time domains. Then, we present the optimal TFPSP design and provide the corresponding pilot scheduling algorithm. Further, we propose a tensor-based information geometry approach (IGA) to estimate the TB domain channel tensors. Leveraging the specific structure of beam matrices and the properties of TFPSPs, we propose a low-complexity implementation of the tensor-based IGA. We validate the efficiency of our proposed channel acquisition approach through extensive simulations. Simulation results demonstrate the superior performance of our approach. The proposed approach can effectively suppress inter-UT interference with low complexity and limited pilot overhead, thereby enhancing channel estimation performance. Particularly in scenarios with a large number of UTs, the channel acquisition method outperforms existing approaches by reducing the normalized mean square error (NMSE) by more than 8 dB.

cs.IT↗