Hao Wang

Deep learning based methods have been increasingly applied to wireless localization and sensing. However, many existing methods suffer from poor generalization, which limits their adaptability to unseen environments. Additionally, retraining these models demands a substantial amount of labeled data, which is both time-consuming and labor-intensive. To address these challenges, we propose GenLoc, a generalization framework that enables accurate mitigation of ranging and angle estimation errors in unseen environments via feature disentanglement, without the need for additional data collection and online training. The key lies in learning fine-grained domain-invariant representations that are first extracted by minimizing feature distribution discrepancies across various domains and then refined through domain-invariant and domain-specific feature decoupling. Extensive experiments were conducted on real-world datasets, covering five distinct scenarios, each with five different obstacles, and a thorough comparison with several existing approaches illustrates that GenLoc improves average distance and angle estimation accuracy by more than 22% and 28%, respectively, while others exhibit negligible improvements or significant degradation.

Ranging error mitigation is key to improving indoor localization accuracy, but diverse environments challenge algorithm generalization. This letter proposes a multi-source domain adaptation framework for ranging error mitigation, where a weighted maximum mean discrepancy (MMD) is design to dynamically balances domain similarity and regression accuracy during feature alignment. By dynamically aligning domain mappings, the proposed method reduces negative transfer effects caused by varying discrepancies between different source and target domains. Experimental results show that the proposed network outperforms existing deep learning approaches in reducing ranging error across diverse indoor scenarios.

Channel knowledge map (CKM) is viewed as a digital twin of wireless channels, providing location-specific channel knowledge for environment-aware communications. A fundamental problem in CKM-assisted communications is how to construct the CKM efficiently. Current research focuses on interpolating or predicting channel knowledge based on error-free channel knowledge from measured regions, ignoring the extraction of channel knowledge. This paper addresses this gap by unifying the extraction and representation of channel knowledge. We propose a novel CKM construction framework that leverages the received signals of the base station (BS) as online and low-cost data. Specifically, we partition the BS coverage area into spatial grids. The channel knowledge per grid is represented by a set of multi-path powers, delays, and angles, based on the principle of spatial consistency. In extracting these channel parameters, the challenges lie in strong intercell interference and non-linear relationships between received signals and channel parameters. To address these issues, we formulate the problem of CKM construction into a problem of Bayesian inference, employing a interference-activity prior model to characterize the path-loss differences of interferers. Under the Bayesian inference framework, we develop a hybrid messagepassing algorithm for the interference-cancellation-based CKM construction. Based on the CKM, we obtain the joint frequency-space covariance of the user channel and design a CKM-assisted Bayesian channel estimator. The computational complexity of the channel estimator is substantially reduced by exploiting the CKM-derived covariance structure. Numerical results show that the proposed CKM provides accurate channel parameters at low signal-to-interference-plus-noise ratio (SINR) and that the CKM-assisted channel estimator significantly outperforms state-of-the-art counterparts.

How to achieve the best compromise between user experience and energy consumption has become a topic worth studying in both terminal and network sides of 6G mobile communication system. To solve this problem, this paper proposes a Deep Reinforcement Learning (DRL) based smart power saving framework. However, the state-of-art RL models face the problem of slow convergence in new environments, which seriously affects the performance of online usage. To address the generalization ability of the RL model for untrained state-action correlations, this paper proposes a Deep Variational Reinforcement Learning (DVRL) model based on Partially Observable Markov Decision Process (POMDP) and Variational Inference theory. Moreover, in order to represent more accurately the user experience, this paper have introduced the E2E-QoE model. Simulation results demonstrate that the proposed scheme exhibits generalization capability on test sets with unknown partially observable states, which achieves a 23.9% power-saving gain under good signal and 26.7% user experience gain under weak signal. Additionally, road test results indicate that the proposed scheme achieves reductions in average normalized power consumption of 0.063 and 0.117 in good and weak scenarios, respectively, compared to traditional MDP-based DRL.

Deep learning methods have been widely applied in indoor localization to mitigate the impact of non-line-of-sight (NLOS) and multipath effects, achieving high-precision results. However, these methods have to integrate large amounts of data for training and often struggle to adapt to new and unseen scenarios. This letter proposes a domain generalization method based on domain adversarial training (DAT) and mutual information (MI) estimation, where cross-environment generalization localization is achieved by extracting domain-invariant features, eliminating the need for training data from new scenes. Experimental results demonstrate that the proposed method outperforms other deep learning approaches in terms of localization accuracy across various indoor environments.

The fifth-generation (5G) wireless communication signal is ideal for positioning by the user equipment (UE) due to its high precision, low cost, low latency, and easy integration. However, UE-side positioning faces significant challenges in complex electromagnetic environments. Obstacles that block the line-of-sight (LOS) path can severely impact localization accuracy. This paper proposes a deep learning architecture with two-stage inference to achieve high-precision positioning at the user side with multipath channels. For the first time, online learning is achieved without the need to retrain the neural network (NN), making it suitable for implementation on UE. Specifically, the variational inference theory is used to sense the environment and improve positioning accuracy using multipath information. This approach significantly reduces range error in non-line-of-sight (NLOS) channels. Furthermore, a new NN with a neural processes regressor (NPR) is proposed to learn the distribution of ranging bias directly from received signals. The proposed learning architecture and networks can be implemented without retraining, even under different circumstances and environments, making it more computationally efficient than most existing NNs and suitable for practical deployment. Simulation results demonstrate that the proposed approach outperforms conventional techniques with various channels.

In this paper, we consider the topic of multiple-input multiple-output (MIMO) detection for coded systems. Tree search based detectors are well-behaved but have high complexity, while low-cost random sampling based detectors may suffer from distribution mismatch and performance loss. To overcome these issues, we propose a data-driven algorithm, called trained parameter based path sampling (TPbPS), to optimize parameters of the sampling distribution for random sampling based detectors. A more robust approach is designed to avoid sampling repetition and missing by derandomizing the path sampling, where paths are determined by the trained sampling distribution as well as fixed and uniformly-distributed numbers. The error probability is derived for the proposed TPbPS, and it shows that the full diversity can be achieved by elaborately designing a set of sampling path for each quantized noise level. Further, a soft-TPbPS algorithm of computing reliable soft outputs for channel decoders is proposed, via applying end-to-end Bayesian optimization to learn parameters of sampling distribution that goes straight at the objective of optimizing the block error rate performance. Combining these techniques yields enhanced soft MIMO detection designs that non-trivially advance the state-of-the-art, and provide significant performance or complexity gains over traditional methods.

Millimeter wave (mmWave) communication stands as a crucial technology for X-reality and cloud gaming due to its extensive bandwidth. Nevertheless, its user experience suffers from degradation caused by obstacles like humans and objects. Unlike base station (BS) solutions, user-side approaches offer greater flexibility in addressing blockage issues. This paper introduces an innovative beam management architecture named reference-signal-received-power (RSRP)-based blockage-prediction-aided compressive sensing (RSRP-BPCS). Unlike existing methods, our proposed algorithm accounts for real communication system constraints and protocol procedures, ensuring practicality. The key advantage lies in deriving channel information based on RSRP without requiring signal phase knowledge. Additionally, we present two novel strategies on the user side: a multiple-beam-sweeping procedure (MBSP) and a proactive blockage prediction approach. These techniques enable the monitoring of different beam pair links and early anticipation of blockage instances. To achieve rapid recovery, we design an explainable deep learning beam tracking scheme which is joint data and model-based to explore the compressive projection of the power spectrum. The best part is that our strategies are standard-compatible, demanding no supplementary side information, which significantly simplifies implementation with minimal complexity. Through extensive simulations, our proposed algorithm demonstrates exceptional performance in mitigating the impact of blockage, proving its efficacy and practicality.

We propose a learning-aided soft multi-input multi-output (MIMO) detection method, consisting of a search path selection module and a symbol distribution fitting module. The search path selection module uses a neural network to generate a set of candidate paths. Different from traditional tree search algorithms, we utilize the soft estimates output by a neural network to calculate and cancel the symbol interference. The symbol distribution fitting module aims to solve the missing counter-hypotheses problem, where an additional neural network is employed to estimate the marginal posterior probability distribution of each symbol. Numerical results demonstrate the effectiveness of the proposed method especially when the allowed computational complexity is highly limited.