BTBowen Tang

Bowen Tang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Bowen Tang has a total of 16 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • GTFA-Net: A Generative Network for High-Resolution Time–Frequency Analysis in Dual-Frequency Continuous-Wave Radar Tracking

    by Yipeng Ding, Minhao Ding, Zhixiang Zhang, Runjin Liu, Bowen Tang

    Dual-frequency continuous wave (DFCW) radar is a promising technology for indoor target tracking, yet its performance is constrained by the low time-frequency resolution (TFR) and frequency ambiguity of traditional time-frequency analysis (TFA) methods, such as the short-time Fourier transform. To overcome these challenges, this paper proposes a novel semi-supervised network named GTFA-Net, which integrates a non-local mechanism for enhanced TFR. The proposed GTFA-Net integrates a Various Noise Maker (VNM), a Multi-Scale Short-Time Module (MDSTM), and an Energy Concentration Module (ECM) with a non-local mechanism. The proposed VNM enhances the robustness of the network by adding different types of noise to the input signal, while the MDSTM improves the richness of time-frequency information by introducing multiple short-time filtering windows. The ECM, incorporating an enhanced non-local attention mechanism, effectively extracts global time-frequency features. Experimental results demonstrate that GTFA-Net significantly outperforms traditional TFA methods in terms of noise immunity, energy concentration, and instantaneous frequency extraction accuracy, making it a promising solution for high-precision indoor target tracking.

  • P2S-Pose: A Physics Informed RF Sensing Framework for Structure Aware 3D Human Pose Estimation

    by Runjin Liu, Yipeng Ding, Bowen Tang, Zhaoyang He

    Human pose estimation (HPE) from radio-frequency (RF) signals is challenged by propagation distortions, inter frequency coupling, and the lack of synergy between physical layer sensing and structural reasoning. In this letter, we propose P2S-Pose, a physics informed RF sensing framework that unifies signal modeling and structure aware inference. A Physics Informed Sensing Enhancement (PISE) stream disentangles frequency coupling, models propagation delays, and calibrates channel responses, yielding stable priors for volumetric inference. Guided by these cues, a Structure Aware Voxel Reasoning (SAVR) stage reconstructs per-bone volumetric fields with part aware supervision, producing interpretable and geometrically consistent intermediates. Compact bone and joint tokens are then fused by a Hierarchical Graph Fusion (HGF) module that performs cross level interaction and topology aware graph reasoning to regress anatomically valid 3D poses. Extensive experiments demonstrate that P2S-Pose achieves high accuracy, strong robustness, and effective generalization across subjects and actions, advancing RF-based 3D human sensing toward human centric wireless perception.

  • Multi-resolution fractional extracting transform for target localization by Doppler radar

    by Yipeng Ding, Guangxin Dongye, Minhao Ding, Bowen Tang

    Accurate estimation of the instantaneous frequency (IF) is essential for target localization using Doppler through-wall radar (TWR). However, conventional time-frequency analysis (TFA) techniques suffer from severe time-frequency aliasing when multiple signal components intersect, leading to inaccurate IF estimation and degraded localization performance. To address this problem, this paper proposes a multi-resolution fractional extracting transform (MRFET). First, the initial rotation angle parameter is estimated via the short-time fractional Fourier transform (STFRFT), based on which a fractional Gaussian window (FGW) is constructed within a local rotation angle range. Subsequently, the short-time Fourier transform (STFT) is performed on the signal by FGWs with different variance parameters, and a multi-resolution superimposition strategy is employed to integrate key features from multiple time-frequency distribution (TFD) results. Finally, a novel local frequency extraction operator is applied to the fused TFD to enhance energy concentration further and effectively suppress time-frequency aliasing interference. Experimental results demonstrate that, compared to conventional methods, MRFET significantly improves both IF estimation and target localization accuracy, while exhibiting high robustness against noise.

  • Continuous Activity Recognition Algorithm for FMCW Radar Integrating Consistency Regularization and Self-Distillation Techniques

    by Minhao Ding, Yipeng Ding, Zhixiang Zhang, Runjin Liu, Bowen Tang

    In recent years, Frequency-Modulated Continuous Wave (FMCW) radar-based continuous human activity recognition (HAR) has attracted significant attention due to its privacy preserving and robust capabilities. However, existing methods often focus on isolated actions, neglecting transitional phases and real world dynamics. To overcome these limitations, a continuous HAR algorithm, DST-Net, is introduced. It combines a two-branch self-distillation multi-stage temporal convolutional network with a spectrum enhancement strategy, incorporating elements like time warping, time masking, and frequency masking. This strategy significantly improves the model's robustness by augmenting the dataset and preventing over-reliance on specific input features. The proposed loss function of DST-Net is comprised of three components: classification loss, consistency loss, and continuous temporal smoothness loss. The consistency loss boosts agreement between the two branches, the continuous temporal smoothness loss helps in minimizing sudden changes during activity transitions, and the classification loss is employed for continuous HAR. Experimental results show that proposed DST-Net achieves a 69.7% F1 score, 74.6 edit distance, and 95.6% classification accuracy, outperforming existing advanced methods. Quantitative analysis reveals a 5% and 4% improvement in F1 score in three different environments, validating its robustness and generalization ability. These innovations highlight DST-Net's effectiveness in addressing the limitations of existing continuous HAR algorithms, offering a comprehensive solution for real-world applications.


  • Doppler Through-Wall Radar target localization based on Local optimum Time-Frequency Chirp Reassignment

    by Bowen Tang Tang, Yipeng Ding, Runjin Liu, Minhao Ding

    Doppler-based Through-Wall Radar (TWR) has shown great potential for non-contact target localization. However, in scenarios involving multiple moving targets, closely spaced or intersecting instantaneous frequency (IF) components pose significant challenges to conventional time-frequency analysis (TFA), resulting in poor localization accuracy. This paper proposes a novel algorithm-Local Optimum Time-Frequency Chirp Reassignment (LPFCR)-to address this issue. The method begins with a Short-Time Fourier transform (STFT) for initial time frequency representation, followed by region segmentation based on energy concentration to extract representative local features. A chirp-based local synchronization strategy is then introduced to perform high-dimensional reassignment in both frequency and chirp-rate domains, enhancing resolution and reducing time frequency aliasing. Experimental results demonstrate that LPFCR significantly improves IF estimation and target localization accuracy compared to traditional methods.

  • Application of STFT–NGRC Joint Algorithm for Doppler Through-Wall Radar Target Tracking

    by Runjin Liu, Yipeng Ding, Yiqun Peng, Bowen Tang, Minhao Ding

    Estimation accuracy of target Instantaneous Frequency (IF) significantly affects the tracking performance of a through wall radar (TWR). In this paper, a novel theoretical approach to estimate IF of human targets is proposed with a united application of the short-time Fourier transform (STFT) and a modified next-generation reservoir computing (NGRC) system, which can be very promising in real-time human tracking applications. For this approach, the STFT is first applied to identify aliasing areas of radar echoes. Then, a NGRC dynamic learning system with adaptive error compensation processing is proposed for target IF prediction on these aliasing areas. Finally, the NGRC output weights are updated and the prediction process is iteratively performed until the IF estimation results of all the aliasing intervals are acquired. Compared with traditional methods based on fixed output weights throughout the entire prediction process, the proposed approach can more effectively suppress the cumulative errors and improve the target tracking performance of a TWR. A series of experimental measurements are presented as a preliminary assessment of the proposed approach.


  • Cross-Attention Augmented End-to-End Architecture for SISO FMCW Radar Activity Recognition With Swin-Transformer

    by Minhao Ding, Bowen Tang, Yiqun Peng, Runjin Liu, Yipeng Ding

    Frequency-Modulated Continuous-Wave (FMCW) radar is extensively used in human activity recognition (HAR) due to its non-intrusive operation, all-weather performance, and anti-interference ability. However, FMCW signals predominantly relies on the Discrete Fourier Transform (DFT) method, which is limited by its fixed basis functions and restricted time-frequency resolution (TFR). Therefore, this paper proposes a cross-attention augmented end-to-end architecture for FMCW radar HAR called CASA-SWIN. The preprocessing module of CASA-SWIN learns the potential Range-Time (R-T), Range-Doppler (R-D), Dopper-Time(D-T) features of rawdata through multiple complex fully connected layers with preset parameters. Furthermore, it utilizes a three-branch cross-attention mechanism (TBCAM) module to extract intrinsic related information for R-T, D-T, R-D branches. For the D-T branch, proposed multi-window filtering module (MWFM) comprises multiple 1D complex convolutions with varying kernel sizes, emulating different window lengths in the Short-Time Fourier Transform (STFT). Finally, a three-branch swin-transformer is used as the HAR classifier. Moreover, proposed CASA-SWIN is evaluated on self-collected HAR, Gait recognition datasets, and open-source gesture recognition datasets. Compared to 12 advanced HAR algorithms, a series of experiments including comparative experiments, cross-environment experiments, open-source datasets experiments, and ablation experiments verify the effectiveness of the proposed module and CASA-SWIN.


  • A novel algorithm for mitigating time–frequency aliasing in Doppler through-wall radar

    by Bowen Tang, Yipeng Ding, Runjin Liu, Minhao Ding

    In recent years, Doppler through-wall radar has been widely used in target localization. However, time–frequency aliasing occurs, which poses a great challenge to the estimation of target parameters. To solve this problem, we propose a novel time–frequency enhancement algorithm. In this paper, we roughly identify the time-frequency features of the interested target component from the time–frequency distribution of the local maximum synchrosqueezing transform and construct a series of discrete demodulation operators that approximate the true properties. Then an optimized window function with adaptive parameters (window width) is designed according to its time–frequency features, which can help effectively compress the echo energy from the selected target component. Finally, an instantaneous frequency (IF) estimator with a local maximum extraction criterion is designed to reduce the noise interference. The experimental results demonstrate that the presented algorithm successfully mitigates time–frequency aliasing, significantly enhancing the accuracy of IF estimation and target localization.


  • Local Maximum Reassignment Chirplet Basis Transform With Doppler Through-Wall Radar Target Localization

    by Bowen Tang, Yipeng Ding, Yiqun Ping, Runjin Liu, Minhao Ding

    In recent years, Doppler through-wall radar (TWR) have shown its potential in Internet of Things (IoT) applications like target positioning, smart security, and health monitoring. However, traditional time-frequency analysis (TFA) struggles with close-range or crossed instantaneous frequency (IF) from multiple targets, reducing positioning accuracy. To address this, we propose a novel algorithm, local maximum reassignment chirplet basis transform (LMRCBT). First, the time-frequency characteristics of the target signal are analyzed using short-time Fourier transform (STFT) and the time-frequency distribution (TFD) is chunked to ensure local consistency. Next, we introduce the chirplet basis function combined with a kurtosis-based local optimal chirp domain theory to capture the optimal TFD and enhance signal energy concentration. Finally, we design a time-frequency reassignment operator (RO) to optimize the redistribution of time frequency coefficients, improving resolution and boosting noise suppression. The main contribution of LMRCBT is its effective handling of nonstationary signals, especially those with close-range, nonproportional, or crossed IF, improving target localization accuracy and suppressing time-frequency aliasing, as shown in experiments.