Xinyue Jiang

Radio maps play a crucial role in optimizing wireless network performance and configuration, providing insights into the spatial distribution of radio frequency signal power. Existing solutions often face challenges in generalizing and adapting across various environments, frequency bands, and vertical dimensions. To overcome these limitations, we propose UniRM, a universal large model designed for constructing multiband 3D radio maps. UniRM leverages large-scale pre-training and prompt learning techniques to accurately generate radio maps across diverse environments, altitudes, and frequency bands. Specifically, UniRM employs a UNet-based encoder-decoder architecture during pre-training to extract universal latent representations that capture shared features across different environmental conditions. A prompt learning module further enhances this by transforming auxiliary inputs, such as environmental descriptions, frequency bands, and altitudes, into discriminative embeddings, thereby enabling effective cross-domain generalization and ensuring robustness in unseen scenarios. Extensive experiments using a large, diverse dataset covering numerous scenarios demonstrate that UniRM outperforms state-of-the-art baselines by over 10% in key metrics, including mean squared error, normalized mean squared error, root mean squared error, and peak signal-to-noise ratio. Notably, zero-shot evaluations highlight UniRM’s strong ability to generalize to new environments without retraining. The code for UniRM is available at: https://github.com/Shirleyue/UniRM.

Jamming signal power detection (JPD) is a crucial step in wireless jamming cognition technology. By detecting the power values of multiple jamming signals, prior information can be provided for the dynamic optimization of anti-jamming strategies, thereby effectively improving the communication system’s resilience to jamming signals. However, in complex electromagnetic countermeasures environments, the radar receivers frequently encounter a multitude of composite jamming signals emitted by multiple jammers, which exhibit significant overlap across temporal and spectral domains. Furthermore, the diversity and uncertainty of multiple jamming signals result in an explosive growth of composite labels, posing challenges to existing deep learning algorithms. Hence, our study presents JPDNet, an innovative framework for detecting multiple jamming signals’ power. The model consists of a deep convolutional encoder/decoder paired with a masks extractor, enabling the reconstruction and power detection of multiple jamming signals. This JPD approach can effectively overcome the limitation of composite labels. The stacked multi-scale temporal convolutional networks (M-TCN) function as the backbone of the masks extractor, aiming to capture the long-term dependencies, and fully extract multi-scale features from the input jamming signals. Great experimental results indicate that, contrasted to the most relevant approaches, JPDNet exhibits superior power detection performance.

It has been demonstrated the advantage of basic stroke recognition algorithm in the field of human-computer interaction (HCI). However, most traditional techniques heavily rely on the touch-contact operations to obtain character information, which limits the further application in non-contact scenario such as germ infection environment, high/low temperature environment or scene for blind human. This paper proposes a non-contact and automatic basic stroke recognition algorithm for handwritten Chinese characters based on frequency modulated continuous wave (FMCW) radar system. First, the radar system collects intermediate frequency (IF) signal of the eight basic strokes given as follows: (horizontal stroke), (dot stroke), (lift stroke), (left falling stroke), (bend stroke), (right falling stroke), (vertical stroke) and (hook stroke). Second, a range-time sequence (RTS) is obtained from IF signal by the window fast Fourier transform (window-FFT) algorithm, and an azimuth-time sequence (ATS) is obtained from IF signal by the frequency domain Capon (FD-Capon) algorithm. Then, the feature area-framing, binarization and open operation (FA-FBO) algorithm is proposed to enhance the features of the above two sequences. After that, a feature map set containing RTS feature map (RTSFM) and ATS feature map (ATSFM) is obtained. Finally, a novel convolutional neural network (CNN) model is customized to perform the strokes classification task with these feature maps as input. Experimental results demonstrate that the proposed scheme is able to effectively recognize the eight basic strokes and achieve an average classification accuracy of 99.25%.

Ground penetrating radar (GPR), as a non-invasive instrument, has been widely used in the civil field. The interpretation of GPR data plays a vital role in underground infrastructures to transfer raw data to the interested information, such as diameter. However, the diameter identification of objects in GPR B-scans is a tedious and labor-intensive task, which limits the further application in the field environment. The paper proposes a deep learning-based scheme to solve the issue. First, an adaptive target region detection (ATRD) algorithm is proposed to extract the regions from B-scans that contain hyperbolic signatures. Then, a Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) framework is developed that integrates Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) network to extract hyperbola region features. It transfers the task of diameter identification into a task of hyperbola region classification. Experimental results conducted on both simulated and field datasets demonstrate that the proposed scheme has a promising performance for diameter identification. The CNN-LSTM framework achieves an accuracy of 99.5% on simulated datasets and 92.5% on field datasets.

Gesture recognition that is based on high-resolution radar has progressively developed in human-computer interaction field. In a radar recognition-based system, it is challenging to recognize various gesture types because of the lacking of gesture transversal feature. In this paper, we propose an integrated gesture recognition system that is based on frequency modulated continuous wave MIMO radar combined with deep learning network for gesture recognition. First, a pre-processing algorithm, which consists of the windowed fast Fourier transform and the intermediate-frequency signal band-pass-filter (IF-BPF), is applied to obtain improved Range Doppler Map. A range FFT based MUSIC (RFBM) two-dimensional (2D) joint super-resolution estimation algorithm is proposed to obtain a Range Azimuth Map to obtain gesture transversal feature. Range Doppler Map and Range Azimuth Map then respectively form a Range Doppler Map Time Sequence (RDMTS) and a Range Azimuth Map Time Sequence (RAMTS) in gesture recording duration. Finally, a Dual stream three-dimensional (3D) Convolution Neural Network combined with Long Short Term Memory (DS-3DCNN-LSTM) network is designed to extract and fuse features from both RDMTS and RAMTS, and then classify gestures with radial and transversal change. The experimental results show that the proposed system could distinguish 10 types of gestures containing transversal and radial motions with an average accuracy of 97.66%.