Impulse radio ultra-wideband (IR-UWB) radar has the advantages of low cost, high resolution, and independence of light and weather conditions. Its potential in human activity recognition (HAR) for IoT device sensing draws interest. One challenge in this domain is effectively representing spatial static and temporal dynamic information in echo sequences. Transformers, used extensively in NLP and CV, have powerful sequence long-range dependency modeling capabilities. However, in the field of radar HAR, the application research of Transformers is still insufficient. In addition, there is currently a lack of publicly available IR-UWB radar human action datasets. To this end, we proposed various fine-grained feature image calculation methods and designed an IR-UWB Radar Human Activity dataset (IURHA2023). This paper presents a Swin-Transformer encoder combining cosine similarity attention and patch overlap to obtain deep spatio-temporal features of human action feature images. Compared with other proposed Transformer models or traditional CNNs and RNNs, the improved Swin-Transformer encoder performs better. To further improve the feature learning capability of the backbone network and the robustness to echo variations, we propose a supervised contrastive learning-enhanced Swin-Transformer (SCL-SwinT). It obtains distinctions and compact embeddings by comparing the similarities of positive and negative examples partitioned according to labels. Experimental results on the IURHA2023 dataset show that SCL-SwinT achieves a recognition rate exceeding 90%, and the inference speed on IoT edge devices satisfies real-time applications. Ablation experiments demonstrate the effectiveness of the proposed components. In addition, SCL-SwinT exhibits good robustness to environmental factors like noise, multipath, and distance.

The satisfaction of various basic requirements of cognitive radar by Ultra-wideband (UWB) signals makes UWB cognitive radar attract extensive attention. The variety and large dynamic range of jamming in the UWB spectrum range make jamming identification critical and challenging. However, the traditional method has low recognition accuracy, high computational complexity, and difficulty in multi-signal recognition. In this paper, we propose a multi-scale attention network (MANet) for carrier-free UWB cognitive radar to identify target signals and nine types of jamming signals. MANet extracts different fine features by multi-scale dilation convolution. The features are stitched together in the channel dimension. The subtle features that are beneficial for recognition are then substantially enhanced using channel attention blocks. The proposed method combines the time domain and frequency domain features to improve the recognition performance by using the powerful feature extraction ability and generalization ability of MANet. Simulation results show that the overall recognition accuracy of the method is 93.1%, with less storage space, shorter FLOPs and inference time than the five recognition methods, and better and more stable recognition performance is also achieved at low jamming-to-noise ratios (JNR).

Unmanned aerial vehicles (UAVs) are used in different scenarios. The need for intelligent and accurate recognition of targets using UAV-based radars is increasingly urgent. As unique information of targets, micro-Doppler characteristics have become an important basis for target recognition. For three typical ground targets including pedestrians, wheeled vehicles and tracked vehicles, the micro-Doppler modulation of the UAV vibration on echo signals is analyzed firstly in this paper. Secondly, singular value decomposition (SVD) is adopted to suppress ground clutter and environmental noise. In order to decrease the computation and increase the contrast, all colored time-frequency diagrams of micro-Doppler signals are transferred into grayscale images. Thirdly, a multi-level target recognition method based on the transfer learning of deep convolutional neural networks (DCNNs) is proposed. The first-level recognition is designed to accurately distinguish pedestrians, wheeled vehicles and tracked vehicles, while the purpose of the second-level of recognition is to determine the three states of pedestrians including stepping, walking and jogging. At last, experiments are carried out using a designed Doppler radar to prove the effectiveness of proposed method. Results indicate that method in this paper not only avoids the complex feature extraction process, but also has higher accuracy and better robustness.

In order to achieve precise operations on specified targets from the unmanned aerial vehicles (UAVs), classifying ground targets correctly is especially important. Micro-Doppler effect which provides unique information of targets has been the basis for targets classification. Due to the effect of ground clutter, noise and complex signal modulation, enhancing micro-Doppler features of UAV-to-ground targets is necessary for accurate classification. This paper firstly establishes the models of UAV-to-ground targets including wheeled vehicles, tracked vehicles and pedestrians to analyze their micro-Doppler differences. Secondly, Principal Components Analysis (PCA) is utilized to remove the ground clutter. Compared with other algorithms, PCA can use a small amount of calculation to remove the ground clutter while retain nearby micro-Doppler signals. Then, micro-Doppler signals are sparsely represented based on Fourier basis. Orthogonal Matching Pursuit (OMP) is chosen to reconstruct micro-Doppler components and refine spectral lines after random projection. The three steps make up Compressed Sensing (CS) together. At last, non-linear transform of Doppler spectrum is conducted to further enhance the distinction of micro-Doppler spectral lines. Distinguishing micro-Doppler features are extracted from pre-processed micro-Doppler signals, which eventually contributes to the accurate targets classification. Comparison with other methods is also made to prove the robustness and anti-noise performance of proposed method.

In this study, we introduce MobileViTX, an enhanced MobileViT architecture for human activity recognition in impulse radio ultra-wideband (IR-UWB) radar applications. MobileViT is a lightweight Vision Transformer mainly consisting of MobileViT blocks and MobileNetv2 blocks. Modifications to the MobileNetv2 block include adding a Drop Path and an SE module and altering activation functions to hard-sigmoid and hard-swish. Additionally, the self-attention in the MobileViT block is transformed to possess linear complexity. These adjustments aim to accelerate inference while preserving high accuracy. We experiment with a dataset from 20 individuals performing 20 distinct actions, using 5-fold cross-validation to assess our model’s performance. Results show MobileViTX outperforms the original MobileViT and other models in both recognition rate and efficiency.

The Chaotic Phase-modulated Linear Frequency modulation Short-range Detection System (CPLF-SrDS) is a new type of composite detection system. Compared with traditional methods, it benefits from the short-term randomness and long-term certainty of chaos, and is more powerful in determining the distance. To design such a system, the key problem lies in robustly recovering the echo signal against externally suppressed interferences. Combining 4th-order cumulant and joint diagonalization, this paper first develops a new signal separation approach to realize an underdetermined separation of the echo signal and interference signal from their mixture. Then, both bispectrum analysis and singular spectrum analysis are adopted to analyze the separated signals and extract their characteristic parameters. Further, the GA-BP network is utilized to classify the echo and interference signals. Finally, the anti-interference analysis of this CPLF-SrDS is realized. Experimental results show the effectiveness of the proposed method in the presence of externally suppressed amplitude modulation (AM) interference and frequency modulation (FM) interference.