31 December 2021 in IEEE Transactions on Neural Networks and Learning Systems
Automatic modulation classification (AMC) is an essential part in a cognitive radio receiver. Benefited from the discriminative constellation characteristics among most modulations, AMC methods based on constellation diagrams usually achieve pleasant performance. However, in noncooperation communication systems, constellation diagrams expressing modulations explicitly are difficult to obtain via blind symbol timing synchronization, especially in complicated wireless channels. Therefore, this article proposes a novel constellation diagram-based AMC architecture called attentive Siamese networks (ASNs) by considering multitiming constellation diagrams (MCDs) and selecting the proper symbol timings at the feature level, which is a more robust way than the conventional signal-level symbol timing synchronization. In detail, convolutional neural networks sharing the same parameters first extract deep feature vectors for MCDs. Then, an attention inference module weights all the deep feature vectors. Finally, AMC is realized based on the weighted feature vectors. Moreover, the ASN architecture can be trained end-to-end. Comparing with the state-of-the-art methods that take diverse representations of received baseband signals as input, experimental results based on the RadioML 2018.01A dataset and non-Gaussian noise dataset demonstrate that ASN achieves a remarkable improvement, whose classification accuracy goes over 99% when the signal-to-noise ratio (SNR) > 10 dB.
26 May 2016 in IEEE Transactions on Multimedia
The “color-plus-depth” format represents a 3D scene using multiple color and depth images captured by an array of closely spaced cameras. Using this format, a novel image as observed from a horizontally shifted virtual viewpoint can be synthesized via depth-image-based rendering (DIBR), using neighboring camera-captured viewpoint images as reference. In this paper, using the same popularized color-plus-depth representation, we propose to construct, in addition, novel images as observed from virtual viewpoints closer to the 3D scene, enabling a new dimension of view navigation. To construct this new image type, we first perform a new DIBR pixel-mapping for z-dimensional camera movement. We then identify expansion holes-a new kind of missing pixels unique in z-dimensional DIBR-mapped images-using a depth layering procedure. To fill expansion holes we formulate a patch-based maximum a posteriori problem, where the patches are appropriately spaced using diamond tiling. Leveraging on recent advances in graph signal processing, we define a graph-signal smoothness prior to regularize the inverse problem. Finally, we design a fast iterative reweighted least square algorithm to solve the posed problem efficiently. Experimental results show that our z-dimensional synthesized images outperform images rendered by a naı̈ve modification of VSRS 3.5 by up to 4.01 dB.
11 July 2023 in IET Radar, Sonar & Navigation
The jamming effect on radar is dominated by the design of the jamming waveform directly. Traditional jamming waveforms are generated using the template‐based method and are not environmentally resilient. Without the precise radar centre frequency support, using the direct guidance of the radar signal power spectrum density (PSD), a novel jamming waveform design method with an online unsupervised generative learning framework is proposed. The proposed framework consists of two modules: the waveform online generation module and the waveform unsupervised optimisation module. For the waveform online generation module, the well‐designed neural network (NN) is used to generate the jamming waveform. While for the waveform unsupervised iterative optimisation module, two loss functions are developed as two learning tasks to guide back‐propagation online. Given transmit power, the jamming PSD is designed adaptively to focus on the radar band and fit the radar PSD well, which improves the receiver in‐band jamming‐to‐signal ratio (JSR). It is worth noting that NN is end‐to‐end trained online without extensive prior training data. The framework can generate jamming waveforms within reasonable physical parameter ranges, and numerical experiments have shown that the designed jamming waveform performed better jamming effects on radar target detection in comparison with the traditional one.
9 March 2021 in IEEE Sensors Letters
This letter investigates the problem of determining the static target in multistatic systems using bistatic range (BR) measurements, which can be obtained by some local sensor signal processing methods. The localization problem is first formulated as a semidefinite programming problem by using the semidefinite relaxation technique. Then, an initial position estimate is given by augmenting the unknown parameter space, which is required for solving the SDP problem. Specifically, this initialization is also a closed-form solution. Compared with existing algorithms, the proposed estimator has the advantage of not relying on the prior knowledge of BR measurement noise power. Simulation results validate the performance of the proposed estimator.