Compressed sensing (CS) reconstruction of nonsparse scenes is one of the difficulties in synthetic aperture radar (SAR) imaging technology. Although the conventional CS method with sparse representation has proven applicable for nonsparse SAR reconstruction, its disadvantages are unsatisfactory imaging quality and high computational complexity under downsampling. In this letter, a novel deep learning approach for nonsparse SAR scene reconstruction is proposed based on sparse representation and the iterative shrinkage threshold algorithm (ISTA). Specifically, we first develop a sparse-representation-based imaging model associated with the $\ell _{1}$ sparse regularizer in nonlinear transform domains. Then, the advantages of the recurrent neural network (RNN) and convolutional neural network (CNN) are incorporated into an ISTA-inspired deep unfolded network (DUN) called SR-ISTA-Net, in which all the parameters are layer-varied rather than handcrafted. The experiments verify that the proposed SR-ISTA-Net can provide high-quality reconstruction results under nonsparse scenes while substantially reducing the imaging time.

Currently, the sparse imaging problem of synthetic aperture radar (SAR) is primarily addressed by compressed sensing (CS) theory, which introduces prior information into image restoration tasks via regularization. However, simple regularization constraints cannot provide the complex structural information of targets, and their denoising performance is unsatisfactory at low signal-to-noise ratios (SNRs) and sampling rates. In this letter, a novel sparse SAR reconstruction network is proposed based on plug-and-play (PnP) and approximated observation. First, a chirp-scaling algorithm (CSA) operator-derived approximated observation model is applied to reduce computational costs. Then, this sparse imaging problem is iteratively solved with a matched filter-based approximate message-passing (MFAMP) method. To overcome the limitations of prior models in existing sparse imaging methods, a PnP prior model is incorporated within the sparse reconstruction framework instead of using the ℓ ₁ sparse regularizer. Finally, the solution procedure is unfolded as a deep imaging network, dubbed as PnP-MFAMP-Net. Experimental results validate its robustness and superiority. Even at a sampling rate of 25%, PnP-MFAMP-Net can achieve a PSNR gain of approximately 15 dB compared to the AMP-Net.

Unmanned aerial vehicle- (UAV-) assisted communication has great potential to provide on-demand wireless services and improve the outdoor link throughput. In this paper, a UAV-based cognitive radio network (CRN) is investigated in which the UAV works as a secondary user (SU). Considering the overlay spectrum sensing mode, the UAV can operate on the licensed spectrum bands of primary user (PU) only when PU is idle. In each working frame structure, both sensing time slot and transmission time slot are analysed in radians. Specifically, our objective is to maximize the spectrum efficiency (SE) of the UAV by jointly optimizing the sensing radian and the number of radians. For the single-radian and multiradian schemes, the dichotomy and alternative iterative optimization (AIO) algorithm are proposed to solve the SE optimization problem. Simulation results show that the proposed multiradian cooperative spectrum sensing (CSS) scheme can achieve better performance on ensuring the quality-of-service (QoS) of the PU, and it can significantly enhance the SE of the UAV especially in the severe channel environments.

The ground moving target (GMT) is defocused due to unknown motion parameters in synthetic aperture radar (SAR) imaging. Although the conventional Omega-K algorithm (Omega-KA) has been proven to be applicable for GMT imaging, its disadvantages are slow imaging speed, obvious sidelobe interference, and high computational complexity. To solve the above problems, a SAR-GMT imaging network is proposed based on trainable Omega-KA and sparse optimization. Specifically, we propose a two-dimensional (2-D) sparse imaging model deducted from the Omega-KA focusing process. Then, a recurrent neural network (RNN) based on an iterative optimization algorithm is built to learn the trainable parameters of Omega-KA by an off-line supervised training method, and the solving process of the sparse imaging model is mapped to each layer of the RNN. The proposed trainable Omega-KA network (Omega-KA-net) forms a new GMT imaging method that can be applied to high-quality imaging under down-sampling and a low signal to noise ratio (SNR) while saving the imaging time substantially. The experiments of simulation data and measured data demonstrate that the Omega-KA-net is superior to the conventional algorithms in terms of GMT imaging quality and time.

As a result of residual phase error caused by motion parameters, ground moving targets (GMTs) are defocused and displaced in conventional synthetic aperture radar (SAR) imaging. Although some refocusing algorithms for GMT have been proposed, these methods are difficult to handle geometric correction and sparse recovery in squint mode simultaneously. Deep learning (DL) technology has been successfully used to solve radar imaging problems in microwave vision, where deep unfolding networks (DUNs) and convolutional neural networks (CNNs) are the most widely applied. In this article, a novel GMT imaging network (GMTIm-Net) is proposed for squint SAR and sparse sampling, whose framework combines DUN and CNN advantages. Specifically, we first incorporate a matched filter-based approximated observation model and a minimum entropy-based motion parameter estimation method within a sparse reconstruction framework. An iterative shrinkage threshold algorithm is adopted to solve this framework, and the solution procedure is unfolded as a GMT refocusing network. Then, we introduce plug-and-play (PnP) technology to replace the $\ell _{1}$ norm-based regularizer for improving its noise immunity. Finally, a CNN-based image transformation network is proposed to perform geometric correction of imaging results in squint mode. By inputting the 2-D sparse complex-valued GMT echo, the trained GMTIm-Net can efficiently output focused and corrected GMT images. The experiments demonstrate that our proposed GMTIm-Net outperforms conventional GMT focusing methods in terms of focusing performance and computing efficiency.