In this paper, we propose a quantized penalty gradient (QPG) detection algorithm for massive multiple-input multiple-output (MIMO) systems with low-resolution analog-to-digital converters (ADCs). To tackle the challenges of maximum likelihood (ML) detection under discrete constraints, we reformulate the detection problem into an unconstrained optimization by introducing two customized penalty functions that promote alignment between the estimated signals and target constellation set. Based on this, the QPG algorithm is developed to efficiently solve the resulting problem, achieving competitive detection performance with only second-order computational complexity. We further provide a theoretical analysis establishing the Lipschitz continuity of the objective function, which guarantees the monotonic descent property of QPG and ensures its convergence. Moreover, we prove that QPG efficiently finds the local minima with an accessible linear convergence rate, thus leading to an explicit trade-off between detection performance and computational complexity. Finally, simulation results confirm the significant performance gains of QPG over the conventional quantized detectors across various channel conditions, while maintaining low computational complexity.

Next-generation wireless communication systems are unifying massive MIMO and integrated sensing and communication (ISAC) to enhance sensing and communication performance simultaneously. In this paper, the signal detection problem for MIMO-ISAC systems is modeled as a mixed-integer least squares problem (MILSP). To solve it in an efficient way, an iterative algorithm combining alternating direction method of multipliers (ADMM) and neighborhood search (NS) technique is proposed, which is named as NS-ADMM. Specially, at each iteration, the output of ADMM serves for the following neighborhood search to achieve the extra performance gain. Moreover, a flexible mechanism of ADMM iterations is also given for a better estimation of the sensing signals. Finally, simulations demonstrate the proposed NS-ADMM algorithm has significant performance advantages with low computational complexity.

In this paper, we propose the decentralized likelihood ascent search (DLAS)-aided detection for the distributed large-scale multiple-input multiple-output (MIMO) systems to achieve more remarkable performance gains. With the help of DLAS, traditional distributed iterative methods are able to achieve better performance than the linear detection schemes such as ZF and MMSE. According to analysis, we derive the equivalent noise and the post-processing SNR for DLAS. More importantly, based on them, we demonstrate that the proposed DLAS-aided detection achieves the full received diversity. To further facilitate its implementation in practice, we design the decentralized effective ring (DER) architecture with significantly reduced bandwidth requirement and better parallel computation. Finally, simulation results demonstrate that the proposed DLAS-aided detection attains the same received diversity as ML detection while surpassing state-of-the-art decentralized schemes in terms of BER performance, with reduced complexity and bandwidth costs.

Hybrid precoding plays an important role in massive MIMO systems for reducing the hardware cost caused by radio frequency (RF) chains. In this paper, an efficient joint hybrid precoding and analog combining (EJHPAC) scheme is proposed for massive MIMO with multiple-antenna user equipment (UE), which applies the phase elimination method to harvest the power gain. Specifically, the problem of analog combining is transformed to a least square problem with constant modulus constraint. Based on it, we adopt the gradient descent projection (GDP) method to the analog combiner and jointly design the related hybrid precoding algorithm, which leads to the proposed EJHPAC algorithm. According to complexity analysis and simulation results, we show that the EJHPAC algorithm has advantages in both spectral efficiency and computational complexity for massive MIMO systems.

Although the multiresolution analysis (MRA)-based pansharpening methods are able to generate high-resolution multispectral (HRMS) images with good spectral retention, they are prone to spatial distortion. To address this problem, this letter combines deep learning (DL) with the MRA methods and proposes a novel detail-injected-based multiscale asymmetric residual network. The difference strategy is combined with MRA methods and the corresponding injection coefficients are obtained using the multiscale residual block (MSRB) to effectively map spatial information to each waveband of the multispectral (MS) images. In addition, asymmetric convolution block (ACB) is embedded in the residual network to obtain more robust features, and an inception feature pyramid network (FPN) is designed to enrich the spatial information of the fusion results while fusing features at different levels. Experimental results show that the method proposed in this letter outperforms the state-of-the-art pansharpening methods.

Compared with ordinary optical images, the situation of remote sensing images is much more complicated. The problems caused by the shooting angles over the Earth’s surface are: 1) some target categories with more complex shooting environments greatly increase the difficulty of detection and 2) the remote sensing images with large and small targets at the same time leading to large changes in the target scale are difficult to handle. In this letter, we designed a novel scenario context-aware-based bidirectional feature pyramid network (SCBi-FPN) to address the above problems. There are two key modules of the proposed network: the scene context-aware module uses pyramid pooling to aggregate contextual information of the different regions to obtain better global contextual information. The bidirectional feature pyramid network (Bi-FPN) module with squeeze and excitation (SE) blocks connects feature layers at different scales in a cross-scale manner and performs weighted feature map fusion before passing through the SE blocks to enable the network to obtain more accurate information. The experiments demonstrate that our designed network has good results compared with the state-of-the-art methods. In particular, we achieved mean average precision (mAP) of 92.92 on the publicly available NWPU VHR-10 dataset.

Many real images are shot under hazy conditions and need to be processed for better quality. So far the majority of the dehazing methods try to faithfully invert the standard hazy image degradation model. In such an approach, not only is the basic problem theoretically unsolvable, but the parameters used in many methods are often hard to estimate accurately. In this paper we make two contributions. We first derive the standard haze model in the HSV color space, which is preferred over the traditional RGB color space for contrast enhancement due to its robustness to color distortion. Specifically, we show that under generally valid assumptions, the H channel is invariant to haze degradation. The S channel can be recovered when the ambient light and the medium transmission coefficients are estimated as in the traditional approach. Only the V channel takes the same form of the haze model as for the RGB channels. Secondly, we show that the standard image dehazing problem can be viewed as a type of image contrast enhancement. It is then possible to place a less stringent requirement on faithful parameter estimation in the framework of contrast enhancement. Based on such insight, we propose an algorithm that combines the image dehazing and contrast enhancement seamlessly. The major advantage of this new approach is that instead of degrading the processed image quality, the inaccurately estimated parameters with intentional bias tend to enhance the image contrast. Experimental results show that the new approach enjoys a large performance advantage over existing methods.

Impulse noise is commonly encountered during image transmission and many methods have been proposed to remove it. Although it is now possible to recover the true image reasonably well, even under severe noise (90% pixel contamination), essentially all methods published so far follow the standard procedure of noisy pixel detection/classification and then noisy pixel value reconstruction, without any further processing. In this paper we show an interesting empirical discovery that the traditionally denoised image tends to have the estimation error with a Laplacian distribution, which makes it possible to add a postprocessing stage to denoise the traditionally obtained result with this new type of noise. We propose a practical algorithm within this new framework and experimental results show that superior results can be obtained over previously published methods.

So far there are two separate approaches to removing salt-and-pepper noise: the median type filtering and the variational formulation. The first approach usually has fast speed, while the latter produces greatly improved result at much slower speed. In this paper we show that the variational approach can be approximated as a region growing process and propose a novel iterative algorithm that combines the strength of these two approaches. When viewed within a single iteration, the algorithm acts like a median type filter. When viewed across iterations, the filter achieves the region growing effect accomplished by the variational approach. Extensive simulations show that the proposed algorithm achieves the state of the art performance with the fastest speed published so far. The insight gained in this paper could have broader applications.

Salt-and-pepper impulse noise is one commonly encountered noise type during image and video communication. So far the state of the art methods can reasonably restore images corrupted by salt-and-pepper noise whose level is up to 90%. We propose a novel quadratic type variational formulation of this noise removal problem. This approach first uses a simple yet fast method to eliminate all salt-and-pepper noise pixels as well as possibly some clean pixels, then the clean image is efficiently reconstructed from the remaining clean pixels by minimizing a carefully designed functional. Because the functional is quadratic type, fast unconditional convergence is guaranteed. Simulation results show that the proposed method outperforms previously published results and can tolerate noise level of as much as 95%.

The state of the art impulse noise removal methods make use of the noise variance, or equivalently the noise mixing probabilityp, and are iterative procedures (e.g., , ). However, so far there has been a lack of effective estimator forp. As a result, true values ofpare often used during simulation, which may not be practical. Furthermore, the optimal stopping criteria for the iterative algorithms have been elusive until recently. In a computationally heavy method is proposed for determining the optimal number of iterations. In this letter we make two contributions. We first develop a robust estimator forpby using the empirical observation that a natural image usually doesn't cover all pixel value range, then we design an efficient linear transformation to replace complicated computation of order statistics. Based on this estimatedpvalue, we further derive the formula for estimating the true image histogram, and use it to formulate a new efficient optimal stopping criterion during the iterative denoising process. This formulation has a simple interpretation of its optimality and yields improved denoising performance.