With great potential in providing global coverage and real-time service, recently, direct satellite-to-device (DS2D) communication has attracted considerable attention. However, how to effectively utilize the costly satellite resources to satisfy the on-demand massive connectivity remains a huge challenge. This paper proposes a cost-effective hetero-granular resource allocation framework that combines the advantages of ground-based scheduling and spaceborne scheduling. In specific, we aim to optimize both the cell-level and user-level scheduling in a beam-hopping system. The formulated optimization problem is first decomposed into a coarse-grained scheduling problem among cells using historical demand information on the ground station, and a fine-grained scheduling problem among users using real-time service demands on satellite. We solve the mixed-integer non-linear programming problem of coarse-grained scheduling with cross-entropy and quantum particle swarm optimization algorithms to find the global optimum, exploiting the adequate ground-based computational resources. The fine-grained scheduling problem is solved with a generalized-benders-decomposition-based algorithm to accommodate the limited spaceborne resources, which decouples power and bandwidth allocation based on a closed-form solution of optimal dual variables in the primal power allocation problem. Simulation results demonstrate that the proposed method effectively reduces the length of the waiting queue by up to 25.05% compared to the existing methods.

Precise unmanned aerial vehicle (UAV) localization is crucial for the successful operation of UAV swarms, particularly in environments where global navigation satellite system (GNSS) signals are denied or unavailable. This paper introduces a cooperative positioning framework for such scenarios. Considering the resource and coverage constraints of UAVs, a multi-objective cooperative positioning algorithm using a projection factor graph (FG) is proposed. This algorithm achieves dimensionality reduction and interaction of the positioning message among UAVs, employing low-complexity techniques for parameter estimation. Simulation results demonstrate that the proposed algorithm achieves linear complexity, significantly lower than cooperative benchmarks like SPAWN with cubic. Furthermore, it provides an 82.5% increase in positioning accuracy over non-cooperative methods, yielding accuracy approaching that of the high-complexity SPAWN algorithm.

Online adaptation to the dynamic channel conditions requires intelligent receivers with online learning ability as well as efficient training samples. In this letter, we propose a new type of reference signal termedlearning reference signal (LRS), which serves as the online training samples for the fast adaptation of the deep neural network (DNN)-aided intelligent receiver. Specifically, we propose a model-agnostic meta-learning (MAML)-based LRS design framework, where the LRS sequence is regarded as the meta parameter and is meta-learned during the offline training. The maximum loss reduction criteria for LRS design is proposed such that the online meta-update based on LRS can maximize the reduction of the symbol error rate (SER). Furthermore, a Matthew effect in gradient-based training of LRS, which causes imbalanced update on different LRS symbols, is identified and then tackled by a novel symbol bundling and multi-stage updating method to ensure convergence. From experiments, we observe that the learned LRS contains both constellation points and non-constellation points, and achieves more than 4dB SER gain compared to using arbitrary constellation points as training samples.

Satellite-aided airborne systems can enable data acquisition in remote areas for the intelligent transportation systems (ITS), leveraging the satellite coverage alongside the mobility and multifunctional capabilities of autonomous aerial vehicles (AAVs). However, due to the harsh environment, ensuring secure and timely task execution is complicated by uncertainties related to both channel conditions and eavesdropping threats. This paper proposes a two-stage optimization method to fully exploit AAVs’ flying, sensing, compressing, and transmitting capabilities for secure data acquisition under dual uncertainties. In the first stage, a deep reinforcement learning strategy is employed to optimize sensing and trajectory planning to explore the eavesdropping environment and balance computational and transmission demands. Building on the sensed information about the eavesdropping environment, the second stage focuses on minimizing task completion time through optimal resource allocation and hierarchical A* path planning, with the channel uncertainty addressed by incorporating an outage probability constraint. Simulations demonstrate that the proposed method can reduce data completion time by 35.3%, validating its effectiveness in uncertain environments.

Space-air-ground integrated network presents a promising solution to the challenge of accommodating large-scale device access while confronting sophisticated interference threats. Existing random access techniques neglect the dynamic interference environment and thus often struggle to realize anti-intelligent interference effectively. This paper proposes a novel approach to address this issue. By employing deep reinforcement learning algorithms, we utilize real-time feedback to adapt to the dynamic environment resulting from the time-varying interference strategy, as well as the involvement of various types of entities. Moreover, we propose a hierarchical reward function to improve the access efficiency. Simulation results show that our method reduces the congestion between users by up to 47% and enhances access efficiency is about 3.2 times compared with random access under malicious jammer intro conclusion finding.

Non-orthogonal multiple access (NOMA) promises to fulfill the fast-growing connectivities in future Internet of Things (IoT) using abundant multiple-access signatures. While explicitly notifying the utilized NOMA signatures causes large signaling cost, blind signature classification naturally becomes a low-cost option. To accomplish signature classification for NOMA, we study both likelihood- and feature-based methods. A likelihood-based method is firstly proposed and showed to be optimal in the asymptotic limit of the observations, despite high computational complexity. While feature-based classification methods promise low complexity, efficient features are non-trivial to be manually designed. To this end, we resort to artificial intelligence (AI) for deep learning-based automatic feature extraction. Specifically, our proposed deep neural network for signature classification, namely DeepClassifier, establishes on the insights gained from the likelihood-based method, which contains two stages to respectively deal with a single observation and aggregate the classification results of an observation sequence. The first stage utilizes an iterative structure where each layer employs a memory-extended network to explicitly exploit the knowledge of signature pool. The second stage incorporates the straight-through channels within a deep recurrent structure to avoid information loss of previous observations. Experiments show that DeepClassifier approaches the optimal likelihood-based method with a reduction of 90% complexity.

Despite the great success of GANs in images translation with different conditioned inputs such as semantic segmentation and edge maps, generating high-fidelity realistic images with reference styles remains a grand challenge in conditional image-to-image translation. This paper presents a general image translation framework that incorporates optimal transport for feature alignment between conditional inputs and style exemplars in image translation. The introduction of optimal transport mitigates the constraint of many-to-one feature matching significantly while building up accurate semantic correspondences between conditional inputs and exemplars. We design a novel unbalanced optimal transport to address the transport between features with deviational distributions which exists widely between conditional inputs and exemplars. In addition, we design a semantic-activation normalization scheme that injects style features of exemplars into the image translation process successfully. Extensive experiments over multiple image translation tasks show that our method achieves superior image translation qualitatively and quantitatively as compared with the state-of-the-art.

Image inpainting aims to complete the missing or corrupted regions of images with realistic contents. The prevalent approaches adopt a hybrid objective of reconstruction and perceptual quality by using generative adversarial networks. However, the reconstruction loss and adversarial loss focus on synthesizing contents of different frequencies and simply applying them together often leads to inter-frequency conflicts and compromised inpainting. This paper presents WaveFill, a wavelet-based inpainting network that decomposes images into multiple frequency bands and fills the missing regions in each frequency band separately and explicitly. WaveFill decomposes images by using discrete wavelet transform (DWT) that preserves spatial information naturally. It applies L1 reconstruction loss to the decomposed low-frequency bands and adversarial loss to high-frequency bands, hence effectively mitigate inter-frequency conflicts while completing images in spatial domain. To address the inpainting inconsistency in different frequency bands and fuse features with distinct statistics, we design a novel normalization scheme that aligns and fuses the multi-frequency features effectively. Extensive experiments over multiple datasets show that WaveFill achieves superior image inpainting qualitatively and quantitatively.

Deep learning has been exploited to tackle the multi-user detection problem of non-orthogonal multiple access (NOMA), for its high detection accuracy and low computational delay. Existing deep learning algorithms adopt an offline training and online deploying method. When the online system configuration, such as the number of the users, differs from that in offline training, existing deep learning algorithms fail to work due to the mismatch of the output dimensions. In this paper, we propose an online reconfigurable deep learning framework for multi-user detection which can adapt to diversified number of the users. Inspired by the factor graph representation of NOMA, the framework is designed as the composition of several interlinked deep neural network branches where each branch is dedicated for the detection of a single user. The connections among the branches are configurable to achieve online dynamic extension or clipping so as to match the varying number of NOMA users. Experiments validate the online reconfigurability and the performance gain of the proposed deep learning framework.

Image inpainting is an underdetermined inverse problem, which naturally allows diverse contents to fill up the missing or corrupted regions realistically. Prevalent approaches using convolutional neural networks (CNNs) can synthesize visually pleasant contents, but CNNs suffer from limited perception fields for capturing global features. With image-level attention, transformers enable to model long-range dependencies and generate diverse contents with autoregressive modeling of pixel-sequence distributions. However, the unidirectional attention in autoregressive transformers is suboptimal as corrupted image regions may have arbitrary shapes with contexts from any direction. We propose BAT-Fill, an innovative image inpainting framework that introduces a novel bidirectional autoregressive transformer (BAT) for image inpainting. BAT utilizes the transformers to learn autoregressive distributions, which naturally allows the diverse generation of missing contents. In addition, it incorporates the masked language model like BERT, which enables bidirectionally modeling of contextual information of missing regions for better image completion. Extensive experiments over multiple datasets show that BAT-Fill achieves superior diversity and fidelity in image inpainting qualitatively and quantitatively.

The rapid booming of future smart city applications and Internet of things (IoT) has raised higher demands on the next-generation radio access technologies with respect to connection density, spectral efficiency (SE), transmission accuracy, and detection latency. Recently, faster-than-Nyquist (FTN) and nonorthogonal multiple access (NOMA) have been regarded as promising technologies to achieve higher SE and massive connections, respectively. In this paper, we aim to exploit the joint benefits of FTN and NOMA by superimposing multiple FTN-based transmission signals on the same physical recourses. Considering the complicated intra- and interuser interferences introduced by the proposed transmission scheme, the conventional detection methods suffer from high computational complexity. To this end, we develop a novel sliding-window detection method by incorporating the state-of-the-art deep learning (DL) technology. The data-driven offline training is first applied to derive a near-optimal receiver for FTN-based NOMA, which is deployed online to achieve high detection accuracy as well as low latency. Monte Carlo simulation results validate that the proposed detector achieves higher detection accuracy than minimum mean squared error-frequency domain equalization (MMSE-FDE) and can even approach the performance of the maximum likelihood-based receiver with greatly reduced computational complexity, which is suitable for IoT applications in smart city with low latency and high reliability requirements.