This paper designs a novel pilot scheme for extralarge massive MIMO (XL-MIMO) systems, by leveraging the spatial non-stationarity introduced by the large aperture of the extremely large aperture arrays (ELAA). The spatial nonstationarity results in distinct visibility for different users at ELAA, particularly those located far apart, which reduces inter-user interference. This property motivates our novel pilot scheme to group users with distinct visibility regions to share the same frequency subcarriers for channel estimation, hence reducing pilot overhead while accommodating numerous users. Specifically, the proposed pilot scheme employs frequencydivision multiplexing for inter-group channel estimation, while intra-group users - benefiting from strong spatial orthogonality from the distinct visibility region - are distinguished by shifted cyclic codes, similar to code-division multiplexing. Additionally, we propose a low-complexity channel estimation algorithm within a turbo Bayesian inference framework, where the channel support of each user at the ELAA features clustered sparsity in the antenna-delay domain and is modeled by a 2 -dimensional (2-D) Markov random field. Simulations show that the proposed pilot scheme and algorithm allow the XL-MIMO system to support more users, and deliver superior channel estimation performance.

Deep learning methods have been proposed for tooth segmentation on high-resolution intra-oral scans (IOS) that plays a crucial role in clinical dental practice. However, they generally segment teeth in a low-resolution data with a fixed receptive field and generate final segmentation by up-sampling interpolation, and neglect teeth's morphology priors: their similar dental arch structures and significantly different curvatures in different parts of each tooth. They thus lack adaptability to different parts of each tooth, and show less accurate segmentation of boundary points between teeth and gums due to the up-sampling computation. Further, cluttered poses of IOS limit their generalization and usability of teeth location and geometric information. To address these limitations, a morphology prior enhanced teeth segmentation framework is proposed in this paper. Firstly, a robust preprocessing is introduced to align poses of different IOS by computing their dental arch orientations, thereby improving segmentation generalization and usability of IOS geometric information. Secondly, a decomposition-merging strategy is designed to avoid the up-sampling limitation, which decomposes an IOS into multiple low-resolution data and merges their segmentation outcomes into a high-resolution result. Thirdly, an innovative module integrating semantic and geometric features is proposed to adaptively select deformable receptive fields. It geometrically samples within a variable probability space to construct receptive fields with varied graph relationships for different points, facilitating adaptive segmentation of different parts of each tooth. Experimental results on 6238 IOS from four centers demonstrate that our method significantly outperforms 11 state-of-the-art methods, achieving a 6.93% enhancement for cross-center testing.

Pixel-based reconfigurable intelligent surfaces (RISs) employ a novel design to achieve high reflection gain at a lower hardware cost by eliminating the phase shifters used in traditional RIS. However, this design presents challenges for channel estimation and passive beamforming due to its non-separable state response, rendering existing solutions ineffective. To address this, we first approximate the non-separable RIS response functions using a kernel-based method and a deep neural network, achieving high accuracy while reducing computational and memory complexity. Next, we propose a simplified cascaded channel model that focuses on dominated scattering paths with limited unknown parameters, along with customized algorithms to estimate short-term and long-term parameters separately. Finally, we introduce a low-complexity passive beamforming algorithm to configure the discrete RIS state vector, maximizing the achievable rate. Our simulation results demonstrate that the proposed solution significantly outperforms various baselines across a wide SNR range.

Pediatric teeth exhibit significant changes in type and spatial distribution across different age groups. This variation makes pediatric teeth segmentation from cone-beam computed tomography (CBCT) more challenging than that in adult teeth. Existing methods mainly focus on adult teeth segmentation, which however cannot be adapted to spatial distribution of pediatric teeth with individual changes (SDPTIC) in different children, resulting in limited accuracy for segmenting pediatric teeth. Therefore, we introduce a novel topology structure-guided graph convolutional network (TSG-GCN) to generate dynamic graph representation of SDPTIC for improved pediatric teeth segmentation. Specifically, this network combines a 3D GCN-based decoder for teeth segmentation and a 2D decoder for dynamic adjacency matrix learning (DAML) to capture SDPTIC information for individual graph representation. 3D teeth labels are transformed into specially-designed 2D projection labels, which is accomplished by first decoupling 3D teeth labels into class-wise volumes for different teeth via one-hot encoding and then projecting them to generate instance-wise 2D projections. With such 2D labels, DAML can be trained to adaptively describe SDPTIC from CBCT with dynamic adjacency matrix, which is then incorporated into GCN for improving segmentation. To ensure inter-task consistency at the adjacency matrix level between the two decoders, a novel loss function is designed. It can address the issue with inconsistent prediction and unstable TSG-GCN convergence due to two heterogeneous decoders. The TSG-GCN approach is finally validated with both public and multi-center datasets. Experimental results demonstrate its effectiveness for pediatric teeth segmentation, with significant improvement over seven state-of-the-art methods.

This paper investigates a fully distributed federated learning (FL) problem, in which each device is restricted to only utilize its local dataset and the information received from its adjacent devices that are defined in a communication graph to update the local model weights for minimizing the global loss function. To incorporate the communication graph constraint into the joint posterior distribution, we exploit the fact that the model weights on each device is a function of its local likelihood and local prior and then, the connectivity between adjacent devices is modeled by a Dirichlet distribution. In this way, the joint distribution can be factorized naturally by a factor graph. Based on the Dirichlet-based factor graph, we propose a novel distributed approximate Bayesian inference algorithm that combines loopy belief propagation (LBP) and variational Bayesian inference (VBI) for distributed FL. Specifically, VBI is used to approximate the non-Gaussian marginal posterior as a Gaussian distribution in local training process and then, the global training process resembles Gaussian LBP where only the mean and variance are passed among adjacent devices. Furthermore, we propose a new damping factor design according to the communication graph topology to mitigate the potential divergence and achieve consensus convergence. Simulation results verify that the proposed solution achieves faster convergence speed with better performance than baselines.

In this paper, we investigate multi-resolution model compression for deep neural networks (DNNs) from a Bayesian perspective. By considering the DNN models with channel masks and proposing a resolution likelihood as well as a two-layer sparse prior for the channel masks, we formulate the multi-resolution model compression as a Bayesian inference problem. To solve this problem, we propose a partial update block variational Bayesian inference (PUB- VBI) algorithm which can infer an approximate posterior for the intractable true posterior. The variational posterior and the updating rules are carefully designed such that the proposed algorithm has a low complexity. Simulation results demonstrate that our proposed method can outperform the baselines on various neural network models and datasets.

In this paper, we propose a Variational Bayesian Autoencoder (VBA)-based channel state information (CSI) compression and feedback scheme for massive multiple-input multiple-output (MIMO) systems. The proposed scheme incorporates the model-assisted knowledge of low-dimensional feedback features and the sparsity of channel to achieve enhanced compression efficiency. We also design a CsiVBA architecture that outputs distributions of the feedback features and the channel at the encoder and decoder, respectively, which facilitates a Bayesian training formulation exploiting the underlying channel sparsity. In addition, we also propose a low-complexity training scheme for new networks of different bit rates, significantly reducing the retraining cost for new compression requirements. Simulation results show that the proposed scheme achieves better rate-distortion trade-offs than the state-of-the-art solutions.

This article investigates the scheduling framework of the federated multitask learning (FMTL) problem with a hard-cooperation structure over wireless networks, in which the scheduling becomes more challenging due to the different convergence behaviors of different tasks. Based on the special model structure, we propose a dynamic user and task scheduling scheme with a block-wise incremental gradient aggregation algorithm, in which the neural network model is decomposed into a common feature-extraction module and$M$task-specific modules. Different block gradients with respect to different modules can be scheduled separately. We further propose a Lyapunov-drift-based scheduling scheme that minimizes the overall communication latency by utilizing both the instantaneous data importance and the channel state information. We prove that the proposed scheme can converge almost surely to a KKT solution of the training problem such that the data-distortion issue is resolved. Simulation results illustrate that the proposed scheme significantly reduces the communication latency compared to the state-of-the-art baseline schemes.

Deep learning (DL)-based channel estimation has achieved remarkable success. However, most existing works focus on the white Gaussian noise which are inapplicable for cell-edge users under inter-cell interference (ICI). In this paper, we address this issue by proposing a novel DL-based channel estimation solution with a cascaded model-based and model-free deep neural network (DNN) structure. Specifically, the model-based module is designed by the variational Bayesian inference (VBI) technique to suppress the time-varying ICI, and the model-free module is designed by the Denoising Sparse Autoencoder (DSAE) structure to further refine the channel estimation. The proposed DNN is firstly pre-trained by offline supervised training, and various channel statistics are encapsulated in the DNN weights with the assist of a hyper-prior net modelling different sparse priors for different training samples. Then, an online Bayesian learning algorithm is proposed to train the model-based VBI module based on real-time pilot samples to track the online channel statistics. Simulation results verify that the proposed solution outperforms various state-of-the-art baseline schemes in a large SINR range with comparable performance to the estimator with genie-aided channel statistics.

This article investigates the analog gradient aggregation (AGA) solution to overcome the communication bottleneck for wireless federated learning applications by exploiting the idea of analog over-the-air transmission. Despite the various advantages, this special transmission solution also brings new challenges to both transceiver design and learning algorithm design due to the nonstationary local gradients and the time-varying wireless channels in different communication rounds. To address these issues, we propose a novel design of both the transceiver and learning algorithm for the AGA solution. In particular, the parameters in the transceiver are optimized with the consideration of the nonstationarity in the local gradients based on a simple feedback variable. Moreover, a novel learning rate design is proposed for the stochastic gradient descent algorithm, which is adaptive to the quality of the gradient estimation. Theoretical analyses are provided on the convergence rate of the proposed AGA solution. Finally, the effectiveness of the proposed solution is confirmed by two separate experiments based on linear regression and the shallow neural network. The simulation results verify that the proposed solution outperforms various state-of-the-art baseline schemes with a much faster convergence speed.

In this paper, we consider a novel Internet of Things (IoT) system in smart city called unmanned aerial vehicle‐ (UAV‐) assisted cognitive backscatter network, where a UAV is employed as both a relay and a radio frequency source to help the data transmission between ground IoT backscatter devices (BDs) and a remote data center (DC). However, since the IoT applications are usually not assigned dedicated spectrum resource in smart cities, these data transmissions from BDs to the DC should share the licensed spectrum of cellular users (CUs). Therefore, we aim to maximize the minimum uplink throughput among all BDs while avoiding severe interference to CUs via joint spectrum management and UAV trajectory design. To solve the problem, we propose an iterative method utilizing block coordinated decent to partition the variables into two blocks. For the spectrum management problem, we first prove its convexity with the transmit power and time scheduling and then propose a two‐step method to solve the two variables sequentially. For the UAV trajectory design problem, we resort to the fractional programming method to handle it. Simulation results demonstrate that the proposed algorithm can significantly increase the average max‐min rate of the BDs while guaranteeing the acceptable interference to CUs with a fast convergence speed.

In this article, a symbiotic radio (SR) system is proposed to support passive Internet of Things (IoT), in which a backscatter device (BD), also called IoT device, is parasitic in a primary transmission. The primary transmitter (PT) is designed to assist both the primary and BD transmissions, and the primary receiver (PR) is used to decode the information from the PT as well as the BD. The symbol period for BD transmission is assumed to be either equal to or much greater than that of the primary one, resulting in parasitic SR (PSR) or commensal SR (CSR) setup. We consider a basic SR system which consists of three nodes: 1) a multiantenna PT; 2) a single-antenna BD; and 3) a single-antenna PR. We first derive the achievable rates for the primary and BD transmissions for each setup. Then, we formulate two transmit beamforming optimization problems, i.e., the weighted sum-rate maximization (WSRM) problem and the transmit power minimization (TPM) problem, and solve these nonconvex problems by applying the semidefinite relaxation (SDR) technique. In addition, a novel transmit beamforming structure is proposed to reduce the computational complexity of the solutions. The simulation results show that for CSR setup, the proposed solution enables the opportunistic transmission for the BD via energy-efficient passive backscattering without any loss in spectral efficiency, by properly exploiting the additional signal path from the BD.

Reconfigurable intelligent surfaces (RIS) is a promising solution to build a programmable wireless environment via steering the incident signal in fully customizable ways with reconfigurable passive elements. In this paper, we consider a RIS-aided multiuser multiple-input single-output (MISO) downlink communication system. Our objective is to maximize the weighted sum-rate (WSR) of all users by joint designing the beamforming at the access point (AP) and the phase vector of the RIS elements, while both the perfect channel state information (CSI) setup and the imperfect CSI setup are investigated. For perfect CSI setup, a low-complexity algorithm is proposed to obtain the stationary solution for the joint design problem by utilizing the fractional programming technique. Then, we resort to the stochastic successive convex approximation technique and extend the proposed algorithm to the scenario wherein the CSI is imperfect. The validity of the proposed methods is confirmed by numerical results. In particular, the proposed algorithm performs quite well when the channel uncertainty is smaller than 10%.

Intelligent reflecting surface (IRS) is a promising solution to build a programmable wireless environment for future communication systems, in which the reflector elements steer the incident signal in fully customizable ways by passive beamforming. This work focuses on the downlink of an IRS-aided multiuser multiple-input single-output (MISO) system. A practical IRS assumption is considered, in which the incident signal can only be shifted with discrete phase levels. Then, the weighted sum-rate of all users is maximized by joint optimizing the active beamforming at the base-station (BS) and the passive beamforming at the IRS. This non-convex problem is firstly decomposed via Lagrangian dual transform, and then the active and passive beamforming can be optimized alternatingly. In addition, an efficient algorithm with closed-form solutions is proposed for the passive beamforming, which is applicable to both the discrete phase- shift IRS and the continuous phaseshift IRS. Simulation results have verified the effectiveness of the proposed algorithm as compared to different benchmark schemes.

In this paper, we are interested in a symbiotic radio (SR) system, in which a passive full-duplex backscatter device (BD) is parasitic in an active primary transmission. The primary transmitter (PT) with multiple antennas is designed to broadcast common messages to the primary receiver (PR) and the BD, as well as to support passive information transmission from the BD to the PR. To do so, the full-duplex BD uses a fraction of the incident signal from the PT to decode the common messages, and simultaneously transmits its own information to the PR by backscattering the remaining part of the incident signal. We formulate a transmit power minimization problem by jointly designing the beamforming vector at the PT and the power splitting factor at the BD. This problem is first solved by the semi-definite relaxation technique together with a one-dimensional linear exhaustive search over the power splitting factor. Then, a suboptimal but low-complexity solution with closed-form expressions is proposed. Simulation results have shown that the proposed suboptimal solution achieves almost the same performance as the one obtained by the exhaustive search.

Cognitive ambient backscatter communication is a novel spectrum sharing paradigm, in which the backscatter system shares not only the same spectrum, but also the same radio-frequency source with the legacy system. Conventional energy detector (ED) suffers from severe error floor problem due to the existence of co-channel direct link interference (DLI) from the legacy system. In this paper, novel error-floor-free detectors are proposed to tackle the DLI using multiple receive antennas at the reader. First, beamforming-assisted ED and likelihood-ratio-based detector are proposed for backscatter symbol detection when the reader has perfect channel state information (CSI). Then a novel statistical clustering framework is proposed for joint CSI feature learning and backscatter symbol detection. Extensive simulation results have shown that the proposed methods can significantly outperform the conventional ED. In addition, the proposed clustering-based methods perform comparably as their counterparts with perfect CSI.

Ambient backscatter communication (AmBC) is a promising solution to energy-efficient and spectrum-efficient Internet of Things with stringent power and cost constraints. In an AmBC system, recovering the tag information at the reader, however, is a challenging task due to the difficulty in acquiring the relevant channel-state information (CSI). To eliminate the need to estimate the CSI, in this paper, we propose a label-assisted transmission framework, in which two known labels are transmitted from the tag before data transmission. By exploring the received signal constellation information, we propose modulation-constrained expectation maximization algorithm, based on which two detection methods are developed. One method, referred to as constellation learning with labeled signals, learns the parameters by clustering the labeled signals and recovers the unlabeled signals by the learnt parameters. The other method, referred to as constellation learning with labeled and unlabeled signals, uses all received signals in clustering. Efficient initialization techniques are provided for the two clustering algorithms. Finally, extensive simulation results show that the proposed constellation learning methods achieve comparable performance as the optimal detector with perfect CSI.

Ambient backscatter communications (AmBC) enable wireless communications riding on ambient radio frequency (RF) signals instead of self-generated RF signals. Therefore, it has been considered as a promising candidate for the future Internet-of-Things with stringent energy and spectrum constraints. In this paper, we investigate a full-duplex-enabled cognitive backscatter network, in which an AmBC system underlays a primary cellular system, and the primary access point can transmit primary signals and receive backscatter signals simultaneously via full-duplex communications. We aim to maximize the throughput of the AmBC system while guaranteeing the minimum rate requirements of the primary system via joint time scheduling, transmit power allocation, and reflection coefficient (RC) adjustment. To solve the problem, we propose an iterative method utilizing block coordinated decent to partition the variables into the time scheduling variable and the joint transmit power allocation and RC adjustment variable. For the time scheduling problem, we first prove its convexity and then utilize the interior-point method to solve it. For the joint power allocation and RC adjustment problem, we resort to the concave-convex procedure to transform it into a sequence of convex optimization problems, and then adopt Lagrange dual decomposition to tackle these convex optimization problems. The simulation results demonstrate that the proposed method can significantly increase the throughput of the AmBC system with a fast convergence speed.

Cognitive ambient backscatter communication (AmBC) is a novel spectrum sharing paradigm for green Internet of Things, in which the backscatter system shares not only the same spectrum, but also the same radio- frequency (RF) source with the legacy system. The conventional energy detector (ED) suffers from a severe error floor problem due to the existence of co-channel direct link interference (DLI) from the legacy system. In this paper, novel error-floor-free detectors are proposed to tackle the DLI through multi-antenna receive beamforming. A novel statistical clustering framework is proposed for joint channel state information (CSI) learning and backscatter symbol detection. Extensive simulation results have shown that the proposed methods can significantly outperform the conventional ED, and verified that the proposed clustering-based method achieves performance comparable to that of the perfect CSI cases.

In ambient backscatter communication (AmBC), it is a challenging task to recover the tag information at the reader due to the difficulty in obtaining the relevant channel state information (CSI). In this paper, we translate the signal detection problem into a clustering problem, for which two known labels are transmitted from the tag as the prior knowledge to assist clustering initialization and signal detection. By exploiting the received signals directly, two clustering-inspired detection methods are proposed, one is called clustering with labeled signals (CLS), and the other is referred to as clustering with labeled and unlabeled signals (CLUS). Both methods are developed based on the proposed modulation-constrained (MC) Gaussian mixture model (GMM). Finally, extensive simulation results show that the proposed methods only have small gaps compared with the optimal detection with perfect CSI.

This paper discusses a typical cooperative sensing model where the local observations are conditionally dependent. The primary user sends a binary phase-shifted keying modulated signal and the hard combining strategy is implemented among second users. Two low-complexity local quantization schemes corresponding to the choice of quantizing either the amplitude or the sign as the one-bit reported message are compared. We show that the optimal fusion rule for quantizing the sign is non-monotone, which is different from the design for quantizing amplitude scheme. Based on large deviation theory, we finish the parameter optimization for both schemes and use the error exponent to evaluate the asymptotical performance. Several closed-form expressions are derived. Numerical results show that to quantize the sign is asymptotically optimal in low-SNR regime, while to quantize the amplitude performs well in high-SNR regime. This work shows that the dependence among local observations may greatly help the final decision in some case, and the non-monotone fusion rule may achieve better performance than the monotone one for dependent observations.