Federated learning (FL) nowadays involves heterogeneous compound learning tasks as cognitive applications’ complexity increases. For example, a self-driving system hosts multiple tasks simultaneously (e.g., detection, classification, segmentation, etc.) and expects FL to retain life-long intelligence involvement. However, our analysis demonstrates that, when deploying compound FL models for multiple training tasks on a GPU, certain issues arise: As different tasks’ skewed data distributions and corresponding models cause highly imbalanced learning workloads, current GPU scheduling methods lack effective resource allocations; Therefore, existing FL schemes, only focusing on heterogeneous data distribution but runtime computing, cannot practically achieve optimally synchronized federation. To address these issues, we propose a full-stack FL optimization scheme to tackle both intra-device GPU scheduling and inter-device FL coordination for multi-task training. Specifically, our works illustrate two key insights in this research domain: Competitive resource sharing is beneficial for parallel model executions, and the proposed concept of “virtual resource” could effectively characterize and guide the practical per-task resource utilization and allocation; Additionally, architectural-level coordination improves FL performance by aligning task workloads with GPU utilization. Our experiments demonstrate that the FL performance could be significantly escalated. Specifically, we observed a 2.16×–2.38× increase in intra-device GPU training throughput and a 2.53×–2.80× boost in inter-device FL coordination efficiency across diverse multi-task scenarios.

Recent advances in distributed swarm learning (DSL) offer a promising paradigm for edge Internet of Things. Such advancements enhance data privacy, communication efficiency, energy saving, and model scalability. However, the presence of non-independent and identically distributed (non-i.i.d.) data pose a significant challenge for multi-access edge computing, degrading learning performance and diverging training behavior of vanilla DSL. Further, there still lacks theoretical guidance on how data heterogeneity affects model training accuracy, which requires thorough investigation. To fill the gap, this paper first study the data heterogeneity by measuring the impact of non-i.i.d. datasets under the DSL framework. This then motivates a new multi-worker selection design for DSL, termed M-DSL algorithm, which works effectively with distributed heterogeneous data. A new non-i.i.d. degree metric is introduced and defined in this work to formulate the statistical difference among local datasets, which builds a connection between the measure of data heterogeneity and the evaluation of DSL performance. In this way, our M-DSL guides effective selection of multiple works who make prominent contributions for global model updates. We also provide theoretical analysis on the convergence behavior of our M-DSL, followed by extensive experiments on different heterogeneous datasets and non-i.i.d. data settings. Numerical results verify performance improvement and network intelligence enhancement provided by our M-DSL beyond the benchmarks.

In distributed learning systems, robustness threat may arise from two major sources. On the one hand, due to distributional shifts between training data and test data, the trained model could exhibit poor out-of-sample performance. On the other hand, a portion of working nodes might be subject to Byzantine attacks, which could invalidate the learning result. In this article, we propose a new research direction that jointly considers distributional shifts and Byzantine attacks. We illuminate the major challenges in addressing these two issues simultaneously. Accordingly, we design a new algorithm that equips distributed learning with both distributional robustness and Byzantine robustness. Our algorithm is built on recent advances in distributionally robust optimization (DRO) as well as norm-based screening (NBS), a robust aggregation scheme against Byzantine attacks. We provide convergence proofs in three cases of the learning model being nonconvex, convex, and strongly convex for the proposed algorithm, shedding light on its convergence behaviors and endurability against Byzantine attacks. In particular, we deduce that any algorithm employing NBS (including ours) cannot converge when the percentage of Byzantine nodes is $(1/3)$ or higher, instead of $(1/2)$ , which is the common belief in current literature. The experimental results verify our theoretical findings (on the breakpoint of NBS and others) and also demonstrate the effectiveness of our algorithm against both robustness issues, justifying our choice of NBS over other widely used robust aggregation schemes. To the best of our knowledge, this is the first work to address distributional shifts and Byzantine attacks simultaneously.

This paper considers the problem of decentralized resource allocation in the presence of Byzantine attacks. Such attacks occur when an unknown number of malicious agents send random or carefully crafted messages to their neighbors, aiming to prevent the honest agents from reaching the optimal resource allocation strategy. We characterize these malicious behaviors with the classical Byzantine attacks model, and propose a class of Byzantine-resilient decentralized resource allocation algorithms augmented with dual-domain defenses. The honest agents receive messages containing the (possibly malicious) dual variables from their neighbors at each iteration, and filter these messages with robust aggregation rules. Theoretically, we prove that the proposed algorithms converge to a neighborhood of the optimal resource allocation strategy, given that the robust aggregation rules are properly designed. Numerical experiments are conducted to corroborate the theoretical results.

This paper studies the decentralized dynamic kernel learning problem where each agent in the network receives continuous streaming local data and works collaboratively to learn a non-linear function "on the fly" in a dynamic environment. We utilize the random feature (RF) mapping method to circumvent the curse of dimensionality issue in conventional kernel methods and reformulate the dynamic kernel learning problem as a dynamic parameter optimization problem, which is then efficiently solved by the Decentralized Dynamic Kernel Learning via ADMM (DDKL) framework. To further improve communication efficiency, we incorporate the quantization and censoring strategies in the communication stage and develop the Quantized and Communication-censored DDKL (QC-DDKL) algorithm. We theoretically prove that QC-DDKL can achieve the optimal sublinear regret $\mathcal{O}(\sqrt T )$ over T time slots. Simulation results also corroborate the learning effectiveness and the communication efficiency of the proposed method.

Federated learning (FL) is a promising distributed learning framework in terms of privacy protection and communication saving. Most existing FL techniques are developed for independent-and-identically-distributed (IID) datasets, but suffer from performance degradation under Non-IID datasets. To cope with this issue, most existing work designs solutions from data perspectives (e.g., sharing some data samples between local devices) to eliminate the heterogeneity of distributed datasets, which causes extra communication overhead and may expose user privacy that contradicts FL's original intention. Unlike the existing data-based methods, we propose a generative adversarial network (GAN) based FL, named as GANFed, which is designed from a feature perspective. Specifically, we embed a discriminator into the FL network, which works with the shallow layers as a generator to form a GAN in FL. By incorporating such a GAN, the output of the shallow layers tends to present more IID features compared with the original Non-IID input data. These extracted features from the shallow layers are then used to train the deep layers of the FL network. In this way, the proposed GANFed reduces the weight divergence of the local models, and hence improves the performance of FL. Without data exchange, our GANFed avoids the leakage of user privacy and reduces the communication overhead. Experimental results show that our GANFed outperforms the standard FedAvg on Non-IID dataset in terms of improved test accuracy.

As deep learning continues to advance and is applied to increasingly complex scenarios, the demand for concurrent deployment of multiple neural network models has arisen. This demand, commonly referred to as multi-tenant computing, is becoming more and more important. However, even the most mature GPU-based computing systems struggle to adequately address the significant heterogeneity and complexity among concurrent models in terms of resource allocation and runtime scheduling. And this usually results in considerable resource utilization and throughput issues. To tackle these issues, this work proposes a set of optimization techniques that advance the granularity of computing management from both the spatial and temporal perspectives, specifically tailored to heterogeneous model compositions for deep learning inference and training. These techniques are further integrated as GACER --- an automated optimization framework that provides high-utilization, high-throughput, and low-latency multi-tenant computing support. And our experiments demonstrate that GACER significantly improves the overall resource utilization and consistently achieves outstanding speedups compared to native GPU computing frameworks and existing state-of-the-art optimization works.

For future intelligent communication systems, radiomap estimation (RME) is essential for acquiring panoramic awareness of spectrum spatial distribution in wireless environments. Recently, deep learning-based RME methods have been developed to reconstruct radiomaps from spectrum measurements collected at distributed sensors. However, these methods rely on gathering all input data at a central fusion center, resulting in large communication overheads, high computation costs, and privacy leakage concerns. To address these challenges, this work proposes a FedRME approach that makes federated learning applicable for distributed RME over a large-scale network, accommodating geographically heterogeneous transmitter locations and propagation environments. Specifically, we partition the large area into smaller regions to reduce the model complexity required for learning the radiomap in each region. Meanwhile, we incorporate the landscape map as an auxiliary input to induce a common learning model that adheres to the same propagation physics across all these heterogeneous regions. In doing so, fusion centers in all regions can collaborate through federated learning to enhance the overall RME performance. Simulation results indicate that our proposed method outperforms existing benchmarks, particularly under limited data, achieving higher learning accuracy with reduced model complexity and lower computational cost.

Abstract This paper investigates the capability of plain Vision Transformers (ViTs) for semantic segmentation using the encoder–decoder framework and introduce SegViTv2 . In this study, we introduce a novel Attention-to-Mask (ATM) module to design a lightweight decoder effective for plain ViT. The proposed ATM converts the global attention map into semantic masks for high-quality segmentation results. Our decoder outperforms popular decoder UPerNet using various ViT backbones while consuming only about $$5\%$$ 5 % of the computational cost. For the encoder, we address the concern of the relatively high computational cost in the ViT-based encoders and propose a Shrunk ++ structure that incorporates edge-aware query-based down-sampling (EQD) and query-based up-sampling (QU) modules. The Shrunk++ structure reduces the computational cost of the encoder by up to $$50\%$$ 50 % while maintaining competitive performance. Furthermore, we propose to adapt SegViT for continual semantic segmentation, demonstrating nearly zero forgetting of previously learned knowledge. Experiments show that our proposed SegViTv2 surpasses recent segmentation methods on three popular benchmarks including ADE20k, COCO-Stuff-10k and PASCAL-Context datasets. The code is available through the following link: https://github.com/zbwxp/SegVit .

This article focuses on online kernel learning over a decentralized network. Each agent in the network receives online streaming data and collaboratively learns a globally optimal nonlinear prediction function in the reproducing kernel Hilbert space (RKHS). To overcome the curse of dimensionality issue in traditional online kernel learning, we utilize random feature (RF) mapping to convert the nonparametric kernel learning problem into a fixed-length parametric one in the RF space. We then propose a novel learning framework, named online decentralized kernel learning via linearized ADMM (ODKLA), to efficiently solve the online decentralized kernel learning problem. To enhance communication efficiency, we introduce quantization and censoring strategies in the communication stage, resulting in the quantized and communication-censored ODKLA (QC-ODKLA) algorithm. We theoretically prove that both ODKLA and QC-ODKLA can achieve the optimal sublinear regret over time slots. Through numerical experiments, we evaluate the learning effectiveness, communication efficiency, and computation efficiency of the proposed methods.

Data-driven machine learning techniques have been advocated for signal detection in complex wireless environments. However, when applied to wideband spectrum sensing scenarios, they face practical challenges including very large data dimensionality, insufficient training data, and implicit inter-band dependencies. Current literature focuses on deep convolutional models, whose inherent model structure is not well suited for representing the diverse spectrum occupancy patterns of practical wideband networks, causing inefficient performance-complexity tradeoff and excessive sensing time. To address these issues, this paper develops a novel Spectrum Transformer with multi-task learning for wideband spectrum sensing at high sample efficiency. Empowered by the multi-head self-attention mechanism, the transformer architecture is designed to effectively learn both the inner-band spectral features and the inter-band spectrum occupancy correlations in the wideband regime. Simulations show that the proposed Spectrum Transformer outperforms the existing methods based on convolutional neural networks especially in the small-data case, by achieving higher sensing accuracy with an 89% reduction in model complexity.

This paper studies distributed online learning under Byzantine attacks. The performance of an online learning algorithm is characterized by (adversarial) regret, and a sublinear bound is preferred. But we prove that, even with a class of state-of-the-art robust aggregation rules, in an adversarial environment and with Byzantine participants, distributed online gradient descent can only achieve a linear adversarial regret bound, which is tight. This is the inevitable consequence of Byzantine attacks, even though we can control the constant of the linear adversarial regret to a reasonable level. Interestingly, when the environment is not fully adversarial so that the losses of the honest participants are i.i.d. (independent and identically distributed), we show that sublinear stochastic regret, in contrast to the aforementioned adversarial regret, is possible. We develop a Byzantine-robust distributed online gradient descent algorithm with momentum to attain such a sublinear stochastic regret bound.

This paper proposes a super-resolution harmonic retrieval method for uncorrelated strictly non-circular signals, whose covariance and pseudo-covariance present Toeplitz and Hankel structures, respectively. Accordingly, the augmented covariance matrix constructed by the covariance and pseudo-covariance matrices is not only low rank but also jointly Toeplitz-Hankel structured. To efficiently exploit such a desired structure for high estimation accuracy, we develop a low-rank Toeplitz-Hankel covariance reconstruction (LRTHCR) solution employed over the augmented covariance matrix. Further, we design a fitting error constraint to flexibly implement the LRTHCR algorithm without knowing the noise statistics. In addition, performance analysis is provided for the proposed LRTHCR in practical settings. Simulation results reveal that the LRTHCR outperforms the benchmark methods in terms of lower estimation errors.

In this paper, we study a communication-efficient distributed learning scheme through a holistic integration of federated learning (FL) and particle swarm optimization, called DSL, which is suitable for the implementation of intelligent IoT applications. Since only one selected optimum from all local devices need to report its local model updates to the parameter server, the communication cost of DSL is much reduced compared to its counterpart of standard FL. However, the DSL is vulnerable to adversarial attackers. To achieve Byzantine-resilient DSL, we propose to introduce a shared dataset for scoring local updates to screen attackers. We further provide the convergence analysis to theoretically demonstrate that CB-DSL is superior than the standard FL. Experiment results show that the learning performance of our proposed CB-DSL outperforms the existing benchmarks with only a small amount of globally shared data. It enjoys higher robustness against Byzantine attacks than the vanilla DSL, and has better communication efficiency than the standard FL11Our code can be found at: https://github.com/fuanxiyin/CB-DSL.git..

Federated learning (FL) methods face major challenges including communication bottleneck, data heterogeneity and security concerns in edge IoT scenarios. In this paper, inspired by the success of biological intelligence (BI) of gregarious organisms, we propose a novel edge learning approach for swarm IoT, called communication-efficient and Byzantine-robust distributed swarm learning (CB-DSL), through a holistic integration of AI-enabled stochastic gradient descent and BI-enabled particle swarm optimization. To deal with non-independent and identically distributed (non-i.i.d.) data issues and Byzantine attacks, a very small amount of global data samples are introduced in CB-DSL and shared among IoT workers, which not only alleviates the local data heterogeneity effectively but also enables to fully utilize the exploration-exploitation mechanism of swarm intelligence. Further, we provide convergence analysis to theoretically demonstrate that the proposed CB-DSL is superior to the standard FL with better convergence behavior. In addition, to measure the effectiveness of the introduction of the globally shared dataset, we also evaluate the model divergence by deriving its upper bound. Numerical results verify that the proposed CB-DSL outperforms the existing benchmarks in terms of faster convergence speed, higher convergent accuracy, lower communication cost, and better robustness against non-i.i.d. data and Byzantine attacks11Our code can be found at:https://github.com/fuanxiyin/CB-DSL.git..

Fairness and robustness are two important goals in the design of modern distributed learning systems. Despite a few prior works attempting to achieve both fairness and robustness, some key aspects of this direction remain underexplored. In this paper, we try to answer three largely unnoticed and unaddressed questions that are of paramount significance to this topic: (i) What makes jointly satisfying fairness and robustness difficult? (ii) Is it possible to establish theoretical guarantee for the dual property of fairness and robustness? (iii) How much does fairness have to sacrifice at the expense of robustness being incorporated into the system? To address these questions, we first identify data heterogeneity as the key difficulty of combining fairness and robustness. Accordingly, we propose a fair and robust framework called H-nobs which can offer certified fairness and robustness through the adoption of two key components, a fairness-promoting objective function and a simple robust aggregation scheme called norm-based screening (NBS). We explain in detail why NBS is the suitable scheme in our algorithm in contrast to other robust aggregation measures. In addition, we derive three convergence theorems for H-nobs in cases of the learning model being nonconvex, convex, and strongly convex respectively, which provide theoretical guarantees for both fairness and robustness. Further, we empirically investigate the influence of the robust mechanism (NBS) on the fairness performance of H-nobs, the very first attempt of such exploration.

We propose a simple yet effective framework for instance and panoptic segmentation, termed CondInst (conditional convolutions for instance and panoptic segmentation). In the literature, top-performing instance segmentation methods typically follow the paradigm of Mask R-CNN and rely on ROI operations (typically ROIAlign) to attend to each instance. In contrast, we propose to attend to the instances with dynamic conditional convolutions. Instead of using instance-wise ROIs as inputs to the instance mask head of fixed weights, we design dynamic instance-aware mask heads, conditioned on the instances to be predicted. CondInst enjoys three advantages: 1) Instance and panoptic segmentation are unified into a fully convolutional network, eliminating the need for ROI cropping and feature alignment. 2) The elimination of the ROI cropping also significantly improves the output instance mask resolution. 3) Due to the much improved capacity of dynamically-generated conditional convolutions, the mask head can be very compact (e.g., 3 conv. layers, each having only 8 channels), leading to significantly faster inference time per instance and making the overall inference time less relevant to the number of instances. We demonstrate a simpler method that can achieve improved accuracy and inference speed on both instance and panoptic segmentation tasks. On the COCO dataset, we outperform a few state-of-the-art methods. We hope that CondInst can be a strong baseline for instance and panoptic segmentation. Code is available at: https://git.io/AdelaiDet.

This paper proposes two novel schemes of wideband compressive spectrum sensing (CSS) via block orthogonal matching pursuit (BOMP) algorithm, for achieving high sensing accuracy in real time. These schemes aim to reliably recover the spectrum by adaptively adjusting the number of required measurements without inducing unnecessary sampling redundancy. To this end, the minimum number of required measurements for successful recovery is first derived in terms of its probabilistic lower bound. Then, a CSS scheme is proposed by tightening the derived lower bound, where the key is the design of a nonlinear exponential indicator through a general-purpose sampling-controlled algorithm (SCA). In particular, a sampling-controlled BOMP (SC-BOMP) is developed through a holistic integration of the existing BOMP and the proposed SCA. For fast implementation, a modified version of SC-BOMP is further developed by exploring the block orthogonality in the form of sub-coherence of measurement matrices, which allows more compressive sampling in terms of smaller lower bound of the number of measurements. Such a fast SC-BOMP scheme achieves a desired tradeoff between the complexity and the performance. Simulations demonstrate that the two SC-BOMP schemes outperform the other benchmark algorithms.

For distributed learning among collaborative users, this paper develops and analyzes a communication-efficient scheme for federated learning (FL) over the air, which incorporates 1-bit compressive sensing (CS) into analog-aggregation transmissions. To facilitate design parameter optimization, we analyze the efficacy of the proposed scheme by deriving a closed-form expression for the expected convergence rate. Our theoretical results unveil the tradeoff between convergence performance and communication efficiency as a result of the aggregation errors caused by sparsification, dimension reduction, quantization, signal reconstruction and noise. Then, we formulate a joint optimization problem to mitigate the impact of these aggregation errors through joint optimal design of worker scheduling and power scaling policy. An enumeration-based method is proposed to solve this non-convex problem, which is optimal but becomes computationally infeasible as the number of devices increases. For scalable computing, we resort to the alternating direction method of multipliers (ADMM) technique to develop an efficient implementation that is suitable for large-scale networks. Simulation results show that our proposed 1-bit CS based FL over the air achieves comparable performance to the ideal case where conventional FL without compression and quantification is applied over error-free aggregation, at much reduced communication overhead and transmission latency.

This paper proposes deep kernel learning networks with multiple learning paths (DKL-MLP) for nonlinear function approximation. Leveraging the random feature (RF) mapping technique, kernel methods can be implemented as a two-layer neural network, at drastically reduced workload on weight training. Motivated by the representation power of the deep architecture in deep neural networks, we devise a vanilla deep kernel learning network (DKL) by applying RF mapping at each layer and learn the last layer only. To improve the learning performance of DKL, we add multiple trainable paths to DKL and develop the DKL-MLP method so that some implicit information from earlier hidden layers to the output layer can be learned. We prove that both DKL and DKL-MLP permit universal representation of a wide variety of interesting functions with arbitrarily small error and have no bad local minimum. Numerical experiments on both regression and classification tasks are provided to demonstrate the learning performance and computational efficiency of the proposed methods.

In this paper, we focus on federated learning (FL) over the air based on analog aggregation transmission in realistic wireless networks. We first derive a closed-form expression for the expected convergence rate of FL over the air, which theoretically quantifies the impact of analog aggregation on FL. Based on that, we further develop a joint optimization model for accurate FL implementation, which allows a parameter server to select a subset of edge devices and determine an appropriate power scaling factor. Such a joint optimization of device selection and power control for FL over the air is then formulated as an mixed integer programming problem. Finally, we efficiently solve this problem via a simple finite-set search method. Simulation results show that the proposed solutions developed for wireless channels outperform a benchmark method, and could achieve comparable performance of the ideal case where FL is implemented over reliable and error-free wireless channels.

We explore the capability of plain Vision Transformers (ViTs) for semantic segmentation and propose the SegViT. Previous ViT-based segmentation networks usually learn a pixel-level representation from the output of the ViT. Differently, we make use of the fundamental component—attention mechanism, to generate masks for semantic segmentation. Specifically, we propose the Attention-to-Mask (ATM) module, in which the similarity maps between a set of learnable class tokens and the spatial feature maps are transferred to the segmentation masks. Experiments show that our proposed SegViT using the ATM module outperforms its counterparts using the plain ViT backbone on the ADE20K dataset and achieves new state-of-the-art performance on COCO-Stuff-10K and PASCAL-Context datasets. Furthermore, to reduce the computational cost of the ViT backbone, we propose query-based down-sampling (QD) and query-based up-sampling (QU) to build a Shrunk structure. With our Shrunk structure, the model can save up to 40% computations while maintaining competitive performance.

We present a simple yet effective fully convolutional one-stage 3D object detector for LiDAR point clouds of autonomous driving scenes, termed FCOS-LiDAR. Unlike the dominant methods that use the bird-eye view (BEV), our proposed detector detects objects from the range view (RV, a.k.a. range image) of the LiDAR points. Due to the range view's compactness and compatibility with the LiDAR sensors' sampling process on self-driving cars, the range view-based object detector can be realized by solely exploiting the vanilla 2D convolutions, departing from the BEV-based methods which often involve complicated voxelization operations and sparse convolutions. For the first time, we show that an RV-based 3D detector with standard 2D convolutions alone can achieve comparable performance to state-of-the-art BEV-based detectors while being significantly faster and simpler. More importantly, almost all previous range view-based detectors only focus on single-frame point clouds since it is challenging to fuse multi-frame point clouds into a single range view. In this work, we tackle this challenging issue with a novel range view projection mechanism, and for the first time demonstrate the benefits of fusing multi-frame point clouds for a range-view based detector. Extensive experiments on nuScenes show the superiority of our proposed method and we believe that our work can be strong evidence that an RV-based 3D detector can compare favourably with the current mainstream BEV-based detectors. Code will be made publicly available.

Conversational recommender systems (CRSs) provide recommendations through interactive conversations. CRSs typically provide recommendations through relatively straightforward interactions, where the system continuously inquires about a user's explicit attribute-aware preferences and then decides which items to recommend. In addition, topic tracking is often used to provide naturally sounding responses. However, merely tracking topics is not enough to recognize a user's real preferences in a dialogue.

As a promising distributed learning technology, analog aggregation-based federated learning over the air (FLOA) provides high communication efficiency and privacy provisioning under the edge computing paradigm. When all edge devices (workers) simultaneously upload their local updates to the parameter server (PS) through commonly shared time-frequency resources, the PS obtains the averaged update only rather than the individual local ones. While such a concurrent transmission and aggregation scheme reduces the latency and communication costs, it unfortunately renders FLOA vulnerable to Byzantine attacks. Aiming at Byzantine-resilient FLOA, this article starts from analyzing the channel inversion (CI) mechanism that is widely used for power control in FLOA. Our theoretical analysis indicates that although CI can achieve good learning performance in the benign scenarios, it fails to work well with limited defensive capability against Byzantine attacks. Then, we propose a novel scheme called the best effort voting (BEV) power control policy that is integrated with stochastic gradient descent (SGD). Our BEV-SGD enhances the robustness of FLOA to Byzantine attacks, by allowing all the workers to send their local updates at their maximum transmit power. Under worst-case attacks, we derive the expected convergence rates of FLOA with CI and BEV power control policies, respectively. The rate comparison reveals that our BEV-SGD outperforms its counterpart with CI in terms of better convergence behavior, which is verified by experimental simulations.

In computer vision, object detection is one of most important tasks, which underpins a few instance-level recognition tasks and many downstream applications. Recently one-stage methods have gained much attention over two-stage approaches due to their simpler design and competitive performance. Here we propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to other dense prediction problems such as semantic segmentation. Almost all state-of-the-art object detectors such as RetinaNet, SSD, YOLOv3, and Faster R-CNN rely on pre-defined anchor boxes. In contrast, our proposed detector FCOS is anchor box free, as well as proposal free. By eliminating the pre-defined set of anchor boxes, FCOS completely avoids the complicated computation related to anchor boxes such as calculating the intersection over union (IoU) scores during training. More importantly, we also avoid all hyper-parameters related to anchor boxes, which are often sensitive to the final detection performance. With the only post-processing non-maximum suppression (NMS), we demonstrate a much simpler and flexible detection framework achieving improved detection accuracy. We hope that the proposed FCOS framework can serve as a simple and strong alternative for many other instance-level tasks. Code is available at: git.io/AdelaiDet.

This paper proposes an efficient direction of departure (DOD) and direction of arrival (DOA) estimation method for multi-input multi-output (MIMO) systems. For uncorrelated scenarios, the redundancy of the covariance matrix is first exploited by establishing its concise representation through redundancy reduction, which transforms the original large-size covariance matrix into a smaller-size matrix without loss of useful angle information. Then, the resulting transformed matrix, which retains a salient structure, permits efficient two-dimensional (2D) angle estimators working on a reduced-size problem for DOD and DOA estimation. Compared with conventional subspace-based methods, the proposed method incorporating an appropriate 2D angle estimator is more computationally efficient and can achieve higher estimation accuracy for small numbers of snapshots and low signal-to-noise ratios, which are verified by simulation results.

In distributed learning and optimization, a network of multiple computing units coordinates to solve a large-scale problem. This article focuses on dynamic optimization over a decentralized network. We develop a communication-efficient algorithm based on the alternating direction method of multipliers (ADMM) with quantized and censored communications, termed DQC-ADMM. At each time of the algorithm, the nodes collaborate to minimize the summation of their time-varying, local objective functions. Through local iterative computation and communication, DQC-ADMM is able to track the time-varying optimal solution. Different from traditional approaches requiring transmissions of the exact local iterates among the neighbors at every time, we propose to quantize the transmitted information, as well as adopt a communication-censoring strategy for the sake of reducing the communication cost in the optimization process. To be specific, a node transmits the quantized version of the local information to its neighbors, if and only if the value sufficiently deviates from the one previously transmitted. We theoretically justify that the proposed DQC-ADMM is capable of tracking the time-varying optimal solution, subject to a bounded error caused by the quantized and censored communications, as well as the system dynamics. Through numerical experiments, we evaluate the tracking performance and communication savings of the proposed DQC-ADMM.

Federated learning (FL) is an attractive paradigm for making use of rich distributed data while protecting data privacy. Nonetheless, non-ideal communication links and limited transmission resources may hinder the implementation of fast and accurate FL. In this paper, we study joint optimization of communications and FL based on analog aggregation transmission in realistic wireless networks. We first derive closed-form expressions for the expected convergence rate of FL over the air, which theoretically quantify the impact of analog aggregation on FL. Based on the analytical results, we develop a joint optimization model for accurate FL implementation, which allows a parameter server to select a subset of workers and determine an appropriate power scaling factor. Since the practical setting of FL over the air encounters unobservable parameters, we reformulate the joint optimization of worker selection and power allocation using controlled approximation. Finally, we efficiently solve the resulting mixed-integer programming problem via a simple yet optimal finite-set search method by reducing the search space. Simulation results show that the proposed solutions developed for realistic wireless analog channels outperform a benchmark method, and achieve comparable performance of the ideal case where FL is implemented over error-free wireless channels.

We propose a fully convolutional multi-person pose estimation framework using dynamic instance-aware convolutions, termed FCPose. Different from existing methods, which often require ROI (Region of Interest) operations and/or grouping post-processing, FCPose eliminates the ROIs and grouping post-processing with dynamic instance-aware keypoint estimation heads. The dynamic keypoint heads are conditioned on each instance (person), and can encode the instance concept in the dynamically-generated weights of their filters. Moreover, with the strong representation capacity of dynamic convolutions, the keypoint heads in FCPose are designed to be very compact, resulting in fast inference and making FCPose have almost constant inference time regardless of the number of persons in the image. For example, on the COCO dataset, a real-time version of FCPose using the DLA-34 backbone infers about 4.5×faster than Mask R-CNN (ResNet-101) (41.67 FPS vs. 9.26 FPS) while achieving improved performance (64.8% APkpvs. 64.3% APkp). FCPose also offers better speed/accuracy trade-off than other state-of-the-art methods. Our experiment results show that FCPose is a simple yet effective multi-person pose estimation framework. Code is available at: https://git.io/AdelaiDet

We present a high-performance method that can achieve mask-level instance segmentation with only bounding-box annotations for training. While this setting has been studied in the literature, here we show significantly stronger performance with a simple design (e.g., dramatically improving previous best reported mask AP of 21.1% [13] to 31.6% on the COCO dataset). Our core idea is to redesign the loss of learning masks in instance segmentation, with no modification to the segmentation network itself. The new loss functions can supervise the mask training without relying on mask annotations. This is made possible with two loss terms, namely, 1) a surrogate term that minimizes the discrepancy between the projections of the ground-truth box and the predicted mask; 2) a pairwise loss that can exploit the prior that proximal pixels with similar colors are very likely to have the same category label.Experiments demonstrate that the redesigned mask loss can yield surprisingly high-quality instance masks with only box annotations. For example, without using any mask annotations, with a ResNet-101 backbone and 3× training schedule, we achieve 33.2% mask AP on COCO test-dev split (vs. 39.1% of the fully supervised counterpart). Our excellent experiment results on COCO and Pascal VOC indicate that our method dramatically narrows the performance gap between weakly and fully supervised instance segmentation.Code is available at: https://git.io/AdelaiDet

The problem of recovering heavy components of a high-dimensional vector from compressed data is of great interest in broad applications, such as feature extraction under scarce computing memory and distributed learning under limited bandwidth. Recently, a compression algorithm called count sketch has gained wide popularity to recover heavy components in various fields. In this paper, we carefully analyze count sketch and illustrate that its default recovery method, namely median filtering, has a distinct error pattern of reporting false positives. To counteract this error pattern, we propose a new scheme called zero checking which adopts a two-step recovery approach to improve the probability of detecting false positives. Our proposed technique builds on rigorous error analysis, which enables us to optimize the selection of a key design parameter for maximum performance gain. The empirical results show that our scheme achieves better recovery accuracy than median filtering and requires less samples to accurately recover heavy components.

Federated learning learns from scattered data by fusing collaborative models from local nodes. However, conventional coordinate-based model averaging by FedAvg ignored the random information encoded per parameter and may suffer from structural feature misalignment. In this work, we propose Fed2, a feature-aligned federated learning framework to resolve this issue by establishing a firm structure-feature alignment across the collaborative models. Fed2 is composed of two major designs: First, we design a feature-oriented model structure adaptation method to ensure explicit feature allocation in different neural network structures. Applying the structure adaptation to collaborative models, matchable structures with similar feature information can be initialized at the very early training stage. During the federated learning process, we then propose a feature paired averaging scheme to guarantee aligned feature distribution and maintain no feature fusion conflicts under either IID or non-IID scenarios. Eventually, Fed2 could effectively enhance the federated learning convergence performance under extensive homo- and heterogeneous settings, providing excellent convergence speed, accuracy, and computation/communication efficiency.

Very recently, a variety of vision transformer architectures for dense prediction tasks have been proposed and they show that the design of spatial attention is critical to their success in these tasks. In this work, we revisit the design of the spatial attention and demonstrate that a carefully devised yet simple spatial attention mechanism performs favorably against the state-of-the-art schemes. As a result, we propose two vision transformer architectures, namely, Twins- PCPVT and Twins-SVT. Our proposed architectures are highly efficient and easy to implement, only involving matrix multiplications that are highly optimized in modern deep learning frameworks. More importantly, the proposed architectures achieve excellent performance on a wide range of visual tasks including image-level classification as well as dense detection and segmentation. The simplicity and strong performance suggest that our proposed architectures may serve as stronger backbones for many vision tasks.

Semantic segmentation requires per-pixel prediction for a given image. Typically, the output resolution of a segmentation network is severely reduced due to the downsampling operations in the CNN backbone. Most previous methods employ upsampling decoders to recover the spatial resolution.Various decoders were designed in the literature. Here, we propose a novel decoder, termed dynamic neural representational decoder (NRD), which is simple yet significantly more efficient. As each location on the encoder's output corresponds to a local patch of the semantic labels, in this work, we represent these local patches of labels with compact neural networks. This neural representation enables our decoder to leverage the smoothness prior in the semantic label space, and thus makes our decoder more efficient. Furthermore, these neural representations are dynamically generated and conditioned on the outputs of the encoder networks. The desired semantic labels can be efficiently decoded from the neural representations, resulting in high-resolution semantic segmentation predictions.We empirically show that our proposed decoder can outperform the decoder in DeeplabV3+ with only $\sim$$30\%$ computational complexity, and achieve competitive performance with the methods using dilated encoders with only $\sim$$15\% $ computation. Experiments on Cityscapes, ADE20K, and Pascal Context demonstrate the effectiveness and efficiency of our proposed method.

This paper studies the decentralized optimization and learning problem where multiple interconnected agents aim to learn an optimal decision function defined over a reproducing kernel Hilbert space by jointly minimizing a global objective function, with access to their own locally observed dataset. As a non-parametric approach, kernel learning faces a major challenge in distributed implementation: the decision variables of local objective functions are data-dependent and thus cannot be optimized under the decentralized consensus framework without any raw data exchange among agents. To circumvent this major challenge, we leverage the random feature (RF) approximation approach to enable consensus on the function modeled in the RF space by data-independent parameters across different agents. We then design an iterative algorithm, termed DKLA, for fast-convergent implementation via ADMM. Based on DKLA, we further develop a communication-censored kernel learning (COKE) algorithm that reduces the communication load of DKLA by preventing an agent from transmitting at every iteration unless its local updates are deemed informative. Theoretical results in terms of linear convergence guarantee and generalization performance analysis of DKLA and COKE are provided. Comprehensive tests on both synthetic and real datasets are conducted to verify the communication efficiency and learning effectiveness of COKE.

The Kalman filter has successful applications in many fields. The applicability of the standard Kalman filter critically hinges on the accurate prior knowledge of all the system model parameters. However, in practical applications, it can be difficult or unrealistic to obtain these parameters, in which case it is a common practice to employ some pre-determined alternatives for the unknown model parameters. This letter investigates the case of unknown initial state priors by assessing how their pre-determined alternatives affect the performance of the Kalman filter. The ranking of three types of mean squared errors is established. It is found that the definiteness of the initial state prior deviation is critical, which determine the ranking of the three types of mean squared errors. Such results provide guideline on the choice of the pre-determined initial state prior. Numerical examples are provided to validate the results.

Many applications such as smart metering and location based services pose strong privacy requirements but achieving privacy protection at the client side is a non-trial problem as payment for the services must be computed by the server at the end of each billing period. In this paper, we propose a privacy preservation and billing scheme termed PPDIR based on delayed information release. PPDIR relies on a novel group signature mechanism and the asymmetric Rabin cryptosystem to protect the privacy of the clients and their requests, to achieve accountability and non-repudiation, and to shift the computational complexity to the server side. It adopts a secret token for anonymity and the token is updated for each client at the beginning of each billing period and securely released only to the server at the end of the billing period. Such a strategy can prevent the server from linking a client's requests made at different billing periods. It also prevents any adversary from linking any request to any client. Note that the server is able to figure out all requests made by a client within a billing period after receiving the delayed token, which is unavoidable for billing purpose. We prove the security properties of the group signature scheme, and analyze the security strength of PPDIR. Our study indicates that PPDIR can achieve privacy-preservation, confidentiality, non-repudiation, accountability, and other security objectives. We also evaluate the performance of our scheme in terms of communication and computational overheads.

Instance segmentation is one of the fundamental vision tasks. Recently, fully convolutional instance segmentation methods have drawn much attention as they are often simpler and more efficient than two-stage approaches like Mask R-CNN. To date, almost all such approaches fall behind the two-stage Mask R-CNN method in mask precision when models have similar computation complexity, leaving great room for improvement. In this work, we achieve improved mask prediction by effectively combining instance-level information with semantic information with lower-level fine-granularity. Our main contribution is a blender module which draws inspiration from both top-down and bottom-up instance segmentation approaches. The proposed BlendMask can effectively predict dense per-pixel position-sensitive instance features with very few channels, and learn attention maps for each instance with merely one convolution layer, thus being fast in inference. BlendMask can be easily incorporate with the state-of-the-art one-stage detection frameworks and outperforms Mask R-CNN under the same training schedule while being faster. A light-weight version of BlendMask achieves 36.0 mAP at 27 FPS evaluated on a single 1080Ti. Because of its simplicity and efficacy, we hope that our BlendMask could serve as a simple yet strong baseline for a wide range of instance-wise prediction tasks.

The success of deep neural networks relies on significant architecture engineering. Recently neural architecture search (NAS) has emerged as a promise to greatly reduce manual effort in network design by automatically searching for optimal architectures, although typically such algorithms need an excessive amount of computational resources, e.g., a few thousand GPU-days. To date, on challenging vision tasks such as object detection, NAS, especially fast versions of NAS, is less studied. Here we propose to search for the decoder structure of object detectors with search efficiency being taken into consideration. To be more specific, we aim to efficiently search for the feature pyramid network (FPN) as well as the prediction head of a simple anchor-free object detector, namely FCOS, using a tailored reinforcement learning paradigm. With carefully designed search space, search algorithms and strategies for evaluating network quality, we are able to efficiently search a top-performing detection architecture within 4 days using 8 V100 GPUs. The discovered architecture surpasses state-of-the-art object detection models (such as Faster R-CNN, RetinaNet and FCOS) by 1.5 to 3.5 points in AP on the COCO dataset, with comparable computation complexity and memory footprint, demonstrating the efficacy of the proposed NAS for object detection.

To date, instance segmentation is dominated by two-stage methods, as pioneered by Mask R-CNN. In contrast, one-stage alternatives cannot compete with Mask R-CNN in mask AP, mainly due to the difficulty of compactly representing masks, making the design of one-stage methods very challenging. In this work, we propose a simple single-shot instance segmentation framework, termed mask encoding based instance segmentation (MEInst). Instead of predicting the two-dimensional mask directly, MEInst distills it into a compact and fixed-dimensional representation vector, which allows the instance segmentation task to be incorporated into one-stage bounding-box detectors and results in a simple yet efficient instance segmentation framework. The proposed one-stage MEInst achieves 36.4% in mask AP with single-model (ResNeXt-101-FPN backbone) and single-scale testing on the MS-COCO benchmark. We show that the much simpler and flexible one-stage instance segmentation method, can also achieve competitive performance. This framework can be easily adapted for other instance-level recognition tasks. Code is available at: git.io/AdelaiDet

Recently Convolutional Neural Networks (CNNs) are widely applied into novel intelligent applications and systems. However, the CNN computation performance is significantly hindered by its computation flow, which computes the model structure sequentially by layers with massive convolution operations. Such a layer-wise sequential computation flow can cause certain performance issues, such as resource under-utilization, huge memory overhead, etc. To solve these problems, we propose a novel CNN structural decoupling method, which could decouple CNN models into "critical paths" and eliminate the inter-layer data dependency. Based on this method, we redefine the CNN computation flow into parallel and cascade computing paradigms, which can significantly enhance the CNN computation performance with both multi-core and single-core CPU processors. Experiments show that, our DC-CNN framework could reduce 24% to 33% latency on multi-core CPUs for CIFAR and ImageNet. On small-capacity mobile platforms, cascade computing could reduce the latency by average 24% on ImageNet and 42% on CIFAR10. Meanwhile, the memory reduction could also reach average 21% and 64%, respectively.

This paper concerns wideband direction of arrival (DoA) estimation with sparse linear arrays (SLAs). We rely on the assumption that the power spectrum of the wideband sources is the same up to a scaling factor, which could in theory allow us to resolve not only more sources than the number of antennas but also more sources than the number of degrees of freedom (DoF) of the difference co-array of the SLA. We resort to the Jacobi-Anger approximation to transform the coarray response matrices of all frequency bins into a single virtual uniform linear array (ULA) response matrix. Based on the obtained model, two super-resolution DoA estimation approaches based on atomic norm minimization (ANM) are proposed, one with and one without prior knowledge of the power spectrum. Simulation results show that our proposed methods outperform the state of the art and are indeed capable of resolving more sources than the number of DoF of the difference co-array.

Higher-order statistics (HOS) estimation hinges on the availability of a huge amount of data records, which causes exceedingly high sampling rates and overwhelming energy consumption for the sampling devices, especially when dealing with wideband signals. To overcome these challenges, this paper develops a novel compressive cumulant slice sensing (CCSS) method that aims to efficiently reconstructing the 1D diagonal slice of higher-order cumulants under the compressive sensing framework. It first parsimoniously collects data with a properly designed sampler based on the minimal sparse ruler principle, and then accurately reconstructs the 1D diagonal cumulant slice from those compressive measurements. The reconstructed HOS can be useful in many signal processing tasks, which is illustrated through the task of line spectrum estimation with resilience to colored Gaussian noise at reduced sampling rates.

This paper develops an enhanced low-rank structured covariance reconstruction (LRSCR) method based on the decoupled atomic norm minimization (D-ANM), for super-resolution two-dimensional (2D) harmonic retrieval with multiple measurement vectors. This LRSCR-D-ANM approach exploits a potential structure hidden in the covariance by transferring the basic LRSCR to an efficient D-ANM formulation, which permits a sparse representation over a matrix-form atom set with decoupled 1D frequency components. The new LRSCR-D-ANM method builds upon the existence of a generalized Vandermonde decomposition of its solution, which otherwise cannot be guaranteed by the basic LRSCR unless a very conservative condition holds. Further, a low-complexity solution of the LRSCR-D-ANM is provided for fast implementation with negligible performance loss. Simulation results verify the advantages of the proposed LRSCR-D-ANM over the basic LRSCR, in terms of the wider applicability and the lower complexity.

While social networking sites gain massive popularity for their friendship networks, user privacy issues arise due to the incorporation of location-based services (LBS) into the system. Preferential LBS takes a user’s social profile along with their location to generate personalized recommender systems. With the availability of the user’s profile and location history, we often reveal sensitive information to unwanted parties. Hence, providing location privacy to such preferential LBS requests has become crucial. However, the current technologies focus on anonymizing the location through granularity generalization. Such systems, although provides the required privacy, come at the cost of losing accurate recommendations. Hence, in this paper, we propose a novel location privacy-preserving mechanism that provides location privacy through k -anonymity and provides the most accurate results. Experimental results that focus on mobile users and context-aware LBS requests prove that the proposed method performs superior to the existing methods.

Both accuracy and efficiency are of significant importance to the task of semantic segmentation. Existing deep FCNs suffer from heavy computations due to a series of high-resolution feature maps for preserving the detailed knowledge in dense estimation. Although reducing the feature map resolution (i.e., applying a large overall stride) via subsampling operations (e.g., polling and convolution striding) can instantly increase the efficiency, it dramatically decreases the estimation accuracy. To tackle this dilemma, we propose a knowledge distillation method tailored for semantic segmentation to improve the performance of the compact FCNs with large overall stride. To handle the inconsistency between the features of the student and teacher network, we optimize the feature similarity in a transferred latent domain formulated by utilizing a pre-trained autoencoder. Moreover, an affinity distillation module is proposed to capture the long-range dependency by calculating the non local interactions across the whole image. To validate the effectiveness of our proposed method, extensive experiments have been conducted on three popular benchmarks: Pascal VOC, Cityscapes and Pascal Context. Built upon a highly competitive baseline, our proposed method can improve the performance of a student network by 2.5% (mIOU boosts from 70.2 to 72.7 on the cityscapes test set) and can train a better compact model with only 8% float operations (FLOPS) of a model that achieves comparable performances.

Recent semantic segmentation methods exploit encoder-decoder architectures to produce the desired pixel-wise segmentation prediction. The last layer of the decoders is typically a bilinear upsampling procedure to recover the final pixel-wise prediction. We empirically show that this oversimple and data-independent bilinear upsampling may lead to sub-optimal results. In this work, we propose a data-dependent upsampling (DUpsampling) to replace bilinear, which takes advantages of the redundancy in the label space of semantic segmentation and is able to recover the pixel-wise prediction from low-resolution outputs of CNNs. The main advantage of the new upsampling layer lies in that with a relatively lower-resolution feature map such as 1/16 or 1/32 of the input size, we can achieve even better segmentation accuracy, significantly reducing computation complexity. This is made possible by 1) the new upsampling layer's much improved reconstruction capability; and more importantly 2) the DUpsampling based decoder's flexibility in leveraging almost arbitrary combinations of the CNN encoders' features. Experiments on PASCAL VOC demonstrate that with much less computation complexity, our decoder outperforms the state-of-the-art decoder. Finally, without any post-processing, the framework equipped with our proposed decoder achieves new state-of-the-art performance on two datasets: 88.1% mIOU on PASCAL VOC with 30% computation of the previously best model; and 52.5% mIOU on PASCAL Context.

This paper aims at accurate channel estimation for millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems under practical limitations, including an arbitrary array geometry and a hybrid hardware structure. Taking on an angle-based approach, this work adopts a generalized array manifold separation approach via the Jacobi-Anger approximation, which transforms a non-ideal, non-uniform array manifold into a virtual array domain with a desired uniform geometric structure to facilitate super-resolution angle estimation and channel acquisition. Accordingly, structure-based optimization techniques are developed to estimate the channel parameters within a short sensing time. In particular, the difference in time-variation of path angles and path gains is capitalized to design a two-step scheme that can quickly sense fading channels. Theoretical results are provided on the fundamental limits of the proposed technique in terms of sample efficiency. Simulations testify the effectiveness of the proposed approaches.

This paper develops efficient super-resolution spatial channel covariance estimation techniques for millimeter-wave (mmWave) massive multiple-input multiple-output (MIMO) systems under the hybrid precoding constraint. Two structure-based optimization techniques including low-rank structured covariance reconstruction and dynamic atomic norm minimization are proposed to accurately estimate the channel covariance matrix. For computational efficiency, a fast iterative algorithm is developed via the alternating direction method of multipliers. The extension of this work to the higher-dimensional cases is also discussed. Simulation results verify the effectiveness of the proposed methods in hybrid mmWave massive MIMO systems.

Orthogonal Matching Pursuit is an iterative greedy algorithm used to find a sparse approximation for high-dimensional signals. The algorithm is most popularly used in Compressive Sensing, which allows for the reconstruction of sparse signals at rates lower than the Shannon-Nyquist frequency, which has traditionally been used in a number of applications such as MRI and computer vision and is increasingly finding its way into Big Data and data center analytics. OMP traditionally suffers from being computationally intensive and time-consuming, this is particularly a problem in the area of Big Data where the demand for computational resources continues to grow. In this paper, the data-level parallelization of OMP through blocking is examined. Traditionally blocking has been used to ac- celerate the performance of OMP reconstruction for big data image analytics. However, as we show in this work, blocking, particularly in the form of vectorizing, introduces significant error in terms of PSNR and SSIM index in the reconstruction quality. In response, we deploy the concept of stitching to recover the lost accuracy. We further examine the influence of the level of blocking and amount of stitching (overlap between each block) with regard to recon- struction time and reconstructed image quality. While stitching boosts up the image reconstruction accuracy significantly, the ob- ject detection count results show anywhere from 11.84% to 140.54% improvement, depending on the cases being compared, it introduces significant overhead with regard to reconstruction time. To address the overhead, we deploy hardware accelerated base solutions. Given the emergence of hardware accelerators in data centers and for big data analytics in form of FPGAs, our solution effectively utilizes this resource to enhance the performance overhead of stitching by 25%. We show the minimum block size required for an FPGA speed-up.

This paper proposes a communication- and computation-efficient algorithm to solve a convex consensus optimization problem defined over a decentralized network. A remarkable existing algorithm to solve this problem is the alternating direction method of multipliers (ADMM), in which at every iteration every node updates its local variable through combining neighboring variables and solving an optimization subproblem. The proposed algorithm, called as communication-censored linearized ADMM (COLA), leverages a linearization technique to reduce the iteration-wise computation cost of ADMM and uses a communication-censoring strategy to alleviate the communication cost. To be specific, COLA introduces successive linearization approximations to the local cost functions such that the resultant computation is first-order and light-weight. Since the linearization technique slows down the convergence speed, COLA further adopts the communication-censoring strategy to avoid transmissions of less informative messages. A node is allowed to transmit only if the distance between the current local variable and its previously transmitted one is larger than a censoring threshold. We establish convergence as well as sublinear and linear rates of convergence of COLA, and demonstrate its satisfactory communication-computation tradeoff with numerical experiments.

We propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to semantic segmentation. Almost all state-of-the-art object detectors such as RetinaNet, SSD, YOLOv3, and Faster R-CNN rely on pre-defined anchor boxes. In contrast, our proposed detector FCOS is anchor box free, as well as proposal free. By eliminating the pre-defined set of anchor boxes, FCOS completely avoids the complicated computation related to anchor boxes such as calculating overlapping during training. More importantly, we also avoid all hyper-parameters related to anchor boxes, which are often very sensitive to the final detection performance. With the only post-processing non-maximum suppression (NMS), FCOS with ResNeXt-64x4d-101 achieves 44.7% in AP with single-model and single-scale testing, surpassing previous one-stage detectors with the advantage of being much simpler. For the first time, we demonstrate a much simpler and flexible detection framework achieving improved detection accuracy. We hope that the proposed FCOS framework can serve as a simple and strong alternative for many other instance-level tasks. Code is available at: https://tinyurl.com/FCOSv1.

Recent progress has been made on developing a unified framework for joint text detection and recognition in natural images, but existing joint models were mostly built on two-stage framework by involving ROI pooling, which can degrade the performance on recognition task. In this work, we propose convolutional character networks, referred as CharNet, which is an one-stage model that can process two tasks simultaneously in one pass. CharNet directly outputs bounding boxes of words and characters, with corresponding character labels. We utilize character as basic element, allowing us to overcome the main difficulty of existing approaches that attempted to optimize text detection jointly with a RNN-based recognition branch. In addition, we develop an iterative character detection approach able to transform the ability of character detection learned from synthetic data to real-world images. These technical improvements result in a simple, compact, yet powerful one-stage model that works reliably on multi-orientation and curved text. We evaluate CharNet on three standard benchmarks, where it consistently outperforms the state-of-the-art approaches [25, 24] by a large margin, e.g., with improvements of 65.33%→71.08% (with generic lexicon) on ICDAR 2015, and 54.0%→69.23% on Total-Text, on end-to-end text recognition. Code is available at: https://github.com/MalongTech/research-charnet.

Compressive sensing (CS) has been widely used in the Internet of Things (IoT) to achieve efficient information collection. However, existing works have mainly focused on utilizing CS to lower the sampling rate or reduce the number of transmissions, without explicitly accounting for the heterogeneity of energy consumption in IoT environments. In this paper, we propose a CS-based prejudiced random sensing strategy (PRSS) that explicitly considers the heterogeneous energy consumption of IoT sensor nodes at different locations, in order to accurately attain a desired tradeoff between the overall energy consumption and the sensing accuracy. Specifically, each sensor node participates in sensing via distributed random access based on an assigned sensing probability, which is determined by its energy consumption in sending the sensed data, data collision rate and its contribution to recovery accuracy. We employ the statistical restricted isometry property as a practical indicator of the recovery accuracy and derive a sufficiently good recovery error bound based on it. Accordingly, we devise a novel convex optimization framework to find the most energy-efficient sensing probability assignment strategy with accuracy guarantee. We evaluate the PRSS using real-world sea surface temperature data traces. Comparative simulations corroborate that the PRSS can significantly reduce energy consumption and prolong network lifetime without sacrificing sensing accuracy.

Health-related Internet of Things (IoT) devices are becoming more popular in recent years. On the one hand, users can access information of their health conditions more conveniently; on the other hand, they are exposed to new security risks. In this paper, we presented, to the best of our knowledge, the first in-depth security analysis on home-use electroencephalography (EEG) IoT devices. Our key contributions are twofold. First, we reverse-engineered the home-use EEG system framework via which we identified the design and implementation flaws. By exploiting these flaws, we developed two sets of novel easy-to-exploit PoC attacks, which consist of four remote attacks and one proximate attack. In a remote attack, an attacker can steal a user's brain wave data through a carefully crafted program while in the proximate attack, the attacker can steal a victim's brain wave data over-the-air without accessing the victim's device on any sense when he is close to the victim. As a result, all the 156 brain-computer interface (BCI) apps in the NeuroSky App store are vulnerable to the proximate attack. We also discovered that all the 31 free apps in the NeuroSky App store are vulnerable to at least one remote attack. Second, we proposed a novel deep learning model of a joint recurrent convolutional neural network (RCNN) to infer a user's activities based on the reduced-featured EEG data stolen from the home-use EEG IoT devices, and our evaluation over the real-world EEG data indicates that the inference accuracy of the proposed RCNN is can reach 70.55%.

Recent trends show that the popularity of online social networks (OSNs) has been increasing rapidly. From daily communication sites to online communities, an average person's daily life has become dependent on these online networks. Hence, it has become evident that protection should be provided to these networks from unwanted intruders. In this paper, we consider the data privacy on OSNs at the network level rather than the user level. This network-level privacy helps us to prevent information leakage to third-party users, such as advertisers. We propose a novel scheme that combines the privacy of all the elements of a social network: node, edge, and attribute privacy by clustering the users based on their attribute similarity. We use an enhanced equi-cardinal clustering (ECC) as a way to achieve k-anonymity. We further improve k-anonymity with l-diversity. Our proposed enhanced ECC ensures that there are at least “k” users in any given network as well as the attributes in each cluster has at least l-distinct values. We further provide proofs on how the proposed ECC ensures k-anonymity and the maximum information loss. We consider a weighted directed social network graph as an input to our method to consider the existing complexities in a social network. With the help of two real-world data sets, we evaluate this method in terms of privacy and efficiency.

Cooperative jamming is deemed as a promising physical layer-based approach to secure wireless transmissions in the presence of eavesdroppers. In this paper, we investigate cooperative jamming in a two-tier 5G heterogeneous network (HetNet), where the macrobase stations (MBSs) at the macrocell tier are equipped with large-scale antenna arrays to provide space diversity and the local base stations (LBSs) at the local cell tier adopt non-orthogonal multiple access (NOMA) to accommodate dense local users (LUs). In the presence of imperfect channel state information, we propose three robust secrecy transmission algorithms that can be applied to various scenarios with different security requirements. The first algorithm employs robust beamforming (RBA) that aims to optimize the secrecy rate of a marcouser (MU) in a macrocell. The second algorithm provides robust power allocation (RPA) that can optimize the secrecy rate of an LU in a local cell. The third algorithm tackles a robust joint optimization (RJO) problem across tiers that seek the maximum secrecy sum rate of a target MU and a target LU robustly. We employ convex optimization techniques to find feasible solutions to these highly non-convex problems. The numerical results demonstrate that the proposed algorithms are highly effective in improving the secrecy performance of a two-tier HetNet.

In this paper, we design an uplink/downlink channel estimation method for millimeter wave (mmWave) massive multiple-input multiple-output (MIMO) systems and investigate the impact of beam squint effect that accompanies large array configuration. Specifically, we adopt the off-grid sparse Bayesian learning (SBL) that directly works on the continuous angle-delay parameter domain and avoids the grid mismatch problem. Hence, the proposed method achieves good channel estimation accuracy and handles the wideband direction of arrival (DOA) estimation problem for mmWave massive MIMO communications, where beam squint effect was previously ignored by many existing literatures. The Cramér-Rao bound for unknown parameters is derived to make the proposed study complete. More importantly, a much simplified downlink channel estimation scheme is designed with the aid of angle-delay reciprocity, which significantly reduces training and feedback overhead. The simulation results are provided to demonstrate the superior performance of the proposed method over existing ones.

Text detection and recognition in natural images have long been considered as two separate tasks that are processed sequentially. Jointly training two tasks is non-trivial due to significant differences in learning difficulties and convergence rates. In this work, we present a conceptually simple yet efficient framework that simultaneously processes the two tasks in a united framework. Our main contributions are three-fold: (1) we propose a novel text-alignment layer that allows it to precisely compute convolutional features of a text instance in arbitrary orientation, which is the key to boost the performance; (2) a character attention mechanism is introduced by using character spatial information as explicit supervision, leading to large improvements in recognition; (3) two technologies, together with a new RNN branch for word recognition, are integrated seamlessly into a single model which is end-to-end trainable. This allows the two tasks to work collaboratively by sharing convolutional features, which is critical to identify challenging text instances. Our model obtains impressive results in end-to-end recognition on the ICDAR 2015 [19], significantly advancing the most recent results [2], with improvements of F-measure from (0.54, 0.51, 0.47) to (0.82, 0.77, 0.63), by using a strong, weak and generic lexicon respectively. Thanks to joint training, our method can also serve as a good detector by achieving a new state-of-the-art detection performance on related benchmarks. Code is available at https://github.com/tonghe90/textspotter.

In this paper, we investigate whether we can distinguish that a subject is making a correct or incorrect behavioral response by analyzing the fMRI data of localized brain regions, obtained from a feature-based attention experiment. For each subject, we first construct the feature vectors for each region of interest (including V1, MT or IPS1) from the fMRI signals. Second, we project the feature vectors onto a lower dimensional subspace using Linear Discriminant Analysis (LDA), where the difference between two classes (correct vs. incorrect response) is maximized. Finally, we apply the Bayesian classifier to the projected data, and find that the classification accuracies corresponding to V1, MT and IPS1 are 87.2%, 90.8% and 81.7%, respectively, when all the trials are considered. Our analysis indicates that: when people make correct or incorrect responses, significant difference exists in the fMRI signals, especially in V1 and MT regions, and the difference can be effectively captured by the LDA-Bayesian classifier. We also prove that: when the original data are normally distributed, LDA, which aims to maximize the difference between different classes, is equivalent to the optimal Maximum Likelihood (ML) based classification method.

This letter presents a fast algorithm named iterative Vandermonde decomposition and shrinkage-thresholding (IVDST), which offers a low-complexity solution to atomic norm minimization (ANM) in off-grid compressed sensing for line spectral estimation from few measurements. It implements the ANM principle via the accelerated proximal gradient (APG) technique, without invoking computationally expensive semidefinite programming (SDP). To approximate the proximal operator in each APG iteration, Vandermonde decomposition is applied to utilize the Toeplitz structure inherent in the line spectral model, and the low-rank property of the Toeplitz-structured matrix is enforced via a simple shrinkage-thresholding operation. The IVDST algorithm effectively reduces the order of computational complexity compared to SDP-based solutions. It also offers an explicit way to bridge the ANM principle with classic super-resolution line spectral estimation algorithms, such as MUSIC.

Dynamic construction of optimal portfolio is investigated. Multiple assets are allocated and rebalanced periodically based on different principles. We develop several dynamic allocation strategies to maximize long-term portfolio value based on Kelly's approach related to mutual information. We show that the resulting asset allocation strategy outperforms the traditional approaches and produces an excellent trade-off between risk and return. Out of sample simulation results are also provided to demonstrate the performance.

This paper presents an efficient optimization technique for super-resolution two-dimensional (2D) direction of arrival (DOA) estimation by introducing a new formulation of atomic norm minimization (ANM). ANM allows gridless angle estimation for correlated sources even when the number of snapshots is far less than the antenna size, yet it incurs huge computational cost in 2D processing. This paper introduces a novel formulation of ANM via semi-definite programming, which expresses the original high-dimensional problem by two decoupled Toeplitz matrices in one dimension, followed by 1D angle estimation with automatic angle pairing. Compared with the state-of-the-art 2D ANM, the proposed technique reduces the computational complexity by several orders of magnitude with respect to the antenna size, while retaining the benefits of ANMin terms of super-resolution performance with use of a small number of measurements, and robustness to source correlation and noise. The complexity benefits are particularly attractive for large-scale antenna systems such as massive MIMO and radio astronomy.

In millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems, channel estimation in the presence of sparse multipath fading boils down to two-dimensional (2D) direction-of-arrival (DOA) estimation followed by path gain estimation. To achieve super-resolution angle estimation at affordable complexity, this paper develops an efficient channel estimation approach by applying a truncated atomic norm minimization (T-ANM) technique, which is implemented via partial antenna activation during training-based channel estimation. This technique makes use of a key observation that the sparse scattering characteristics of mmWave MIMO channel gives rise to a low-rank two-level Toeplitz structure in the angular domain. Because of the low-rank property, only a subset of the transceiver antennas needs to be activated to save training resources. Meanwhile, the Toeplitz structure enables ANM-based gridless 2D DOA estimation via reduced-size semidefinite programming. Simulation results show that the proposed reduced-size method can achieve comparable spectral efficiency as the full-size benchmark method at much lower computational complexity and shorter sensing time.

In wideband spectrum sensing, an unlicensed user determines which portions of a given band have been left idle by the licensed users. A historical deficiency of wideband spectrum sensing, the inability to detect signals with low duty cycle, was addressed in a recent paper, where wideband temporal spectrum sensing was introduced. We propose an algorithm for reliable detection of low duty cycle signals in noisy environments. We leverage this recent advance in wideband spectrum sensing, and apply a well-known edge detection algorithm to determine channel boundaries. Numerical results are presented which show performance improvements over the original wideband temporal spectrum sensing algorithm, particularly in low signal-to-noise ratio scenarios.

In this paper, we propose a weighted alternating direction method of multipliers (ADMM) to solve the consensus optimization problem over a decentralized network. Compared with the conventional ADMM that is popular in decentralized network optimization, the weighted ADMM is able to tune its weight matrices for the purpose of reducing the communication cost spent in the optimization process. We first prove convergence and establish linear convergence rate of the weighted ADMM. Second, we maximize the derived convergence speed and obtain the best weight matrices on a given topology. Third, observing that exchanging information with all the neighbors is expensive, we maximize the convergence speed while limit the number of communication arcs. This strategy finds a subgraph within the underlying topology to fulfill the optimization task and leads to a favorable tradeoff between the number of iterations and the communication cost per iteration. Numerical experiments demonstrate advantages of the weighted ADMM over its conventional counterpart in expediting the convergence speed and reducing the communication cost.

In millimeter-wave (mmWave) multiple-input multiple-output (MIMO) systems, channel estimation is a challenging task in terms of acquiring the instantaneous channel state information (CSI), because both the estimation complexity and the overhead required for pilot symbols and feedback grow drastically as the number of antennas increases. Alternatively, some channel statistics in the form of partial CSI are sufficient for transceiver optimization and noncoherent detection in time-varying wireless environments. To obtain such useful statistical information accurately and efficiently, this paper proposes a new channel statistics estimation method using the compressive covariance sensing technique, which directly estimates the desired second-order statistics of the channel while bypassing the intermediate recovery of the instantaneous channel matrix itself. A diagonal-search orthogonal matching pursuit (DS-OMP) algorithm is developed for fast channel estimation. The proposed algorithm has low computational complexity and reduced overhead in training and feedback, owing to its proper utilization of the joint sparsity structure of the channel covariance matrix.

Abstract For spectrum sensing, energy detection has the advantages of low complexity, rapid analysis, and requires no knowledge of the transmission signal, which makes it suitable for a wide range of applications. However, under low signal‐to‐noise ratio conditions, the required window length (or the time‐bandwidth product) for energy detection to achieve a desired detection performance is large. In addition, conventional energy detection assumes that the detection tests are independent, that is, there is no overlap between individual detection tests. These properties significantly reduce the detection speed when energy detection is used for the continuous monitoring over a communication channel for the detection of signal transmission activities. In this paper, we propose a sliding window detection analysis with overlap among multiple tests. Algorithms for effective performance analysis of the proposed sliding window energy detection are proposed. The impact of window length on distribution of detection time is investigated. Simulation results on the proposed sliding window energy detection are also compared with the theoretically predicted and conventional energy detection performance estimates. Copyright © 2015 John Wiley & Sons, Ltd.

Mobile social networks have been increasingly popular with the explosive growth of mobile devices. Mobile users are allowed to interact with potential friends within a certain distance. Motivated by this feature, many exciting applications have been developed, yet the challenge of privacy protection is thus aroused. In this paper, we propose an efficient customized privacy preserving friend discovery mechanism, which not only protects the privacy of users' profile, but also establishes a verifiable secure communication channel between matched users. Besides, the initiator has the freedom to set a customized request profile by choosing the interested attributes and giving each attribute a specific value. Moreover, the request profile's privacy protection level is customized by the initiator according to his/her own privacy requirements. We also consider the collusion attacks among unmatched users. To the best of our knowledge, this is the first work to address such a security threat. Our protocol guarantees that only exactly matched users are able to communicate with the initiator securely, while little information can be obtained by other participants. To increase the matching efficiency, our design adopts the Bloom filter to efficiently exclude most unmatched users. As a result, our design effectively protects the profile privacy and efficiently decreases the computational overhead. Security analysis and performance evaluation are conducted to justify the superiority of our protocol.

In this paper, we consider monitoring multiple events in a sensing field using a large-scale wireless sensor network (WSN). The goal is to develop communication-efficient algorithms that are scalable to the network size. Exploiting the sparse nature of the events, we formulate the event monitoring task as an `1 regularized nonnegative least squares problem where the optimization variable is a sparse vector representing the locations and magnitudes of events. Traditionally the problem can be reformulated by letting each sensor hold a local copy of the event vector and imposing consensus constraints on the local copies, and solved by decentralized algorithms such as the alternating direction method of multipliers (ADMM). This technique requires each sensor to exchange their estimates of the entire sparse vector and hence leads to high communication cost. Motivated by the observation that an event usually has limited influence range, we develop two communication-efficient decentralized algorithms, one is the partial consensus algorithm and the other is the Jacobi approach. In the partial consensus algorithm that is based on the ADMM, each sensor is responsible for recovering those events relevant to itself, and hence only consent with neighboring nodes on a part of the sparse vector. This strategy greatly reduces the amount of information exchanged among sensors. The Jacobi approach addresses the case that each sensor cares about the event occurring at its own position. Jacobi-like iterates are shown to be much faster than other algorithms, and incur minimal communication cost per iteration. Simulation results validate the effectiveness of the proposed algorithms and demonstrate the importance of proper modelling in designing communication-efficient decentralized algorithms.

This paper considers the problem of covariance matrix estimation from the viewpoint of statistical signal processing for high-dimensional or wideband random processes. Due to limited sensing resources, it is often desired to accurately estimate the covariance matrix from a small number of sample observations. To make up for the lack of observations, this paper leverages the structural characteristics of the random processes by considering the interplay of three widely-available signal structures: stationarity, sparsity and the underlying probability distribution of the observed random signal. New problem formulations are developed that incorporate both compressive sampling and sparse covariance estimation strategies. Tradeoff study is provided to illustrate the design choices when estimating the covariance matrices using a handful of sample observations.

In this paper, we investigate the possibility of improving the blind detection performance of direct sequence spread spectrum (DSSS) signals for cognitive radios (CRs). We consider a scenario where a wide range of frequency spectrum needs to be monitored by a single CR, and the presence of spread spectrum signals need to be identified reliably in a cost effective manner. We employ compressive sensing to achieve realistic sensing time without imposing excessive sampling-rate requirements on the analog-to-digital converter of the CR. We assume that the number, the center frequencies, and the spreading codes of the DSSS signals are unknown, but only the spread signal's bandwidth is known. We propose a three-step algorithm for the CR. First, the power spectral density (PSD) of the wideband spectrum is estimated using compressed samples, and then detection is performed by thresholding to detect spectrum occupancy based on the estimated PSD. In the third step, knowledge of the spread signal's bandwidth, and the estimated PSD are used to perform fusion by looking at spectrum occupancy at adjacent frequency bins using a sliding window. We present some simulation results illustrating the performance gain in detection achieved by introducing the fusion step. This is a useful result since it allows us to detect with improved performance the presence of multiple DSSS signals distributed over a very wideband spectrum, without requiring the knowledge of each signal's spreading code.

Compressive sampling techniques can effectively reduce the acquisition costs of high-dimensional signals by utilizing the fact that typical signals of interest are often sparse in a certain domain. For compressive samplers, the number of samples Mrneeded to reconstruct a sparse signal is determined by the actual sparsity order Snzof the signal, which can be much smaller than the signal dimension N. However, Snzis often unknown or dynamically varying in practice, and the practical sampling rate has to be chosen conservatively according to an upper bound Smaxof the actual sparsity order in lieu of Snz, which can be unnecessarily high. To circumvent such wastage of the sampling resources, this paper introduces the concept of sparsity order estimation, which aims to accurately acquire Snzprior to sparse signal recovery, by using a very small number of samples Meless than Mr. A statistical learning methodology is used to quantify the gap between Mrand Mein a closed form via data fitting, which offers useful design guideline for compressive samplers. It is shown that Me≥ 1.2Snzlog(N/Snz+ 2) + 3 for a broad range of sampling matrices. Capitalizing on this gap, this paper also develops a two-step compressive spectrum sensing algorithm for wideband cognitive radios as an illustrative application. The first step quickly estimates the actual sparsity order of the wide spectrum of interest using a small number of samples, and the second step adjusts the total number of collected samples according to the estimated signal sparsity order. By doing so, the overall sampling cost can be minimized adaptively, without degrading the sensing performance.

In cognitive radio (CR) networks, multi-CR cooperation is required during spectrum sensing in order to cope with wireless fading effects and the hidden terminal problem. User cooperation offers not only channel diversity gain against fading, but also complexity gain in terms of reduced sampling costs per CR. The latter is particularly useful when the monitored spectrum has very wide bandwidth and yet individual CRs only have limited hardware capability. To jointly collect both diversity gain and complexity gain, this paper develops a novel cooperative spectrum sensing technique based on matrix rank minimization. Subject to sampling-rate limitations, CRs individually collect digital measurements from a segment of the wide spectrum via coordinated selective filtering, with optional compressive sampling to further reduce the sampling rates. The solutions representing the measurements of all users are modeled to possess a low-rank property, and the rank order is the same as the size of the nonzero support of the monitored wide spectrum. Accordingly, a nuclear norm minimization problem is formulated to jointly identify the nonzero support and hence the overall wideband spectrum occupancy. Both tradeoff evaluation and simulation results corroborate that the proposed cooperative sensing technique outperforms traditional averaging-based cooperative schemes given the same sampling costs, because the low-rank property enables efficient utilization and tradeoff of the user diversity in the absence of any channel knowledge.

This paper considers the problem of finding sparse solutions from multiple measurement vectors (MMVs) with joint sparsity. The solutions share the same sparsity structure, and the locations of the common nonzero support contain important information of signal features. When the measurement vectors are collected from spatially distributed users, the issue of decentralized support detection arises. This paper develops a decentralized row-based Lasso (DR-Lasso) algorithm for the distributed MMV problem. A penalty term on row-based total energy is introduced to enforce joint sparsity for the MMVs, and consensus constraints are formulated such that users can consent on the total energy, and hence the common nonzero support, in a decentralized manner. As an illustrative example, the problem of cooperative spectrum occupancy detection is solved in the context of wideband cognitive radio networks.

In cognitive radio (CR) networks, multi-CR cooperation typically takes place during spectrum sensing, to cope with wire less fading effects and the hidden terminal problem. The user cooperation gain not only offers channel diversity against fading, but also allows for reduced sampling costs per CR, which is particularly relevant when the monitored spectrum has very wide bandwidth. To attain a desired tradeoff between channel diversity and sampling costs, this paper develops a new cooperative spectrum sensing technique based on matrix rank minimization. In the absence of channel knowledge, CRs individually collect digital measurements from a region of the wide spectrum via selective filtering, with optional compressive sampling to further reduce sampling rates. The solutions representing all the measurements are modeled to possess a low rank property, with the rank order being the same as the size of the nonzero support of the monitored wide spectrum. Accordingly, a nuclear norm minimization problem is formulated to jointly identify the nonzero support and hence the overall spectrum occupancy. Simulations show that the pro posed technique outperforms traditional averaging-based co operative schemes, given the same sampling costs.

Dynamic spectrum access has emerged as a promising paradigm for improving the Dynamic spectrum utilization efficiency of wireless networks. To enable this new paradigm, fast and accurate spectrum sensing has to be performed over very wide bandwidth in noisy channel environments under energy constraints. Cyclic feature based sensing approach works well under noise uncertainty, but requires very high sampling rates in the wideband regime, and hence incurs high energy consumption and hardware costs. This paper aims to alleviate the sampling requirements of cyclic detectors by utilizing the compressive sampling principle and exploiting the sparsity structure in the two-dimensional cyclic spectrum domain. A technical challenge lies in the fact that the compressive samples collected in the time domain does not have a direct linear relationship with the two dimensional sparse cyclic spectrum of interest, which is a major departure from existing sparse signal recovery techniques for linear sampling systems. This paper solves this challenge by reformulating the vectorized cyclic spectrum into a linear form of the autocorrelation of the compressed samples. Further, based on the recovered cyclic spectrum, new cyclic-based detectors are developed to estimate the spectrum occupancy when multiple sources are present. Simulation shows that the proposed spectrum sensing algorithms can substantially reduce sampling rate with little performance loss, and is robust to the unpredictable noise uncertainty in wireless networks.

This paper investigates the issue of dynamic resource allocation (DRA) in the context of multi-user cognitive radio networks. We present a general framework adopting generalized signal expansion functions for representation of physical-layer radio resources as well as for synthesis of transmitter and receiver waveforms, which allow us to join DRA with waveform adaptation, two procedures that are currently carried out separately. Based on the signal expansion framework, we develop noncooperative games for distributed DRA, which seek to improve the spectrum utilization on a per-user basis under both transmit power and cognitive spectral mask constraints. The proposed DRA games can handle many radio platforms such as frequency, time or code division multiplexing (FDM, TDM, CDM), and even agile platforms with combinations of different types of expansion functions. To avoid the complications of having too many active expansion functions after optimization, we also propose to combine DRA with sparsity constraints. Generally, the sparsity-constrained DRA approach improves convergence of distributed games at little performance loss, since the effective resources required by a cognitive radio are in fact sparse. Finally, to acquire the channel and interference parameters needed for DRA, we develop compressed sensing techniques that capitalize on the sparse properties of the wideband signals to reduce the number of samples used for sensing and hence the sensing time.

For cognitive radio (CR) networks with user hierarchy, the sensing strategy with "listen-before-talk" (LBT) policy plays a key role in spectrum utilization and primary user (PU) protection. However, existing sensing strategies do not handle satisfactorily the randomness of both user locations and channel conditions in the network environment, resulting in inefficient spectrum utilization. To cope with such randomness, this paper develops three dynamic sensing strategies that can adaptively schedule the sensing slots/cycles according to the online link conditions without assuming knowledge of the PU traffic model. The proposed strategies can improve the efficiency of spectrum utilization while being robust with respect to the uncertainty in the PU traffic pattern. To maximize spectrum utilization, the proposed strategies are formulated through closed-form expressions. Efficient methods are introduced to compute the optimal values of the parameters used in the strategies, such as sensing time and sensing threshold. Simulations verify that the proposed sensing strategies offer an evident improvement on the spectrum utilization of the CR network.

For ultra-wideband (UWB) impulse radios, noncoherent energy detectors are motivated for their simple circuitry and effective capture of multipath energy. A major performance-degrading factor in energy detection is the noise floor, which is aggravated for low-duty-cycle UWB signals with a large time-bandwidth product. In this paper, weighted energy detection (WED) techniques are developed for effective noise suppression. The received signal is processed by a set of parallel integrators, each corresponding to a different integration time-window within a symbol period. The outputs of these integrators are weighted and linearly combined to generate decision statistics, while the weights are determined by the signal power collected from the corresponding integrators to improve the effective signal to noise ratio. The WED principle is applied to all phases of receiver processing, including signal detection, timing synchronization and data demodulation. For each phase, the optimal linear detector parameters, including decision thresholds and weighting coefficients, are derived analytically. Simulations show that the proposed noncoherent WED receiver enhances the bit-error-rate performance compared to conventional energy detectors.

For cognitive radios (CRs), compressive sampling (CS) techniques have been utilized for spectrum sensing in order to alleviate the high signal acquisition costs in the wideband regime. Given the desired sensing performance, the fundamental limit on the sampling rates is determined by the actual sparsity order Snzof the signal spectrum, which can be considerably lower than the Nyquist sampling rate. However, Snzis time-varying and hence unknown a priori for a dynamic CR network, and is typically available in the form of its statistical upper bound Smax. When the practical sampling rate is chosen according to Smaxin lieu of Snz, this rate is unnecessarily high for accurate spectrum sensing and hence wasteful of the sensing resources. To circumvent this problem, this paper develops a two-step compressed spectrum sensing (TS-CSS) scheme for efficient wideband sensing. The first step quickly estimates the actual sparsity order of the wide spectrum of interest using a small number of samples, and the second step adjusts the total number of samples collected according to the estimated signal sparsity order. By doing so, the overall sampling rate is minimized adaptively. The cornerstone of this work is the fact that the number of measurements required for estimating the signal sparsity order is (much) smaller than that for reconstructing the sparse signal itself. The gap between these two numbers is delineated in this paper in closed form, which offers valuable design guideline. Validated by simulations, the proposed TS-CSS scheme achieves the desired sensing performance at considerably reduced sensing costs, using a lowered average sampling rate compared with traditional one-step compressive sampling schemes.

This paper addresses the problem of decentralized event detection in large-scale wireless sensor networks (WSNs). Compared with centralized or hierarchical solutions, decentralized algorithms are superior in terms of scalability and robustness. However, traditional decentralized optimization tools, such as consensus optimization, entail intensive information exchange of high-dimensional decision vectors and multipliers. This paper exploits the phenomenon of limited influence, namely, the influence of one event only affects its neighboring area. For this scenario, we let each sensor make decisions for its local area rather than for the entire network, and individual decisions seek to collaboratively reach the global optimum through iterative local communications at low network costs. An optimal solution based on the alternating direction method of multipliers (ADMM) is developed. To further reduce the network communication load, we also propose a heuristic decentralized linear programming (DLP) algorithm, which is shown to be efficient via simulations.

This paper develops a distributed compressed spectrum sensing approach for cooperative wideband multi-hop cognitive radio (CR) networks where both primary users and CR users are active during the sensing stage. Due to the multi-hop nature, each CR needs to sense its individual spectral map that consists of both common spectral components from primary users and individualized spectral innovations arising from emissions of other CRs or interference in its local one-hop region. These CR-dependent spectral innovation components complicate the task of user cooperation for primary user detection. To cope with these difficulties, we adopt the compressed sensing approach at local CRs to attain high-resolution signal recovery at lower-than-Nyquist sampling rates. Each CR alternatively estimates the spectral occupancy of primary and CR users, and exchanges proper information with neighboring CRs to reach global fusion and consensus on the estimated primary user spectrum. The spectral orthogonality between primary users and CR users is exploited to improve the spectral estimation accuracy. Using only one-hop local communications, the proposed distributed algorithm converges fast to the globally optimal solution at low communication and computation load scalable to the network size.

This letter studies cooperative spectrum sensing (CSS) in which secondary users efficiently cooperate to achieve superior detection accuracy with minimum cooperation overhead. We consider each cooperative user only spends "1 bit" on reporting its own sensing decision to data fusion center as the total cooperation overhead. However, this "1 bit" CSS could not gain better sensing outcomes in current data fusion rules (DFRs). To ameliorate this issue, we derive theorems to reveal the optimum threshold of general DFR. Then we propose three novel DFRs and related three algorithms to efficiently obtain the optimum decision threshold for different objectives. By simulations, the proposed DFRs indicate evident improvement on CSS performance.

In this paper, we solve the problem of detecting the entries of a sparse finite-alphabet signal from a limited amount of data, for instance obtained by compressive sampling. While existing methods either rely on the sparsity property, the finite-alphabet property, or none of those properties to solve the under-determined system of linear equations, we capitalize on both the sparsity and the finite-alphabet features of the signal. The problem is first formulated in a Bayesian framework to incorporate the prior knowledge of sparsity, which is then shown to be solvable using sphere decoding (SD) or semi-definite relaxation (SDR) for efficient Boolean programming. A few toy simulations show how our method can outperform existing works.

This letter derives the Cramer-Rao bound (CRB) for source location estimation using joint time-/direction-of-arrival (TOA/DOA) measurements collected in a wireless sensor network setup. Each sensor is capable of measuring both TOA and DOA from a signal source, which enables it to estimate the source's location individually. Data fusion is then employed to reach a global position estimate. For both optimal measurement fusion and linear state fusion, we derive the CRBs to assess the attainable positioning accuracy, which shed light on the impact of network topology and sensor selection on localization accuracy.

In emerging cognitive radio (CR) networks with spectrum sharing, the first cognitive task preceding any dynamic spectrum access is the sensing and identification of spectral holes in wireless environments. This paper develops a distributed compressed spectrum sensing approach for (ultra-)wideband CR networks. Compressed sensing is performed at local CRs to scan the very wide spectrum at practical signal-acquisition complexity. Meanwhile, spectral estimates from multiple local CR detectors are fused to collect spatial diversity gain, which improves the sensing quality especially under fading channels. New distributed consensus algorithms are developed for collaborative sensing and fusion. Using only one-hop local communications, these distributed algorithms converge fast to the globally optimal solutions even for multi-hop CR networks, at low communication and computation load scalable to the network size.

This paper discusses the issue of dynamic resource allocation (DRA) in the context of cognitive radio (CR) networks. We present a general framework adopting generalized transmitter and receiver signal-expansion functions, which allow us to join DRA with waveform adaptation, two procedures that are currently carried out separately. Moreover, the proposed DRA can handle many types of expansion functions or even combinations of different types of functions. An iterative game approach is adopted to perform multi-player DRA, and the best-response strategies of players are derived and characterized using convex optimization. To reduce the implementation costs of having too many active expansion functions after optimization, we also propose to combine DRA with sparsity constraints for dynamic function selection. Generally, it incurs little rate-performance loss since the effective resources required by a CR are in fact sparse.

In this paper we develop a fine synchronization algorithm for multiband OFDM transmission in the presence of frequency selective channels. This algorithm is based on maximum a posteriori (MAP) joint timing and channel estimation that incorporates channel statistical information, leading to considerable performance enhancement relative to existing maximum likelihood (ML) approaches. We carry out a thorough performance analysis of the fine timing algorithm, and link the diversity concept widely used in data communications to the timing performance. We show that the probability of the timing offset equal to or larger than Δ taps has a diversity order of NBmin(Δ,L) in Rayleigh fading channels, where NBis the number of subbands and L is the number of channel taps. This result reveals that the timing estimate is very much concentrated around the true timing as the signal to noise ratio (SNR) increases. Our simulations confirm the theoretical analysis, and also demonstrate the robustness of the proposed timing algorithm against model mismatches in a realistic UWB indoor channel.

Multiple symbol differential detection (MSDD) offers high-performance symbol recovery and bypasses training or channel estimation, which are highly desired features in low- power ultra-wideband (UWB) communications. However, UWB impulse radios entail distinct signaling structures and stringent performance-complexity requirements, giving rise to the need for a new MSDD scheme capable of coping with dense multipath UWB channels and detecting a large block of symbols at practical complexity. To this end, this paper develops a novel MSDD-based UWB receiver that attains the desired performance advantages by jointly detecting blocks of received symbols based on the autocorrelation principle. To enable practical implementations at desired performance versus complexity tradeoffs, new optimization formulations are introduced to derive fast implementation algorithms inspired by powerful signal processing tools including sphere decoding and Viterbi algorithm, in both soft- and hard-decision versions. Extensive simulations testify the realistic performance of the proposed detectors in the presence of multiple access interference, timing synchronization errors and low-resolution digital-to-analog conversion.

Rapid and reliable timing synchronization is a demanding task in the receiver design for ultra-wideband (UWB) communication systems, especially when concurrent multiple access is called for. This paper derives a maximum likelihood based timing synchronization scheme for UWB multiple access. Relying on a set of user-specific training sequences, the novel concept of weighted average template is introduced, leading to a multiple access timing synchronizer that works in the absence of any channel knowledge and is resilient to both multiple access interference and noise. Simulation runs testify the effectiveness of the proposed scheme in terms of mean square timing estimation errors and bit-error-rate detection performance when operating in multiple access and dense multipath scenarios.

Cognitive radio (CR) networks have emerged as attractive candidates to enhance radio spectral utilization efficiency on both licensed bands and license-free bands. This paper investigates the achievable sum capacity of spectrum-sharing CR networks, taking into account of channel multipath profiles, transmit power constraints, as well as the outage probability requirements from primary users holding spectrum licenses. Two transmission formats, narrowband versus ultra-wideband, are compared for adoption in CR networks. Capacity analysis indicates that ultra-wideband CRs are in general more suitable for networks operating over licensed bands, when strict outage requirements are imposed for protecting primary users. On license-free bands, on the other hand, narrowband CRs employing orthogonal channelization offer higher network capacity, while its ultra-wideband counterparts become competitive only when the multipath effect is moderate. In the presence of primary users, the interference temperature constraint limits the narrowband network more than its ultra-wideband counterpart.

In a homogeneous wireless multihop network, the transmission range of nodes is an essential design parameter that critically affects the global design of the network. This paper investigates the relationship between the transmission range and two fundamental characteristics of wireless multihop networks: minimum node degree and k-connectivity. Conventional analysis assumes boundless network deployment area, which suffers from undesired border effects in practical applications based on bounded areas. To circumvent the border effect, this paper provides new analysis to accurately assess the network characteristics, including the upper bound and lower bound of both the minimum node degree and the k-connectivity. The analytical expressions hold for any arbitrary two-dimensional deployment area, when the nodes are densely deployed in a large and regular area. Simulation results corroborate with the derived analytical expressions.

In the emerging paradigm of open spectrum access, cognitive radios dynamically sense the radio-spectrum environment and must rapidly tune their transmitter parameters to efficiently utilize the available spectrum. The unprecedented radio agility envisioned, calls for fast and accurate spectrum sensing over a wide bandwidth, which challenges traditional spectral estimation methods typically operating at or above Nyquist rates. Capitalizing on the sparseness of the signal spectrum in open-access networks, this paper develops compressed sensing techniques tailored for the coarse sensing task of spectrum hole identification. Sub-Nyquist rate samples are utilized to detect and classify frequency bands via a wavelet-based edge detector. Because spectrum location estimation takes priority over fine-scale signal reconstruction, the proposed novel sensing algorithms are robust to noise and can afford reduced sampling rates.

The design of ultra-wideband (UWB) impulse radio receivers has to cope with complex signal propagation environments with dense multipath fading, which complicate channel estimation and multipath energy capture. As the date transmission rate increases, additional interference between adjacent symbols arises which may further degrade the detector performance. To address these problems, this paper presents a multi-symbols differential detector that is capable of mitigating inter-symbol interference (ISI) without requiring training or costly channel estimation. A reduced-complexity implementation based on the Viterbi algorithm is proposed, which offers adjustable parameters to balance between performance and complexity. Simulations under typical indoor operating environments demonstrate that the proposed UWB detection scheme is remarkably robust against the effects of both channel noise and ISI.

For cognitive radio networks, a popular approach to dynamic spectrum allocation (DSA) is game theoretic, which improves spectrum efficiency in a distributed manner. In a doubly selective fading channel, a conventional game-based DSA can be cumbersome to implement due to channel variation, which requires to re-train the channel estimator and re-calculate DSA decisions for every transmission burst within the channel coherence time. To enable DSA intelligence at affordable costs, this paper proposes novel adaptive DSA algorithms based on channel tracking. The locally predicted channels are employed to update DSA decisions, thus reducing signaling and computational overheads. Two algorithms are derived: an extended Kalman filter (EKF) assisted game (EKFG) and an EKF-updated game (EKFUG). Simulation shows that EFKUG is competitive for applications with low rate control channels by saving most communication burden and maintaining small data-rate loss, whereas EKFG is suitable for high data-rate applications that can afford moderate rate control channels.

For cognitive wireless networks, this paper analyzes the achievable sum capacity of spectrum-sharing cognitive radio (CR) networks, taking into account of a number of factors including channel multipath profile, transmit power constraints, as well as the outage probability at a primary legacy user. Two transmission formats, narrowband versus ultra-wideband, are compared for adoption in CRs. Capacity analysis indicates that ultra-wideband CRs are in general more suitable for networks operating over licensed bands, when strict outage requirements are imposed for protecting primary users. On license-free bands, on the other hand, narrowband CRs employing orthogonal channelization offer higher network capacity, while its ultra-wideband counterparts become competitive only when the multipath effect is moderate. In the presence of primary users, the interference temperature constraint deteriorates a narrowband cognitive network more than its ultra-wideband counterpart.

Ubiquitous wireless networking calls for efficient dynamic spectrum allocation (DSA) among heterogeneous users with diverse transmission types and bandwidth demands. To meet user-specific quality-of-service (QoS) requirements, the power and spectrum allocated to each user should lie inside a bounded region in order to be meaningful for the targeted application. Most existing DSA methods aim at enhancing the total system utility. As such, spectrum wastage may arise when the system-wise optimal allocation falls outside the desired region for QoS provisioning. The goal of this paper is to develop QoS-aware distributed DSA schemes using the game-theoretic approach. We derive DSA solutions that respect QoS and avoid naively boosting or sacrificing some users' utilities to maximize the network spectrum utilization. Specifically, we propose two game-theoretic DSA techniques: one resorts to proper scaling of the transmission power according to each user's useful utility range, and the other embeds the QoS factor into the utility function used for dynamic gaming. In addition, we introduce two new metrics to evaluate DSA schemes from a practical QoS perspective, namely "system useful utility" and "fraction of QoS satisfied users." Simulations confirm that the proposed DSA techniques outperform existing QoS-blind game models in terms of spectrum sharing efficiency in heterogeneous networks.

Ultra-wideband (UWB) wireless channels are typically characterized by dense multipath scattering, which on the one hand provides very large multipath diversity, but on the other hand complicates UWB impulse radio receiver design as far as channel estimation and multipath energy capture are concerned. In this paper, we present a multi-symbols differential detector (MSDD) in which blocks of M received symbols are jointly processed based on the autocorrelation detection principle, bypassing training or costly channel estimation. As the block size M increases, the detection performance improves, but maximum- likelihood MSDD quickly becomes computationally impractical when M is over 10. To circumvent high complexity, we derive a new implementation of the MSDD decision rule based on the sphere decoding (SD) algorithm. The proposed SD approach to MSDD offers attractive detection performance based on a simple receiver structure and at affordable computational complexity polynomial in M.

This paper considers the problem of decentralized data fusion (DDF) for large wireless sensor networks with stringent bandwidth requirements. To reduce the power and bandwidth costs of wireless transmissions, each sensor node is confined to quantize its sensing data and send 1-bit information only. Under this setting, we derive the maximum likelihood (ML) data fusion rule for decentralized parameter estimation, and analyze its Cramer-Rao lower bound (CRLB) of the fusion performance in the sense of mean square distortion. Depending on the underlying noise characteristics, our 1-bit DDF scheme can achieve estimation performance competitive to or even surprisingly better than that of centralized fusion over unquantized data. There is considerable saving in communication costs, which in turn reduces network energy consumption. Furthermore, we investigate network optimization, for which a worst-case robust design methodology is adopted to formulate a well-behaved min/max optimization problem. From the information processing viewpoint, the resulting optimized network offers robust fusion performance at minimal costs of communication resources.

Abstract — For pulsed ultra-wideband (UWB) radios, a major challenge in receiver design is to collect sufficient energy from ultra-short pulses exposing to strong multipath scattering. We develop a UWB receiver structure along with low-complexity timing synchronization and data demodulation schemes base on noisy templates (NT). The NT receiver design enables sufficient energy capture with full multipath diversity, and achieves asymptotically optimal detection performance with robustness to mis-timing. To alleviate the noise effect, a decision directed (DD) scheme is presented to lower the noise variance of the template. The detection error performance of the NT receiver is analyzed and compared with that of RAKE receivers, under realistic channel and timing estimation errors. Insights on the design tradeoffs of NT versus RAKE reception are provided, using unifying metrics that capture the relative importance of various performance-critical factors of individual receivers in the UWB regime. Both analysis and simulations confirm that the NT receiver outperforms the RAKE with a limited number of fingers under practical operating conditions. Index Terms — pulsed UWB, synchronization, detection and receiver structure, noisy template, RAKE, asymptotic optimality I.

The need for effective capture of multipath energy presents a key challenge to receiver design for pulsed ultra-wideband (UWB) systems operating in non-line-of-sight propagation environments. Conventional RAKE receivers can capture only a fraction of the received signal energy under practical implementation constraints, and have to deal with stringent synchronization and channel estimation requirements. Transmit-reference and autocorrelation receivers can effectively collect energy from all the received multipath components without explicit channel estimation, but the detection performance is limited by noise enhancement effects and the data rate drops by 50% because of pilot symbol overhead. In this paper, we develop decision-directed autocorrelation (DDA) receivers for effective multipath energy capture at low complexity. Operating in an adaptive decision-directed mode, the proposed DDA methods can considerably lower the noise level in the self-derived template waveform, thus improving overall detection performance. There is little loss in energy efficiency since no reference pilots are required during adaptation. Analytical performance analysis along with corroborating simulations is performed to evaluate the error performance of the proposed receivers in indoor lognormal fading channels

Realizing the large user capacity planned for ultra-wideband (UWB) systems motivates multiuser detection (MUD). However, it is impractical to implement conventional chip-rate MUD methods, because UWB signaling gives rise to high detection complexity and difficulty in capturing energy scattered by dense multipath. In this paper, we develop a reception model for UWB multiple access based on frame-rate sampled signals in lieu of chip-rate samples. This model enables low-complexity MUD, of which we examine a reduced-rank Wiener filter for blind symbol detection. We show that frame-rate UWB samples have a small number of distinct eigenvalues in the data covariance matrix, resulting in warp convergence of reduced-rank filtering. The proposed MUD method exhibits good performance at low complexity, even in the presence of strong frequency-selective multipath fading.

Two popular UWB receivers, the RAKE correlator and noisy-template (NT) auto-correlator, are evaluated and compared under realistic channel and timing estimation errors. A unified performance analysis framework is developed under the notion of receiver operating efficiency (ROE), which measures the effective received energy of a practical receiver against that of an ideal receiver. ROE expressions show that a RAKE receiver is limited by its energy capture capability, which is reflected not only in the number of RAKE fingers employed, but also in the timing offset estimator errors occurring at these fingers. An NT receiver is fairly robust to timing errors but is subject to noise enhancement effects, which can be alleviated by using transmissions with higher duty cycle and more training symbols. Through ROE, these critical system parameters are evaluated in terms of their contributions to overall detection accuracy. Such results are useful in identifying the preferred operating regimes of each receiver. Conversely, given practical operating conditions and system constraints, they help to select the more effective receiver structure and guide the design of key system parameters.

In this paper, we propose a new two-stage (TS) structure for computationally efficient maximum-likelihood decoding (MLD) of linear block codes. With this structure, near optimal MLD performance can be achieved at low complexity through TS processing. The first stage of processing estimates a minimum sufficient set (MSS) of candidate codewords that contains the optimal codeword, while the second stage performs optimal or suboptimal decoding search within the estimated MSS of small size. Based on the new structure, we propose a decoding algorithm that systematically trades off between the decoding complexity and the bounded block error rate performance. A low-complexity complementary decoding algorithm is developed to estimate the MSS, followed by an ordered algebraic decoding (OAD) algorithm to achieve flexible system design. Since the size of the MSS changes with the signal-to-noise ratio, the overall decoding complexity adaptively scales with the quality of the communication link. Theoretical analysis is provided to evaluate the potential complexity reduction enabled by the proposed decoding structure.

Multiple-input multiple-output (MIMO) techniques are applied to ultrawideband (UWB) systems to achieve high-rate communications over indoor wireless channels. The receiver employs a zero-forcing (ZF) scheme to separate N parallel transmitted data streams for each resolvable multipath component. A RAKE is then applied to combine the ZF paths carrying information of the same symbol to form the decision variable. Analytical error rate expression of an (N,M,L) system (N transmit antennas, M receive antennas, and L paths combined) over a pragmatic indoor log-normal fading channel is derived, which captures the diversity order by a single degree-of-freedom parameter.

Realizing the great potential of impulse radio communications depends critically on the success of timing acquisition. To this end, optimum data-aided (DA) timing offset estimators are derived in this paper based on the maximum likelihood (ML) criterion. Specifically, generalized likelihood ratio tests (GLRTs) are employed to detect an ultrawideband (UWB) waveform propagating through dense multipath and to estimate the associated timing and channel parameters in closed form. Capitalizing on the pulse repetition pattern, the GLRT boils down to an amplitude estimation problem, based on which closed-form timing acquisition estimates can be obtained without invoking any line search. The proposed algorithms only employ digital samples collected at a low symbol rate, thus reducing considerably the implementation complexity and acquisition time. Analytical acquisition performance bounds and corroborating simulations are also provided.

The overall system efficiency of impulse radio communications relies critically on judicious allocation of transmission resources, a portion of which should be used to ensure successful timing acquisition. In data-aided mode, optimum timing offset estimation depends not only on the mechanism used for energy capture and the acquisition algorithm employed to recover timing information, but also on the training sequence (TS) pattern from which the timing information is to be extracted. Furthermore, the transmission resources used for timing have to be balanced with that for conveying information messages in order to strike desirable tradeoffs between timing accuracy and information rate. In Part I of this paper, data-aided timing offset estimation is derived based on the maximum likelihood (ML) criterion, where only symbol-rate samples are needed for low-complexity receiver processing. To minimize the mean-square timing errors of these ML synchronizers while at the same time maximizing the average system capacity, TS design and transmit power allocation are investigated in this paper. The optimum training pattern and the number, placement, and power distribution between training and information-bearing symbols are formulated as a resource allocation optimization problem whose solution optimizes system-level performance with the minimum amount of resources consumed.

Timing acquisition for ultrawideband (UWB) communication systems operating in dense multipath environments faces major challenges due to the stringent requirements to resolve and capture ultrashort transmitted pulses. This paper develops low-complexity maximum-likelihood (LC-ML) acquisition methods that offer explicit design options to trade off acquisition accuracy and complexity. The proposed schemes are based on a tapped delay line (TDL) model whose tap spacing is set in accordance with a low frame-level rate. By avoiding subpulse rate sampling, the LC-ML methods achieve low complexity and fast acquisition speed and at the same time retain good estimation accuracy due to the underlying ML principle. Both the data-aided (DA) and nondata-aided (NDA) versions are derived. It is also demonstrated by simulations that the proposed synchronizers are markedly robust with respect to the effects of both multipath channel and multiple access interference.

The overall system performance of data-aided ultra-wideband (UWB) communications relies critically on the accuracy of synchronization and channel estimation during the training phase. The total transmission resources should be properly allocated between training and information symbols in order to strike a desired balance between performance and information rate. To this end, this paper derives optimum transmission schemes that judiciously allocate the limited transmit power and pulse numbers for signal reception tasks performing optimal timing acquisition, channel estimation, as well as symbol detection. The resulting selection of transmitter parameters not only enables both the timing and channel estimators to attain the minimum mean-square estimation errors, but also maximizes the average system capacity.

Realizing the benefits of ultra-wideband (UWB) communications hinges critically on judicious pulse shape design to enable UWB spectral mask compatibility, and co-existence with and adaptation to other wireless devices. To this end, we propose a convex optimization based waveform design method for UWB radios. By casting the pulse design problem as a (convex) semidefinite program (SDP) over the pulse autocorrelation, globally optimal waveform designs can be efficiently obtained. While the focus of this paper is on the design of waveforms that optimally utilize the bandwidth and power allowed by the spectral mask, the flexibility of the SDP framework also allows the optimization of several other system objectives.

The overall system efficiency of impulse radio communications relies critically on judicious allocation of transmission resources, a portion of which should be used to ensure successful timing acquisition. Data-aided timing offset estimation has been derived by the authors based on the maximum likelihood (ML) criterion, where only symbol-rate samples are needed for low-complexity receiver processing. To minimize the mean-square timing errors of these ML synchronizers while at the same time maximizing the average system capacity, the training sequence design and the transmit power allocation are investigated in this paper. The optimum training pattern, as well as the number, placement, and power allocation of training vs. information-bearing symbols, are formulated as a resource allocation optimization problem, whose solution offers the optimum system-level performance with the minimum amount of resources consumed.

This work presents a maximum-likelihood (ML) decoding algorithm for linear block codes. In this algorithm, the optimal performance is achieved at low computational complexity through a two-stage processing. At the first stage, a minimum sufficient set S that includes the optimal solution is estimated. With the minimum sufficient set, the decoding complexity can be greatly reduced without performance degradation. At the second stage, ordered processing is performed within the estimated minimum sufficient set S to obtain the optimal solution. During the ordered processing, S is adoptively updated to minimize the computational complexity, and an effective stopping criterion is used to decide whether the optimal solution is found. Ordered processing not only helps to find the optimal solution quickly, but also enables simplified sub-optimal solutions with bounded block error rates. The proposed algorithm is also extended to decode block turbo codes. Finally, simulation results are given to show that this algorithm achieves optimal performance with a low average computational complexity.

A cross-layer, near-optimal soft-decision equalization approach for frequency selective multiinput multioutput (MIMO) communication systems is proposed in this paper. With the help of iterative processing, two detection and estimation schemes based on second-order statistics are harmoniously put together to yield a two-part receiver structure: local multiuser detection (MUD) using soft-decision probabilistic data association (PDA) detection, and dynamic noise-interference tracking using Kalman filtering. The proposed Kalman-PDA approach performs local MUD within a subblock of the received data instead of over the whole data set, to reduce the computational load. At the same time, all the interference affecting the local subblock, including both multiple access and intersymbol interference, is properly modeled as the state vector of a linear system, and dynamically tracked by Kalman filtering. The performance of Kalman-PDA is further enhanced by applying a simple automatic repeat request (ARQ) protocol. The ARQ-aided MIMO detection algorithm can adjust to any preset retransmission rate to ensure a desired performance and data-rate trade-off.

This letter presents a robust decision-feedback equalization design that mitigates the error-propagation problem for multiuser direct-sequence code-division multiple-access systems under multipath fading. Explicit constraints for signal energy preservation are imposed on the filter weight vector to monitor and maintain the quality of the hard decisions in the nonlinear feedback loop. Such a measure protects the desired signal power against the detrimental effect of erroneous past decisions, thus providing the leverage to curb error propagation.

In this paper, a soft-decision based detection technique is developed for BLAST MIMO systems with less receive (Rx) antennae than transmit (Tx) antennae. The existing nulling and cancellation algorithm (NC) is unable to work for such systems theoretically. Avoiding any zero-forcing mechanism, the proposed approach works for rank-deficient channels via soft-decision based interference cancellation. Moreover, it outperforms NC in BLAST systems with more Rx than Tx antennae. The computational cost of the proposed approach is only on the 3rd order of the number of symbols to be recovered and linear in the signal constellation size, which is lower than the complexity of NC. This paper establishes the soft-decision approach as an attractive candidate to achieve detection performance better than that of NC. Simulation comparisons of the soft-decision approach with the NC algorithm are presented.

The unique advantages of ultrawideband (UWB) technology are somewhat encumbered by its stringent timing tolerances. To help to appreciate how critical timing offset estimation is for UWB, the bit-error-rate (BER) sensitivity to timing offsets is investigated in this paper for correlation-based UWB receivers. Focusing on UWB transmissions with time hopping, the BER expressions are derived for various operating conditions and system setups, including AWGN and frequency flat channels, dense multipath fading channels, and various receiver types including sliding correlators and RAKE combiners. It is demonstrated through analyses that time-hopping based multiple access systems exhibit little tolerance to acquisition errors, while the energy capture capability of a RAKE combiner can be severely compromised by mistiming.

Low-complexity rapid timing acquisition constitutes a major challenge in realizing the high potential that ultra-wideband (UWB) wireless technology promises for indoor communications. We derive and test two such timing acquisition algorithms which capitalize on the cyclostationarity that is naturally present in UWB transmissions. Our novel schemes are blind, they do not require multiple antennas or oversampling, and rely on frame-rate sampling which reduces complexity and acquisition delay considerably.

This article presents a beamspace version of the class of polynomial parameterized weighted subspace fitting (WSF) methods, including the well-known methods iterative quadratic maximum likelihood and method of direction estimation, for direction-of-arrival estimation with a uniform linear array. The difficulty with beamspace operation is that the Vandermonde structure of the array manifold may not preserve during a beamspace transformation. We tackle this obstacle using discrete Fourier transform beams and derive a parameterized polynomial representation of the signal subspace in the beamspace dimension. The resulting beamspace iterative quadratic WSF algorithm is not only computationally attractive, but also enjoys improved resolution threshold over element-space estimation.

This paper investigates the output signal-to-interference-and-noise ratio (SINR) behaviors of various linear and nonlinear adaptive RLS filters, including MMSE and DFE filters with known data input, MOE detector, and constrained blind MMSE and DFE filters. New analytic results on the optimal and steady state SINR performance of the constrained adaptive filters are derived. These analyses facilitate comparisons of various filtering methods under different operation conditions. We demonstrate that computer simulation results match well with our theoretical analyses.

A new iterative MUSIC algorithm is proposed based on the successive cancellation idea. It can be cast as an alternative of the expectation maximization (EM) algorithm, in which the maximization step is carried out by one-dimensional MUSIC estimation. This algorithm is computationally attractive because of the use of high-efficiency MUSIC at each iteration. Moreover, it is able to resolve coherent signals through the signal decomposition. A theoretical performance evaluation procedure is described to analyze the estimation variance of the iterative MUSIC algorithm. Both analytic and experimental results indicate that the iterative MUSIC method is effective for coherent signals.

Stochastic Bayesian detection has recently emerged as a competitive receiver design paradigm for wireless communications applications. It uses Monte Carlo simulations to perform Bayesian inference of a probabilistically modeled communication system so as to obtain the maximum a posterior (MAP) symbol detection and/or channel estimation results. The Monte Carlo concept is attractive in that it is intuitive, optimal, and works for generic Bayesian network structures. However, conventional Monte Carlo methods suffer from poor convergence especially when there is less likely evidence in the collected samples, in which case the simulations are wasted in sample spaces that contribute little to the inference estimates. We present an adaptive sampling method in which the sample allocation process is optimized for efficient MAP detection. It is then demonstrated that this optimized adaptive sampling method can be applied to wireless communication systems for high-efficiency symbol detection and channel estimation. The effectiveness of the derived blind Bayesian multiuser detection is verified by computer simulations.

Decision feedback equalization (DFE) is a very effective receiver component that cancels intersymbol interference without inducing noise amplification. However, it suffers from error propagation in the start-up phase and when the environment undergoes a sudden change. For DS-CDMA wireless systems, we develop robust measures to mitigate the catastrophic error propagation effect using appropriate constraints on the receiver weight vectors. These constraints are constructed from the spreading codes of the desired user to preserve its output energy. By doing so, a desirable level of signal output power is maintained to keep the error probability low, thus preventing decision errors from occurring or propagating. Moreover, the introduction of signal-preserving constraints offers capacity of multiple access interference suppression. Efficient adaptive implementations are developed, and simulations performed to verify the effectiveness of the proposed constrained DFE detectors. It is demonstrated that the constructed constraints can eliminate the training overhead, offering complete blind implementations for multiuser decision-directed DFE filters.

In signal processing, there are problems where the processed signal output energy is maximized while the noise component is minimized. This gives rise to a max/min problem, which is equivalent to a generalized eigenvalue problem. Exemplary applications of the max/min formulation have been seen in Capon's blind beamforming method and the blind minimum output energy (MOE) detection in CDMA wireless communications. The solution to such a problem involves eigen-decomposition of a transformed data covariance matrix inverse, which is computationally expensive to implement. This paper offers an adaptive RLS solution to the optimal minimum power filtering problem without involving eigen-decompositions. It is based on a new recursive least square updating procedure that works for multiple linear constraints, and uses a one-dimensional subspace tracking method to update the filter weights. The performance is comparable with that of using the direct eigen-decomposition and matrix inversion.

Hexagonal arrays are widely used in practice but have received less attention in the optimum array processing literature. In this paper, we show how unitary ESPRIT can be applied to hexagonal arrays for direction-of-arrival (DOA) estimation. The resulting estimates exhibit good threshold behavior, and are close to the Cramer-Rao bound above threshold. We also show how to use spatial smoothing in hexagonal arrays for DOA estimation in the presence of coherent signals.