In this work, we present an in-depth study to measure, analyze and enhance aerial performance (here, uplink throughput) for drones flying in the low sky over two operational 5G networks in the US (AT&T and T-Mobile). Different from prior aerial 5G measurement studies, we have made three new endeavors. First, through extensive experiments in the low sky (below 120 m), we not only characterize aerial performanceobservedover operational 5G networks, but also quantitively assess performance potentialsnot observed but missedin the sky. We have several new findings that have not been reported before: higher 5G performance potentials are realized in the sky than on the ground (say, faster data speed in the sky); But surprisingly, more performance potentials are also missed in the sky (namely, 5G could have been even much faster but such potentials are not fully utilized in the sky). Second, we delve into root causes behind missed performance potentials and find that current 5G cell selection should take the blame despite the impacts of radio resource allocation in the underlying physical layer. Cell selection is designed for terrestrial scenarios and misses good 5G cells under aerial radio channel conditions. Third, we thus devise a data-driven solution called5GAir++to patch cell selection in practice.5GAir++is promising to pursue more 5G performance potentials in the low sky. We have validated its effectiveness over real-world traces with two applications of bulky file upload and video live streaming. Datasets and codes are released.

5G is much faster than 4G, offering faster data transfer and better user experience overall. Intuitively, 5G should be used as much as possible. However, in this study, we unveil a surprising finding in operational 5G networks: 5G radio access may be in a persistent ON-OFF loop which repeatedly turns 5G on and then off. We conduct extensive measurement experiments with three US operators (T-Mobile, AT&T, and Verizon) in two US cities to characterize and analyze 5G ON-OFF loop instances in the wild. Surprisingly, we find that such 5G ON-OFF loops are not rare. They are widely observed at many places, significantly hurting data performance (from several hundreds of Mbps to tens of or even zero Mbps). We further dive into their causes and uncover that inconsistent triggers to turn 5G on and off co-exist in real-world settings, repeatedly releasing 5G radio access after getting 5G back. We identify three loop types each with distinct triggering events/causes (sub-types). Inconsistent policies and mechanisms on both network and device sides, as well as ''improper'' use of certain frequency channels, are responsible for the loops observed in this study. Our datasets and artifacts have been released on Github and MI-LAB.

5G advances cellular network technologies and significantly boosts mobile data performance. However, we find that mobile devices may not always access such high performance. More precisely, we observe high performance variance in 5G, where actual data performance fluctuates intensively and the actual performance that the mobile device gets is much lower than what the device could have got at best. In this paper, we aim to characterize performance variance in 5G, quantify its impacts and understand why. We conduct extensive experiments in one US city to measure performance variations with a major carrier. Our study shows new findings different from prior studies over 4G networks. In particular, 5G performance variance changes dramatically in two aspects: (i) The performance variance between different cellsets still commonly exists in 5G non-standalone (NSA), but not in 5G Standalone (SA). (ii) 5G SA uses one single cellset at most locations but still experiences with significant performance variance. Our cause analysis shows the performance variance in 5G NSA is attributed to varying numbers of cells in carrier aggregation, diverse resource allocation, and Secondary Cell Group (SCG) failures. For 5G SA, the primary cause lies in high dynamic radio conditions of the serving cells. These findings shed new insights to boost actual performance that mobile device can get in the wild.

Ultrasonography plays an essential role in breast cancer diagnosis. Current deep learning based studies train the models on either images or videos in a centralized learning manner, lacking consideration of joint benefits between two different modality models or the privacy issue of data centralization. In this study, we propose the first decentralized learning solution for joint learning with breast ultrasound video and image, called FedBCD. To enable the model to learn from images and videos simultaneously and seamlessly in client-level local training, we propose a Joint Ultrasound Video and Image Learning (JUVIL) model to bridge the dimension gap between video and image data by incorporating temporal and spatial adapters. The parameter-efficient design of JUVIL with trainable adapters and frozen backbone further reduces the computational cost and communication burden of federated learning, finally improving the overall efficiency. Moreover, considering conventional model-wise aggregation may lead to unstable federated training due to different modalities, data capacities in different clients, and different functionalities across layers. We further propose a Fisher information matrix (FIM) guided Layer-wise Aggregation method named FILA. By measuring layer-wise sensitivity with FIM, FILA assigns higher contributions to the clients with lower sensitivity, improving personalized performance during federated training. Extensive experiments on three image clients and one video client demonstrate the benefits of joint learning architecture, especially for the ones with small-scale data. FedBCD significantly outperforms nine federated learning methods on both video-based and image-based diagnoses, demonstrating the superiority and potential for clinical practice. Code is released at https://github.com/tianpeng-deng/FedBCD.

The characteristics of 3D nodules on Computed Tomography (CT), including size, location, shape, and attenuation, are primary medical clues for distinguishing between benign and malignant nodules. To support evidence-based decision-making for lung cancer screening in clinical practice, we present a 3D Nodule Content-based Metric Learning (3D-NCML) network to retrieve subsolid-benign, subsolid-malignant, solid-benign, and solid-malignant nodules similar to the indeterminate ones. The inputs of 3D-NCML are 3D patches that exactly contain the whole nodule to ensure all visual information is included. A spatial position and size coding module, a shape encoder module, and an attenuation extraction module are designed based on medical clues for guiding the network to learn important characteristics of nodules. Experiments on the LIDC-IDRI dataset and a private dataset demonstrate that 3D-NCML outperforms other methods by quantitative and qualitative analysis, with more similar nodules retrieved and ranked ahead.

In this work, we present our measurement study to characterize and analyze operational 5G performance potentials for cellular-connected drones that fly in the low sky. We not only measure aerial performance observed over an operational 5G network (here, T-Mobile, one major 5G operator in the US), but also quantitively assess potentials missed in the sky. Different from prior measurement studies, we compare 5G performance potentials realized and missed in the low sky and on the ground. We have several new findings that have not been reported before: higher 5G performance potentials are realized in the sky than on the ground (say, faster data speed in the sky); But surprisingly, more performance potentials are also missed in the sky (namely, 5G can have been even much faster but such potentials are not fully utilized in the sky). We delve into root causes behind missed potentials and find that current 5G cell selection is designed for terrestrial scenarios and misses good candidate cells under aerial radio channel conditions. We thus devise a patch solution called 5Gair to pursue more 5G potentials in the low sky and validate its effectiveness over real-world traces (released at [1]).

With 5G deployment gaining momentum, the control-plane traffic volume of cellular networks is escalating. Such rapid traffic growth motivates the need to study the mobile core network (MCN) control-plane design and performance optimization. Doing so requires realistic, large control-plane traffic traces in order to profile and debug the mobile network performance under real workload. However, large-scale control-plane traffic traces are not made available to the public by mobile operators due to business and privacy concerns. As such, it is critically important to develop accurate, scalable, versatile, and open-to-innovation control traffic generators, which in turn critically rely on an accurate traffic model for the control plane. Developing an accurate model of control-plane traffic faces several challenges: (1) how to capture the dependence among the control events generated by each User Equipment (UE), (2) how to model the inter-arrival time and sojourn time of control events of individual UEs, and (3) how to capture the diversity of control-plane traffic across UEs. We present a novel two-level hierarchical state-machine-based control-plane traffic model. We further show how our model can be easily adjusted from LTE to NextG networks (e.g., 5G) to support modeling future control-plane traffic. We experimentally validate that the proposed model can generate large realistic control-plane traffic traces. We have open-sourced our traffic generator to the public to foster MCN research.

Existing multi-view learning methods based on the information bottleneck principle exhibit impressing generalization by capturing inter-view consistency and complementarity. They leverage cross-view joint information (consistency) and view-specific information (complementarity) while discarding redundant information. By fusing visual features, multi-view learning methods help medical image processing to produce more reliable predictions. However, multi-views of medical images often have low consistency and high complementarity due to modal differences in imaging or different projection depths, thus challenging existing methods to balance them to the maximal extent. To mitigate such an issue, we improve the information bottleneck (IB) loss function with a balanced regularization term, termed IBB loss, reassembling the constraints of multi-view consistency and complementarity. In particular, the balanced regularization term with a unique trade-off factor in IBB loss helps minimize the mutual information on consistency and complementarity to strike a balance. In addition, we devise a triplet multi-view network named TM net to learn the consistent and complementary features from multi-view medical images. By evaluating two datasets, we demonstrate the superiority of our method against several counterparts. The extensive experiments also confirm that our IBB loss significantly improves multi-view learning in medical images.

It turns out that convolutional neural networks (CNNs) have excellent medical image processing capabilities. Hence, effectively and efficiently deploying CNNs on devices with varying computing power to make computer-aided diagnosis puts on the agenda. However, it is a dilemma to balance the limited computing resources and model complexity. Previously, we proposed factorized convolution with spectral normalization (FConvSN) to mitigate the bottleneck of deploying CNNs for 2D medical images. But due to the cube structure of 3D convolutional kernels, it does not work well for 3D medical images. Directly flattening 3D kernels to 2D weights for matrix factorization may undermine the learning ability along depth-wise, resulting in the loss of depth information and the decline of model performance. To this end, we factorize a 3D convolutional kernel to 2D weight matrices with depth-wise dimensions, then assign an attentive score for each 2D weight matrix by a depth-wise dependencies-induced adaptive attention block (AA). AA with a temperature hyper-parameter helps convolution kernel to better capture depth-wise dependencies in 3D medical images, improving its learning ability along the depth direction. We term this novel factorized convolution as FConvAA used for compressing model complexity without impairing the depth-wise expressivity. We also impose spectral normalization (SN) for FConvAA to constrain spectral norm-wise weights. We conduct extensive experiments on the public lung CT dataset LUNA16 and the private retina OCT dataset to demonstrate the effectiveness and feasibility of our FConvAA.

As the basis for developing glaucoma treatment strategies, Anterior Chamber Angle (ACA) evaluation is usually dependent on experts' Judgements. However, experienced ophthalmologists needed for these Judgements are not widely available. Thus, computer-aided ACA evaluations become a pressing and efficient solution for this issue. In this paper, we propose a novel end-to-end frame-work GCNet for automated Glaucoma Classification based on ACA images or other Glaucoma-related medical images. We first collect and label an ACA image dataset with some pixel-level annotations. Next, we introduce a segmentation module and an embedding module to enhance the performance of classifying ACA images. Within GCNet, we design a Cross-Module Aggregation Net (CMANet) which is a weakly-supervised metric learning network to capture contextual information exchanging across these modules. We conduct experiments on the ACA dataset and two public datasets REFUGE and SIGF. Our experimental results demonstrate that GCNet outperforms several state-of-the-art deep models in the tasks of glaucoma medical image classifications. The source code of GCNet can be found at https://github.com/Jingqi-H/GCNet.

In this paper, we propose iCellSpeed, an on-device solution to increase data access speed by substantiating unrealized performance potentials. We find that performance potentials are missed in today's mobile networks, as the data speed a user device gets is much lower than what the device could get. The issue is rooted in the current cell selection practice, which misses good candidate cells that offer faster access speed, thus under-utilizing the available capabilities in mobile networks. We design iCellSpeed to facilitate network-controlled cell selection with proactive device-side assistance towards more desirable cells. Our evaluation over AT&T and Verizon confirms its effectiveness. iCellSpeed increases data access speed by more than 10 Mbps at 79% of test locations (> 25Mbps at 29% of locations, up to 80.6 Mbps). It doubles access speed at 62.5% of locations with the gain up to 28.4x. Datasets are available at [9].

Millimeter-wave (mmWave) networking represents a core technology to meet the demanding bandwidth requirements of emerging connected vehicles. However, the feasibility of mmWave vehicle-to-everything (V2X) connectivity has long been questioned. One major doubt lies in how the highly directional mmWave links can sustain under high mobility. In this paper, we present the first comprehensive reality check of mmWave V2X networks. We deploy an experimental testbed to mimic a typical mmWave V2X scenario, and customize a COTS mmWave radio to enable microscopic investigation of the channel and the link. We further construct a high-fidelity 3D ray-tracer to reproduce the mmWave characteristics at scale. With this toolset, we study the mmWave V2X coverage, mobility and blockage, codebook/beam management, and spatial multiplexing. Our measurement debunks some common misperceptions of mmWave V2X networks. In particular, due to the constrained roadway network structures, we find the beam management can be handled easily by the often-denounced beam scanning schemes, as long as the codebook is properly designed. Blockage can be almost eliminated through proper basestation deployment and cooperation. Highly effective spatial multiplexing can be realized even without sophisticated MIMO radios. Our work points to possible ways to realize efficient and reliable mmWave networks under high mobility, while maintaining the simplicity of standard network protocols.

Millimeter-wave (mmWave) networks are conventionally considered to bear a fundamental coverage limitation, due to the directional beams and limited field-of-view (FoV) of the phased array antennas. In this paper, we explore an array of phased arrays (APA) architecture, which aggregates co-located phased arrays with complementary FoVs to approximate WiFi-like omni-directional coverage. We found that straightforwardly activating all the arrays may even hamper network performance. To fully exploit the APA's potential, we propose X-Array, which jointly selects the arrays and beams, and applies a dynamic co-phasing mechanism to ensure different arrays' signals enhance each other. X-Array also incorporates a link recovery mechanism to identify alternative arrays/beams that can efficiently recover the link from outage. We have implemented X-Array on a commodity 802.11ad APA radio. Our experiments demonstrate that X-Array can approach omni-directional coverage and maintain high performance in spite of link dynamics.

The emerging millimeter-wave (mmWave) networking technology promises to unleash a new wave of multi-Gbps wireless applications. However, due to high directionality of the mmWave radios, maintaining stable link connection remains an open problem. Users' slight orientation change, coupled with motion and blockage, can easily disconnect the link. In this paper, we propose RoMil, a robotic mmWave relay that optimizes network coverage through wireless sensing and autonomous motion/rotation planning. The robot relay automatically constructs the geometry/reflectivity of the environment, by estimating the geometries of all signal paths. It then navigates itself along an optimal moving trajectory, and ensures continuous connectivity for the client despite environment/human dynamics. We have prototyped RoMil on a programmable robot carrying a commodity 60 GHz radio. Our field trials demonstrate that RoMil can achieve nearly full coverage in dynamic environment, even with constrained speed and mobility region.

The emerging millimeter-wave (mmWave) networking technology promises to unleash a new wave of multi-Gbps wireless applications. However, due to high directionality of the mmWave radios, maintaining stable link connection remains an open problem. Users' slight orientation change, coupled with motion and blockage, can easily disconnect the link. In this paper, we propose miDroid, a robotic mmWave relay that optimizes network coverage through wireless sensing and autonomous motion/rotation planning. The robot relay automatically constructs the geometry/reflectivity of the environment, by estimating the geometries of all signal paths. It then navigates itself along an optimal moving trajectory, and ensures continuous connectivity for the client despite environment/human dynamics. We have prototyped miDroid on a programmable robot carrying a commodity 60 GHz radio. Our field trials demonstrate that miDroid can achieve nearly full coverage in dynamic environment, even with constrained speed and mobility region.

In a Multi-User Multiple Input Multiple Output (MU-MIMO) system, an access point (AP) equipped with multiple antennas can serve multiple users simultaneously (i.e., support concurrent multi-streams) and thus achieves multi-fold through- put gain. In practice, however, the gain is significantly comprised by frame-size diversity,i.e., shorter frames need to wait for the finish of the longest frame, which leads to low channel utilization and thus throughput degradation. Frame padding (i.e., more than one short frames are grouped together to fill in the idle channel) has been proposed to solve the problem, but existing approaches are based on heuristic and cannot fully exploit the potential of padding. In this paper, we propose Knapsack Padding (KPad), a novel model-driven frame padding design to maximize MU-MIMO channel utilization. We first formally formulate the frame padding problem as amulti-stream knapsackmodel, and then design a stream decoupling mechanism to handle the unique and complicated inter-stream interference underlying the model, so as to derive the optimal padding schedule efficiently. We evaluate KPad using trace-driven emulation. Extensive evaluation results demonstrate remarkable throughput gain (up to 42%) compared with the state-of-the-art.

Global navigation satellite system (GNSS) can provide autonomous geo-spatial positioning and time synchronization services for both civil and military uses. Satellite links in GNSS are used to transmit signal for constellation management and other applications. In this work, we focus on solving the satellite link scheduling problem over dynamic satellite network with the aim of minimizing the number of participant ground-based management stations and the cost of communication between satellites in the background of GNSS networking. Firstly, we assume the navigation constellation has finite states and cope the dynamic topology with Finite State Automation method. Secondly, a Fireworks algorithm (FWA) is designed according to the characteristic of the scheduling problem. Finally, the FWA is compared with ant colony optimization (ACO). The performance analysis of different scenarios is given. The study in this paper provides technical reference for the management of future large-scale satellite network.