Search

Gait-based age estimation is a key technique for many applications. It is well known that age estimation uncertainty is highly dependent on age (i.e., small for children and large for adults), and it is important to know the uncertainty for the above-mentioned applications. Therefore, we propose a method for uncertainty-aware gait-based age estimation by introducing a label distribution learning framework. Specifically, we design a network that takes an appearance-based gait feature as input and outputs discrete label distributions in the integer age domain. We then train the network to minimize a loss function, which is defined as the dissimilarity between the estimated age distribution and the ground-truth age distribution, in addition to the conventional mean absolute error for the estimated age. Additionally, we demonstrate that uncertainty-aware gait-based age estimation is beneficial for two applications: person search by age query and people counting by age group. Experiments on the world’s largest gait database, OULP-Age, demonstrated that the proposed method can successfully represent age estimation uncertainty, and outperforms or is comparable with state-of-the-art methods in terms of age estimation accuracy. Moreover, we demonstrated the effectiveness of the uncertainty-aware framework in applications to person search and people counting through experiments on the database.

The $p$ -GaN gate technology is commonly implemented to achieve normally OFF gallium nitride (GaN) devices. Nonetheless, the threshold voltage instability under OFF-state stress remains a concern. In this article, the characterization technique of threshold voltage shift is proposed and the $V_{\text {th}}$ shift’s impact on device performance is investigated. Specifically, a fast $V_{\mathrm {th}}$ the measurement circuit is introduced and validated to successfully characterize both the magnitude and the time-constant of the $V_{\mathrm {th}}$ shift. According to the experimental results, the $V_{\mathrm {th}}$ increases by more than 50% within several $\mu \text{s}$ after the high drain voltage is applied. In contrast, the $V_{\mathrm {th}}$ slowly (tens of seconds) gets back to the static value after the stress is removed. Following the characterization, the threshold voltage instability’s impact on the device’s static/switching performance is studied experimentally. A knee point shift is observed at low gate voltage, and the static ON-resistance value remains unchanged. Regarding the switching performance, the turn-ON loss increases by >20% after the high drain voltage stress. The change in turn-ON loss can be reduced when the gate resistance is decreased from 20 to $0~\Omega $ . In terms of the turn-OFF loss, the impact of the $V_{\mathrm {th}}$ shift is negligible. It is concluded that time-dependent threshold voltage shift needs to be considered in $p$ -GaN device’s modeling, and high gate drive voltage together with low gate resistance is recommended to mitigate the $V_{\mathrm {th}}$ instability’s effect on $p$ -GaN high-electron-mobility transistor’s performance.

Avalanche ruggedness is one of the key factors for safe and reliable power converters deployed in various automotive subsystems. In this article, the avalanche capability of power MOSFETs is tested under different repetitive avalanche conditions to differentiate the die degradation and package degradation. While examining the degradation mechanisms, the corresponding electrical parameter shifts are analyzed to better understand the structural changes and root causes and identify potential precursors for condition monitoring. For this purpose, a high-resolution and cost-effective nanosecond current pulse generator (CPG) is designed. Due to the high-resolution pulsewidth modulation function of a digital signal processor (DSP), short current pulses can be generated in the order of 100 ps where the magnitude and duration of the current pulsewidth can be adjusted precisely. During the experiments, power MOSFETs are stressed under two different pulsewidths. It is observed that the ON-state resistance gradually increases in both cases, yet the roots of the degradations are different. The die degradation and package degradation are responsible for the changes in short pulse and long pulse, respectively. In the short pulse case, some devices show saturation in drain leakage current and ON-state resistance, while some others show a significant threshold voltage drop, which leads to a noticeable shift of transfer and output characteristics. The electrical parameter shifts indicate a possible gate degradation after the device aging. At the end of the tests, failure analyses are conducted on both devices under test (DUTs) under different stress conditions, revealing different failure mechanisms.

In this article, 3-D vector-based model predictive current control (MPCC) for the open-end winding permanent magnet synchronous generator (OEW-PMSG) system driven by a dual inverter with a common dc bus is proposed to suppress the zero-sequence current. First, the 3-D voltage vector distribution of this system is analyzed and their zero-sequence voltages are investigated. Then, in order to reduce the computation burden of prediction control caused by vector enumeration, an improved current prediction control is developed, in which candidate voltage vectors for the dual inverter are determined according to the position of the zero-sequence reference voltage vector and the sector location of the $\alpha \beta $ -axis reference voltage vector. Moreover, to further improve the steady-state control performance of the whole generation system, a separate vector-selection method for the dual inverter is proposed. In this method, the first inverter applies one voltage vector in a control period and the voltage vector of the second inverter is synthesized by two vectors and selected in 3-D space. Finally, experiments are carried out to validate proposed methods, and the comparison experiments indicate that the proposed method has better control performance compared to conventional methods.

Herein, the influence of platinum (Pt) catalyst on the interface between SiO 2 nanofiller and addition cure silicone rubber (ALSR) matrix is investigated with different concentrations of Pt catalyst. It is found through various tests that Si–O–Si covalent bonds and hydrogen bonds form at the interface with a moderate amount of Pt catalyst. Based on these observations, a model for the interfacial structure of ALSR/SiO 2 nanocomposites is proposed. In particular, it is demonstrated, without changing the mass fraction and surface modifier of the filler, a more compact interface is obtained by increasing the amount of added Pt catalyst by a few ppm, resulting in a 25% increase of breakdown strength, 10% increase in thermal conductivity, 50% increase in tensile strength, and a 31% increase in the elongation at break.

This article proposes a new complete condition monitoring method which can independently monitor both the threshold voltage drift and the packaging degradation accurately by monitoring only the reverse body diode voltage drop at different gate bias levels. The SiC MOSFETs are aged through an accelerated aging method and the corresponding changes in electrical parameters are periodically measured to assess their correlation with the state of the device's health. It has been revealed that the on -state resistance reveals the combination of gate oxide and package-related degradation while the threshold voltage mainly depicts gate oxide-related issues. Unlike these two parameters, the body diode voltage drop is found to independently indicate the state of device health both for package and gate oxide. This is caused by a unique secondary conduction mode of SiC MOSFETs in third-quadrant operation which is clearly disclosed with transition boundaries by a proposed body diode transfer characterization curve. In order to have a detailed condition information of the device with a simple circuit, monitoring the reverse body diode voltage drop at 0 and $-$ 5 V gate bias is proposed. The proposed condition monitoring method is implemented on a gate drive circuit and experimental results are given for two artificially degraded devices. The experimental results confirm that the proposed method can independently monitor both the gate oxide and packaging degradations accurately by monitoring a single precursor parameter. The proposed method can be integrated into the gate driver or converter itself to monitor SiC devices.

Online junction temperature $({{T_j}})$ measurement enables robust power converter operations by providing overtemperature protection and condition monitoring of the power devices. For SiC MOSFETs, the real-time ${T_j}$ information is especially critical as limited field data are available regarding the reliability. In this article, utilizing the turn -on delay time as temperature sensitive electrical parameter, an online ${T_j}$ measurement is realized through an intelligent gate drive. Specifically, the turn -on delay time is translated into the pulsewidth of a digital signal through the conditioning/logic circuits. During ${T_j}$ measurements, the adjustable gate resistance circuit is activated to improve the measurement sensitivity beyond 600 ps/°C. Using the high-resolution capture module (300-ps resolution) in the system microcontroller, this pulsewidth is measured and then converted to junction temperature with a resolution of ${T_j}$ measurement method. The switching test results show that the circuit is able to precisely measure ${T_{d,{\rm{on}}}}$ and offers a good linearity/sensitivity for ${T_j}$ estimation. In the continuous operation, the junction temperature of a decapsulated device using an infrared camera and ${T_j}$ obtained from the circuit match well with ${T_{d,{\rm{on}}}}$ is considered for SiC MOSFETs, and an aging compensation scheme is discussed to maintain the measurement accuracy throughout the device's lifetime. It is shown that the proposed circuit provides an accurate real-time ${T_j}$ measurement for SiC MOSFETs, which can be deployed to improve the power converters’ reliability.

Virtualization has attained mainstream status in enterprise IT industry. Despite its widespread adoption, it is known that virtualization also introduces non-trivial overhead when tasks are executed on a virtual machine (VM). In particular, a combined effect from device virtualization overhead and CPU scheduling latency can cause performance degradation when computation intensive tasks and I/O intensive tasks are co-located on a VM. Such an interference also causes extra energy consumption. In this paper, we present Hylics , a novel solution that enables efficient data traverse paths for both I/O and computation intensive workloads. This is achieved with the provision of in-memory file system and network service at the hypervisor level. Several important design issues are pinpointed and addressed during our prototype implementation, including efficient intermediate data sharing, network service offloading, and QoS-aware memory usage management. Based on our real-world deployment on KVM, we show that Hylics can significantly improve computation and I/O performance for hybrid workloads. Moreover, this design also alleviates the existing virtualization overhead and naturally optimizes the overall energy efficiency.

With the deep integration of information, communication and operation technologies, industrial wireless control networks (IWCNs) are playing key roles to drive industrial revolution. This article proposes a 5G-based new radio (NR) for IWCNs to realize ultra-reliable low-latency communication over unlicensed bands, namely WIA-NR. We first define a hierarchical star network topology, which employs a hybrid centralized and distributed system management architecture. Then, a three-layer protocol stack is defined by integrating some fundamental functions of 5G and proposing new functions to enhance reliability and timeliness. The key technologies include dynamic multi-channel listen before talk, adaptive channel hopping, over-the-air time synchronization, industrial data priority scheduling, aggregation and disaggregation. Finally, system-level simulations are performed to evaluate the performance of WIA-NR. The results show that WIA-NR can achieve less than 1 ms latency under 99.999 percent reliability constraint, which can well satisfy critical industrial control. Meanwhile, WIA-NR can harmoniously coexist with WiFi over unlicensed bands.

To further improve the fault-tolerant operation capacity of the open-winding permanent magnet synchronize motor system after second-time fault occurring, a novel fault-tolerant topology and a fault-tolerant control strategy for single phase open-circuit is proposed in this article. The fault-tolerant topology proposed in this article can operate stably in case of second-time fault that occurs after one-phase open-circuit fault by the way of sharing a bridge arm. Thus, the fault-tolerant ability of the driving system is further improved. In the terms of control strategy, the amplitude and the phase of the remaining two-phase current are redesigned to keep the system operating steadily according to the conventional expression of motor torque. Moreover, considering that the electromagnetic torque of the system will contain the third harmonic component when the three-phase current is unbalanced, the torque fluctuation will affect the stability of the system. A torque optimization strategy for suppressing the third harmonic component of the torque by injecting corresponding harmonics into the current is proposed. Finally, the experimental results validate the effectiveness of the proposed topology and strategy.

We present a novel self-supervised training framework with 3D displacement (3DD) module for accurately estimating per-pixel depth maps from single laparoscopic images. Recently, several self-supervised learning based monocular depth estimation models have achieved good results on the KITTI dataset, under the hypothesis that the camera is dynamic and the objects are stationary, however this hypothesis is often reversed in the surgical setting (laparoscope is stationary, the surgical instruments and tissues are dynamic). Therefore, a 3DD module is proposed to establish the relation between frames instead of ego-motion estimation. In the 3DD module, a convolutional neural network (CNN) analyses source and target frames to predict the 3D displacement of a 3D point cloud from a target frame to a source frame in the coordinates of the camera. Since it is difficult to constrain the depth displacement from two 2D images, a novel depth consistency module is proposed to maintain depth consistency between displacement-updated depth and model-estimated depth to constrain 3D displacement effectively. Our proposed method achieves remarkable performance for monocular depth estimation on the Hamlyn surgical dataset and acquired ground truth depth maps, outperforming monodepth, monodepth2 and packnet models.

Model-based gait recognition is considered to be promising due to the robustness against some variations, such as clothing and baggage carried. Although model-based gait recognition has not been fully explored due to the difficulty of human body model fitting and the lack of a large-scale gait database, recent progress in deep learning-based approaches to human body model fitting and human pose estimation is mitigating the difficulty. In this paper, we, therefore, address the remaining issue by presenting a large-scale human pose-based gait database, OUMVLP-Pose, which is based on a publicly available multi-view large-scale gait database, OUMVLP. OUMVLP-Pose has many unique advantages compared with other public databases. First, OUMVLP-Pose is the first gait database that provides two datasets of human pose sequences extracted by two standard deep learning-based pose estimation algorithms, OpenPose and AlphaPose. Second, it contains multi-view large-scale data, i.e., over 10,000 subjects and 14 views for each subject. In addition, we also provide benchmarks in which different kinds of gait recognition methods, including model-based methods and appearance-based methods, have been evaluated comprehensively. The model-based gait recognition methods have shown promising performances. We believe this database, OUMVLP-Pose, will greatly promote model-based gait recognition in the next few years.

In this article, an in situ SiC mosfet degradation monitoring method using readily available converter sensors is presented. Device's drain-source resistance in saturation and ohmic regions are employed as aging precursors to, respectively, indicate die and package-related degradations. The presented findings in this article are essential for low-cost and practical early warning systems in power electronics applications. To verify the effectiveness of the proposed method, a number of power-cycling tests are conducted to analyze aging process and fatigue mechanism. Device static parameters are measured throughout accelerated aging tests and the physical causes behind them are discussed in detail. It is shown that the drain-source resistance within saturation region at low gate bias can be used to indicate die-related degradation. Similarly, at full positive gate bias, device on -state drain-source resistance can be used to detect package-related degradation. In order to find out optimal variables such as applied gate bias, pulse duration, etc., a comprehensive single pulse test is conducted and the results are reported. Finally, the proposed degradation monitoring method is experimentally validated in the context of a three-phase inverter considering different sensor configurations and load types.

Multipath TCP (MPTCP) has recently been suggested as a promising transport protocol to boost the utilization of underlaying datacenter networks, yet it also increases the host CPU power consumption. It remains unclear whether datacenters can indeed benefit from using MPTCP from the perspective of energy efficiency. Through realworld measurement of MPTCP, we show that the energy efficiency of MPTCP is largely related to the flow completion time and the existence of link-sharing subflows. In particular, we find that the link-sharing subflows in MPTCP will significantly elevate the CPUs’ power consumption on hosts. To make the matter worse, it will also reduce the transmission efficiency for both throughput-sensitive long flows and latency-sensitive short flows. To address such a problem, we present MPTCP-D, an energy-efficient enhancement of MPTCP in datacenter networks. MPTCP-D incorporates a novel congestion control algorithm that improves energy efficiency by minimizing the flow completion time. It also has a build-in subflow elimination mechanism that precludes link-sharing subflows from increasing the host CPU power consumption. We implement MPTCP-D in the Linux kernel, analyze the parameter selection in the algorithm and study its performance through packet-level simulation and on Amazon EC2. Our results show that, without degrading the performance of the long flow throughput and the short flow completion time, MPTCP-D reduces the long flow energy consumption by up to 72% compared to DCTCP for data transfers, and reduces the short flow power consumption by up to 46% compared to MPTCP with link-sharing subflows.

Clothing and carrying status variations are the two key factors that affect the performance of gait recognition because people usually wear various clothes and carry all kinds of objects, while walking in their daily life. These covariates substantially affect the intensities within conventional gait representations such as gait energy images. Hence, to properly compare a pair of input gait features, an appropriate metric for joint intensity is needed in addition to the conventional spatial metric. We therefore propose a unified joint intensity transformer network for gait recognition that is robust against various clothing and carrying statuses. Specifically, the joint intensity transformer network is a unified deep learning-based architecture containing three parts: a joint intensity metric estimation net, a joint intensity transformer, and a discrimination network. First, the joint intensity metric estimation net uses a well-designed encoder-decoder network to estimate a sample-dependent joint intensity metric for a pair of input gait energy images. Subsequently, a joint intensity transformer module outputs the spatial dissimilarity of two gait energy images using the metric learned by the joint intensity metric estimation net. Third, the discrimination network is a generic convolution neural network for gait recognition. In addition, the joint intensity transformer network is designed with different loss functions depending on the gait recognition task (i.e., a contrastive loss function for the verification task and a triplet loss function for the identification task). The experiments on the world’s largest datasets containing various clothing and carrying statuses demonstrate the state-of-the-art performance of the proposed method.

The dynamic on -resistance in gallium nitride (GaN) devices is problematic as it can impair the converter's efficiency due to the increased conduction loss. In this paper, the hard-switching transient's effect on the dynamic on -resistance is, for the first time, evaluated experimentally on a commercial high-voltage GaN high electron mobility transistor. A new ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ measurement circuit with fast sensing speed is designed, and an accurate measurement of ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ can be realized experimentally within 49.6 ns after the device's current reaches to the load current. A double-pulse-test setup is designed to comprehensively evaluate the switching transient's effect on ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ under different operating conditions. From the experimental results, it is found that the turn- on and turn- off gate resistance have a significant impact on the dynamic on -resistance whereas the cross-talk effect on ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ is negligible. Specifically, at 400 V/25 A, more than 27% (28.2%) increase in ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ is observed when the external turn -on (turn- off ) gate resistance increases from 0 to 20 Ω. Detailed discussion and quantitative analysis are provided to explain the experimental results. In terms of the turn -on process, it is concluded that the ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ variation is mainly caused by the different numbers of generated hot electrons. For the turn -off transient, it is confirmed the variation of drain current at different dv/dt slew rate leads to the ${R_{{\rm{dyn}} - {\rm{ds}},{\rm{on}}}}$ difference.

A 1 × 2 integrated four-mode power splitter based on multimode interference is proposed and experimentally demonstrated for the first time. A three-dimensional simulation is used to optimize the core/cladding refractive index difference and the input waveguide position for the TE00, TE01, TE10, and TE11 input waveguide modes. Minimum insertion loss (IL) and mode-dependent loss (MDL) of 0.1 and 0.2 dB are predicted, respectively. IL and MDL less than 1 dB can be maintained over a bandwidth of approximately 50 and 25 nm, respectively. The device was fabricated on SiO2 and experimentally characterized. The variation of the power splitting ratio for each mode is less than 0.4 dB over the full C-band, and the excess losses of the splitter for all four modes exceed the IL of a straight waveguide by less than 1.85 dB.

Multiple network interface card (NIC) configuration is the key to achieving service isolation and link aggregation in enterprise networks. Despite the increasing deployments of multiple virtual NICs ( multi-vNIC ) in cloud environments, the performance and hidden cost of such multi-vNIC configurations are yet to be understood. Our measurements in both Xen and KVM environments reveal that the network workloads on one virtual NIC can negatively affect the transmission performance on other virtual NICs. Such a bottleneck however does not exist in conventional physical machines or to a much lower degree. We therefore explore the possible approaches to detect and resolve such a problem in the virtualization architecture design. In particular, a mutual information-based model is developed to monitor the stability of virtual NIC pairs. Based on this model, we further design a bottleneck avoidance mode to enhance the network performance. Different from the previous research that merely optimizes virtualized network I/O stack, our solution focuses on the entire application I/O workflow in the cloud environments. Extensive evaluations show that our enhancements can mitigate the performance degradation of multi-vNIC configuration, making it a practically workable solution for cloud environments.

Networking and machine virtualization play critical roles in the success of modern cloud computing. The energy consumption of physical machines has been carefully examined in the past, including the impact from network traffic. When it comes to virtual machines (VMs) in cloud data centers, the interplay between energy consumption and network traffic, however, becomes much more complicated. Through real-world measurement on both Xen- and KVM-based platforms, we show that these state-of-the-art virtualization designs noticeably increase the demand of CPU resources when handling network transactions, generating excessive interrupt requests with ceaseless context switching, which in turn, increases energy consumption. Even when a physical machine is in an idle state, its VM's network transactions will incur nontrivial energy consumption. More interestingly, the energy consumption significantly varies with traffic allocation strategies and virtual CPU affinity conditions , which was not seen in conventional physical machines. Looking closely into the virtualization architectures, we then pinpoint the root causes and examine that our measurement results can be extended for various network configurations. Moreover, we also provide initial solutions toward optimizing energy consumption in virtualized environments.

This paper studies a green paradigm for the underlay coexistence of primary users (PUs) and secondary users (SUs) in energy harvesting cognitive radio networks (EH-CRNs), wherein battery-free SUs capture both the spectrum and the energy of PUs to enhance spectrum efficiency and green energy utilization. To lower the transmit powers of SUs, we employ multi-hop transmission with time division multiple access, by which SUs first harvest energy from the RF signals of PUs, and then, transmit data in the allocated time concurrently with PUs, all in the licensed spectrum. In this way, the available transmit energy of each SU mainly depends on the harvested energy before the turn to transmit, namely energy causality. Meanwhile, the transmit powers of SUs must be strictly controlled to protect PUs from harmful interference. Thus, subject to the energy causality constraint and the interference power constraint, we study the end-to-end throughput maximization problem for optimal time and power allocation. To solve this nonconvex problem, we first equivalently transform it into a convex optimization problem and then propose the joint optimal time and power allocation (JOTPA) algorithm that iteratively solves a series of feasibility problems until convergence. Extensive simulations evaluate the performance of EH-CRNs with JOTPA in three typical deployment scenarios and validate the superiority of JOTPA by making comparisons with two other resource allocation algorithms.

The scientific landscape of manufacturing systems continues to change hand in hand with the evolution of the field itself. It is anticipated that new scientific insights will impact further growth in this area. However, the number of personnel with major specialization in complex systems, those that get hired by academia and industry, remains small; yet, the professional requirements are high and include a thorough education in one or more of the following

This paper develops space-delay adaptive processing (SDAP) for multiple-input, multiple-output (MIMO) arrays with application to classifying human motion in wide-area indoor environments. The proposed adaptive method resolves motion returns using joint spatial-temporal processing on the received signals. Results with real data demonstrate that MIMO SDAP can mitigate ghost motion returns due to indoor multipath scattering and bandwidth limitation for better separation of micro-Doppler signatures from multiple targets. Human activities such as bed turnovers and arm or leg movements are then classified based on micro-Doppler analysis for patient pressure ulcer prevention.

This letter studies the outage performance of multihop cognitive relay networks with energy harvesting in underlay paradigms, wherein the secondary users are powered by a dedicated power beacon (PB) and their transmit powers are subject to the harvested energy from PB and the interference constraint from the primary user. We derive the exact outage probability for Rayleigh block fading and prove that the outage probability is monotonically decreasing with respect to the transmit power of PB. Furthermore, we derive the asymptotic outage probability to study the outage saturation phenomenon and propose an iterative algorithm that jointly optimizes the transmit power of PB and the harvest-to-transmit ratio to approximate the minimum outage probability. Simulation results validate the theoretical results.

In this paper, a decoupled controlled triport dc/dc converter is derived by combining two bidirectional single-phase buck–boost converters and one isolated full-bridge converter for multiple input source applications. The power density is improved, and the circuit structure is simplified because the power devices are completely shared in the primary side. Furthermore, the pulsewidth modulation plus phase-shift control strategy is introduced to provide two control freedoms and achieve the decoupled voltage regulation within a certain operating range. The duty cycle of the bidirectional buck–boost converters is adopted to balance the voltage between the two primary input terminals, while their phase angle is applied to regulate the accurate secondary voltage. Furthermore, zero-voltage-switching soft-switching operation is provided for all of the primary power switches due to the inherent phase-shift control scheme. Moreover, the two filter inductors in the bidirectional buck–boost converters and the isolated transformer in the full-bridge topology are integrated and replaced by the winding-cross-coupled inductors to reduce the component numbers and simplify the magnetic structure. Finally, a 1-kW prototype is built to verify all theoretical considerations, and it is shown that the proposed topology is particularly advantageous in the distributed power generation system with multiple energy sources.

Precision ultra-wideband ranging requires accurate estimation of time of arrival (TOA). The threshold-based TOA estimation has attracted considerable interest due to its simplicity. In this paper, we propose a technique to set the threshold as a function of delay instead of a single value as in conventional methods. Simulation results show that the proposed method is superior to conventional methods.