Many deep convolutional neural network (CNN) inference accelerators on the field-programmable gate array (FPGA) platform have been widely adopted due to their low power consumption and high performance. In this paper, we develop the following to improve performance and power efficiency. First, we use a high bandwidth memory (HBM) to expand the bandwidth of data transmission between the off-chip memory and the accelerator. Second, a fully-pipelined manner, which consists of pipelined inter-layer computation and a pipelined computation engine, is implemented to decrease idle time among layers. Third, a multi-core architecture with shared-dual buffers is designed to reduce off-chip memory access and maximize the throughput. We designed the proposed accelerator on the Xilinx Alveo U280 platform with in-depth Verilog HDL instead of high-level synthesis as the previous works and explored the VGG-16 model to verify the system during our experiment. With a similar accelerator architecture, the experimental results demonstrate that the memory bandwidth of HBM is 13.2× better than DDR4. Compared with other accelerators in terms of throughput, our accelerator is 1.9×/1.65×/11.9× better than FPGA+HBM2 based/low batch size (4) GPGPU/low batch size (4) CPU. Compared with the previous DDR+FPGA/DDR+GPGPU/DDR+CPU based accelerators in terms of power efficiency, our proposed system provides 1.4-1.7×/1.7-12.6×/6.6-37.1× improvement with the large-scale CNN model.

To simultaneously enhance data rate and physical layer security (PLS) for low-luminance smartphone screen to camera uplink communication, space division multiplexing using high-luminance cell-size reduction arrangement is numerically analyzed and experimentally verified. The uplink consists of a low-luminance smartphone screen and an indoor telephoto camera at a long distance of 3.5 meters. The high-luminance cell-size reduction arrangement avoids the influence of spatial inter-symbol interference (ISI) and ambient light to obtain a stable low-luminance screen. To reduce the screen luminance without decreasing the screen pixel value, the arrangement reduces only the high-luminance cell area while keeping the cell spacing. In this study, two technical issues related to high-luminance cell-size reduction arrangement are solved. First, a numerical analysis and experimental results show that the high-luminance cell-size reduction arrangement is more effective in reducing the spatial ISI at low luminance than the conventional low-luminance cell arrangement. Second, in view point of PLS enhancement at wide angles, symbol error rate should be low in front of the screen and high at wide angles. A numerical analysis and experimental results show that the high-luminance cell-size reduction arrangement is more suitable for enhancing PLS at wide angles than the conventional low-luminance cell arrangement.

This paper summarizes the modulation configurations of phase locked loops (PLLs) and their integration in semiconductor circuits, e.g., the input modulation for cellular phones, direct-modulation for low power wireless sensor networks, feedback-loop modulation for high-speed transmission, and two-point modulation for short-range radio transceivers. In this survey, basic configuration examples of integrated circuits for wired and wireless applications which are using the PLL modulation configurations are explained. It is important to select the method for simply and effectively determining the characteristics corresponding to the specific application. The paper also surveys technologies for future PLL design for digitizing of an entire PLL to reduce the phase noise due to a modulation by using a feedback loop with a precise digital phase comparison and a numerically controlled oscillator with high linearity.

This paper proposes a deep neural network named BayesianPUFNet that can achieve high prediction accuracy even with few challenge-response pairs (CRPs) available for training. Generally, modeling attacks are a vulnerability that could compromise the authenticity of physically unclonable functions (PUFs); thus, various machine learning methods including deep neural networks have been proposed to assess the vulnerability of PUFs. However, conventional modeling attacks have not considered the cost of CRP collection and analyzed attacks based on the assumption that sufficient CRPs were available for training; therefore, previous studies may have underestimated the vulnerability of PUFs. Herein, we show that the application of Bayesian deep neural networks that incorporate Bayesian statistics can provide accurate response prediction even in situations where sufficient CRPs are not available for learning. Numerical experiments show that the proposed model uses only half the CRP to achieve the same response prediction as that of the conventional methods. Our code is openly available on https://github.com/bayesian-puf-net/bayesian-puf-net.git.

Reconfigurable intelligent surface (RIS) has the capability of boosting system performance by manipulating the wireless propagation environment. This paper investigates a downlink RIS-aided non-orthogonal multiple access (NOMA) system, where a RIS is deployed to enhance physical-layer security (PLS) in the presence of an eavesdropper. In order to improve the main link's security, the RIS is deployed between the source and the users, in which a reflecting element separation scheme is developed to aid data transmission of both the cell-center and the cell-edge users. Additionally, the closed-form expressions of secrecy outage probability (SOP) are derived for the proposed RIS-aided NOMA scheme. To obtain more deep insights on the derived results, the asymptotic performance of the derived SOP is analyzed. Moreover, the secrecy diversity order is derived according to the asymptotic approximation in the high signal-to-noise ratio (SNR) and main-to-eavesdropper ratio (MER) regime. Furthermore, based on the derived results, the power allocation coefficient and number of elements are optimized to minimize the system SOP. Simulations demonstrate that the theoretical results match well with the simulation results and the SOP of the proposed scheme is clearly less than that of the conventional orthogonal multiple access (OMA) scheme obviously.

Multi-band transmission technologies promise to cost-effectively expand the capacity of optical networks by exploiting low-loss spectrum windows beyond the conventional band used in already-deployed fibers. While such technologies offer a high potential for capacity upgrades, available capacity is seriously restricted not only by the wavelength-continuity constraint but also by the signal-to-noise ratio (SNR) constraint. In fact, exploiting more bands can cause higher SNR imbalance over multiple bands, which is mainly due to stimulated Raman scattering. To relax these constraints, we propose wavelength-selective band switching-enabled networks (BSNs), where each wavelength channel can be freely switched to any band and in any direction at any optical node on the route. We also present two typical optical node configurations utilizing all-optical wavelength converters, which can realize the switching proposal. Moreover, numerical analyses clarify that our BSN can reduce the fiber resource requirements by more than 20% compared to a conventional multi-band network under realistic conditions. We also discuss the impact of physical-layer performance of band switching operations on available benefits to investigate the feasibility of BSNs. In addition, we report on a proof-of-concept demonstration of a BSN with a prototype node, where C+L-band wavelength-division-multiplexed 112-Gb/s dual-polarization quadrature phase-shift keying signals are successfully transmitted while the bands of individual channels are switched node-by-node for up to 4 cascaded nodes.

To understand the radiation mechanism of an electrically small spherical helix antenna, we develop a theory on the radiation characteristics of the antenna. An analytical model of the antenna presuming a current on the wire to be sinusoidally distributed is proposed and analyzed with the spherical wave expansion. The radiation efficiency, radiation resistance, and radiation patterns are obtained in closed-form expression. The radiation efficiency evidently varies with the surface area of the wire and the radiation resistance depends on the square of the length of the wire. The obtained result for the radiation pattern illustrates the tilt of the pattern caused by the modes asymmetric to the z-axis. The radiation efficiency formula indicates a good agreement between the simulation and measurement result. In addition, the radiation resistance of the theoretical and simulation results exhibits good agreement. Considering the effect of the feeding structure of the fabricated antenna, the radiation resistance of the analytical model can be treated as a reasonable result. The result of radiation pattern also shows good agreement between the simulation and measurement results excluding a small contribution from the feeding cable acting as a scatterer.

We have designed a near-infrared wavelength Si/CaF2 DFB quantum cascade laser and investigated the possibility of single-mode laser oscillation by analysis of the propagation mode, gain, scattering time of Si quantum well, and threshold current density. As the waveguide and resonator, a slab-type waveguide structure with a Si/CaF2 active layer sandwiched by SiO2 on a Si (111) substrate and a grating structure in an n-Si conducting layer were assumed. From the results of optical propagation mode analysis, by assuming a λ/4-shifted bragg waveguide structure, it was found that the single vertical and horizontal TM mode propagation is possible at the designed wavelength of 1.70µm. In addition, a design of the active layer is proposed and its current injection capability is roughly estimated to be 25.1kA/cm2, which is larger than required threshold current density of 1.4kA/cm2 calculated by combining analysis results of the scattering time, population inversion, gain of quantum cascade lasers, and coupling theory of a Bragg waveguide. The results strongly indicate the possibility of single-mode laser oscillation.

Fine-grained image classification is one of the key basic tasks of computer vision. The appearance of traditional deep convolutional neural network (DCNN) combined with attention mechanism can focus on partial and local features of fine-grained images, but it still lacks the consideration of the embedding mode of different attention modules in the network, leading to the unsatisfactory result of classification model. To solve the above problems, three different attention mechanisms are introduced into the DCNN network (like ResNet, VGGNet, etc.), including SE, CBAM and ECA modules, so that DCNN could better focus on the key local features of salient regions in the image. At the same time, we adopt three different embedding modes of attention modules, including serial, residual and parallel modes, to further improve the performance of the classification model. The experimental results show that the three attention modules combined with three different embedding modes can improve the performance of DCNN network effectively. Moreover, compared with SE and ECA, CBAM has stronger feature extraction capability. Among them, the parallelly embedded CBAM can make the local information paid attention to by DCNN richer and more accurate, and bring the optimal effect for DCNN, which is 1.98% and 1.57% higher than that of original VGG16 and Resnet34 in CUB-200-2011 dataset, respectively. The visualization analysis also indicates that the attention modules can be easily embedded into DCNN networks, especially in the parallel mode, with stronger generality and universality.

Car navigation systems provide traffic jam information. In this study, we attempt to provide more detailed traffic jam information that considers the lane in which a traffic jam is in. This makes it possible for users to avoid long waits in queued traffic going toward an unintended destination. Lane-specific traffic jam detection utilizes image processing, which incurs long processing time and high cost. To reduce these, we propose a “suddenness index (SI)” to categorize candidate areas as sudden or periodic. Sudden traffic jams are prioritized as they may lead to accidents. This technology aggregates the number of connected cars for each mesh on a map and quantifies the degree of deviation from the ordinary state. In this paper, we evaluate the proposed method using actual global positioning system (GPS) data and found that the proposed index can cover 100% of sudden lane-specific traffic jams while excluding 82.2% of traffic jam candidates. We also demonstrate the effectiveness of time savings by integrating the proposed method into a demonstration framework. In addition, we improved the proposed method's ability to automatically determine the SI threshold to select the appropriate traffic jam candidates to avoid manual parameter settings.

The data augmentation method is known as a helpful technique to generate a dataset with a large number of images from one with a small number of images for supervised training in deep learning. However, a low validity augmentation method for image recognition was reported in a recent study on artificial intelligence (AI). This study aimed to clarify the optimal data augmentation method in deep learning model generation for the recognition of white blood cells (WBCs). Study Design: We conducted three different data augmentation methods (rotation, scaling, and distortion) on original WBC images, with each AI model for WBC recognition generated by supervised training. The subjects of the clinical assessment were 51 healthy persons. Thin-layer blood smears were prepared from peripheral blood and subjected to May-Grünwald-Giemsa staining. Results: The only significantly effective technique among the AI models for WBC recognition was data augmentation with rotation. By contrast, the effectiveness of both image distortion and image scaling was poor, and improved accuracy was limited to a specific WBC subcategory. Conclusion: Although data augmentation methods are often used for achieving high accuracy in AI generation with supervised training, we consider that it is necessary to select the optimal data augmentation method for medical AI generation based on the characteristics of medical images.

In recent years, driver's visual attention has been actively studied for driving automation technology. However, the number of models is few to perceive an insight understanding of driver's attention in various moments. All attention models process multi-level image representations by a two-stream/multi-stream network, increasing the computational cost due to an increment of model parameters. However, multi-level image representation such as optical flow plays a vital role in tasks involving videos. Therefore, to reduce the computational cost of a two-stream network and use multi-level image representation, this work proposes a single stream driver's visual attention model for a critical situation. The experiment was conducted using a publicly available critical driving dataset named BDD-A. Qualitative results confirm the effectiveness of the proposed model. Moreover, quantitative results highlight that the proposed model outperforms state-of-the-art visual attention models according to CC and SIM. Extensive ablation studies verify the presence of optical flow in the model, the position of optical flow in the spatial network, the convolution layers to process optical flow, and the computational cost compared to a two-stream model.

Danmaku commenting has become popular for co-viewing on video-sharing platforms, such as Nicovideo. However, many irrelevant comments usually contaminate the quality of the information provided by videos. Such an information pollutant problem can be solved by a comment classifier trained with an abstention option, which detects comments whose video categories are unclear. To improve the performance of this classification task, this paper presents Nicovideo-specific language representations. Specifically, we used sentences from Nicopedia, a Japanese online encyclopedia of entities that possibly appear in Nicovideo contents, to pre-train a bidirectional encoder representations from Transformers (BERT) model. The resulting model named Nicopedia BERT is then fine-tuned such that it could determine whether a given comment falls into any of predefined categories. The experiments conducted on Nicovideo comment data demonstrated the effectiveness of Nicopedia BERT compared with existing BERT models pre-trained using Wikipedia or tweets. We also evaluated the performance of each model in an additional sentiment classification task, and the obtained results implied the applicability of Nicopedia BERT as a feature extractor of other social media text.

Convolutional neural networks (CNNs) have become popular in medical image segmentation. The widely used deep CNNs are customized to extract multiple representative features for two-dimensional (2D) data, generally called 2D networks. However, 2D networks are inefficient in extracting three-dimensional (3D) spatial features from volumetric images. Although most 2D segmentation networks can be extended to 3D networks, the naively extended 3D methods are resource-intensive. In this paper, we propose an efficient and accurate network for fully automatic 3D segmentation. Specifically, we designed a 3D multiple-contextual extractor to capture rich global contextual dependencies from different feature levels. Then we leveraged an ROI-estimation strategy to crop the ROI bounding box. Meanwhile, we used a 3D ROI-attention module to improve the accuracy of in-region segmentation in the decoder path. Moreover, we used a hybrid Dice loss function to address the issues of class imbalance and blurry contour in medical images. By incorporating the above strategies, we realized a practical end-to-end 3D medical image segmentation with high efficiency and accuracy. To validate the 3D segmentation performance of our proposed method, we conducted extensive experiments on two datasets and demonstrated favorable results over the state-of-the-art methods.

There are demands for high-speed and stable ground-to-train optical communication as a network environment for trains. The existing ground-to-train optical communication system developed by the authors uses a camera and a QPD (Quadrant photo diode) to capture beacon light. The problem with the existing system is that it is impossible to identify the ground station. In the system proposed in this paper, a beacon light modulated with the ID of the ground station is transmitted, and the ground station is identified by demodulating the image from the dual-port camera on the opposite side. In this paper, we developed an actual system and conducted experiments using a car on the road. The results showed that only one packet was lost with the ping command every 1 ms near handover. Although the communication device itself has a bandwidth of 100 Mbps, the throughput before and after the handover was about 94 Mbps, and only dropped to about 89.4 Mbps during the handover.

A drone swarm is a robotic architecture having multiple drones cooperate to accomplish a mission. Nowadays, heterogeneous drone swarms, in which a small number of gateway drones (GWs) act as protocol translators to enable the mixing of multiple swarms that use independent wireless protocols, have attracted much attention from many researchers. Our previous work proposed Path Optimizer — a method to minimize the number of end-to-end path-hops in a remote video monitoring system using heterogeneous drone swarms by autonomously relocating GWs to create a shortcut in the network for each communication request. However, Path Optimizer has limitations in improving communication quality when more video sessions than the number of GWs are requested simultaneously. Path Coordinator, which we propose in this paper, achieves a uniform reduction in end-to-end hops and maximizes the allowable hop satisfaction rate regardless of the number of sessions by introducing the cooperative and synchronous relocation of all GWs. Path Coordinator consists of two phases: first, physical optimization is performed by geographically relocating all GWs (relocation phase), and then logical optimization is achieved by modifying the relaying GWs of each video flow (rerouting phase). Computer simulations reveal that Path Coordinator adapts to various environments and performs as well as we expected. Furthermore, its performance is comparable to the upper limits possible with brute-force search.

In recent years, physical layer security (PLS), which is based on information theory and whose strength does not depend on the eavesdropper's computing capability, has attracted much attention. We have proposed a chaos modulation method as one PLS method that offers channel coding gain. One alternative is based on polar codes. They are robust error-correcting codes, have a nested structure in the encoder, and the application of this mechanism to PLS encryption (PLS-polar) has been actively studied. However, most conventional studies assume the application of conventional linear modulation such as BPSK, do not use encryption modulation, and the channel coding gain in the modulation is not achieved. In this paper, we propose a PLS-polar method that can realize high-quality transmission and encryption of a modulated signal by applying chaos modulation to a polar-coding system. Numerical results show that the proposed method improves the performance compared to the conventional PLS-polar method by 0.7dB at a block error rate of 10-5. In addition, we show that the proposed method is superior to conventional chaos modulation concatenated with low-density parity-check codes, indicating that the polar code is more suitable for chaos modulation. Finally, it is demonstrated that the proposed method is secure in terms of information theoretical and computational security.

In this study, we discuss a structure that realizes a wideband polarization splitter comprising fiber 1 with a single core and fiber 2 with circular pits, which touch the top and bottom of a single core. The refractive index profile of the W type was adopted in the core of fiber 1 to realize the wideband. We compared the maximum bandwidth of BW-15 (bandwidth at an extinction ratio of -15dB) for the W type obtained in this study with those (our previous results) of BW-15 for the step and graded types with cores and pits at the same location; this comparison clarified that the maximum bandwidth of BW-15 for the W type is 5.22 and 4.96 times wider than those of step and graded types, respectively. Furthermore, the device length at the maximum bandwidth improved, becoming slightly shorter. The main results of the FPS in this study are all obtained by numerical analysis based on our proposed MM-DM (a method that combines the multipole method and the difference method for the inhomogeneous region). Our MM-DM is a quite reliable method for high accuracy analysis of the FPS composed of inhomogeneous circular regions.

An approach to dedicated computers is discussed in this study as a possibility for portable, low-cost, and low-power consumption high-performance computing technologies. Particularly, dataflow architecture dedicated computer of the finite integration technique (FIT) for 2D magnetostatic field simulation is considered for use in industrial applications. The dataflow architecture circuit of the BiCG-Stab matrix solver of the FIT matrix calculation is designed by the very high-speed integrated circuit hardware description language (VHDL). The operation of the dedicated computer's designed circuit is considered by VHDL logic circuit simulation.

In real world applications of multiclass classification models, misclassification in an important class (e.g., stop sign) can be significantly more harmful than in other classes (e.g., no parking). Thus, it is crucial to improve the recall of an important class while maintaining overall accuracy. For this problem, we found that improving the separation of important classes relative to other classes in the feature space is effective. Existing methods that give a class-sensitive penalty for cross-entropy loss do not improve the separation. Moreover, the methods designed to improve separations between all classes are unsuitable for our purpose because they do not consider the important classes. To achieve the separation, we propose a loss function that explicitly gives loss for the feature space, called class-sensitive additive angular margin (CAMRI) loss. CAMRI loss is expected to reduce the variance of an important class due to the addition of a penalty to the angle between the important class features and the corresponding weight vectors in the feature space. In addition, concentrating the penalty on only the important class hardly sacrifices separating the other classes. Experiments on CIFAR-10, GTSRB, and AwA2 showed that CAMRI loss could improve the recall of a specific class without sacrificing accuracy. In particular, compared with GTSRB's second-worst class recall when trained with cross-entropy loss, CAMRI loss improved recall by 9%.