To realize Society 5.0, edge AI techniques have attracted attention. On the other hand, security issues of edge AI have been reported. In addition, in the field of hardware security, the threat of hardware Trojan (HT) is emphasized. To defend the AI device from malicious attacks, it is important to check the vulnerability against various attacks. Therefore, this study proposes a new HT for AI inference devices. The proposed HT falsifies the inference result with respect to an arbitrary trigger input. The proposed HT concentrates on the Lookup Table (LUT) structure, and can be achieved by rewriting the LUT table information. As a result, the proposed HT does not need additional trojan trigger and payload circuits, that is, it can be implemented without the circuit overhead. Experiments by field programable gate array show the validity of the proposed HT.
Network Traversal with Mobility (NTMobile), which provides both IP mobility and connectivity in a mixed IPv4/IPv6 environment, can control whether or not encrypted UDP tunnel communication is allowed per node according to an access control list. However, in the case of NTMobile nodes where communication is allowed, not only authorized application communication but also malware communication can pass through NATs and firewalls. This paper proposes a new mechanism to control the communication per process. With the proposed mechanism, even if an NTMobile node is allowed to communicate, it can identify the relevant application process from the sending and receiving packets, and control the packet passing or dropping in accordance with the rules. As a result of implementing and verifying a prototype of the proposed method, we confirmed that the communication availability can be controlled for each process. We also evaluated the throughput performance and confirmed that the proposed method can achieve the performance without any practical problems by utilizing the cache function.
Several studies have been conducted on measurements based on the self-coupling effect of semiconductor lasers. In these studies, the modulation of the optical output caused by the scattering of light from an object to the active layer is considered. This sensor based on the self-coupling effect can measure distance and velocity simultaneously. In this study, the theoretical equations for the measurement resolutions of a sensor that could measure distance and velocity simultaneously were obtained. In addition, the influence of the distance and velocity resolutions on the accuracy of the velocity measurement was investigated. The distance resolution of the sensor was determined based on the wavelength band of the semiconductor laser, whereas the velocity resolution was determined based on the modulation frequency.
In this study, we developed a new method to detect unknown low-height obstacles using 3D point clouds from stereo cameras. Conventional semantic segmentation methods using a depth image by Deep Neural Network (DNN) can detect road surfaces with a high accuracy. However, it is difficult to detect unknown low-height obstacles not included in the training data. Methods that use geometric information such as normal and height face difficulty to find objects with a surface that parallels to the road surface and low objects, respectively. Therefore, we deal with the difficult problem of detecting unknown low-height obstacles. To solve the problem, we focused on the difference in difficult detection between DNN and geometric methods. Based on the confidence from the output of DNN, we help difficult obstacle detection for DNN by using geometric information, and vice versa. When tested on a robot equipped with a stereo camera, the IoU, which indicates the detection accuracy of unknown obstacles, was improved by 18.1 percentage points compared to DNN. Moreover, our method enabled the robot to safely avoid three types of unknown low-height obstacles.
In this letter, a synthesis method of the stabilized observers with a function of estimating unmeasurable inputs (OFE) to second-order lag systems is proposed. The basic idea is to synthesize the compensator gain of OFE such that one pole of OFE can be freely given while another can be chosen from the stable area. Usefulness of the proposed method is illustrated by a numerical example with the stabilization of the OFE and the estimation performance of disturbances.
We propose a method for generating a self-hearing voice by applying the voice conversion technique to a recorded voice. We sometimes feel unnatural when we hear our own recorded voice. One of the reasons is the difference in the acoustic transmission path. Our proposed method imitates voice changes caused by the acoustic path with the voice conversion technique. In the experiments, two different models were used to generate a self-hearing voice. The generated self-hearing voices were subjectively evaluated whether they were suitable as self-hearing voices. The results addressed that self-hearing voice can be generated by the voice conversion technique.
We have recently developed a several milli-joule, sub-nanosecond laser source for ophthalmic therapy application which allows to lower a laser breakdown threshold, less than half, and more stably/effectively yield the breakdown phenomenon compared to that based on conventionally employed source, implying the developed source has a potential to realize more minimally invasive ophthalmic treatment for secondary cataract. For further laser performance, the use of a photonic crystal device as an output coupler of the laser source has been also demonstrated. With this system, pulse width shortening and polarization control of the laser output could be achieved simultaneously. This paper presents these recent ophthalmic laser developments.
For Internet-of-things (IoT) terminals, wiring or battery replacement are not effective power supplying methods due to excess installation and maintenance cost. Optical wireless power transmission (OWPT) is a promising candidate for easy charging with the advantages of long transmission distances, good directionality, and portable size. In this research, a light emitting diode (LED) based OWPT system that achieved safety light beam, portable size, and large electricity power supply was designed and demonstrated through simulations and experiments. In addition, a novel design of the LED-array OWPT system, which has arrayed multiple LEDs as the entire light source, has been proposed to improve the final output above 380 mW. In particular, the tolerant distances and alignment deviations of both single-LED and LED-array OWPT systems were analyzed in detail.
Recently, research to achieve narrow emission from GaN-based VCSELs has been of great interest. Sony's VCSELs with curved mirror has feature of strong lateral optical condiment, in which, causes wider emission angle. The authors obtained narrow emission angle of 3.9 degree by adopting concave mirror with large radius of curvature.
In this paper, we propose design of low-power digital expanded-pulse width modulation (E-PWM) using adiabatic dynamic CMOS logic (ADCL) for 12-step dimming control circuit of an LED lighting system. As results of HSPICE simulation, power consumption of E-PWM was reduced by approximately 90% compared to conventional CMOS 12-step dimming PWM circuits. Furthermore, the layout of E-PWM was designed using 0.18µm standard CMOS process and circuit area of 12-step dimming PWM circuit was compared between previous studies and this study. The area of the design layout is not only significantly reduced compared to the dimming PWM of previous studies using ADCL, but also reduced approximately 68% compared to the conventional CMOS 12-step dimming PWM circuit.
The column readout circuit in CMOS image sensor is highly integrated analog circuit arranged for each column of pixels in general. Due to the degree of integration, crosstalk between columns is likely to occur, resulting in deterioration in image quality. However, the integration of circuit components is limited in column floor plan, so that countermeasures unique to high integration is required. In this paper, various causes of smear, which is a typical image noise, in the readout circuit and its countermeasures are discussed.
This paper proposes parasitic compensated 2-phase 3rd and 5th-order switched capacitor low-pass filters using single operational amplifier. The effect of parasitic capacitance can be theoretically removed under the assumption that ratio of capacitance and parasitic capacitance is constant. Simulation results show good agreement with the theory.
Signal-to-quantization-noise ratio (SQNR) of discrete-time delta-sigma modulators (DTΔΣMs) are degraded when low-gain operational amplifiers (Op-Amps) are used. Correlated level shifting (CLS) technique has been proposed to compensate gain of Op-Amp in pipeline analog-to-digital converters. In this paper, we show that gain of the Op-Amps are improved by applying CLS technique to Op-Amps in the integrators of DTΔΣM, and SQNR is improved even if low-gain Op-Amps are used. The simulation results show that SQNR is improved by 6 dB and SQNR using 374x gain Op-Amp is equivalent to that using 30x gain with CLS technique, and considering nonlinearity of Op-Amps, harmonic distortions are suppressed by over 15 dB.
In edge computing environments, edge servers are placed on network edges in addition to centralized cloud servers in order to reduce latency and avoid congestion. For designing edge computing systems, in this paper, we propose an optimization method that solves a joint optimization problem of edge server placement and virtual machine allocation. In this problem, we determine the placement of edge servers in a network and allocate virtual machines to those edge servers. To do so, we first provide mathematical formulations that obtain optimal solution of the problem, and then a metaheuristic algorithm based on the simulated annealing algorithm that obtains approximation solution in short time. Through numerical experiments, we show the performance of the proposed optimization method.
Understanding information processing in peripheral somatosensory pathway is essential for revealing mechanisms account for sensory impairments. To investigate cellular functions under in vivo-like conditions, many cell-culture systems for mimicking the structure of somatosensory fibers have been developed. However, there remains a difficulty in detecting electrophysiological properties of individual axons from the axon bundle. To address this, we proposes two analytical methods of evaluating conduction properties of axons using a culture device which allows for multi-site recording of extracellular potentials from an axon bundle. First, we tried to estimate velocity distribution of axons by electrical stimulations. Using the relationships between excitability and conduction velocity, we classified axons into two groups (0.57 m/s and 0.43 m/s) based on analysis of response waveforms evoked by electrical stimulations. Second, we investigated propagation properties of capsaicin-induced action potentials in individual axons. Localized application of capsaicin into the distal part of axons elicited antidromic propagation (68% of total propagations) and induced high-frequency activity (> 10 Hz) which leads to slowing of conduction velocity (12% of initial velocity). These phenomena are related to in vivo somatosensory processing, indicating that the methods will be a useful tool for studying somatosensory processing at the cellular level.
Model biological membranes are a useful tool to study the molecular properties and functions of membrane proteins. We develop a strategy to directly reconstitute mammalian membrane proteins from the cell membrane into a model biological membrane to bypass the technically challenging solubilization and purification processes. We expressed dopamine D2 receptor (D2R), a G-protein coupled receptor (GPCR), in Chinese hamster ovary (CHO) cells and produced cell membrane blebs by chemical induction. By introducing blebs into a patterned framework of lithographically-polymerized lipid bilayer on the substrate surface, we could form a planar bilayer and observe single molecules of D2R. Interestingly, a much higher density of D2R molecules were reconstituted in a nanometric cleft between the substrate and a poly-dimethylsiloxane (PDMS) elastomer sheet. This methodology should enable to evaluate the physicochemical properties and functions of a wide range of mammalian membrane proteins.
In this paper, we propose a method to perform partial re-colorization of an image that has been miscolored by fully automatic color restoration. This makes it possible to make partial colorization without changing the areas that were correctly colorized by the fully automated method.
Authors previously proposed a closed loop identification method based on Laguerre functions for an industrial controller with non-linear procedures. In this method, the controlled and manipulated variable are presented in the frequency domain using the orthonormality of the Laguerre functions. The estimated process model is described as their ratio. However, the estimated model has the relative order of 0, which is not suitable for simulation. This paper proposes a method to correct the estimated model to strictly proper model by partial fraction decomposition. The effectiveness of the proposed method is confirmed by PID control simulations for the corrected model and actual model.
The robust controller is often required for suppressing modeling errors and disturbances. In order to design the robust controller analytically, the H∞ control problem and the µ-synthesis are proposed as the model-based controller design techniques. These two methods need the nominal plant model and the uncertainty. However, it is difficult to obtain the accurate plant model with limited complexity. The inappropriate plant model leads to large uncertainty. As a result, the designed controller is conservative. This paper proposes designing the robust controller with only frequency responses. In addition, the proposed method designs the controller automatically by the numerical optimization. In the method, the robust performance condition is addressed as the design problem. The robust performance condition is formulated as a iterative convex optimization using the linear approximation by the Taylor expansion. However, since the Taylor expansion is used for the formulation, the initial controller is needed and the designed controller depends on the initial controller. To solve the problem, the particle swarm optimization (PSO) is introduced to obtain a reasonable initial controller. The PSO reduces dependence on initial controllers because many initial controllers are given randomly as candidates of the solution. Simulation and experimental results are shown to confirm the effectiveness of the proposed method.
Inspection of the occurrence of underground cavities is important for preventing sudden road collapse accidents. As a non-excavation inspection method, there is a ground penetrating radar technology that observes the backscattering response when electromagnetic waves are directed into the ground. It is expected that accurate estimation of the underground structure will be possible by devising the analysis method of the response data obtained by the radar. In this paper, we approximate the underground structure as a parallel multilayer structure with layers stacked in the direction perpendicular to the ground surface, and construct two mathematical models that give the backscattering response when waves are radiated into the ground. The underground structure can be estimated by optimizing the structural parameters of the model so that the backscattering response given by the model matches the measured response. After comparing the accuracy of the two models, we propose a structure estimation method using Particle Swarm Optimization (PSO). PSO is one of the metaheuristic optimization algorithms with excellent global exploration performance in a multidimensional data space. In this paper, the existence of a cavity is estimated by optimizing the medium constants of many layers, assuming an underground structure model consisting of 21 layers. The possibility of precise inspection of underground cavities is demonstrated.
The vehicle routing problems for services of collecting home garbages are hard to optimize due to not only their combinatorial scale but also their intrinsic constraints. This problem is becoming a serious issue from viewpoints of cost savings of the public service of local governments. We try to develop the optimization method incorporating realistic functions in order to evaluate and compare various systems of garbage collection services. In this optimization method, the problem is formulated as a set covering problem which enables to decompose the problem into routes generation and routes selection. The routes can be generated by adopting dynamic programming techniques on the road network of the city. On the other hand, the optimal set of routes is selected by solving the set covering problem of the candidate route collection. Through some numerical evaluation for the realistic services of garbage collection, the potential for application of our method is examined.
Urban-scale traffic simulations using a digital map of Kobe City showed a power-law behaviors in the distribution of traffic volume relative to the number of road segments. In order to clarify the origin of this behavior, simulations were conducted using artificially generated random and Manhattan-type road networks. The behavior was observed in the former, but not in the latter. Similar results were obtained when the simulation was performed with uniform OD (Origin and Destination) distributions on the map of Kobe City. We also found similar results only with shortest path search without simuation for each uniformly randomized OD distribution. These suggest that the road network itself is causing the behavior. To understand its origin, we assumed a mathematical toy model in which a tree structure is embedded in the city of Kobe and random networks and confirmed the behavior appears on Cayley trees and a road network with bottlenecks which generated from the lattice networks based on the bond percolation theory. We also confirmed that such tree structures and bottlenecks are embedded in the simulation results above and each produces a power-law like distribution. The relationship between these facts and the origin of the power law is discussed.
Non-linearity characteristics of equipment, such as Cogeneration system (CGS), would complicate to optimally operate microgrid system (MGS) composed of it and growth the amount of calculation (time) to make an operation planning. In this paper, the authors use particle swarm optimization (PSO), which is known as a heuristic, to solve complex problems while maintaining non-linearity, and propose an optimal operation planning method that combines 1-week calculations and 24-hour calculations to obtain high accuracy solution by consideration of multiple days in appropriate calculation time. Through numerical experiments, we evaluate calculation accuracy, time, the effects of prediction error.
Inverse reinforcement learning is used for complex control tasks by using experts. However, since the learning results depend on the expert, it is impossible to imitate ungiven policies from expert when there are multiple optimal polices for the same goal, or when the environment changes from the training. The problems can be solved by giving multiple experts and representing their features in the latent space. the proposed method extends information maximizing generative adversarial imitation learning with adversarial inverse reinforcement learning to deal with such environment. Experiments show that the proposed method can not only imitate multiple experts, but also estimate ungiven polices.
We investigated the effects of incoherent visual motion information on impression of walking using Virtual Reality. We defined and modulated the Gain which was the ratio of the visual velocity and the walking velocity. We conducted an experiment in which participants walked while estimating six kinds of impressions of walking: feeling of body lightness, feeling of vigor, enjoyment of walking, difficulty of walking, feeling of speed from physical sensation, and felling of exhaustion after walking as functions of the Gain. The results showed that the impressions of walking depend on the Gain in VR environment. The feeling of body lightness, the feeling of vigor, the enjoyment of walking, and the feeling of speed from physical sensation were enhanced with increasing the Gain but those feeling were saturated at high Gains. The difficulty of walking and the feeling of exhaustion after walking did not depend on the Gain. These results indicate that we can control the impressions of walking by modulating the Gain.
We digitized the paintings of the Genji-e cultural property and studied how to express their photorealistic appearance using virtual reality technology. In order to capture the striking reflections of the gold leaf in the paintings, we applied BRDF (Bidirectional Reflectance Distribution Function) data, an index of the spectral reflectance of gold, to the game engine’s shader to obtain a texture close to the appearance of gold leaf. We have also developed an interactive viewing system that allows users to appreciate Genji-e paintings in the space of the era in which they were painted. We conducted an evaluation experiment of the content and verified its effectiveness.
In recent years, the number of devices that have the ability to use speech as a user interface has been increasing. This function has the defect that its power consumption is large. It is not a problem for devices that use an external power supply, but it becomes a serious problem for battery-powered devices such as internet of things (IoT) devices. To solve this, we propose an analog front-end circuit that combines speech detection and position detection. By using these two detections as triggers to turn on the speech recognition, we can identify that a person is nearby. This makes it possible to minimize the operation time of the speech recognition system and to improve the power consumption.
This paper presents the validation of the zero-crossing method and the linear predictive coding (LPC) analysis method as the speech detection methods used in our proposed system. The analysis results show the relationship between power consumption, detection time, and detection accuracy.
The genetic algorithms (GA) are evolutionary algorithms that imitate the evolution process of many living things. GA operates multipoint search, therefore the parallelization of the genetic operators are effective to shorten computational time. The island model GA is one of extended model of GA such that introducing parallel computation. This paper develops new algorithm of island model GA by introducing the concept of TCPSO (Two-swarm Cooperative Particle Swarm Optimization). In the proposed method, the individuals are updated by using different procedures which are determined depending on the categorized groups. Additionally, this paper conducts numerical experiments and the result indicates the effectiveness of the proposed algorithm.
We have developed the fill-in workbook system with the aim of understanding students’ learning attitudes and supporting teachers in improving lessons. In the lessons using this system, students take the lessons while filling in the blanks using the digital textbook. This system collects the logs. By filling in the blanks, students in class are forced to interact with the system, and it is possible to collect learning behaviors of students who are not active. Teachers can adjust the pace of the lesson by referring to the information in real time during the lesson. In addition, by presenting the trends of the entire lesson to teachers and students after the lesson, they can use it for retrospective review of the lesson.
It is expected that data will be shared with high reliability among multiple sites to improve operations entire the supply chain. In this research, we propose the dynamic distribution method for containerized data processing functions to collect and analyze data at multiple locations. When collecting and analyzing data from entire the supply chain, communication between the data processing function and the volume becomes a bottleneck, causing a problem of degrading the data processing. By the proposed method that dynamically determines the location of data processing functions based on the traffic volume and the communication delay time, we confirmed data processing speed is improved.
Smart Community (SC) uses IoT sensors to provide smart grid control, traffic management, and similar IoT services. These services expect to run at the network edge or fog layer to provide low latency network services, encapsulate citizens’ private information, support low-cost IoT terminals from cyber-attacks, and support other cutting-edge fog and edge services. SC edge is a service platform that supports edge/fog services for IoT terminals by using Docker containers. Initially, SC edge/fog computing nodes did not support the function of service migration. However, service migration is necessary to support remote deployment and service distribution in SC networks. The existing container migration techniques focus mainly on resource utilization. However, SC services should handle loss-free data stream processing and order-preservation of network packets to gather IoT sensor data after migration. In addition, SC services require one-to-many migration to support high throughput loads when required. Therefore, this paper focuses on enhancing SC service flexibility by introducing migration for relocatable and network consistency guaranteed containerized services. SC edge proposes multiple container migration techniques that are suitable for network services. The proposed techniques can improve resource consumption, guarantee network traffic consistency, and apply one-to-many migration patterns. Layer leveraging migration (LLM) reduces the overall migration time by 10.8% for an elastic search Docker container than available Docker migration methods. Additionally, consistency guaranteed migration (CGM) is proposed to guarantee network consistency. However, CGM consumes additional resources compared to LLM for IoT data management. Finally, One to N Consistency Guaranteed Migration (O2NCGM) is proposed to support one-to-many migration with data consistency that shows similar performance to CGM.
In various industrial fields such as chemistry and steel, efforts are being made to improve productivity by visualizing and analyzing the operating status of data obtained from various manufacturing facilities and their control systems. Therefore, in the future, we believe that there will be more cases where new applications are added to manufacturing facilities and control systems, and the application is applied to different manufacturing facilities and control systems. So we want to avoid duplication of work in data integration between applications and control systems. To solve this problem, we propose a data management platform “Context-based Data Management System” that semi-automatically resolves differences in semantics of data using the context of application data and data-source data. Based on the case where the proposed platform was applied to a chemical plant, it was confirmed that man-hours were reduced by more than 1/3.
In recent years, the wearable computing equipped with several sensors built into a small microcomputer has been attracting people’s attention, and various information and data are acquired from people. In this paper, a novel behavior suggestion system using a small microcomputer connected with an ECG (Electrocardiograph) monitor and an accelerometer is proposed.
In this system, the small microcomputer is connected with the server machine via wireless network. This server machine can assess a person’s heart rate, the status of autonomic nervous system, a person’s location information, and a person’s activity. In this way, the system for a person’s behavior suggestion is realized. And, optimized information for behavior suggestion is generated on some scenario base from several threshold values, and this useful information is presented to a user.
In order to evaluate the effectiveness of this system, first, basic experiments was performed to acquire the finding of optimal parameters. Then, suggestion experiments were performed using this system. These results confirmed that this system proposes useful information by experiments of two scenarios.
In this paper, the system architecture, its mechanism, and several experiments to confirm its efficacy are shown. Several possible improvements of this system in the future are also shown.
There is a growing need to use photovoltaic (PV) technology to mitigate global warming and the depletion of fossil fuels while also enhancing energy security. However, the high network penetration of PVs has various negative effects on electrical power systems. In recent years, automated operating systems have been introduced to enable rapid and automatic restoration after a power outage. However, PVs shut down in the case of a power outage and are not automatically restored. Therefore, it is necessary to compensate the electric power output of the PVs by using an extra power supply at the time of power recovery. For this compensation process, it is necessary for a system operator to monitor the unmeasurable power outputs of the PVs, including those belonging to other organizations and individuals, to ensure that the automated operating system is used properly. Based on these aspects, in this paper, we proposed an estimation method of PV power output utilizing only power consumption data obtained from smart meters and confirmed the high accuracy of the good temporal resolution estimates using measured data.
We propose the control system for driving robot using Hierarchical Reinforcement Learning. Driving Robots are playing an active role in test driving for evaluating fuel consumption and exhaust gas of automobiles. We can consider Reinforcement Learning as one of the control methods for driving robot. The control system using Reinforcement Learning has the advantage that there is no need to adjust parameters manually. However, Reinforcement Learning suffer from poor sample efficiency because it requires a lot of trials. In this research, we propose the control system for driving robot using the algorithm for learning hierarchical policy. Moreover, we introduce State Abstraction in Hierarchical Reinforcement Learning. By using abstract state, each low-level policy specialize in distinct behavior. The advantage of this method is that we can improve the sample efficiency by transferring low-level policies learned using multiple vehicles. The experimental result shows that the proposed method improve the sample efficiency in vehicle velocity tracking task.
We have developed a normal-dispersion passively mode-locked ytterbium-doped bismuth-based fiber (YDBF) laser with a 1 GHz repetition frequency at around 1062 nm wavelength using a short linear cavity design. We confirmed a pulse train at a high-repetition frequency of 1 GHz with a pulse repetition interval of 1 ns with pump power of 390 mW. Dispersion compensation optics are not needed in the cavity. After adding dispersion to obtained spectrum data as the phase term, we Fourier-transformed them to obtain a pulse waveform. For that waveform, we roughly estimated the full width at half maximum (FWHM) of the pulse broadened by the chirp effect caused by normal dispersion of YDBF. It was about 860 fs.
In this paper, a stability analysis of a fast convergence adaptive algorithm with square sum of correlation function as a cost function and the application to whitening of the input wideband signal of adaptive notch filter are discussed. Whitening is achieved by using the autocorrelation function of a signal with a spectrum that is inversely related to the power density spectrum of the broadband signal. The whitening function of the adaptive algorithm is analyzed and the convergence properties of the adaptive notch filter are verified by computer simulation.
The background of this research is to reduce the burden on users for password management with the spread of information devices such as smartphones. The purpose of this research is to generate a unique password character string that reflects the change in color tone in the image using the RGB component. The proposed method extracts the DCT sign of the RGB component by zigzag scan and generates a password character string. As a result, it was found that the proposed method is an image-specific password character string that reflects the difference in color tone.