Showing papers by "Mitsubishi Electric published in 2016"

Single-Channel Multi-Speaker Separation using Deep Clustering

Abstract: Deep clustering is a recently introduced deep learning architecture that uses discriminatively trained embeddings as the basis for clustering. It was recently applied to spectrogram segmentation, resulting in impressive results on speaker-independent multi-speaker separation. In this paper we extend the baseline system with an end-to-end signal approximation objective that greatly improves performance on a challenging speech separation. We first significantly improve upon the baseline system performance by incorporating better regularization, larger temporal context, and a deeper architecture, culminating in an overall improvement in signal to distortion ratio (SDR) of 10.3 dB compared to the baseline of 6.0 dB for two-speaker separation, as well as a 7.1 dB SDR improvement for three-speaker separation. We then extend the model to incorporate an enhancement layer to refine the signal estimates, and perform end-to-end training through both the clustering and enhancement stages to maximize signal fidelity. We evaluate the results using automatic speech recognition. The new signal approximation objective, combined with end-to-end training, produces unprecedented performance, reducing the word error rate (WER) from 89.1% down to 30.8%. This represents a major advancement towards solving the cocktail party problem.

Improved MVDR beamforming using single-channel mask prediction networks

Abstract: Recent studies on multi-microphone speech databases indicate that it is beneficial to perform beamforming to improve speech recognition accuracies, especially when there is a high level of background noise. Minimum variance distortionless response (MVDR) beamforming is an important beamforming method that performs quite well for speech recognition purposes especially if the steering vector is known. However, steering the beamformer to focus on speech in unknown acoustic conditions remains a challenging problem. In this study, we use singlechannel speech enhancement deep networks to form masks that can be used for noise spatial covariance estimation, which steers the MVDR beamforming toward the speech. We analyze how mask prediction affects performance and also discuss various ways to use masks to obtain the speech and noise spatial covariance estimates in a reliable way. We show that using a single mask across microphones for covariance prediction with minima-limited post-masking yields the best result in terms of signal-level quality measures and speech recognition word error rates in a mismatched training condition.

PWM Carrier Harmonic Iron Loss Reduction Technique of Permanent-Magnet Motors for Electric Vehicles

Abstract: This paper investigates a method to increase fuel economy of motors for an electric vehicle (EV) traction system. The low-torque region of an EV motor is frequently used in everyday urban driving and in fuel economy measurement. Pulsewidth modulation (PWM) carrier harmonic iron loss accounts for a high percentage of total loss for EV motors in this region. This paper reveals the relationship between phase current and carrier harmonic iron loss for a permanent-magnet synchronous motor (PMSM) in the low-torque region. Thus, a novel carrier harmonic loss reduction technique is proposed and applied to two sets of three-phase windings for a concentrated winding PMSM. The loss is successfully reduced in a 70-kW EV motor as well as a two-dimensional finite-element method (2-D-FEM) calculation. The efficiency of the developed motor is calculated and compared with that of a conventional motor in the JC08 mode driving cycle.

Single-Channel Multi-Speaker Separation using Deep Clustering

Abstract: Deep clustering is a recently introduced deep learning architecture that uses discriminatively trained embeddings as the basis for clustering. It was recently applied to spectrogram segmentation, resulting in impressive results on speaker-independent multi-speaker separation. In this paper we extend the baseline system with an end-to-end signal approximation objective that greatly improves performance on a challenging speech separation. We first significantly improve upon the baseline system performance by incorporating better regularization, larger temporal context, and a deeper architecture, culminating in an overall improvement in signal to distortion ratio (SDR) of 10.3 dB compared to the baseline of 6.0 dB for two-speaker separation, as well as a 7.1 dB SDR improvement for three-speaker separation. We then extend the model to incorporate an enhancement layer to refine the signal estimates, and perform end-to-end training through both the clustering and enhancement stages to maximize signal fidelity. We evaluate the results using automatic speech recognition. The new signal approximation objective, combined with end-to-end training, produces unprecedented performance, reducing the word error rate (WER) from 89.1% down to 30.8%. This represents a major advancement towards solving the cocktail party problem.

More Efficient Privacy Amplification With Less Random Seeds via Dual Universal Hash Function

Abstract: We explicitly construct random hash functions for privacy amplification (extractors) that require smaller random seed lengths than the previous literature, and still allow efficient implementations with complexity $O(n\log n)$ for input length $n$ . The key idea is the concept of dual universal2 hash function introduced recently. We also use a new method for constructing extractors by concatenating $\delta $ -almost dual universal2 hash functions with other extractors. Besides minimizing seed lengths, we also introduce methods that allow one to use non-uniform random seeds for extractors. These methods can be applied to a wide class of extractors, including dual universal2 hash function, as well as to the conventional universal2 hash functions.

Development of DC/DC Converter for Battery Energy Storage Supporting Railway DC Feeder Systems

Abstract: This paper describes the development of a bilateral dc/dc converter rated at 1500 V with a peak power of 500 kW for battery energy storage systems supporting railway dc feeder systems. The dc/dc converter converts regenerated power from a braking train and charges the batteries. The dc/dc converter discharges the stored energy to feed the train during powering. The converter main circuit and the control system are described, and successful test results obtained at the factory are reported. For the main circuit test, a special test method was applied. For the control system test, a real-time digital simulator was used for hardware-in-loop tests. In addition to the converter tests, the results from dc feeder system tests with an actual rectifier and battery bank are also presented.

Demonstration of SiC-MOSFET embedding Schottky barrier diode for inactivation of parasitic body diode

Abstract: External Schottky barrier diodes (SBD) are generally used to suppress the conduction of the body diode of MOSFET. A large external SBD is required for a high voltage module because of its high specific resistance, while the forward voltage of SBD should be kept smaller than the built-in potential of the body diode. Embedding SBD into MOSFET with short cycle length increases maximum source-drain voltage where body diode remains inactive, resulting in high current density of SBD current. We propose a MOSFET structure where an SBD is embedded into each unit cell and an additional doping is applied, which allows high current density in reverse operation without any activation of body diode. The proposed MOSFET was successfully fabricated and much higher reverse current density was demonstrated compared to the external SBD. We can expect to reduce total chip size of high voltage modules using the proposed MOSFET embedding SBD.

On the Performance of Full-Duplex Two-Way Relay Channels With Spatial Modulation

Abstract: In this paper, the spatial modulation (SM) technique is employed at the source and relay nodes in a full-duplex two-way relay channel (FD-TWRC) to support spectral-efficient bi-directional communications while guaranteeing a low cost implementation. Maximum likelihood detectors are employed at each node that is subject to an intrinsic self-loop interference. We first propose a tight upper bound on the average bit error probability (ABEP). Then, based on the ABEP upper bound, an asymptotic ABEP expression is derived in the high signal-to-noise ratio (SNR) regime. Exploiting the asymptotic ABEP, an exact SNR threshold for the selection between FD-TWRC-SM and half-duplex (HD)-TWRC-SM is derived in a closed form, which sheds light on when it is beneficial to select the FD (or HD) mode. In addition, the power allocation (PA) among sources and relay is investigated, through which an optimal PA factor in terms of ABEP is obtained. All analytical results derived in this paper are verified by Monte Carlo simulations, from which some new insights are obtained on the performance of FD-TWRC-SM.

Mobile communications system and mobile terminal

Abstract: A mobile terminal judges whether or not the mobile terminal is able to make a transition to a DTX period during Active, and, when judging that the mobile terminal is able to make a transition to a DTX period during Active, notifies a base station to that effect. If the base station judges that the mobile terminal is able to make a transition to a DRX period during Active when triggered by the notification from the mobile terminal, the base station temporarily stops supply of electric power to the data transmission processing units and the data reception processing units of the mobile terminal.

Exemplar learning for extremely efficient anomaly detection in real-valued time series

Abstract: We investigate algorithms for efficiently detecting anomalies in real-valued one-dimensional time series. Past work has shown that a simple brute force algorithm that uses as an anomaly score the Euclidean distance between nearest neighbors of subsequences from a testing time series and a training time series is one of the most effective anomaly detectors. We investigate a very efficient implementation of this method and show that it is still too slow for most real world applications. Next, we present a new method based on summarizing the training time series with a small set of exemplars. The exemplars we use are feature vectors that capture both the high frequency and low frequency information in sets of similar subsequences of the time series. We show that this exemplar-based method is both much faster than the efficient brute force method as well as a prediction-based method and also handles a wider range of anomalies. We compare our algorithm across a large variety of publicly available time series and encourage others to do the same. Our exemplar-based algorithm is able to process time series in minutes that would take other methods days to process.

Adaptively Attribute-Hiding (Hierarchical) Inner Product Encryption

Abstract: This paper proposes the first inner product encryption (IPE) scheme that is adaptively secure and fully attribute-hiding (attribute-hiding in the sense of the definition by Katz, Sahai and Waters), while the existing IPE schemes are either fully attribute-hiding but selectively secure or adaptively secure but weakly attribute-hiding. The proposed IPE scheme is proven to be adaptively secure and fully attribute-hiding under the decisional linear assumption in the standard model. The IPE scheme is comparably as efficient as the existing attribute-hiding IPE schemes. We also present a variant of the proposed IPE scheme with the same security that achieves shorter public and secret keys. A hierarchical IPE scheme can be constructed that is also adaptively secure and fully attribute-hiding under the same assumption. In this paper, we extend the dual system encryption technique by Waters into a more general manner, in which new forms of ciphertext and secret keys are employed and new types of information theoretical tricks are introduced along with several forms of computational reduction.

Upgrade of small angle x-ray scattering beamline BL-6A at the photon factory

Abstract: BL-6A has been operational since 2011 as a small angle X-ray scattering (SAXS) beamline at the Photon Factory (PF), and beginning in 2013 its old components and systems, which were mainly inside the experimental hutch, have been extensively updated. Both the vacuum-passes located between the sample stage and the detector and the fixed surface plate have been replaced by a new semi-automatic diffractometer. These upgrades allow simultaneous SAXS/WAXS experiments and grazing-incidence small angle X-ray scattering (GISAXS) measurements to be conducted. The hybrid pixel detector PILATUS3 1M is installed for SAXS, and PILATUS 100K is available as a WAXS detector. Additionally, a pinhole equipped with a micro-ion chamber is available to realize a lower-background and higher-resolution of low angles. Moreover, in a simultaneous SAXS/WAXS experiment, we developed a new beam stop with an embedded photodiode. Thus, BL-6A has evolved into a multipurpose beamline capable of dealing with various types of samples and experimental techniques.

Power conversion device

Abstract: A power conversion device includes an alternating current rotating machine having a multiple of multi-phase windings, a multiple of power conversion means that convert direct current voltage of a direct current power supply based on a multiple of switching signals and apply voltage to the multiple of multi-phase windings, first current detection means that detects a first bus current, which is current flowing between the power conversion means that applies voltage to one multi-phase windings of the multiple of multi-phase windings and the direct current power supply, and first phase current calculation means that calculates the current flowing through the one multi-phase windings based on the first bus current, wherein the first current detection means detects the first bus current at a timing at which a multiple of voltage vectors based on the multiple of switching signals neighbor or coincide.

Iteration-Aware LDPC Code Design for Low-Power Optical Communications

Abstract: Recent low-density parity-check (LDPC) codes have shown capacity-approaching performance for various communications systems. However, their promising performance cannot always be obtained due to practical constraints, such as finite codeword length, finite iteration, finite memory, and finite precision. In this paper, we focus on a practical design method of high-performance LDPC codes under a constraint of finite-iteration decoding for low-power optical communications. We introduce an iteration-aware LDPC code design approach, which is based on decoding trajectory in extrinsic information transfer chart analysis. It is demonstrated that an LDPC code designed by the conventional curve-fitting method exhibits nearly 2 dB of penalty, when the maximum number of iterations is limited. The results suggest that the LDPC code should be adaptively changed, e.g., when the number of decoding iterations is decreased to save power consumption. We also extend our design method to a multi-objective optimization concept by taking average degrees into account, so that the threshold and the computational complexity are minimized at the same time. The proposed Pareto-optimal codes can achieve additional 2-dB gain or 50% complexity reduction at maximum, in low-power decoding scenarios.

Photocurrent enhancement of graphene phototransistors using p–n junction formed by conventional photolithography process

Abstract: A p–n junction was developed in a graphene transistor by a simple photolithography process used in typical semiconductor processes. The p- and n-type regions were formed by coating photoresist on part of the graphene channel and immersion of the uncovered graphene region in alkali developer, respectively. A 3-fold enhancement of the photocurrent was observed at the maximum field effect mobility. It is therefore important to maximize the field effect mobility by doping to maximize the photocurrent. The results obtained here are an important step toward the production of high-sensitivity graphene-based phototransistors compatible with conventional industrial procedures.

Dialog state tracking with attention-based sequence-to-sequence learning

Abstract: We present an advanced dialog state tracking system designed for the 5th Dialog State Tracking Challenge (DSTC5). The main task of DSTC5 is to track the dialog state in a human-human dialog. For each utterance, the tracker emits a frame of slot-value pairs considering the full history of the dialog up to the current turn. Our system includes an encoder-decoder architecture with an attention mechanism to map an input word sequence to a set of semantic labels, i.e., slot-value pairs. This handles the problem of the unknown alignment between the utterances and the labels. By combining the attention-based tracker with rule-based trackers elaborated for English and Chinese, the F-score for the development set improved from 0.475 to 0.507 compared to the rule-only trackers. Moreover, we achieved 0.517 F-score by refining the combination strategy based on the topic and slot level performance of each tracker. In this paper, we also validate the efficacy of each technique and report the test set results submitted to the challenge.

Real-Time Parameter Identification and Integration on Deadbeat-Direct Torque and Flux Control (DB-DTFC) Without Inducing Additional Torque Ripple

Abstract: This paper is dedicated to exploring real-time parameter identification approaches and their integration with deadbeat-direct torque and flux control (DB-DTFC) without additional torque ripple. Using the voltage model at medium- and high-speed ranges, DB-DTFC is insensitive to parameters. Performance degrades at low speeds with the rotor time constant errors, because the current model is utilized for flux linkage estimation. Two different real-time parameter estimation approaches are developed, including a flux observer-based model reference adaptive system (MRAS) and a pulsating flux injection-based method. Both are experimentally evaluated for how well they mitigate DB-DTFC performance degradation at low speed. For the MRAS-based method, the fundamental components are used for parameter convergence, which does not induce additional torque ripple. For the injection-based method, traditional d -axis pulsating voltage vector injection used in indirect field-oriented control (IFOC) yields significant torque ripple, particularly at high speed. The proposed pulsating flux injection scheme used in DB-DTFC excites the same magnitude of current harmonics and induces no additional torque ripple over a wide speed range.

Characterization and Modeling of a 1.2-kV 30-A Silicon-Carbide MOSFET

Abstract: This paper describes a novel compact model for a SiC-MOSFET. The model is useful to achieve accurate simulation of output characteristics from a linear region to a saturation region, selecting both gate–source voltage and temperature as parameters. In order to construct the model systematically, attention is paid to a physics-based modeling procedure with channel mobility as an adjustable parameter. The model also features characterization and modeling of an internal drain–gate capacitor. The model shows fairly good agreement in the output characteristics and the dynamic behavior of both gate drive circuit and main power circuits between the experimental and simulated results. This successful validation indicates that this model offers a promising circuit-based simulation tool for designing whole power conversion systems using SiC-MOSFETs.

Acoustic carrier transportation induced by surface acoustic waves in graphene in solution

Abstract: The acoustic charge transportation induced by surface acoustic wave (SAW) propagation in graphene in solution was investigated. The sign of acoustic current (I A) was found to switch when crossing the Dirac point because the major carrier was transitioned from holes to electrons by the change in electrolyte-gate voltage. I A also exhibited a peak value under conditions of both hole and electron conduction. These results can be explained on the basis of a change in the type of major carrier in graphene, as well as a change in the carrier mobility of graphene.

Optimized power modules for silicon carbide MOSFET

Abstract: An Integrated Power Board technology was used to construct a 3D power module. This packaging is suitable for use of WBG devices as it reduces the inductive parasitics to the strict minimum, with a 2nH loop inductance in our 1.2kV/80A SiC prototype using SiC MOSFETs. The packaging presents virtually no parasitics or necessity for slowing down the commutation, which is oscillation-free. The conducted emissions of the 3D module are more than halved in comparison to those of a bespoke wire-bonded, EMI-optimised module.

Online state of charge estimation for Lithium-ion batteries using Gaussian process regression

Abstract: This paper presents an application of Gaussian process regression (GPR) to estimate a state of charge (SoC) of Lithium-ion (Li-ion) batteries with different kernel functions. One of the practical advantages of using GPR is that uncertainties in the estimates can be quantified, which enables reliability assessment of the SoC estimate. The inputs of GPR are voltage, current and temperature measurements of the battery and the output is an estimate of SoC. First, training is performed in which optimal hyperparameters of a kernel function are determined to model data properties. Then, the battery SoC is estimated online based on the trained model. The kernel function is the key element in the GPR model since it encodes the prior assumptions about the properties of the function being modeled. Therefore, the impact of kernel function selection on the estimation performance is analyzed using both simulated data and experimental data collected from a LiMn2O4/hard-carbon battery with a nominal capacity of 4.93Ah operating under constant charge and discharge currents.

A Soft-Switched Asymmetric Flying-Capacitor Boost Converter With Synchronous Rectification

Abstract: The multilevel flying-capacitor boost converter was analyzed for asymmetric voltage operation—this permits loss optimization that takes advantage of different voltage class MOSFETs. A cost-motivated design of a suitable zero-voltage zero-current switching snubber is then developed that permits a great reduction of the inductor size. With the proposed snubber, synchronous rectification operation is compared to that of diode boost, with a particular attention to the contribution of nonlinear MOSFET parasitics. Experimental results from a 2-kW/30-kHz prototype justify the effectiveness of this solution with a conversion efficiency around $99\% \pm 0.1\%$ for a wide load range.

Semiconductor device and method of manufacturing semiconductor device

Abstract: The present invention is a semiconductor device including an IGBT and an FWD formed in a silicon carbide semiconductor region. The IGBT is provided with an emitter electrode, a base region, an emitter region, a collector region, a collector electrode, a gate insulating film, and a gate electrode. The emitter electrode is formed on one principal surface side of the silicon carbide semiconductor region and the collector region is formed on one principal surface side of the silicon carbide semiconductor region. The gate insulating film is in contact with the silicon carbide semiconductor region, the emitter region, and the base region and the gate electrode is opposite to the gate insulating film. The FWD is provided with a base contact region and a cathode region. The base contact region is adjacent to the emitter region and is electrically connected to the emitter electrode. The cathode region is arranged in an upper layer portion on the other principal surface side of the silicon carbide semiconductor region, is provided adjacent to the collector region, and is electrically connected to the collector electrode. The IGBT is further provided with a carrier trap-reducing region. The carrier trap-reducing region is arranged at a main current-conducting region in an upper silicon carbide semiconductor region in the collector region and has a carrier trap reduced more than in the upper silicon carbide semiconductor region in the collector region.

Fabrication of photonic crystal structures by tertiary-butyl arsine-based metal-organic vapor-phase epitaxy for photonic crystal lasers

Abstract: The fabrication of air/semiconductor two-dimensional photonic crystal structures by air-hole-retained crystal regrowth using tertiary-butyl arsine-based metal–organic vapor-phase epitaxy for GaAs-based photonic crystal lasers is investigated. Photonic crystal air holes with filling factors of 10–13%, depths of ~280 nm, and widths of 120–150 nm are successfully embedded. The embedded air holes exhibit characteristic shapes due to the anisotropy of crystal growth. Furthermore, a low lasing threshold of ~0.5 kA/cm2 is achieved with the fabricated structures.