Showing papers by "Mitsubishi Electric published in 2015"

Speech enhancement and recognition using multi-task learning of long short-term memory recurrent neural networks

Abstract: Long Short-Term Memory (LSTM) recurrent neural network has proven effective in modeling speech and has achieved outstanding performance in both speech enhancement (SE) and automatic speech recognition (ASR). To further improve the performance of noise-robust speech recognition, a combination of speech enhancement and recognition was shown to be promising in earlier work. This paper aims to explore options for consistent integration of SE and ASR using LSTM networks. Since SE and ASR have different objective criteria, it is not clear what kind of integration would finally lead to the best word error rate for noise-robust ASR tasks. In this work, several integration architectures are proposed and tested, including: (1) a pipeline architecture of LSTM-based SE and ASR with sequence training, (2) an alternating estimation architecture, and (3) a multi-task hybrid LSTM network architecture. The proposed models were evaluated on the 2nd CHiME speech separation and recognition challenge task, and show significant improvements relative to prior results.

Characteristics of 600, 1200, and 3300 V Planar SiC-MOSFETs for Energy Conversion Applications

Abstract: SiC-MOSFETs provide superior performance for next generation power electronics systems. High threshold voltage 600 V SiC-MOSFETs were realized utilizing a reoxidation process, which drastically improves a tradeoff between an ON-resistance and a threshold voltage. Low-loss SiC-MOSFETs with a 1200 V/100-A rating have been developed. Using the developed SiC-MOSFETs, 1200 V/800-A high-power full SiC module with an ON-resistance as low as 2.9 m $\Omega $ at 150 °C was successfully fabricated. The high-power module markedly reduces power loss especially at high carrier frequency. Large-area 3300 V SiC-MOSFETs with an ON-resistance of 52 m $\Omega $ at 175 °C exhibit an adequate reverse bias safe operating area and 3300 V SiC-MOSFETs screened by applying a body diode current stress show stable characteristics under a continuous current through their body diode for 1000 h.

Electricity theft detection using smart meter data

Abstract: Electricity theft is a major concern for the utilities. With the advent of smart meters, the frequency of collecting household energy consumption data has increased, making it possible for advanced data analysis, which was not possible earlier. We have proposed a temperature dependent predictive model which uses smart meter data and data from distribution transformer to detect electricity theft in an area. The model was tested for varying amounts of power thefts and also for different types of circuit approximations. The results are encouraging and the model can be used for real world application.

Adaptive Estimation of the State of Charge for Lithium-Ion Batteries: Nonlinear Geometric Observer Approach

Abstract: This paper considers the state of charge (SoC) and parameter estimation of lithium-ion batteries. Different from various prior arts, where estimation is based on local linearization of a nonlinear battery model, nonlinear geometric observer approach is followed to design adaptive observers for the SoC and parameter estimation based on nonlinear battery models. A major advantage of the proposed approach is the possibility to establish the exponential stability of the resultant error dynamics of state and parameter estimation. The proposed adaptive observers are shown to be robust with respect to unmodeled process uncertainties. Analysis also shows the design tradeoff between the convergence rate and the robustness of the estimation error dynamics with respect to the measurement noise. Simulation and experimental results validate the effectiveness and main advantages of the proposed approach. Error analysis is presented and explains the experimental results.

Power conversion system

Abstract: When a plurality of storage batteries is used by connecting them in parallel, since the progress of degradation differs among the storage batteries, a power conversion system includes a degradation information acquisition device for acquiring the degradation information of the storage batteries, a temperature information acquisition device for detecting the temperature information of the storage batteries, and a control device for controlling the storage battery power converter based on the degradation information of the storage batteries by the degradation information acquisition device and the temperature information of the storage batteries by the temperature information acquisition device so that the degradation states of the plurality of the storage batteries can be matched.

Silicon carbide semiconductor device and silicon carbide semiconductor device manufacturing method

Abstract: A silicon carbide semiconductor device capable of achieving a decrease in ON resistance and an increase in breakdown voltage and a method for manufacturing a silicon carbide semiconductor device. A silicon carbide semiconductor device includes a silicon carbide substrate and a drift layer. The drift layer includes a breakdown voltage holding layer extending from a point where a doping concentration has a predetermined value to a surface of the drift layer. The doping concentration in the breakdown voltage holding layer continuously decreases from the point where the doping concentration has the predetermined value to a modulation point located further toward the surface of the drift layer than a midpoint in a film thickness direction of the breakdown voltage holding layer. The doping concentration in the breakdown voltage holding layer continuously increases from the modulation point to the surface of the drift layer.

Bayesian Speech and Language Processing

Abstract: With this comprehensive guide you will learn how to apply Bayesian machine learning techniques systematically to solve various problems in speech and language processing. A range of statistical models is detailed, from hidden Markov models to Gaussian mixture models, n-gram models and latent topic models, along with applications including automatic speech recognition, speaker verification, and information retrieval. Approximate Bayesian inferences based on MAP, Evidence, Asymptotic, VB, and MCMC approximations are provided as well as full derivations of calculations, useful notations, formulas, and rules. The authors address the difficulties of straightforward applications and provide detailed examples and case studies to demonstrate how you can successfully use practical Bayesian inference methods to improve the performance of information systems. This is an invaluable resource for students, researchers, and industry practitioners working in machine learning, signal processing, and speech and language processing.

Moving image coding apparatus and moving image decoding apparatus

Abstract: A macro block size determining unit 1 determines the size of each macro block on a frame-by-frame basis. A macro block dividing unit 2 divides an inputted image into macro blocks each having the size determined by the macro block size determining unit 1. A macro block coding unit 3 determines a coding mode for each of the macro blocks divided by the macro block dividing unit 2, and codes pixel values in each of the macro blocks in the determined coding mode.

Vehicle energy management device

Abstract: A vehicle energy management device includes: an energy consumption related information acquisition unit which acquires energy consumption related information that is information related to energy consumption of an own vehicle in each of traveling sections; an energy consumption related information correction unit which corrects the energy consumption related information; and a control plan production unit which produces a control plan for vehicle instruments based on an estimated value of an energy consumption amount of each of the vehicle instruments calculated by using the corrected energy consumption related information. The energy consumption related information correction unit corrects the energy consumption related information based on a result of comparing an actually measured value of traveling characteristics of the own vehicle in each of the road categories and a statistical result of traveling characteristics of a plurality of general vehicles in each of the traveling sections.

Achieving short ciphertexts or short secret-keys for adaptively secure general inner-product encryption

Abstract: In this paper, we present two non-zero inner-product encryption (NIPE) schemes that are adaptively secure under a standard assumption, the decisional linear (DLIN) assumption, in the standard model. One of the proposed NIPE schemes features constant-size ciphertexts and the other features constant-size secret-keys. Our NIPE schemes imply an identity-based revocation (IBR) system with constant-size ciphertexts or constant-size secret-keys that is adaptively secure under the DLIN assumption. Any previous IBR scheme with constant-size ciphertexts or constant-size secret-keys was not adaptively secure in the standard model. This paper also presents two zero inner-product encryption (ZIPE) schemes each of which has constant-size ciphertexts or constant-size secret-keys and is adaptively secure under the DLIN assumption in the standard model. They imply an identity-based broadcast encryption system with constant-size ciphertexts or constant-size secret-keys that is adaptively secure under the DLIN assumption. We also extend the proposed ZIPE schemes in two directions, one is a fully-attribute-hiding ZIPE scheme with constant-size secret-keys, and the other a hierarchical ZIPE scheme with constant-size ciphertexts.

A fast short-circuit protection method using gate charge characteristics of SiC MOSFETs

Abstract: This paper describes a fast protection circuit for silicon carbide metal-oxide semiconductor field-effect transistors (SiC MOSFETs) subjected to hard-switching faults (HSFs). In terms of reliability of power converters, the protection of power semiconductors against short-circuit failures is of great concern. The reverse transfer capacitance increases with decreasing drain-source voltage during normal turn-on transient. Under HSF conditions, on the other hand, it hardly changes because the drain-source voltage remains high. As a consequence, quite a significant difference appears in gate charge characteristics between under HSF conditions and normal turn-on operation. Hence, an HSF can be detected by monitoring both the gate- source voltage and the amount of gate charge. The proposed protection circuit has high noise tolerance because it monitors not only the gate-source voltage but also the amount of gate charge. The validity of the protection circuit is verified by experiment. The proposed protection circuit can detect the HSF within only one microsecond.

Surface light source device and liquid crystal display device

Abstract: An angular intensity distribution shaping member includes a first surface for receiving a light beam emitted from a light source, the first surface having a length in a width direction that is longer than that in a thickness direction; second surfaces forming a plate-shaped light path on which the light beam incident from the first surface propagates by total reflection, the second surfaces including at least one adjustment surface for spreading a full angle of an angular intensity distribution, in the width direction, of the light beam incident from the first surface so that the full angle becomes wider than that of the angular intensity distribution of the light beam just after emission from the light source; and a third surface through which the light beam exits, the full angle of its angular intensity distribution in the width direction having been widened by the at least one adjustment surface.

3.3 kV/1500 A power modules for the world’s first all-SiC traction inverter

Abstract: We have successfully developed 4H-SiC devices including metal–oxide–semiconductor field-effect transistors (MOSFETs) and Schottky barrier diodes (SBDs) with a rated voltage of 3.3 kV. The conduction loss of the SiC-MOSFET was reduced to as low as that of the Si-insulated gate bipolar transistor (IGBT) by the n-type doping of the junction field-effect transistor region (JFET doping). The JFET doping technique is effective in reducing the temperature coefficient of resistance in the JFET region, leading to the decreased on-resistance of the SiC-MOSFET at high temperatures. These devices have been applied to 3.3 kV/1500 A modules for the world's first all-SiC traction inverter. The switching loss of the new traction inverter system is approximately 55% less than that of a conventional inverter system incorporating Si modules.

Performance Analysis of Cooperative Wireless Networks With Unreliable Backhaul Links

Abstract: A cooperative wireless network, where a cluster of $K$ single-antenna transmitters jointly serve a single-antenna receiver, is considered. Each transmitter is connected to the control unit (CU) via independent but unreliable backhaul links. The CU sends a common message to each transmitter over backhaul links, which upon successful reception, jointly transmit this message to the intended receiver. To facilitate analysis, a general expression is derived for the complementary cumulative distribution function of a sum of $K$ independent random variables, where each random variable is a product of an exponential and a Bernoulli random variable. This result is applied to find a simple closed-form expression that characterizes the system outage performance as a function of network parameters and node geometry. The analytical model is validated using numerical simulations. As an application, the derived expression is also used for investigating the impact of backhaul assignment on the system performance.

Power supply system

Abstract: A power supply system includes a plurality of uninterruptible power supplies provided for a load in parallel. The uninterruptible power supplies each include a power supply unit configured to supply the load with power and being larger in capacitance than the load, and a switch provided between the power supply unit and the load. The power supply system further includes a control unit selecting a first uninterruptible power supply of the plurality of uninterruptible power supplies, and setting the switch of the first uninterruptible power supply to the on state.

Virtual image display device

Abstract: A virtual image display device includes a light source unit, a polarization switching unit, an image generation unit, an optical path unit and a projection unit. The polarization switching unit switches a polarization direction of a light beam emitted from the light source unit. The optical path unit includes a first optical path through which the light beam having the first polarization direction travels and a second optical path through which the light beam having the second polarization direction travels. A first virtual image is displayed at a first distance with the light beam that has traveled through the first optical path. A second virtual image is displayed at a second distance farther than the first distance with the light beam that has traveled through the second optical path.

Concurrent Multiband Digital Outphasing Transmitter Architecture Using Multidimensional Power Coding

Abstract: All-digital outphasing transmitter architecture using multidimensional power coding (MDPC) is proposed for noncontiguous concurrent multiband transmission with a high power efficiency. MDPC transforms multiband digital baseband signals into multibit low-resolution digital signals that drive switching-mode PAs. A prototype digital outphasing transmitter consists of two 1-GHz bandwidth GaN Class-D PAs and a Chireix power combiner. The two GaN PAs are driven by bipolar radio frequency (RF) pulse-width modulation (PWM) signals, which are transformed from a concurrent dual-band LTE signal by MDPC. The dual-band LTE signal with 15-MHz aggregate channel bandwidth at 240 and 500 MHz frequency band is transmitted with $-$ 30 and $-$ 37 dBc out-of-band emissions, respectively. Digital outphasing achieves more than two times higher coding efficiency than conventional concurrent dual-band digital transmitters with the same PAs in Class-S operation. Measured power coding efficiencies of 35.4% and 47.1% are observed with outphasing bipolar and 3-level RF PWM signals respectively, which are encoded from the dual-band LTE signal.

Method of manufacturing semiconductor device

Abstract: A method of manufacturing a semiconductor device, includes forming a laser section on a portion of a substrate, the laser section including an active layer, an upper semiconductor layer on the active layer, and a mask on the upper semiconductor layer; forming a compound semiconductor layer of an indium-containing material in contact with a side of the laser section, the compound semiconductor layer having a projection immediately adjacent the laser section; and wet etching and removing the projection with an etchant containing hydrobromic acid and acetic acid, planarizing the compound semiconductor layer, and producing a (111)A surface in the upper semiconductor layer, under the mask

A Rayleigh Quotient-Based Recursive Total-Least-Squares Online Maximum Capacity Estimation for Lithium-Ion Batteries

Abstract: The maximum capacity, the amount of maximal electric charge that a battery can store, not only indicates the state of health, but also is required in numerous methods for state-of-charge estimation. This paper proposes an alternative approach to perform online estimation of the maximum capacity by solving the recursive total-least-squares (RTLS) problem. Different from prior art, the proposed approach poses and solves the RTLS as a Rayleigh quotient optimization problem. The Rayleigh quotient-based approach can be readily generalized to other parameter estimation problems including impedance estimation. Compared with other capacity estimation methods, the proposed algorithm enjoys the advantages of existing RTLS-based algorithms for instance, low computation, simple implementation, and high accuracy, and thus is suitable for use in real-time embedded battery management systems. The proposed method is compared with existing methods via simulations and experiments.

Rear-emitter Si heterojunction solar cells with over 23% efficiency

Abstract: We have developed highly crystallized n-type microcrystalline Si layers as window layers for rear emitter Si heterojunction solar cells. We introduce a seed layer between an n-type microcrystalline Si layer and an intrinsic amorphous Si layer to improve the crystallinity of the n-type microcrystalline Si layer. By using this stacked layer instead of an n-type amorphous Si layer, the contact resistance between the n-type thin layer and In2O3:H is reduced without Al-doped ZnO. As a result, we obtain a high short-circuit current and a high fill factor simultaneously, and achieve a solar cell efficiency of 23.43%.

Uncertainty propagation through deep neural networks

Abstract: In order to improve the ASR performance in noisy environments , distorted speech is typically pre-processed by a speech enhancement algorithm, which usually results in a speech estimate containing residual noise and distortion. We may also have some measures of uncertainty or variance of the estimate. Uncertainty decoding is a framework that utilizes this knowledge of uncertainty in the input features during acoustic model scoring. Such frameworks have been well explored for traditional probabilistic models, but their optimal use for deep neural network (DNN)-based ASR systems is not yet clear. In this paper, we study the propagation of observation uncertainties through the layers of a DNN-based acoustic model. Since this is intractable due to the nonlinearities of the DNN, we employ approximate propagation methods, including Monte Carlo sampling , the unscented transform, and the piecewise exponential approximation of the activation function, to estimate the distribution of acoustic scores. Finally, the expected value of the acoustic score distribution is used for decoding, which is shown to further improve the ASR accuracy on the CHiME database, relative to a highly optimized DNN baseline.

Simulation and comparison of leakage characteristics of R290 in rolling piston type rotary compressor

Abstract: This paper presents a comparative analysis on leakage losses of a rolling piston type rotary compressor with refrigerants of R290, R22 and R410a. In particular, the difference of leakage losses characteristics between R290 compressor and that of other refrigerant compressors has been studied. Leakage losses through radial clearance, clearances between rolling piston faces and the cylinder head walls, clearances between the sliding vane faces and the cylinder head walls, as well as the clearances between the sides of the sliding vane and the side walls of the chutes inside the cylinder have been studied in the present methodology. Radial leakage caused by the wall velocity has been taken into account in the present study. Simulation results indicate that smaller radial clearance is required for R290 compressor to obtain relatively high efficiency comparing to that of R22 and R410a compressors under the same condition.

Development of a 1500Vdc photovoltaic inverter for utility-scale PV power plants

Abstract: The increase in size of large-scale photovoltaic plants increases the relative impact of ohmic losses in the dc and ac transmission. On the other hand, the amount of strings also increases, along with the number of combiner boxes and related equipment. This results in increased losses and costs that impose a limit to the competitiveness of PV technology for large power plants. PV plants today are rated at 1000V, which is considered low voltage in IEC and ANSI standards, but in fact, the IEC standard covers up to 1500V in the low voltage category. Going to this maximum voltage will allow for a considerable reduction in current, reducing the system losses on the DC side. On the other hand, a longer string can be achieved, reducing the number of combiner boxes and allowing a rise in the ac voltage. This paper presents the development of a 2.3MW inverter with a maximum DC system voltage of 1500V. A neutral point switch type three-level inverter configuration, so-called T-type three-level inverter, is employed for better conversion efficiency. Simulation results confirm the performance of the 1500V rated inverter.

Full PRF-Secure Message Authentication Code Based on Tweakable Block Cipher

Abstract: We propose a new message authentication code MAC based on a tweakable block cipher TBC. We prove that the new MAC is a pseudo-random function PRF upi¾?to $$O2^n$$ queries, that is, full PRF-security, where the output length of the TBC is n bits. We note that although Yasuda proposed a full PRF-secure MAC based on a compression function CF, that does not offer a full PRF-secure TBC-based MAC due to the PRF/PRF switch. Hence our MAC is the first full PRF-secure one based on a TBC.

Power control system, healthcare instrument, activity measurement apparatus, and power instruction apparatus

Abstract: A power control system includes an activity measurement apparatus configured to measure an amount of activity, a healthcare instrument configured to measure metabolic data, and a power instruction apparatus including a healthcare processing unit configured to acquire these pieces of measurement data. The power instruction apparatus includes an environment detection unit configured to detect ambient environment of each of the activity measurement apparatus and the healthcare instrument. The power instruction apparatus is configured to store, in its internal memory, the measurement data acquired from each of the healthcare instrument and the activity measurement apparatus, display the measurement data on its display screen, measure, with the environment detection unit, the ambient environment of each of the activity measurement apparatus and the healthcare instrument, determine environment suitability based on the measurement data, and display determination result on a display device to notify an occupant of the result.

Method for manufacturing semiconductor device

Abstract: An insulating film (2) is formed on a main surface of a semiconductor substrate (1) that includes an active region and a termination region. The insulating film (2) in the active region is etched to form an opening (3). The insulating film (2) is used as a mask and an impurity is implanted into the semiconductor substrate (1) in a direction tilted by 20° or more from a direction normal to the main surface of the semiconductor substrate (1) while rotating the semiconductor substrate (1) to form a diffusion layer (7) in the active region. The diffusion layer (7) extends wider than the opening (3) up to below the insulating film (2) on the termination region side.

Bidirectional non-contact power supply device and bidirectional non-contact power supply system

Abstract: Provided is a bidirectional non-contact power supply device having a high power transmission efficiency with a simple configuration. A bidirectional non-contact power supply device (1) is provided with a coil (2) that supplies and receives power by means of magnetic field coupling to another coil, and in the device, the coil (2) and a capacitor (3) are connected in series, and are connected to an input/output terminal of an inverter circuit (4), and a bidirectional step up/down converter (5) is connected between the inverter circuit (4) and a direct current power supply (7).

Outsourcing the Re-encryption Key Generation: Flexible Ciphertext-Policy Attribute-Based Proxy Re-encryption

Abstract: In this paper, we introduce a new proxy re-encryption (PRE) in that the re-encryption key generation can be outsourced, in attribute-based encryption. We call this new notion flexible ciphertext-policy attribute-based proxy re-encryption (flexible CP-AB-PRE). In ordinary PRE scheme, re-encryption keys are generated by using user’s decryption key and an access structure. So, whenever the access structure is changed, a PRE user has to generate new different re-encryption keys. In order to overcome this disadvantage of the ordinary PRE, the re-encryption key generation of the proposed scheme is divided into the following two steps. First, a user generates universal re-encryption key urk S which indicates delegator’s attributes set S. Second, an authority who has re-encryption secret key rsk generates ordinary re-encryption key \({\sf rk}_{S\rightarrow{\Bbb M}'}\) by using urk S , rsk, and an access structure \({\Bbb M}'\). The user has only to generate single urk S for all re-encryption keys. By this “outsourcing”, the task of re-encryption key generation for a user is reduced only to generate one urk S . Furthermore, supposing a Private Key Generator (PKG) generates urk simultaneously at the time of decryption key generation, the load of re-encryption key generation for users almost vanishes.