Speech captured by an in-ear microphone placed inside an occluded ear has a high signal-to-noise ratio; however, it has different sound characteristics compared to normal speech captured through air conduction. In this study, a method for blind speech quality enhancement is proposed that can convert speech captured by an in-ear microphone to one that resembles normal speech. The proposed method estimates an input-dependent enhancement function by using a neural network in the feature domain and enhances the captured speech via time-domain filtering. Subjective and objective evaluations confirm that the speech enhanced using our proposed method sounds more similar to normal speech than that enhanced using conventional equalizer-based methods.

Due to improvements in hardware and software performance, deep learning algorithms have been used in many areas and have shown good results. In this paper, we propose a noise reduction framework based on a convolutional neural network (CNN) with deconvolution and a modified residual network (ResNet) to remove image noise. Simulation results show that the proposed algorithm is superior to the conventional noise eliminator in subjective and objective performance analyses.

This paper presents recognition of anomalously deformed Kana sequences in Japanese historical documents, for which a contest was held by IEICE PRMU 2017. The contest was divided into three levels in accordance with the number of characters to be recognized: level 1: single characters, level 2: sequences of three vertically written Kana characters, and level 3: unrestricted sets of characters composed of three or more characters possibly in multiple lines. This paper focuses on the methods for levels 2 and 3 that won the contest. We basically follow the segmentation-free approach and employ the hierarchy of a Convolutional Neural Network (CNN) for feature extraction, Bidirectional Long Short-Term Memory (BLSTM) for frame prediction, and Connectionist Temporal Classification (CTC) for text recognition, which is named a Deep Convolutional Recurrent Network (DCRN). We compare the pretrained CNN approach and the end-to-end approach with more detailed variations for level 2. Then, we propose a method of vertical text line segmentation and multiple line concatenation before applying DCRN for level 3. We also examine a two-dimensional BLSTM (2DBLSTM) based method for level 3. We present the evaluation of the best methods by cross validation. We achieved an accuracy of 89.10% for the three-Kana-character sequence recognition and an accuracy of 87.70% for the unrestricted Kana recognition without employing linguistic context. These results prove the performances of the proposed models on the level 2 and 3 tasks.

This paper describes the results of outdoor mobility measurements and high-speed vehicle tests that clarify the 4-by-8 multiple-input multiple-output (MIMO) throughput performance when applying distributed MIMO with narrow antenna-beam tracking in a 28-GHz frequency band in the downlink of a 5G cellular radio access system. To clarify suitable transmission point (TP) deployment for mobile stations (MS) moving at high speed, we examine two arrangements for 3TPs. The first sets all TPs in a line along the same side of the path traversed by the MS, and the other sets one TP on the other side of the path. The experiments in which the MS is installed on a moving wagon reveal that the latter deployment case enables a high peak data rate and high average throughput performance exhibiting the peak throughput of 15Gbps at the vehicle speed of 3km/h. Setting the MS in a vehicle travelling at 30km/h yielded the peak throughput of 13Gbps. The peak throughput of 11Gbps is achieved at the vehicle speed of 100km/h, and beam tracking and intra-baseband unit hand over operation are successfully demonstrated even at this high vehicle speed.

Recently, ZigBee has been attracting attention as a low-power and short-range wireless communication standard. In ZigBee networks, it is necessary to suppress frame transfer load because ZigBee needs to operate within severe capacity constraints and with low power consumption. However, in the typical environments in which ZigBee is used, such as smart home networks, WLAN (Wireless LAN) generally coexists, and radio interference occurs between the two networks. Existing studies focused on only interference avoidance. On the other hand, in this paper, we focus on adaptive cooperation between ZigBee network and WLAN. Specifically, from the viewpoints of WLANs that have wider communication range but have many idle periods in some environments like homes, we propose and study a hop count reduction method of ZigBee frame transfer by partially employing WLAN communication to transfer ZigBee frames. To the best of our knowledge, this is the first paper that considers the adaptive cooperation between ZigBee network and WLAN, where some ZigBee frames are transferred via WLAN to the sink. This is a completely new approach different from existing interference avoidance approaches. Then, we evaluate the hop count reduction by considering the number and the positions of relay points to transfer ZigBee frames to WLAN, and ZigBee tree topology for frame transfer routes. Through the evaluation, two realistic deployment policies of the relay points are derived. Finally, as specific advantages from the hop count reduction, we demonstrate the performance improvement about sink arrival ratio and end-to-end transfer delay of ZigBee frames, and energy consumption.

To minimize the annotation costs associated with training semantic segmentation models and object detection models, weakly supervised detection and weakly supervised segmentation approaches have been extensively studied. However most of these approaches assume that the domain between training and testing is the same, which at times results in considerable performance drops. For example, if we train an object detection network using only web images showing a large object at the center, it can be difficult for the network to detect multiple small objects. In this paper, we focus on training a CNN with only web images and achieve object detection in the wild. A proposal-based approach can address the problem associated with differences in domains because web images are similar to images of the proposal. In both domains, the target object is located at the center of the image and the ratio of the size of the target object to the size of the image is large. Several proposal methods have been proposed to detect regions with high “object-ness.” However, many of these proposals generate a large number of candidates to increase the recall rate. Considering the recent advent of deep CNNs, methods that generate a large number of proposals exhibit problems in terms of processing time for practical use. Therefore, we propose a CNN-based “food-ness” proposal method in this paper that requires neither pixel-wise annotation nor bounding box annotation. Our method generates proposals through backpropagation and most of these proposals focus only on food objects. In addition, we can easily control the number of proposals. Through experiments, we trained a network model using only web images and tested the model on the UEC FOOD 100 dataset. We demonstrate that the proposed method achieves high performance compared to traditional proposal methods in terms of the trade-off between accuracy and computational cost. Therefore, in this paper, we propose an intermediate approach between the traditional proposal approach and the fully convolutional approach. In particular, we propose a novel proposal method that generates high“food-ness” regions using fully convolutional networks based on the backward approach by training food images gathered from the web.

Despite the widespread use of deep learning for speech emotion recognition, they are severely restricted due to the information loss in the high layer of deep neural networks, as well as the degradation problem. In order to efficiently utilize information and solve degradation, attention-based dense long short-term memory (LSTM) is proposed for speech emotion recognition. LSTM networks with the ability to process time series such as speech are constructed into which attention-based dense connections are introduced. That means the weight coefficients are added to skip-connections of each layer to distinguish the difference of the emotional information between layers and avoid the interference of redundant information from the bottom layer to the effective information from the top layer. The experiments demonstrate that proposed method improves the recognition performance by 12% and 7% on eNTERFACE and IEMOCAP corpus respectively.

This paper presents a conifer and broad-leaf tree classification scheme that processes high resolution polarimetric synthetic aperture data above X-band. To validate the proposal, fully polarimetric measurements are conducted in a precisely controlled environment to examine the difference between the scattering mechanisms of conifer and broad-leaf trees at 15GHz. With 3.75cm range resolution, scattering matrices of two tree types were measured by a vector network analyzer. Polarimetric analyses using the 4-component scattering power decomposition and alpha-bar angle of eigenvalue decomposition yielded clear distinction between the two tree types. This scheme was also applied to an X-band Pi-SAR2 data set. The results confirm that it is possible to distinguish between tree types using fully polarimetric and high-resolution data above X-band.

This paper proposes a pulse-tail-feedback VCO, in which the tail transistor is driven using pulse-shaped voltage signals with rail-to-rail swing. The proposed pulse-tail-feedback (PTFB) VCO relies on reducing the current conduction period of the tail transistor and operating the tail transistors in triode region for reducing the flicker and thermal noise from the active elements. Mathematical analysis and circuit level simulations of the phase noise mechanism in the proposed PTFB-VCO is also presented in this paper for validating the effectiveness of the proposed technique. A prototype LC-VCO with the proposed PTFB technique is fabricated in a standard 180nm CMOS. Laboratory measurement shows a power consumption of 1.35mW from a 1.2V supply at 4.6GHz. The proposed PTFB-VCO achieves a flicker corner of 700Hz, which enables the VCO to maintain a fairly constant figure-of-merit (FoM) of -195dB within a wide offset frequency range of 1kHz-10MHz.

A low voltage stochastic flash ADC (analog-to-digital converter) is presented, with an inverter-based comparative unit which is used to replace comparator for comparison. Aiming at the low voltage and low power consumption, a key of our design is in the simplicity of the structure. The inverter-based comparative unit replacing a comparator enables us to decrease the number of transistors for area saving and power reduction. We insert the inverter-chain in front of the comparative unit for the signal stability and discuss an appropriate circuit structure for the resolution by analyzing three different ones. Finally, we design the whole stochastic flash ADC for verifying our idea, where the supply voltage can go down to 0.6V on the 65nm CMOS process, and through post-layout simulation result, we can observe its advantage visually in voltage, area and power consumption.

360-degree video delivery in Virtual Reality is very challenging due to the fact that 360-degree videos require much higher bandwidth than conventional videos. To overcome this problem, viewport-adaptive streaming has been introduced. In this study, we propose a new adaptation method for tiling-based viewport-adaptive streaming of 360-degree videos. For content preparation, the Cubemap projection format is used, where faces or parts of a face are encoded as tiles. Also, the problem is formulated as an optimization problem, in which each visible tile is weighted based on how that tile overlaps with the viewport. To solve the problem, an approximation algorithm is proposed in this study. An evaluation of the proposed method and reference methods is carried out under different tiling schemes and bandwidths. Experiments show that the Cubemap format with tiling provides a lot of benefits in terms of storage, viewport quality across different viewing directions and bandwidths, and tolerance to prediction errors.

In order to optimize the performance of thermoelectric devices, we have fabricated and characterized the micrometer-scaled Si thermopile preserving the phonon-drag effect, where the Si thermopile consists of p- and n-type Si wire pairs. The measured Seebeck coefficient of the p-type Si wire was found to be higher than the theoretical value calculated only from the carrier transport, which indicates the contribution of phonon-drag part. Moreover, the measured Seebeck coefficient increased with increasing the width of Si wire. This fact is considered due to dependency of phonon-drag part on the wire width originating from the reduction of phonon-boundary scattering. These contributions were observed also in measured output voltage of Si-wire thermopile. Hence, the output voltage of Si-wire thermopile is expected can be enhanced by utilizing the phonon-drag effect in Si wire by optimizing its size and carrier concentration.

For the three-user X channel, its degree of freedom (DoF) 9/5 has been shown achievable theoretically through asymptotic model with infinite resources, which is impractical. In this article, we explore the propagation delay (PD) feature among different links to maximize the achievable DoF with the minimum cost. Since perfect interference alignment (IA) is impossible for 9 messages within 5 time-slots, at least one extra time-slot should be utilized. By the cyclic polynomial approach, we propose a scheme with the maximum achievable DoF of 5/3 for 10 messages within 6 time-slots. Feasibility conditions in the Euclidean space are also deduced, which demonstrates a quite wide range of node arrangements.

With the existing of coverage holes, the Quality of Service (such as event response, package delay, and the life time et al.) of a Wireless Sensor Network (WSN) may become weaker. In order to recover the holes, one can locate them by identifying the boundary nodes on their edges. Little effort has been made to distinguish the boundary nodes in a model where wireless sensors are randomly deployed on a three-dimensional surface. In this paper, we propose a distributed method which contains three steps in succession. It first projects the 1-hop neighborhood of a sensor to the plane. Then, it sorts the projected nodes according to their angles and finds out if there exists any ring formed by them. At last, the algorithm validates a circle to confirm that it is a ring surrounding the node. Our solution simulates the behavior of rotating a semicircle plate around a sensor under the guidance of its neighbors. Different from the existing results, our method transforms a three-dimensional problem into a two-dimensional one and maintaining its original topology, and it does not rely on any complex Hamiltonian Cycle finding to test the existence of a circle in the neighborhood of a sensor. Simulation results show our method outperforms others at the correctness and effectiveness in identifying the nodes on the edges of a three-dimensional WSN.

Rotation symmetric Boolean functions which are invariant under the action of cyclic group have been used in many different cryptosystems. This paper presents a new construction of balanced odd-variable rotation symmetric Boolean functions with optimum algebraic immunity. It is checked that, at least for some small variables, such functions have very good behavior against fast algebraic attacks. Compared with some known rotation symmetric Boolean functions with optimum algebraic immunity, the new construction has really better nonlinearity. Further, the algebraic degree of the constructed functions is also high enough.

Wireless Full-Duplex (FD) communication can double the point-to-point throughput. To obtain the full benefits of the FD technique in multi-hop networks, its potential throughput performance in multi-hop networks should be clarified qualitatively and quantitatively. Developing an analytical model for FD multi-hop networks is effective and useful for not only clarifying such network dynamics but also developing the optimal protocol design. However, generalized analytical expression for the end-to-end throughput of FD multi-hop networks has not been proposed. This paper proposes analytical expressions for the end-to-end throughput of string-topology wireless FD multi-hop networks. Our approach is to integrate with the analytical model of the airtime expression, which is an effective analytical approach of the throughput analysis for Half-Duplex (HD) multi-hop networks, and the Markov-chain model considering the FD MAC operation. The proposed model clarify the detailed effect of the FD MAC operation on the throughput performance in multi-hop networks. In particular, it can obtain the end-to-end throughput of FD multi-hop networks for arbitrary number of hops, arbitrary payload size and arbitrary value of the minimum contention window. The analytical expressions verified by comparisons with the simulation results. From the comparisons with the results in HD multi-hop networks, we confirm the effectiveness of the FD communication in multi-hop networks.

The electric properties of a field-effect transistor not only depend on gate surface sidewall but also on channel orientation when applying channel stain engineering. The change of the gate surface and channel orientation through the rotated FinFETs provides the capability to compare the orientation dependence of performance and reliability. This study characterized the <100> and <110> channels of FinFETs on the same wafer under tensile and compressive stresses by cutting the wafer into rectangular silicon pieces and evaluated their piezoresistance coefficients. The piezoresistance coefficients of the <100> and <110> silicon under tensile and compressive stresses were first evaluated based on the current setup. Tensile stresses enhance the mobilities of both <100> and <110> channels, whereas compressive stresses degrade them. Electrical characterization revealed that the threshold voltage variation and drive current degradation of the {100} surface were significantly higher than those of {110} for positive bias temperature instability and hot carrier injection with equal gate and drain voltage (VG=VD). By contrast, insignificant difference is noted for the subthreshold slope degradation. These findings imply that a higher ratio of bulk defect trapping is generated by gate voltage on the <100> surface than that on the <110> surface.

We investigated the low temperature formation of Pd2Si on Si(100) with TiN encapsulating layer formed at 500°C/1 min. Furthermore, the dopant segregation process was performed with ion dose of 1x1015 cm-2 for B+. The uniform Pd2Si was successfully formed with low sheet resistance of 10.4 Ω/sq. Meanwhile, the PtSi formed on Si(100) showed rough surface morphology if the silicidation temperature was 500°C. The estimated Schottky barrier height to hole of 0.20 eV (qφBp) was realized for n-Si(100).

This article proposes a prosody correction method based on partial model adaptation for Chinese-accented Japanese hidden Markov model (HMM)-based text-to-speech synthesis. Although text-to-speech synthesis built from non-native speech accurately reproduces the speaker's individuality in synthetic speech, the naturalness of the synthetic speech is strongly degraded. In the proposed model, to improve the naturalness while preserving the speaker individuality of Chinese-accented Japanese text-to-speech synthesis, we partially utilize HMM parameters of native Japanese speech to synthesize prosody-corrected synthetic speech. Results of an experimental evaluation demonstrate that duration and F0 correction are significantly effective for improving naturalness.

In this study, under the assumption that a robot (1) has a remotely controllable yawing camera and (2) moves in a uniform linear motion, we propose and investigate how to improve the target recognition rate with the camera, by using wireless feedback loop control. We derive the allowable data rate theoretically, and, from the viewpoint of error and delay control, we propose and evaluate QoS-Hybrid ARQ schemes under data rate constraints. Specifically, the theoretical analyses derive the maximum data rate for sensing and control based on the channel capacity is derived with the Shannon-Hartley theorem and the path-loss channel model inside the human body, i.e. CM2 in IEEE 802.15.6 standard. Then, the adaptive error and delay control schemes, i.e. QoS-HARQ, are proposed considering the two constraints: the maximum data rate and the velocity of the camera's movement. For the performance evaluations, with the 3D robot simulator GAZEBO, we evaluated our proposed schemes in the two scenarios: the static environment and the dynamic environment. The results yield insights into how to improve the recognition rate considerably in each situation.