This paper proposes and analyzes a pipelining scheme for a hardware squarer that can square unsigned integers of up to 12 bits. Each stage is designed and adjusted such that stage delays are well balanced and that the critical path delay of the design does not exceed the reference value which is set up based on the analysis. The resultant design has the critical path delay of approximately 3.5 times a full-adder delay. In an implementation using an Intel Stratix V FPGA, the design operates at approximately 23% higher frequency than the comparable pipelined squarer provided in the Intel library.

A monaural speech enhancement method combining deep neural network (DNN) with low rank analysis and speech present probability is proposed in this letter. Low rank and sparse analysis is first applied on the noisy speech spectrogram to get the approximate low rank representation of noise. Then a joint feature training strategy for DNN based speech enhancement is presented, which helps the DNN better predict the target speech. To reduce the residual noise in highly overlapping regions and high frequency domain, speech present probability (SPP) weighted post-processing is employed to further improve the quality of the speech enhanced by trained DNN model. Compared with the supervised non-negative matrix factorization (NMF) and the conventional DNN method, the proposed method obtains improved speech enhancement performance under stationary and non-stationary conditions.

Previous research has shown that transcripts generated by automatic speech recognition (ASR) technologies can improve the listening comprehension of non-native speakers (NNSs). However, we still lack a detailed understanding of how ASR transcripts affect listening comprehension of NNSs. To explore this issue, we conducted two studies. The first study examined how the current presentation of ASR transcripts impacted NNSs' listening comprehension. 20 NNSs engaged in two listening tasks, each in different conditions: C1) audio only and C2) audio+ASR transcripts. The participants pressed a button whenever they encountered a comprehension problem, and explained each problem in the subsequent interviews. From our data analysis, we found that NNSs adopted different strategies when using the ASR transcripts; some followed the transcripts throughout the listening; some only checked them when necessary. NNSs also appeared to face difficulties following imperfect and slightly delayed transcripts while listening to speech - many reported difficulties concentrating on listening/reading or shifting between the two. The second study explored how different display methods of ASR transcripts affected NNSs' listening experiences. We focused on two display methods: 1) accuracy-oriented display which shows transcripts only after the completion of speech input analysis, and 2) speed-oriented display which shows the interim analysis results of speech input. We conducted a laboratory experiment with 22 NNSs who engaged in two listening tasks with ASR transcripts presented via the two display methods. We found that the more the NNSs paid attention to listening to the audio, the more they tended to prefer the speed-oriented transcripts, and vice versa. Mismatched transcripts were found to have negative effects on NNSs' listening comprehension. Our findings have implications for improving the presentation methods of ASR transcripts to more effectively support NNSs.

Microblog data contains rich information of real-world events with great commercial values, so microblog-oriented natural language processing (NLP) tasks have grabbed considerable attention of researchers. However, the performance of microblog-oriented Chinese Word Segmentation (CWS) based on deep neural networks (DNNs) is still not satisfying. One critical reason is that the existing microblog-oriented training corpus is inadequate to train effective weight matrices for DNNs. In this paper, we propose a novel active learning method to extend the scale of the training corpus for DNNs. However, due to a large amount of partially overlapped sentences in the microblogs, it is difficult to select samples with high annotation values from raw microblogs during the active learning procedure. To select samples with higher annotation values, parameter λ is introduced to control the number of repeatedly selected samples. Meanwhile, various strategies are adopted to measure the overall annotation values of a sample during the active learning procedure. Experiments on the benchmark datasets of NLPCC 2015 show that our λ-active learning method outperforms the baseline system and the state-of-the-art method. Besides, the results also demonstrate that the performances of the DNNs trained on the extended corpus are significantly improved.

The next generation wireless LAN standard IEEE 802.11ax aims to provide improved throughput performance in dense environments. We have proposed an efficient channel sounding mechanism for DL-MU-MIMO that has been adopted as a new sounding protocol in the 802.11ax standard. In this paper, we evaluate the overhead reduction in the 802.11ax sounding protocol compared with the 802.11ac sounding protocol. Sounding is frequently performed to obtain accurate channel information from the associated stations in order to improve overall system throughput. However, there is a trade-off between accurate channel information and the overhead incurred due to frequent sounding. Therefore, the sounding interval is an important factor that determines system throughput in DL-MU-MIMO transmission. We also evaluate the effect of sounding interval on the system throughput performance using both sounding protocols and provide a comparative analysis of the performance improvement.

IEEE 802.11ah is a new wireless standard for large-scale wireless connectivity in IoT and M2M applications. One of the major requirements placed on IEEE 802.11ah is the energy-efficient communication of several thousand stations with a single access point. This is especially difficult to achieve during network initialization, because the several thousand stations must rely on the rudimentary approach of random channel access, and the inevitable increase in channel access contention yields a long association delay. IEEE 802.11ah has introduced an authentication control mechanism that classifies stations into groups, and only a small number of stations in a group are allowed to access the medium at a time. Although the grouping strategy provides fair channel access to a large number of stations, the presence of several thousand stations and limitation that only a group can use the channel at a time, causes the association time to remain excessive. In this paper, we propose a novel block association method that enables simultaneous association of all groups. Our experiments verify that our block association method decreases the total association time by many folds.

In this paper, we propose a Mobile Edge Internet of Things (MEIoT) architecture by leveraging the fiber-wireless access technology, the cloudlet concept, and the software defined networking framework. The MEIoT architecture brings computing and storage resources close to Internet of Things (IoT) devices in order to speed up IoT data sharing and analytics. Specifically, the IoT devices (belonging to the same user) are associated to a specific proxy Virtual Machine (VM) in the nearby cloudlet. The proxy VM stores and analyzes the IoT data (generated by its IoT devices) in real-time. Moreover, we introduce the semantic and social IoT technology in the context of MEIoT to solve the interoperability and inefficient access control problem in the IoT system. In addition, we propose two dynamic proxy VM migration methods to minimize the end-to-end delay between proxy VMs and their IoT devices and to minimize the total on-grid energy consumption of the cloudlets, respectively. Performance of the proposed methods is validated via extensive simulations.

Peer-to-peer overlay networks can easily achieve a large-scale content sharing system on the Internet. Although unstructured peer-to-peer networks are suitable for finding entire partial-match content, flooding-based search is an inefficient way to obtain target content. When the shared content is semantically specified by a great number of attributes, it is difficult to derive the semantic similarity of peers beforehand. This means that content search methods relying on interest-based locality are more advantageous than those based on the semantic similarity of peers. Existing search methods that exploit interest-based locality organize multiple peer groups, in each of which peers with common interests are densely connected using short-cut links. However, content searches among multiple peer groups are still inefficient when the number of incident links at each peer is limited due to the capacity of the peer. This paper proposes a novel content search method that exploits interest-based locality. The proposed method can organize an efficient peer-to-peer network similar to the semantic small-world random graph, which can be organized by the existing methods based on the semantic similarity of peers. In the proposed method, topology transformation based on local link replacement maintains the numbers of incident links at all the peers. Simulation results confirm that the proposed method can achieve a significantly higher ratio of obtainable partial-match content than existing methods that organize peer groups.

Deep neural network (DNN)-based speech synthesis can produce more natural synthesized speech than the conventional HMM-based speech synthesis. However, it is not revealed whether the synthesized speech quality can be improved by utilizing a multi-speaker speech corpus. To address this problem, this paper proposes DNN-based speech synthesis using speaker codes as a method to improve the performance of the conventional speaker dependent DNN-based method. In order to model speaker variation in the DNN, the augmented feature (speaker codes) is fed to the hidden layer(s) of the conventional DNN. This paper investigates the effectiveness of introducing speaker codes to DNN acoustic models for speech synthesis for two tasks: multi-speaker modeling and speaker adaptation. For the multi-speaker modeling task, the method we propose trains connection weights of the whole DNN using a multi-speaker speech corpus. When performing multi-speaker synthesis, the speaker code corresponding to the selected target speaker is fed to the DNN to generate the speaker's voice. When performing speaker adaptation, a set of connection weights of the multi-speaker model is re-estimated to generate a new target speaker's voice. We investigated the relationship between the prediction performance and architecture of the DNNs through objective measurements. Objective evaluation experiments revealed that the proposed model outperformed conventional methods (HMMs, speaker dependent DNNs and multi-speaker DNNs based on a shared hidden layer structure). Subjective evaluation experimental results showed that the proposed model again outperformed the conventional methods (HMMs, speaker dependent DNNs), especially when using a small number of target speaker utterances.

Modern FPGAs have been integrated in computing systems as accelerators for long running applications. This integration puts more pressure on the fault tolerance of computing systems, and the requirement for dependability becomes essential. As in the case of CPU-based system, checkpoint/restart techniques are also expected to improve the dependability of FPGA-based computing. Three issues arise in this situation: how to checkpoint and restart FPGAs, how well this checkpoint/restart model works with the checkpoint/restart model of the whole computing system, and how to build the model by a software tool. In this paper, we first present a new checkpoint/restart architecture along with a checkpointing mechanism on FPGAs. We then propose a method to capture consistent snapshots of FPGA and the rest of the computing system. Third, we provide “fine-grained” management for checkpointing to reduce performance degradation. For the host CPU, we also provide a stack which includes API functions to manage checkpoint/restart procedures on FPGAs. Fourth, we present a Python-based tool to insert checkpointing infrastructure. Experimental results show that the checkpointing architecture causes less than 10% maximum clock frequency degradation, low checkpointing latencies, small memory footprints, and small increases in power consumption, while the LUT overhead varies from 17.98% (Dijkstra) to 160.67% (Matrix Multiplication).

To efficiently use network resources, internet service providers need to conduct traffic engineering that dynamically controls traffic routes to accommodate traffic change with limited network resources. The performance of traffic engineering (TE) depends on the accuracy of traffic prediction. However, the size of traffic change has been drastically increasing in recent years due to the growth in various types of network services, which has made traffic prediction difficult. Our approach to tackle this issue is to separate traffic into predictable and unpredictable parts and to apply different control policies. However, there are two challenges to achieving this: dynamically separating traffic according to predictability and dynamically controlling routes for each separated traffic part. In this paper, we propose a macroflow-based TE scheme that uses different routing policies in accordance with traffic predictability. We also propose a traffic-separation algorithm based on real-time traffic analysis and a framework for controlling separated traffic with software-defined networking technology, particularly OpenFlow. An evaluation of actual traffic measured in an Internet2 network shows that compared with current TE schemes the proposed scheme can reduce the maximum link load by 34% (at the most congested time) and the average link load by an average of 11%.

This paper proposes a novel method for two-dimensional (2-D) direction-of-arrival (DOA) estimation of multiple signals employing a sparse L-shaped array structured by a sparse linear array (SLA), a sparse uniform linear array (SULA) and an auxiliary sensor. In this method, the elevation angles are estimated by using the SLA and an efficient search approach, while the azimuth angle estimation is performed in two stages. In the first stage, the rough azimuth angle estimates are obtained by utilizing a noise-free cross-covariance matrix (CCM), the estimated elevation angles and data from three sensors including the auxiliary sensor. In the second stage, the fine azimuth angle estimates can be achieved by using the shift-invariance property of the SULA and the rough azimuth angle estimates. Without extra pair-matching process, the proposed method can achieve automatic pairing of the 2-D DOA estimates. Simulation results show that our approach outperforms the compared methods, especially in the cases of low SNR, snapshot deficiency and multiple sources.

Concurrent transmission (CT) is a revolutionary multi-hop protocol that significantly improves the MAC- and network-layer efficiency by allowing synchronized packet collisions. Although its superiority has been empirically verified, there is still a lack of studies on how the receiver survives such packet collisions, particularly in the presence of the carrier frequency offsets (CFO) between the transmitters. This work rectifies this omission by providing a comprehensive evaluation of the physical-layer receiver performance under CT, and a theoretical analysis on the fading duration of the beating effect resulting from the CFO. The main findings from our evaluations are the following points. (1) Beating significantly affects the receiver performance, and an error correcting mechanism is needed to combat the beating. (2) In IEEE 802.15.4 systems, the direct sequence spread spectrum (DSSS) plays such a role in combatting the beating. (3) However, due to the limited length of DSSS, the receiver still suffers from the beating if the fading duration is too long. (4) On the other hand, the basic M-ary FSK mode of IEEE 802.15.4g is vulnerable to CT due to the lack of error correcting mechanism. In view of the importance of the fading duration, we further theoretically derive the closed form of the average fading duration (AFD) of the beating under CT in terms of the transmitter number and the standard deviation of the CFO. Moreover, we prove that the receiver performance can be improved by having higher CFO deviations between the transmitters due to the shorter AFD. Finally, we estimate the AFD in the real system by actually measuring the CFO of a large number of sensor nodes.

We propose a framework to detect lug position and orientation in robotics that is insensitive to the lug orientation, incorporating a proposed optimization based on the artificial bee colony genetic algorithm. Experimental results show that the proposed optimization method outperformed traditional artificial bee colony and other meta-heuristics in the considered cases and was up to 3 times faster than the traditional approach. The proposed detection framework provided excellent performance to detect lug objects for all test cases.

A ripple-free dual-rate control system is designed for a single-input single-output dual-rate system, in which the sampling interval of a plant output is longer than the holding interval of a control input. The dual-rate system is converged to a multi-input single-output single-rate system using the lifting technique, and a control system is designed based on an error system using the steady-state variable. Because the proposed control law is designed so that the control input is constant in the steady state, the intersample output as well as the sampled output converges to the set-point without both steady-state error and intersample ripples when there is neither modeling nor disturbance. Furthermore, in the proposed method, a two-degree-of-freedom integral compensation is designed, and hence, the transient response is not deteriorated by the integral action because the integral action is canceled when there is neither modeling nor disturbance. Moreover, in the presence of the modeling error or disturbance, the integral compensation is revealed, and hence, the steady-state error is eliminated on both the intersample and sampled response.

For a pre-trained deep convolutional neural network (CNN) for an embedded system, a high-speed and a low power consumption are required. In the former of the CNN, it consists of convolutional layers, while in the latter, it consists of fully connection layers. In the convolutional layer, the multiply accumulation operation is a bottleneck, while the fully connection layer, the memory access is a bottleneck. The binarized CNN has been proposed to realize many multiply accumulation circuit on the FPGA, thus, the convolutional layer can be done with a high-seed operation. However, even if we apply the binarization to the fully connection layer, the amount of memory was still a bottleneck. In this paper, we propose a neuron pruning technique which eliminates almost part of the weight memory, and we apply it to the fully connection layer on the binarized CNN. In that case, since the weight memory is realized by an on-chip memory on the FPGA, it achieves a high-speed memory access. To further reduce the memory size, we apply the retraining the CNN after neuron pruning. In this paper, we propose a sequential-input parallel-output fully connection layer circuit for the binarized fully connection layer, while proposing a streaming circuit for the binarized 2D convolutional layer. The experimental results showed that, by the neuron pruning, as for the fully connected layer on the VGG-11 CNN, the number of neurons was reduced by 39.8% with keeping the 99% baseline accuracy. We implemented the neuron pruning CNN on the Xilinx Inc. Zynq Zedboard. Compared with the ARM Cortex-A57, it was 1773.0 times faster, it dissipated 3.1 times lower power, and its performance per power efficiency was 5781.3 times better. Also, compared with the Maxwell GPU, it was 11.1 times faster, it dissipated 7.7 times lower power, and its performance per power efficiency was 84.1 times better. Thus, the binarized CNN on the FPGA is suitable for the embedded system.

Many text entry methods are available in the use of touch interface devices when using a screen reader, and blind smartphone users and their supporters are eager to know which one is the easiest to learn and the fastest. Thus, we compared the text entry speeds and error counts for four combinations of software keyboards and character-selecting gestures over a period of five days. The split-tap gesture on the Japanese numeric keypad was found to be the fastest across the five days even though this text entry method produced the most errors. The two entry methods on the QWERTY keyboard were slower than the two entry methods on the numeric keypad. This difference in text entry speed was explained by the differences in key pointing and tapping times and their repitition numbers among different methods.

A compact uniplanar antenna design for tablet/laptop applications is proposed. The main design principle of this antenna is the use of the coupling-feed mechanism. The proposed antenna is composed of an inverted L-shaped parasitic element, T-shaped feeding strip, parasitic shorted strip, and a step tuning stub. With its small size of 55mm × 15mm × 0.8mm, the proposed antenna is able to excite dual wideband transmission over the full LTE/WWAN operation ranges of 698-960MHz and 1710-2690MHz. Furthermore, the proposed antenna also exhibits reduced ground effects, such that reducing the ground size of the proposed antenna will not affect its performance.

Precise determination of antenna phase centers is crucial to reduce the uncertainty in gain when employing the three-antenna method, particularly when the range distances are short-such as a 3-m radio anechoic chamber, where the distance between the phase centers and the open ends of an aperture antenna (the most commonly-used reference) is not negligible compared with the propagation distance. An automatic system to determine the phase centers of aperture antennas in a radio anechoic chamber is developed. In addition, the absolute gain of horn antennas is evaluated using the three-antenna method. The phase centers of X-band pyramidal horns were found to migrate up to 18mm from the open end. Uncertainties in the gain were evaluated in accordance with ISO/IEC Guide 93-3: 2008. The 95% confidence interval of the horn antenna gain was reduced from 0.57 to 0.25dB, when using the phase center location instead of the open end. The phase centers, gains, polarization, and radiation patterns of space-borne antennas are measured: low and medium-gain X-band antennas for an ultra small deep space probe employing the polarization pattern method with use of the horn antenna. The 95% confidence interval in the antenna gain decreased from 0.74 to 0.47dB.

Two different types of representations, such as an image and its manually-assigned corresponding labels, generally have complex and strong relationships to each other. In this paper, we represent such deep relationships between two different types of visible variables using an energy-based probabilistic model, called a deep relational model (DRM) to improve the prediction accuracies. A DRM stacks several layers from one visible layer on to another visible layer, sandwiching several hidden layers between them. As with restricted Boltzmann machines (RBMs) and deep Boltzmann machines (DBMs), all connections (weights) between two adjacent layers are undirected. During maximum likelihood (ML) -based training, the network attempts to capture the latent complex relationships between two visible variables with its deep architecture. Unlike deep neural networks (DNNs), 1) the DRM is a totally generative model and 2) allows us to generate one visible variables given the other, and 2) the parameters can be optimized in a probabilistic manner. The DRM can be also fine-tuned using DNNs, like deep belief nets (DBNs) or DBMs pre-training. This paper presents experiments conduced to evaluate the performance of a DRM in image recognition and generation tasks using the MNIST data set. In the image recognition experiments, we observed that the DRM outperformed DNNs even without fine-tuning. In the image generation experiments, we obtained much more realistic images generated from the DRM more than those from the other generative models.