New media processing applications such as image recognition and AR (Augment Reality) have become into practical on embedded systems for automotive, digital-consumer and mobile products. Many-core processors have been proposed to realize much higher performance than multi-core processors. We have developed a low-power many-core SoC for multimedia applications in 40nm CMOS technology. Within a 210mm2 die, two 32-core clusters are integrated with dynamically reconfigurable processors, hardware accelerators, 2-channel DDR3 I/Fs, and other peripherals. Processor cores in the cluster share a 2MB L2 cache connected through a tree-based Network-on-Chip (NoC). Its total peak performance exceeds 1.5TOPS (Tera Operations Per Second). The high scalability and low power consumption are accomplished by parallelized software for multimedia applications. In case of face detection, the performance scales up to 64 cores and the SoC consumes only 2.21W. Moreover, it can execute the 1080p 48fps H.264 decoding about 520mW by 28 cores and the 4K2K 15fps super resolution about 770mW by 32 cores in one cluster. Exploiting parallelism by low power processor cores, the many-core SoC provides several tens of times better energy efficiency than that of a high performance desk-top quad-core processor.

As data volumes explode, data storage costs become a large fraction of total IT costs. We can reduce the costs substantially by using compression. However, it is generally known that database compression is not suitable for write-intensive workloads. In this paper, we provide a comprehensive solution to improve the performance of compressed databases for write-intensive OLTP workloads. We find that storing data too densely in compressed pages incurs many future page splits, which require exclusive locks. In order to avoid lock contention, we reduce page splits by sacrificing a couple of percent of space savings. We reserve enough space in each compressed page for future updates of records and prevent page merges that are prone to incur page splits in the near future. The experimental results using TPC-C benchmark and MySQL/InnoDB show that our method gives 1.5 times higher throughput with 33% space savings compared with the uncompressed counterpart and 1.8 times higher throughput with only 1% more space compared with the state-of-the-art compression method developed by Facebook.

A new digital controller for a single-phase boost power factor correction (PFC) converter operating at a discontinuous conduction mode (DCM), is presented to achieve high input power factor over wide input voltage and load range. A method of duty cycle modulation is proposed to reduce the line harmonic distortion and improve the power factor. The loop regulation scheme is adopted to further improve the system stability and the power factor simultaneously. Meanwhile, a novel digital pulse width modulator (DPWM) based on the delay lock loop technique, is realized to improve the regulation linearity of duty cycle and reduce the regulation deviation. The single-phase DCM boost PFC converter with the proposed digital controller based on the field programmable gate array (FPGA) has been implemented. Experimental results indicate that the proposed digital controller can achieve high power factor more than 0.99 over wide input voltage and load range, the output voltage deviation is less than 3V, and the peak conversion efficiency is 96.2% in the case of a full load.

In this paper, we apply a mutation operation based on a multivariate Cauchy distribution to fast evolutionary programming and analyze its effect in terms of various function optimizations. The conventional fast evolutionary programming in-cooperates the univariate Cauchy mutation in order to overcome the slow convergence rate of the canonical Gaussian mutation. For a mutation of n variables, while the conventional method utilizes n independent random variables from a univariate Cauchy distribution, the proposed method adopts n mutually dependent random variables that satisfy a multivariate Cauchy distribution. The multivariate Cauchy distribution naturally has higher probabilities of generating random variables in inter-variable regions than the univariate Cauchy distribution due to the mutual dependence among variables. This implies that the multivariate Cauchy random variable enhances the search capability especially for a large number of correlated variables, and, as a result, is more appropriate for optimization schemes characterized by interdependence among variables. In this sense, the proposed mutation possesses the advantage of both the univariate Cauchy and Gaussian mutations. The proposed mutation is tested against various types of real-valued function optimizations. We empirically find that the proposed mutation outperformed the conventional Cauchy and Gaussian mutations in the optimization of functions having correlations among variables, whereas the conventional mutations showed better performance in functions of uncorrelated variables.

The rapid variation of wireless channels and feedback delay make the available channel state information (CSI) outdated in dynamic wireless multi-hop networks, which significantly degrades the accuracy of cross-layer resource allocation. To deal with this problem, a cross-layer resource allocation scheme is proposed for wireless multi-hop networks by taking the outdated CSI into account and basing compensation on the results of channel prediction. The cross-layer resource allocation is formulated as a network utility maximization problem, which jointly considers congestion control, channel allocation, power control, scheduling and routing with the compensated CSI. Based on a dual decomposition approach, the problem is solved in a distributed manner. Simulation results show that the proposed algorithm can reasonably allocate the resources, and significantly improve the throughput and energy efficiency in the network.

A new type of the affine projection (AP) algorithms which incorporates the sparsity condition of a system is presented. To exploit the sparsity of the system, a weighted l1-norm regularization is imposed on the cost function of the AP algorithm. Minimizing the cost function with a subgradient calculus and choosing two distinct weightings for l1-norm, two stochastic gradient based sparsity regularized AP (SR-AP) algorithms are developed. Experimental results show that the SR-AP algorithms outperform the typical AP counterparts for identifying sparse systems.

Analysis of the trust network proves beneficial to the users in Online Social Networks (OSNs) for decision-making. Since the construction of trust propagation paths connecting unfamiliar users is the preceding work of trust inference, it is vital to find appropriate trust propagation paths. Most of existing trust network discovery algorithms apply the classical exhausted searching approaches with low efficiency and/or just take into account the factors relating to trust without regard to the role of distrust relationships. To solve the issues, we first analyze the trust discounting operators with structure balance theory and validate the distribution characteristics of balanced transitive triads. Then, Maximum Indirect Referral Belief Search (MIRBS) and Minimum Indirect Functional Uncertainty Search (MIFUS) strategies are proposed and followed by the Optimal Trust Inference Path Search (OTIPS) algorithms accordingly on the basis of the bidirectional versions of Dijkstra's algorithm. The comparative experiments of path search, trust inference and edge sign prediction are performed on the Epinions data set. The experimental results show that the proposed algorithm can find the trust inference path with better efficiency and the found paths have better applicability to trust inference.

A joint continuous phase frequency shift keying (CPFSK) modulation and physical-layer network coding (PNC), i.e., CPFSK-PNC, is proposed for two-way relay channels (TWRCs). This letter discusses the signal detection of the CPFSK-PNC scheme with emphasis on the maximum-likelihood sequence detection (MLSD) algorithm for the relay receiver. The end-to-end error performance of the proposed CPFSK-PNC scheme is evaluated through simulations.

To achieve high quality output speech synthesis systems, data-driven grapheme-to-phoneme (G2P) conversion is usually used to generate the phonetic transcription of out-of-vocabulary (OOV) words. To improve the performance of G2P conversion, this paper deals with the problem of conflicting phonemes, where an input grapheme can, in the same context, produce many possible output phonemes at the same time. To this end, we propose a two-stage neural network-based approach that converts the input text to phoneme sequences in the first stage and then predicts each output phoneme in the second stage using the phonemic information obtained. The first-stage neural network is fundamentally implemented as a many-to-many mapping model for automatic conversion of word to phoneme sequences, while the second stage uses a combination of the obtained phoneme sequences to predict the output phoneme corresponding to each input grapheme in a given word. We evaluate the performance of this approach using the American English words-based pronunciation dictionary known as the auto-aligned CMUDict corpus[1]. In terms of phoneme and word accuracy of the OOV words, on comparison with several proposed baseline approaches, the evaluation results show that our proposed approach improves on the previous one-stage neural network-based approach for G2P conversion. The results of comparison with another existing approach indicate that it provides higher phoneme accuracy but lower word accuracy on a general dataset, and slightly higher phoneme and word accuracy on a selection of words consisting of more than one phoneme conflicts.

Compressive sensing is a promising technique in data acquisition field. A central problem in compressive sensing is that for a given sparse signal, we wish to recover it accurately, efficiently and stably from very few measurements. Inspired by mathematical analysis, we introduce a combining scheme between stability and robustness in reconstruction problems using compressive sensing. By choosing appropriate parameters, we are able to construct a condition for reconstruction map to perform properly.

Recently, we proposed a new data-path architecture, named a large-scale reconfigurable data-path (LSRDP), based on single-flux-quantum (SFQ) circuits, to establish a fundamental technology for future high-end computers. In this architecture, a large number of SFQ floating-point units (FPUs) are used as core components, and their high performance and low power consumption are essential. In this research, we implemented an SFQ half-precision bit-serial floating-point multiplier (FPM) with a target clock frequency of 50GHz, using the AIST 10kA/cm2 Nb process. The FPM was designed, based on a systolic-array architecture. It contains 11,066 Josephson junctions, including on-chip high-speed test circuits. The size and power consumption of the FPM are 6.66mm × 1.92mm and 2.83mW, respectively. Its correct operation was confirmed at a maximum frequency of 93.4GHz for the exponent part and of 72.0GHz for the significand part by on-chip high-speed tests.

In this paper, a new evolutionary approach to recommender systems is presented. The aim of this work is to develop a new recommendation method that effectively adapts and immediately responds to the user's preference. To this end, content-based filtering is judiciously utilized in conjunction with interactive evolutionary computation (IEC). Specifically, a fitness-based truncation selection and a feature-wise crossover are devised to make full use of desirable properties of promising items within the IEC framework. Moreover, to efficiently search for proper items, the content-based filtering is modified in cooperation with data grouping. The experimental results demonstrate the effectiveness of the proposed approach, compared with existing methods.

This letter deals with angular position classification using the synthesized active sonar returns from targets. For the synthesis of active sonar returns, we synthesized active sonar returns based on ray tracing algorithm for 3D highlight models. Then, a fractional Fourier transform (FrFT) was applied to the sonar returns to extract the angular position information depending on the target aspect by utilizing separation capability of the time-delayed combination of linear frequency modulated (LFM) signals in the FrFT domain. With the FrFT-based features, three different target angular positions were classified using neural networks.

This paper proposes a vanishing point-based road detection method. Firstly, a vanishing point is detected using a texture-based method proposed in a recent study. After that, a histogram is generated for detecting two road borders. The road area is defined as the region between the two road borders and below the vanishing point. The experimental results demonstrate that our method performs well in general road images.

In this letter, we propose a simple but effective technique that improves statistical model-based voice activity detection (VAD) by both reducing computational complexity and increasing detection accuracy. The improvements are made by applying Taylor series approximations to the exponential and logarithmic functions in the VAD algorithm based on an in-depth analysis of the algorithm. Experiments performed on a smartphone as well as on a desktop computer with various background noises confirm the effectiveness of the proposed technique.

Broadband wireless channels are frequency selective in nature. In this paper, a novel precoder with finite impulse response (FIR) structure is proposed to maximize the throughput of the multiple-input multiple-output (MIMO) frequency-selective multicast channel. An iteration mechanism is investigated to obtain the desired FIR precoding matrix. In the iterative process, two associated parameters, namely the innovation orientation and the iteration step size, are jointly derived by the convex optimization program and the traditional Gauss-Newton algorithm. Convergence and complexity analyses are presented, and the numerical simulations indicate that the proposed method outperforms the existing schemes in the moderate to high signal to noise ratio (SNR) regime.

Detection of human respiration and heartbeat is an essential demand in medical monitoring, healthcare vigilance, as well as in rescue activities after earthquakes. Radar is an important tool to detect human respiration and heartbeat. Compared to body-attached sensors, radar has the advantage of conducting detection without contacting the subject, which is favorable in practical usage. In this paper, we conduct fundamental studies on ultra-wideband (UWB) radar for detection of the respiration and heartbeat by computer simulations. The main achievement of our work is the development of a UWB radar simulation system. Using the developed simulation system, three UWB frequency bands, i.e., 3.4-4.8GHz, 7.25-10.25GHz, as well as 3.1-10.6GHz, are compared in terms of their respiration and heartbeat detection performance. Our results show that the first two bands present identical performance, while the third one presents much better performance. The effects of using multiple antennas are also evaluated. Our results show that increasing the number of antennas can steadily increase the detection ability.

One of the most important requirements for a routing protocol in wireless body area networks (WBANs) is to lower the network's temperature increase. The temperature of a node is closely related to its activities. The proactive routing approach, which is used by existing routing protocols for WBANs, tends to produce a higher temperature increase due to more frequent activities, compared to the on-demand reactive routing approach. In this paper, therefore, we propose a reactive routing protocol for WBANs called priority-based temperature-aware routing (PTR). In addition to lowering the temperature increase, the protocol also recognizes vital nodes and prioritizes them so they are able to achieve higher throughput. Simulation results show that the PTR protocol achieves a 50% lower temperature increase compared to the conventional temperature-aware routing protocol and is able to improve throughput of vital nodes by 35% when the priority mode is enabled.

In this paper, we present an average-case efficient algorithm to resolve the problem of determining whether two Boolean functions in trace representation are identical. Firstly, we introduce a necessary and sufficient condition for null Boolean functions in trace representation, which can be viewed as a generalization of the well-known additive Hilbert-90 theorem. Based on this condition, we propose an algorithmic method with preprocessing to address the original problem. The worst-case complexity of the algorithm is still exponential; its average-case performance, however, can be improved. We prove that the expected complexity of the refined procedure is O(n), if the coefficients of input functions are chosen i.i.d. according to the uniform distribution over F2n; therefore, it performs well in practice.

This paper is concerned with a method to reduce the computation time of the Discrete Ray Tracing Method (DRTM) which was proposed to numerically analyze electromagnetic fields above Random Rough Surfaces (RRSs). The essence of DRTM is firstly to search rays between source and receiver and secondly to compute electric fields based on the traced rays. In the DRTM, the method discretizes not only RRSs but also ray tracing procedure. In order to reduce computation time for ray searching, the authors propose to modify the conventional algorithm discretizing RRSs with equal intervals to a new one which discretizes them with unequal intervals according to their profiles. The authors also use an approximation of Fresnel function which enables us to reduce field computation time. The authors discuss the reduction rate for computation time of the DRTM from the numerical view points of ray searching and field computation. Finally, this paper shows how much computation time is reduced by the new method.