The difference between training and testing environments is the major reason of performance degradation of speech recognition. In this paper, to further decrease the mismatch, we apply temporal filtering, Auto-Regression and Moving-Average (ARMA) filtering or RelAtive SpecTrAl (RASTA) filtering, as a post-processor for the log-Energy dynamic Range Normalization-Cepstral Mean and Variance Normalization (ERN-CMVN) based speech features, referred to as [EC]-ARMA and [EC]-RASTA. From experimental results conducted on Aurora 2.0 database, the integrated approaches with temporal filtering are shown the best performance among the several integrated approaches.
A novel masker-less based method to track a 2D articulated body under self-occlusion and ambiguity in monocular image sequences is proposed. The proposed method applies SMC to both color and motion features. We employ a self-updated binary occlusion mask to increase accuracy in tracking. To alleviate the effect of illumination, each color body part is formed by a Gaussian mixture model in HS space, and the distribution intersection is used in distance measures of two probability distributions. Moreover, a motion cue is used to prune spurious solutions. Our technique can track the target reliably, especially in occlusion and ambiguous cases.
Due to energy resources constraint in wireless sensor networks, balancing energy consumption is very important in increasing the network life time. In this paper, the new hierarchical and dynamic routing algorithm is proposed to balance energy consumption among the nodes and to prevent from energy holing problem. Simulation results show that the proposed algorithm prolongs the network lifetime about 40% compared to the LEACH protocol.
In this paper a universal fuzzy flip-flop is proposed that can be reconfigured as a fuzzy SR, D, JK, or T flip-flop. When integrated with a multi layer neural network, the resulting reconfigurable fuzzy-neural structure showed excellent learning ability. The sigmoid activation function of neurons in the hidden layers of the multilayer neural network was replaced by the quasi-sigmoidal transfer characteristics of the universal fuzzy flip-flop in the reconfigurable fuzzy-neural structure. Experimental results showed that the reconfigurable fuzzy-neural structure can be effectively trained using either a large or sparse set of data points to closely approximate nonlinear input functions.
This paper proposes a novel concept of Local AutoCorrelation (LAC), which is converted to LAC images for robust face detection under varying illumination conditions. We have applied LAC images to face detection based on AdaBoost algorithm. MIT CBCL images are used for training, while CMU PIE face database including a variety of illumination directions is used for detection. As a result, the LAC is found to be so effective that face detection rate is improved by up to 4.7%.
This work presents a low complexity irregular topology generation algorithm for Netwok-on-Chip (NoC) design flow. The objective of the algorithm is minimizing the energy consumption of the final design. We tested the effectiveness of our algorithm by comparing it with its earlier counterparts on several multimedia applications.
This paper presents an improved side-channel attack based on multivariate regression analysis. The proposed method effectively refines measured side-channel traces using profiled information. Our method is advantageous to conventional profiling attacks as it is robust against attack configurations; it requires less traces while profiling and is less sensitive to the number of interesting points. We demonstrate the effectiveness of our method through concrete experiments in comparison with conventional methods. In the experiments, the proposed method is examined under various configurations to investigate its performance.
The design and physical implementation of a low-power SRAM with 4T CMOS latch bit-cell is presented. The memory cells in this work are composed of two cross-coupled inverters without any access transistors. They are accessed by totally novel read and write methods that result in low operating power dissipation in the nature. A 1.8V SRAM test chip has been fabricated in a 0.18µm CMOS technology, which demonstrated the functionality of the memory cell. This new SRAM operates with 30% reduction in read power and 42% reduction in write power compared to the standard 6T SRAM.
In this paper, we propose a low-complexity and low-latency method of FFT based channel estimation for Mobile OFDM systems. Mobile WiMAX system and LTE are next generation communication systems which adopt OFDM systems to achieve high mobility and a high data rate. Especially for high mobility communication systems, efficient channel estimation, technology should be developed. The proposed channel estimation calculates threshold value of channel impulse response (CIR) using discrete wavelet transform (DWT) in a frequency domain. In the experimental results, the proposed method increased performance with low-complexity and low-latency. Implementaion and simulation results show that the proposed method reduced the latency about 43% compared with conventional DFT based channel estimators.
We have proposed a novel circuit design with the phase correction waveguides in part of the arrayed waveguides to improve the property of an AWG with a narrow spacing and a large channel count without the additional processes such as a UV trimming method. Employing this novel circuit design, we have successfully demonstrated a flat-top 50GHz-88ch with a low crosstalk experimentally.
We present a 60Gbit/s polarization-multiplexed coherent optical OFDM transmission at 5Gsymbol/s by employing 64 QAM subcarrier modulation. We adopted a frequency-stabilized fiber laser and an optical PLL to achieve this high multiplicity in the subcarrier modulation. As a result, we successfully transmitted 60Gbit/s data over 160km with a demodulation bandwidth of only 5.3GHz.
The dynamic backscattering analysis of impedance loaded wire antennas for passive RFID applications is presented where the two loading impedances represent the corresponding logic states in the transmitting data of RFID tags. The analysis is based on the equivalent circuit of receiving antennas from moment method and the time-varying load calculation from the conversion matrix technique in the frequency domain. Simulation studies on 915MHz with resistive or capacitive loads are shown where the reactive load indicates more apparent discrimination for the two logic states, thereby improving the receiving sensitivity of RFID readers. The analysis methods and results can provide useful information for antenna and IC designs of RFID tags.