We developed a method for extracting negative examples when only positive examples are given as supervised data. This method calculates the probability of occurrence of an input example, which should be judged to be positive or negative. It considers an input example that has a high probability of occurrence but does not appear in the set of positive examples as a negative example. We used this method for one of important tasks in natural language processing: automatic detection of misspelled Japanese expressions. The results showed that the method is effective. In this study, we also described two other methods we developed for the detection of misspelled expressions: a combined method and a "leaving-one-out" method. In our experiments, we found that these methods are also effective.

We propose a buffer management mechanism, called V-WFQ (Virtual Weighted Fair Queueing), for achieving an approximately fair allocation of bandwidth with a small amount of hardware in a high-speed network. The basic process for the allocation of bandwidth uses selective packet dropping that compares the measured input rate of the flow with an estimated fair share of bandwidth. Although V-WFQ is a hardware-efficient FIFO-based algorithm, it achieves almost ideal fairness in bandwidth allocation. V-WFQ can be implemented in the high-speed core routers of today's IP backbone networks to provide various high-quality services. We have investigated V-WFQ's performance in terms of fairness and link utilization through extensive simulation. The results of simulation show that V-WFQ achieves a good balance between fairness and link utilization under various simulation conditions.

In this paper, we present an efficient VLSI architecture for the stand-alone application of a speech recognition system based on discriminative Bayesian neural network (DBNN). Regarding the recognition phase, the architecture of the Bayesian distance unit (BDU) is constructed first. In association with the BDU, we propose a template-serial architecture for the path distance accumulation to perform the recognition procedure. A corresponding architecture is also developed to accelerate the discriminative training procedure. It contains the intelligent look-up table for the sigmoid function. In comparison to the traditional one-table method, the memory size reduces drastically with only slight loss of accuracy. Combining the proposed hardware accelerators with the cost efficient programmable core, we took the most out of both programmable and application-specific architectures, including performance, design complexity, and flexibility.

This paper presents a novel low-energy memory design technique based on variable analysis for on-chip data memory (RAM) in application-specific systems, which called VAbM technique. It targets the exploitation of both data locality and effective data width of variables to reduce energy consumed by data transfer and storage. Variables with higher access frequency and smaller effective data width are assigned into a smaller low-energy memory with fewer bit lines and word lines, placed closer the processor. Under constraints of the number of memory banks, VAbM technique use variable analysis results to perform allocating and assigning on-chip RAM into multiple banks, which have different size with different number of word lines and different number of bit lines tailored to each application requirements. Experimental results with several real embedded applications demonstrate significant energy reduction up to 64.8% over monolithic memory, and 27.7% compared to memory designed by memory banking technique.

An effective spatial resolution enhancement method for distributed strain measurement by BOTDR is proposed. An optical fiber is glued to a structure by a length less than the spatial resolution defined by the pulse width, and the Brillouin spectrum of the light scattered from the glued optical fiber is investigated theoretically. The apparent strain xp observed in the fiber is found to be proportional to the accurate strain a. The ratio r=xp/a coincides with the ratio of the glued length to the spatial resolution. Spatial resolution as small as 0.2 m is demonstrated experimentally for small strains of less than 10-3.

In this paper, we propose a novel higher order time-frequency distribution (GDH) for a discrete time signal. This distribution is defined over the original discrete time-frequency grids through a delicate discretization of an equivalent expression of a higher order distribution, for a continuous time signal, in [4]. We also present a constructive design method, for the kernel of the GDH, by which the distribution satisfies (i) the alias free condition as well as (ii) the marginal conditions. Numerical examples show that the proposed distributions reasonably suppress the artifacts which are observed severely in the Wigner distribution and its simple higher order generalization.

In this paper, we propose topic extraction models based on statistical relevance scores between topic words and words in articles, and report results obtained in topic extraction experiments using continuous speech recognition for Japanese broadcast news utterances. We attempt to represent a topic of news speech using a combination of multiple topic words, which are important words in the news article or words relevant to the news. We assume a topic of news is represented by a combination of words. We statistically model mapping from words in an article to topic words. Using the mapping, the topic extraction model can extract topic words even if they do not appear in the article. We train a topic extraction model capable of computing the degree of relevance between a topic word and a word in an article by using newspaper text covering a five-year period. The degree of relevance between those words is calculated based on measures such as mutual information or the χ2-method. In experiments extracting five topic words using a χ2-based model, we achieve 72% precision and 12% recall for speech recognition results. Speech recognition results generally include a number of recognition errors, which degrades topic extraction performance. To avoid this, we employ N-best candidates and likelihood given by acoustic and language models. In experiments, we find that extracting five topic words using N-best candidate and likelihood values achieves significantly improved precision.

A noise suppression algorithm with high speech quality based on weighted noise estimation and MMSE STSA is proposed. The proposed algorithm continuously updates the estimated noise by weighted noisy speech in accordance with an estimated SNR. The spectral gain is modified with the estimated SNR so that it can better utilize the improvement in noise estimation. With a better noise estimate, a more correct SNR is obtained resulting in the enhanced speech with low distortion. Subjective evaluation results show that five-grade mean opinion scores of the new algorithm with and without a speech codec are improved by as much as 0.35 and 0.40 respectively, compared with either the original MMSE STSA or the EVRC noise suppression algorithm.

This paper presents a comparison of the throughput performance employing hybrid automatic repeat request (HARQ) with packet combining, such as Type-I with packet combining (simply Chase combining hereafter) and Type-II (Incremental redundancy hereafter), using turbo coding in a multipath fading channel in high speed downlink packet access (HSDPA). We apply a multipath interference canceller (MPIC) to remove the influence of severe multipath interference. Link level simulation results show that the maximum throughput using Incremental redundancy with 64QAM is improved by approximately 5-8% compared to that using Chase combining, and that the required average received signal energy of 12 code channels per chip-to-background noise spectrum density (Ec/N0) at the throughput of 4 Mbps with Incremental redundancy is decreased by approximately 1.0 dB rather than that with Chase combining when the vehicular speed is higher than approximately 30 km/h. Furthermore, we elucidate based on the system level simulation that although no improvement is obtained in a slow mobility environment such as the average vehicular speed of 3 km/h, the achieved throughput of Incremental redundancy is increased by approximately 5-6% and 13% for the average vehicular speed of 30 km/h and 120 km/h, respectively, compared to that with Chase combining.

We consider bandwidth-efficient modulation schemes for use on non-linear channels, such as that due to a non-linear high power amplifier (HPA) in a wireless system. Continuous Phase Modulation (CPM) schemes are known to perform well on such channels, because they have constant amplitude, but their bandwidth efficiency is low. N-MSK improves this by superposing two or more such signals, but this results in a non-constant amplitude. In this paper we investigate the performance of N-MSK on a non-linear channel, modelled using a travelling wave tube (TWT) non-linearity. We first consider the spectrum and the BER performance of N-MSK on a non-linear channel, making use of the Euclidean distance spectrum of the modulated signal. We then consider the effect of non-linear amplification on these properties. Signal spectrum was determined by simulation, since no closed-form expression is available when the effect of AM-PM conversion are included. We find that the spectrum is remarkably little affected, being only slightly broadened. BER is also evaluated by direct simulation, as well as from the Euclidean distance spectrum. The latter now exhibits a series of clusters, instead of discrete lines, and we find that at least the whole of the first cluster must be considered in calculating the BER, not just the minimum distance or the centroid of the cluster. The detector used in the simulation applies an inverse distortion function, then uses maximum-likelihood sequence estimation (MLSE) set up for the linear channel. This is no longer optimum, because the noise is distorted, and therefore it is also compared with a true MLSE detector. We find that the BER performance is, however, somewhat degraded compared to the linear channel. We determine back-off levels from saturation to optimise overall power efficiency.

Experiments were conducted to examine an approach from language modeling side to improving noisy speech recognition performance. By adopting appropriate word strings as new units of processing, speech recognition performance was improved by acoustic effects as well as by test-set perplexity reduction. Three kinds of word string language models were evaluated, whose additional lexical entries were selected based on combinations of part of speech information, word length, occurrence frequency, and log likelihood ratio of the hypotheses about the bigram frequency. All of the three word string models reduced errors in broadcast news speech recognition, and also lowered test-set perplexity. The word string model based on log likelihood ratio exhibited the best improvement for noisy speech recognition, by which deletion errors were reduced by 26%, substitution errors by 9.3%, and insertion errors by 13%, in the experiments using the speaker-dependent, noise-adapted triphone. Effectiveness of word string models on error reduction was more prominent for noisy speech than for studio-clean speech.

In this paper, we propose a simplified transform-based and variable-rate vocoder, which is evolved from the code-excited linear prediction (CELP) coding structure. With pre-emphasis and de-emphasis filters, the transformed-based CELP vocoder incorporates a long-term predictor, a discrete cosine transform (DCT), and pre-filters and postfilters for achieving perceptually weighted quantization. The proposed transform-based vocoder requires less computational complexity with slightly worse quality than the CELP coders. Furthermore, the proposed DCT-based coding structure easily figured with additional DCT coefficients could simultaneously offer low, middle, and high bit rates to adapt the variation of bandwidth for modern Internet or wireless communications.

Traditional watermarking methods based on spectrum manipulation use watermark bit as a colored noise. However, those methods show poor robustness for filtering and audio compression because of only a few spectrum coefficients being modified. In this paper, we propose a new algorithm for robust audio watermarking by using spectrum warping. Our watermark-embedding scheme is not based on spectrum coefficient manipulation, but based on warping spectrum that shifts a block of spectrum coefficients nonlinearly. The host audio signal is divided appropriately into small segments and then a non-orthogonal transform, such as bi-linear transform, is used for each segment to embed as well as to extract watermarks. Because the proposed scheme uses small segment length, we can afford to embed signature information many times into audio signal without severe degradation of SNR. Our multiple embedding scheme increases robustness against abnormal distortion of watermarked signal and has no bit error rate in MPEG1 Layer-3 128 kbps and 96 kbps compression as well as low/high pass filtering.

The problem of finding a path with two additive constraints, in particular finding a path that satisfies both the cost and the delay constraints, is called multi-constrained path (MCP) problem in the literature. In this paper, we explore the MCP problem based on the idea of single mixed weight --a mixed weight for each link is first obtained by combining its delay and cost, and then Dijkstra's algorithm is used to find the corresponding shortest path. Given two infeasible paths, it can be theoretically proved that a better path can possibly be found if we choose an appropriate parameter to construct the mixed weight. An approximate algorithm is thus proposed to solve the MCP problem. Theoretical analysis demonstrates that this algorithm can make a correct judgment whether there is a feasible path or not with a very high probability even in the strict case where the delay bound is between the delays of the least delay path and the least cost path, while the cost bound is between the costs of the two paths. On the other hand, the time complexity of this algorithm is very small since it only needs to execute Dijkstra's algorithm a limited number of times. The excellent performance of the proposed algorithm is verified by a large number of experiments on networks of different sizes.

The prospects of electron spectroscopy of working organic electronic device structures are discussed. The experimental consideration and the result of actual measurement are presented.

In this paper, we present a new approach to the construction of a set of binary sequences with a zero-correlation zone. The set consists of n pairs of binary sequences and the length of each binary sequence is n2(m+2) for some integers m and n. The Hadamard sequences with length n are used to construct the set. Any sequence in the set has 2(m+1) subsequences, each of length 2n. The author proves that any two subsequences are orthogonal if they belong to different pairs of binary sequences in the set.

Wyner and Ziv considered the rate-distortion function for source coding with side information at the decoder (we call the Wyner-Ziv problem). In this paper we show an information-spectrum approach to the Wyner-Ziv problem for general class of nonstationary and/or nonergodic sources with side information at the decoder, where the distortion measure is arbitrary and may be nonadditive. We show that a general formula for the rate-distortion function of the Wyner-Ziv problem for general sources with the maximum distortion criterion under fixed-length coding by using the information spectrum approach.

The adsorption behavior of cytochrome c was investigated using slab optical waveguide (SOWG) absorption spectroscopy at the near ultraviolet region utilizing thin quartz plates as planar waveguides. SOWG absorption spectra of cytochrome c measured at constant time intervals showed significant influence of surface hydrophilicity and solution chemistry on the adsorption of this important heme protein in quartz surface. Being polar and typically amphoteric, the protein preferred adsorption on hydrophilic surface than on hydrophobic surface as implied by the lower absorbance data obtained in the latter than in the former. At lower ionic strength and in the absence of buffer, the protein molecules tend to adsorb on the quartz surface. Plots of near steady-state absorbance versus protein concentration follow hyperbolic pattern in the absence of buffer or at low ionic strength and become more linear as the buffer concentration is increased. The results presented here are explained in terms of the general qualitative understanding of protein adsorption at solid-aqueous interfaces and further aids in elucidating the properties of protein monolayers and films.

A novel power spectral density accumulation (PSDA) method for estimating the bandwidth of the clutter spectra is proposed, based on a priori knowledge of the shape of the clutter spectra. The comparison of the complexity and the performance between the PSDA method and the general ones is presented. It is shown that the PSDA method is effective for the short-time clutter data in the practical application.

This paper addresses a classic question about whether transmit power control (TPC) can increase the spectrum efficiency of a TDMA system and an FDMA cellular system as in the case of a DS-CDMA cellular system. Two types of TPC schemes are considered; one is slow TPC that regulates the distance dependent path loss and shadowing loss, while the other is fast TPC that regulates multipath fading as well as path loss and shadowing loss. In addition to TPC, antenna diversity reception is considered. The allowable interference rise factor χ, which is defined as the interference plus background noise-to-background noise power ratio, is introduced. The simple expressions for the signal-to-interference plus background noise power ratio (SINR) at the diversity combiner output using maximal-ratio combining (MRC) are derived to obtain the reuse distance by computer simulations. The impact of joint use of TPC and antenna diversity reception on the spectrum efficiency is discussed. It is found that the joint use of fast TPC and antenna diversity is advantageous and larger spectrum efficiency can be achieved than with no TPC. On the other hand, the use of slow TPC is found advantageous only for small values of standard deviation of shadowing loss; however, the improvement in the spectrum efficiency is quite small.