We propose a domain specific model for statistical machine translation. It is well-known that domain specific language models perform well in automatic speech recognition. We show that domain specific language and translation models also benefit statistical machine translation. However, there are two problems with using domain specific models. The first is the data sparseness problem. We employ an adaptation technique to overcome this problem. The second issue is domain prediction. In order to perform adaptation, the domain must be provided, however in many cases, the domain is not known or changes dynamically. For these cases, not only the translation target sentence but also the domain must be predicted. This paper focuses on the domain prediction problem for statistical machine translation. In the proposed method, a bilingual training corpus, is automatically clustered into sub-corpora. Each sub-corpus is deemed to be a domain. The domain of a source sentence is predicted by using its similarity to the sub-corpora. The predicted domain (sub-corpus) specific language and translation models are then used for the translation decoding. This approach gave an improvement of 2.7 in BLEU score on the IWSLT05 Japanese to English evaluation corpus (improving the score from 52.4 to 55.1). This is a substantial gain and indicates the validity of the proposed bilingual cluster based models.

It is well known that cooperative transmission among the single antenna wireless nodes and a proper combining at destination can obtain spatial diversity. In this paper, we introduce a new form of combining technique in cooperative communication. For a coded transmission scheme code-combining can obtain a near optimal low rate code by combining repeated codewords. Instead of MRC (maximal ratio combining) based combining of received coded packets from source and relays, we propose a simple code-combining at destination. For same data rate and power consumption code-combining offers better or similar performance with less complexity than MRC. Moreover using a puncturing technique at the relay we can get a same diversity order as MRC with reduced packet relaying time; equivalently, with higher data rate for over all system. This reduction of transmission time at relay allows us to increase the diversity order by using more than one relay for one source; where each relay forwards a punctured portion of received data. Alternatively, when the relays are not available to improve diversity order, we can use only one relay to cooperate M source nodes where all sources obtain a diversity order of 2 with a higher data rate.

In this paper, we propose a recursive filtering algorithm to restore monochromatic images which are corrupted by general dependent additive noise. It is assumed that the equation which describes the image field is not available and a filtering algorithm is obtained using the information provided by the covariance functions of the signal, noise that affects the measurement equation, and the fourth-order moments of the signal. The proposed algorithm is obtained by an innovation approach which provides a simple derivation of the least mean-squared error linear estimators. The estimation of the grey level in each spatial coordinate is made taking into account the information provided by the grey levels located on the row of the pixel to be estimated. The proposed filtering algorithm is applied to restore images which are affected by general signal-dependent additive noise.

In recent decades, there has been a great deal of research into the problem of bilingual speech recognition - to develop a recognizer that can handle inter- and intra-sentential language switching between two languages. This paper presents our recent work on the development of a grammar-constrained, Mandarin-English bilingual Speech Recognition System (MESRS) for real world music retrieval. Two of the main difficult issues in handling the bilingual speech recognition systems for real world applications are tackled in this paper. One is to balance the performance and the complexity of the bilingual speech recognition system; the other is to effectively deal with the matrix language accents in embedded language. In order to process the intra-sentential language switching and reduce the amount of data required to robustly estimate statistical models, a compact single set of bilingual acoustic models derived by phone set merging and clustering is developed instead of using two separate monolingual models for each language. In our study, a novel Two-pass phone clustering method based on Confusion Matrix (TCM) is presented and compared with the log-likelihood measure method. Experiments testify that TCM can achieve better performance. Since potential system users' native language is Mandarin which is regarded as a matrix language in our application, their pronunciations of English as the embedded language usually contain Mandarin accents. In order to deal with the matrix language accents in embedded language, different non-native adaptation approaches are investigated. Experiments show that model retraining method outperforms the other common adaptation methods such as Maximum A Posteriori (MAP). With the effective incorporation of approaches on phone clustering and non-native adaptation, the Phrase Error Rate (PER) of MESRS for English utterances was reduced by 24.47% relatively compared to the baseline monolingual English system while the PER on Mandarin utterances was comparable to that of the baseline monolingual Mandarin system. The performance for bilingual utterances achieved 22.37% relative PER reduction.

We present an adaptive dynamic load balancing scheme for 3D texture based sort-last parallel volume rendering on a PC cluster equipped with GPUs. Our scheme exploits not only task parallelism but also data parallelism during rendering by combining the hierarchical data structures (octree and parallel BSP tree) in order to skip empty regions and distribute proper workloads to rendering nodes. Our scheme can also conduct a valid parallel rendering and image compositing in visibility order by employing a 3D clustering algorithm. To alleviate the imbalance when the transfer function is changed, a load rebalancing is inexpensively supported by exchanging only needed data. A detailed performance analysis is provided and scaling characteristics of our scheme are discussed. These show that our scheme can achieve significant performance gains by increasing parallelism and decreasing synchronizing costs compared to the traditional static distribution schemes.

Image segmentation is an essential processing step for many image analysis applications. In this paper, a novel image segmentation algorithm using fuzzy C-means clustering (FCM) with spatial constraints based on Markov random field (MRF) via Bayesian theory is proposed. Due to disregard of spatial constraint information, the FCM algorithm fails to segment images corrupted by noise. In order to improve the robustness of FCM to noise, a powerful model for the membership functions that incorporates local correlation is given by MRF defined through a Gibbs function. Then spatial information is incorporated into the FCM by Bayesian theory. Therefore, the proposed algorithm has both the advantages of the FCM and MRF, and is robust to noise. Experimental results on the synthetic and real-world images are given to demonstrate the robustness and validity of the proposed algorithm.

Robust automatic language identification (LID) is the task of identifying the language from a short utterance spoken by an unknown speaker. The mainstream approaches include parallel phone recognition language modeling (PPRLM), support vector machine (SVM) and the general Gaussian mixture models (GMMs). These systems map the cepstral features of spoken utterances into high level scores by classifiers. In this paper, in order to increase the dimension of the score vector and alleviate the inter-speaker variability within the same language, multiple data groups based on supervised speaker clustering are employed to generate the discriminative language characterization score vectors (DLCSV). The back-end SVM classifiers are used to model the probability distribution of each target language in the DLCSV space. Finally, the output scores of back-end classifiers are calibrated by a pair-wise posterior probability estimation (PPPE) algorithm. The proposed language identification frameworks are evaluated on 2003 NIST Language Recognition Evaluation (LRE) databases and the experiments show that the system described in this paper produces comparable results to the existing systems. Especially, the SVM framework achieves an equal error rate (EER) of 4.0% in the 30-second task and outperforms the state-of-art systems by more than 30% relative error reduction. Besides, the performances of proposed PPRLM and GMMs algorithms achieve an EER of 5.1% and 5.0% respectively.

This letter mainly focuses on improving current noise reduction methods to solve the critical speech distortion problems with robust noise reduction in noisy speech signals for speech enhancement over IP networks. For robust noise reduction with packet loss recovery, we propose a novel optimized Wiener filtering technique that uses the estimated SNR (Signal-to-Noise Ratio) with packet loss recovery method which is applied as post-filtering over IP-networks. Simulation results demonstrate that the proposed scheme provides better reduction and recovery rates with considering packet loss and SNR environment than other methods.

The Earth will require sustainable electricity sources equivalent to 3 to 5 times the commercial power presently produced by 2050. Solar Power Satellite (SPS) is one option for meeting the huge future energy demand. SPS can send enormous amounts of power to the Earth as the form of microwave (MW). A highly efficient microwave power transmission (MPT) system is needed for SPS. A critical goal of SPS is to maintain highest Beam Efficiency (BE) because the microwaves from SPS will be converted to utility power unlike the MW from communication satellites. Another critical goal of SPS is to maintain Side Lobe Levels (SLL) as small as possible to reduce interference to other communication systems. One way to decrease SLL and increase BE is the edge tapering of a phased array antenna. However, tapering the excitation requires a technically complicated system. Another way of achieving minimum SLL is with randomly spaced element position but it does not guarantee higher BE and the determination of random element position is also a difficult task. Isosceles Trapezoidal Distribution (ITD) edge tapered antenna was studied for SPS as an optimization between full edge tapering and uniform amplitude distribution. The highest Beam Collection Efficiency (BCE) and lowest SLL (except maximum SLL) are possible to achieve in ITD edge tapering and ITD edge tapered antenna is technically better. The performance of ITD is further improved from the perspective of both Maximum Side Lobe Level (MSLL) and BE by using unequal spacing of the antenna elements. A remarkable reduction in MSLL is achieved with ITD edge tapering with Unequal element spacing (ITDU). BE was also highest in ITDU. Determination of unequal element position for ITDU is very easy. ITDU is a newer concept that is experimented for the first time. The merits of ITDU over ITD and Gaussian edge tapering are discussed.

We address pilot-aided channel estimation for a flat-fading spatially-correlated MIMO system, which employing Uniform Linear Arrays (ULA) on dual side and working in sparse scattering (multipath) environment. In case of sparse scattering, channel matrix and spatial correlation of flat-fading MIMO systems are parameterized by structure of multipaths, which is represented as Direction of Arrivals (DOAs), Direction of Departures (DODs) and complex path fading of each path. Based on this and block-fading property of channel, we design a channel estimation method via estimating multipath parameters using ESPRIT-like DOA-Matrix method which exploits shift-invariance property of ULA. The proposed method is able to obtain improved Mean-Square-Error performance than Least-Square method without prior information of spatial correlation.

This paper describes 10-bit, 80-MSample/s pipelined A/D converters for wireless-communication terminals. To reduce power consumption, we employed the I/Q amplifier sharing technique [1] in which an amplifier is used for both I and Q channels. In addition, common-source, pseudo-differential (PD) amplifiers are used in all the conversion stages for further power reduction. Common-mode disturbances are removed by the proposed common-mode feedforward (CMFF) technique without using fully differential (FD) amplifiers. The converter was implemented in a 90-nm CMOS technology, and it consumes only 24 mW/ch from a 1.2-V power supply. The measured SNR and SNDR are 58.6 dB and 52.2 dB, respectively.

In this paper, horizontal (where an opamp is shared in two adjacent stages) and vertical (where an opamp is shared across two paths) opamp sharing techniques for a two-path band-pass (BP) ΔΣ modulator are described, and input-feedforward two-path fourth-order BP ΔΣ modulators that have only two opamps are proposed. The proposed modulators are based on the horizontal or vertical opamp sharing technique. They can be realized with both a summation circuit using a switched capacitor (SC) network and a second-order high-pass filter (HPF) with a horizontal shared opamp or a double-sampling first-order HPF with a vertical shared opamp, which are based on an SC first-order HPF with an opamp. These techniques can reduce the number of opamps with no additional component and the chip area as well as realize lower power consumption.

In this review paper we showcase recent activities on silicon photonics science and technology research in Hong Kong regarding two important topical areas--microresonator devices and optical nonlinearities. Our work on silicon microresonator filters, switches and modulators have shown promise for the nascent development of on-chip optoelectronic signal processing systems, while our studies on optical nonlinearities have contributed to basic understanding of silicon-based optically-pumped light sources and helium-implanted detectors. Here, we review our various passive and electro-optic active microresonator devices including (i) cascaded microring resonator cross-connect filters, (ii) NRZ-to-PRZ data format converters using a microring resonator notch filter, (iii) GHz-speed carrier-injection-based microring resonator modulators and 0.5-GHz-speed carrier-injection-based microdisk resonator modulators, and (iv) electrically reconfigurable microring resonator add-drop filters and electro-optic logic switches using interferometric resonance control. On the nonlinear waveguide front, we review the main nonlinear optical effects in silicon, and show that even at fairly modest average powers two-photon absorption and the accompanied free-carrier linear absorption could lead to optical limiting and a dramatic reduction in the effective lengths of nonlinear devices.

Real-time content analysis is typically a bottleneck in Web filtering. To accelerate the filtering process, this work presents a simple, but effective early decision algorithm that analyzes only part of the Web content. This algorithm can make the filtering decision, either to block or to pass the Web content, as soon as it is confident with a high probability that the content really belongs to a banned or an allowed category. Experiments show the algorithm needs to examine only around one-fourth of the Web content on average, while the accuracy remains fairly good: 89% for the banned content and 93% for the allowed content. This algorithm can complement other Web filtering approaches, such as URL blocking, to filter the Web content with high accuracy and efficiency. Text classification algorithms in other applications can also follow the principle of early decision to accelerate their applications.

The HITS algorithm proposed by Kleinberg is one of the representative methods of scoring Web pages by using hyperlinks. In the days when the algorithm was proposed, most of the pages given high score by the algorithm were really related to a given topic, and hence the algorithm could be used to find related pages. However, the algorithm and the variants including Bharat's improved HITS, abbreviated to BHITS, proposed by Bharat and Henzinger cannot be used to find related pages any more on today's Web, due to an increase of spam links. In this paper, we first propose three methods to find "linkfarms," that is, sets of spam links forming a densely connected subgraph of a Web graph. We then present an algorithm, called a trust-score algorithm, to give high scores to pages which are not spam pages with a high probability. Combining the three methods and the trust-score algorithm with BHITS, we obtain several variants of the HITS algorithm. We ascertain by experiments that one of them, named TaN+BHITS using the trust-score algorithm and the method of finding linkfarms by employing name servers, is most suitable for finding related pages on today's Web. Our algorithms take time and memory no more than those required by the original HITS algorithm, and can be executed on a PC with a small amount of main memory.

A new load balanced channel sharing method (CSM), namely Heuristic Traffic Load Balanced (HTLB) CSM, is proposed for metro-wavelength division multiple access (WDMA) networks. In particular, HTLB CSM is designed to be effective for pre-allocation based medium access control (MAC) protocols by balancing traffic loads corresponding to pre-assigned destinations per time slot. As a result, HTLB CSM is shown to provide lower time complexity than the well-known sub-optimal load balanced CSM, MULTIFIT CSM. Furthermore, the Jain Index of the HTLB CSM is shown to be higher and more consistent than the MULTIFIT CSM and other pre-fixed CSMs under diverse traffic conditions.

We propose and demonstrate a simple one-bus two-ring configuration where the two rings are mutually coupled that has advantages over the one-ring structure. Unlike a one cavity system, it can exhibit near critically-coupled transmission with a broader range of loss. It can also significantly enhance the cavity finesse by simply making the second ring twice the size of the bus-coupled one, with the enhancement proportional to the intensity buildup in the second ring.

A design of a low-power CMOS ring oscillator for an application to a 13.56 MHz clock generator in an implantable RFID tag is proposed. The circuit is based on a novel voltage inverter, which is an improved version of the conventional current-source loaded inverter. The proposed circuit enables low-power operation and low sensitivity of the oscillation frequency, fOSC, to decay of the power supply VDD. By employing a gm-boosting subcircuit, power dissipation is decreased to 49 µW at fOSC=13.56 MHz. The sensitivity of fOSC to VDD is reduced to -0.02 at fOSC=13.56 MHz thanks to the use of composite high-impedance current sources.

This paper proposes useful circuit structures for achieving a low-voltage/low-power pipelined ADC based on switched-opamp architecture. First, a novel unity-feedback-factor sample-and-hold which manipulates the features of switched-opamp technique is presented. Second, opamp-sharing is merged into switched-opamp structure with a proposed dual-output opamp configuration. A 0.8-V, 9-bit, 10-Msample/s pipelined ADC is designed to verify the proposed circuit. Simulation results using a 0.18-µm CMOS 1P6M process demonstrate the figure-of-merit of this pipelined ADC is only 0.71 pJ/step.

In this paper, we present a clean and simple formulation of survey propagation (SP) for constraint-satisfaction problems as "probabilistic token passing". The result shows the importance of extending variable alphabets to their power sets in designing SP algorithms.