The Fashion industry revolves around Fashion trends. Designers and merchandisers plan products based on predicted trends, and online catalogs sell new trends to customers. An important part of tracking trends is tracking the ever-evolving fashion terminology. We tackle this problem by expanding a given fashion entity set using automatic synonym discovery in an external fashion related corpus, thus adding new terms to the fashion vocabulary. This paper presents supervised methods for synonym discovery that modify two unsupervised methods based on linear and non-linear contexts, respectively. We demonstrate that the task of synonym discovery benefits from using non-linear context as opposed to linear contexts for learning word embeddings.

We propose an adaptive reliability enhancement structure for deeply-scaled CMOS and future devices that exhibit nondeterministic behavior. This structure forms the basis of a confidence-driven computing model that can be implemented in either a rollback recovery or an iterative dual modular redundancy method incorporating synchronous handshake schemes. The performance and cost of the computing model are estimated using a 45 nm CMOS technology and the functionality is verified by FPGA-based emulation. The confidence-driven computing model is demonstrated using a 16-bit, 12-stage CORDIC processor operating under random, transient errors. The confidence-driven computing model adapts to the fluctuating error rates at the device substrate level to guarantee the reliability of computation at the system level. This computing model costs 4.2 times smaller area and 2.7 times less energy overhead than triple modular redundancy to guarantee a system-level mean time to failure of two years.

A half-duplex relay channel consisting of two relays in a diamond topology is studied. In contrast to full-duplex systems, it is shown that time should be optimally allocated among different states. In contrast to the classical three-node relay channel, it is shown that spatial reuse can be utilized to allow communication without interference. Both achievable rate and upper bound on capacity are studied. Several communication schemes such as multihop with spatial reuse, scale-forward, broadcast-multiaccess with common message, compress-forward, as well as hybrid ones are characterized, some of which are novel. It is also shown that simple multihop with spatial reuse achieves the capacity of certain symmetric diamond channels.

The prior work on power-bandwidth tradeoff analysis in fading environments has assumed the ergodicity of the channel statistics along with infinite code block-lengths, which allow defining energy and spectral efficiency measures as a function of the Shannon capacity. We generalize the power-bandwidth tradeoff analysis framework to draw insights for a specific form communication under a non-ergodic fading model. In particular, we study the power-bandwidth tradeoff in a one-dimensional slow-fading multi-hop relay network. We characterize the impact of multi-hopping on the statistical properties of the end-to-end conditional mutual information (conditioned on the specific values of the channel fading parameters and therefore treated as a random variable) and on the energy and spectral efficiency measures computed from the conditional mutual information. Our non-ergodic power-bandwidth tradeoff analysis focuses on two asymptotic regimes: (i) Power-limited (low signal-to-noise ratio (SNR)) regime and (ii) bandwidth-limited (high SNR) regime. In both regimes, we investigate the performance of open-loop (fixed-rate) and closed-loop (rate-adaptive) multi-hop relaying mechanisms in terms of their power-band width tradeoffs and end-to-end outage probabilities. Interestingly, our analysis reveals the new notion of "multi-hop diversity" toward enhancing link reliability in diversity-limited fading environments

Achieving higher link rates in support of newer services and increasing range are the two main goals for next generation wireless LAN networks. This work primarily contributes to a high rate solution, based on a novel extension to the rate-1 space-frequency block code (SFBC) design in L. Shao et al. (Dec. 2003) to full rate case. To further improve performance without increasing complexity, we propose unequal power allocation at the transmitter with no requirement of channel knowledge. Compared with standard 802.11a transmitter which uses convolutional code with a single transmit antenna and antenna selection diversity among 2 receive antennas with soft Viterbi decoding, we show that our HR-SFBC code with 2 transmit antenna not only doubles the throughput (as expected) but also provides additional link margin (and hence improves range).

A novel rate-1, space-frequency (SF) code for an orthogonal frequency division multiplexing (OFDM) system with transmit (M) and receive (N) antennas that achieves the maximum diversity attainable over frequency-selective channels is proposed. Since the SF code symbol is transmitted in one OFDM block duration, it has smaller processing delay than previously proposed space-time-frequency (STF) block codes.

Optimal signal processing of M receiver antennas requires digitization of each of the M RF signals. For example, MRC (maximal ratio combining) of M receiver antennas requires M ADCs (analog to digital converters). Since the ADC is a highly power-consumptive component in current receiver chain architectures, we investigate suboptimal signal processing schemes that combine M analog signals before digitization, thus enabling a much cheaper implementation with a single ADC. We focus on IEEE 802.11a like OFDM LAN systems with one transmit and multiple receiver antennas (SIMO - single input multiple output), with perfect channel knowledge at the receiver. Our scheme uses scalar weights to combine the receiver antennas in the time-domain. These weights are a function of the spatio-temporal channel matrix. The results are very promising: 4-antenna linear combining (LC) outperforms 2-antenna selection diversity by 5-7 dB at an uncoded BER of 0.01, and this margin is even wider at lower BERs.

A number of space-time block codes have been proposed for the quasi-static, flat-fading channel with coherent receiver. All of these block codes are linear codes, i.e., the encoded codeword is a linear function of the input scalar symbols. Here we propose new non-linear space-time block codes, i.e., the codewords are non-linear functions of the input scalar symbols. We demonstrate a non-linear code that outperforms the corresponding linear code by 0.6 to 1.2 dB. We draw parallels between optimal non-linear code design and the well-known simplex conjecture for multi-dimensional A WGN codes. Finally we show that for certain non-uniformly distributed input symbols, the optimal signal set cannot be designed with linear modulation and non-linear modulation is essential for optimality. This has applications in the design of space-time trellis codes and concatenated space-time coding schemes.

Space-time coding for the flat fading, quasi-static MIMO wireless channel has received widespread attention. This paper treats space-time code design for frequency selective channels from a single carrier modulation perspective. We present a general framework for analyzing space-time codes for delay spread channels based on PEP analysis. It is shown that as in the flat fading case, diversity gain is driven by the minimum rank of the pairwise codeword difference matrices. Delay spread, however, imposes a block Hankel structure on the codewords which may prevent certain codes designed for the flat fading case from exploiting full spatio-temporal diversity when used over frequency selective channels. We explicitly show this diversity loss for the delay diversity code. We show that a properly generalized delay diversity (GDD) code achieves full diversity over frequency selective channels. Finally we propose an extension to the GDD which fits naturally in a multi-carrier setting.

A number of space-time codes have recently been proposed for the quasi-static, flat-fading, multiple-antenna wireless channel with coherent receiver. We focus on space-time codes that are linear in the scalar input symbols. We provide design criteria that take into account both the error probability and the channel capacity. We consider the error probability via the union bound instead of the conventional pairwise error probability and provide conditions on unitary codes to minimize the union bound. Then, we provide conditions on linear codes to maximize channel capacity, i.e. capacity-efficient codes. Finally we provide design rules that optimize the error performance of capacity-efficient codes, and demonstrate performance improvements via simulations.

The design of practical coding techniques for the multiple antenna wireless channel is a challenging problem. A number of interesting solutions have been proposed ranging from block codes to trellis codes for the MIMO (multiple input, multiple output) channel. We consider linear block codes for the quasi-static, flat-fading, coherent MIMO channel. A linear code refers to an encoder that is linear with respect to scalar input symbols. We assume maximum likelihood decoding at the receiver. We provide a cohesive framework for analysis of linear codes in terms of a union bound on the conditional probability of symbol error. The error bound is a function of the instantaneous channel realization and does not make any assumptions on channel statistics. We show that the orthogonal block codes proposed by Tarokh, Jafarkhani and Calderbank, (see IEEE Trans. Information Theory, vol.45, no.5, p.1456-67, 1999) achieve the lowest error bound among all unitary codes and are in fact optimal.

Two outstanding examples of transmit diversity schemes for the multiple-antenna flat-fading channel are space-time block coding (STBC) and space-time trellis coding (STTC). We compare the performance of STBC and STTC in terms of the frame error rate keeping the transmit power, spectral efficiency and number of trellis states fixed. We discover that a simple concatenation of space-time block codes with traditional AWGN (additive white Gaussian noise) trellis codes outperforms some of the best known space-time trellis codes at SNRs (signal to noise ratios) of interest. Our result holds for a small number of trellis states with one or two receive antennas, and is useful for the design and implementation of multiple-antenna wireless systems.

The performance of large-vocabulary automatic speech recognition (ASR) systems deteriorates severely in mismatched training and testing conditions. Signal processing techniques based on the human auditory system have been proposed to improve ASR performance, especially under adverse acoustic conditions. The paper compares one such scheme, the ensemble interval histogram (EIH), with the conventional mel cepstral analysis (MEL). These two spectral feature extraction methods were implemented as front ends to a state-of-the-art continuous speech recognizer and evaluated on the TIMIT database (male). To characterize the influence of signal distortion on the representation of different sounds, phone classification experiments were conducted for three acoustic conditions-clean speech, speech through a telephone channel and speech under room reverberations (the last two are simulations). Classification was performed for static features alone and for static and dynamic features, to observe the relative contribution of time derivatives. The performance is displayed as percentage of phones correctly classified. Confusion matrices were also derived from phone classification to provide diagnostic information.