Modular Arrays (MAs) are a promising architecture to enable multi-user communications in next-generation multiple-input multiple-output (MIMO) systems based on extra-large (XL) or gigantic MIMO (gMIMO) deployments, trading off improved spatial resolution with characteristic interference patterns associated with grating lobes. In this work, we analyze whether MAs can outperform conventional collocated deployments, in terms of achievable sum-spectral efficiency (SE) and served users in a multi-user downlink set-up. First, we provide a rigorous analytical characterization of the inter-user interference for modular gMIMO systems operating in the near field. Then, we leverage these results to optimize the user selection and precoding mechanisms, designing two algorithms that largely outperform existing alternatives in the literature, with different algorithmic complexities. Results show that the proposed algorithms yield over 70% improvements in achievable sum-spectral efficiencies compared to the state of the art. We also illustrate how MAs allows us to serve a larger number of users thanks to their improved spatial resolution, compared to the collocated counterpart.

In this work we describe a cognitive model to infer the more likely user's states in data-intensive contexts. Stress, mental fatigue, or even inaptitude, are selected to be inferred by the model based two sources of information: context and psycho-physiological sensors network. As long as a complex, high demanding context will predict those cognitive states that, in turn, will be diagnosed by the set of sensors (EEG and ECG). All these input variables are represented in a probabilistic model in which links are defined based on the literature. The outcome of the model is a probability of being inapt to perform in a suitable way. In case of inaptitude, assistance should be delivered to the user to normalize the current user's state.

Recognizing lines of unconstrained handwritten text is a challenging task. The difficulty of segmenting cursive or overlapping characters, combined with the need to exploit surrounding context, has led to low recognition rates for even the best current recognizers. Most recent progress in the field has been made either through improved preprocessing or through advances in language modeling. Relatively little work has been done on the basic recognition algorithms. Indeed, most systems rely on the same hidden Markov models that have been used for decades in speech and handwriting recognition, despite their well-known shortcomings. This paper proposes an alternative approach based on a novel type of recurrent neural network, specifically designed for sequence labeling tasks where the data is hard to segment and contains long-range bidirectional interdependencies. In experiments on two large unconstrained handwriting databases, our approach achieves word recognition accuracies of 79.7 percent on online data and 74.1 percent on offline data, significantly outperforming a state-of-the-art HMM-based system. In addition, we demonstrate the network's robustness to lexicon size, measure the individual influence of its hidden layers, and analyze its use of context. Last, we provide an in-depth discussion of the differences between the network and HMMs, suggesting reasons for the network's superior performance.

On-line handwriting recognition is unusual among sequence labelling tasks in that the underlying generator of the observed data, i.e. the movement of the pen, is recorded directly. However, the raw data can be difficult to interpret because each letter is spread over many pen locations. As a consequence, sophisticated pre-processing is required to obtain inputs suitable for conventional sequence labelling algorithms, such as HMMs. In this paper we describe a system capable of directly transcribing raw on-line handwriting data. The system consists of a recurrent neural network trained for sequence labelling, combined with a probabilistic language model. In experiments on an unconstrained on-line database, we record excellent results using either raw or pre-processed data, well outperforming a benchmark HMM in both cases.

Many real-world sequence learning tasks require the prediction of sequences of labels from noisy, unsegmented input data. In speech recognition, for example, an acoustic signal is transcribed into words or sub-word units. Recurrent neural networks (RNNs) are powerful sequence learners that would seem well suited to such tasks. However, because they require pre-segmented training data, and post-processing to transform their outputs into label sequences, their applicability has so far been limited. This paper presents a novel method for training RNNs to label unsegmented sequences directly, thereby solving both problems. An experiment on the TIMIT speech corpus demonstrates its advantages over both a baseline HMM and a hybrid HMM-RNN.

In many industrial processes, an incorrect operation can lead to irreparable damage to people, equipment, or the environment. In order to reduce risks, the electronic control systems used in this kind of processes must comply with international standard safety requirements. The safety market solutions are mainly based in the use of high-end programmable logic controllers (PLCs) and/or application specific integrated circuits (ASICs) for concrete processes or machines. The solutions proposed in this paper are based on the implementation of safety redundant control systems in programmable logic devices (PLDs and FPGAs). The proposed methodology for safe controllers design is based on the combination of dynamic logic and redundant circuits with voting capabilities. The use of programmable logic devices (PLDs) instead of high-end programmable logic controllers (PLCs) adds flexibility to the design of safety-related control systems and reduces costs, while maintaining high reliability and a good degree of failure detection.

The perceptual interaction between adjacent CV segments is studied in Fricative-Vowel syllables from a coarticulatory point of view. The results of a perceptual experiment with conflictingcue stimuli are analyzed under the assumption, supported by the results, of the possible influence of F-To-V carryover coarticulation on the integration process. The DAC scale was used to estimate the degree of F-To-V carryover coarticulation in the original syllables. The magnitude of the coarticulatory effect permitted us to derive a prediction of the perceptual results based on the articulatory compatibility between the fricative and the vowel. The correlation between actual and predicted decrease in perceptual identification caused by the insertion of a conflicting transition was computed. The results show that the coarticulatory processes cannot explain the outcome of the perceptual experiment. Nevertheless, the perceptual role played by the /i/ transition can be effectively explained as a consequence of the F-To-V coarticulation.

The perceptual interaction between the consonant and the vowel in fricative+vowel syllables is evaluated. A set of conflicting cue stimuli was used to measure the relative importance of: (a) the influence of the vowel in the previous consonant, and (b) the influence of the fricative in the following vowel. It is concluded that the perceptual interaction between the consonant and the vowel in fricative-vowel syllables can not be explained only by the coarticulatory influence of the consonant or vowel on adjacent segments. The influence of the vowel in the previous fricative is perceptually irrelevant, while the influence of the fricative in the following vowel is more important for the identification of //spl theta// and /f/ than for /s/ and //spl int//. Actually it is perceptually irrelevant for //spl int// and a little important for /s/. Besides the influence of the fricative in the following vowel is notably dependent on the particular vowel.

Perceptual and acoustic spaces obtained using MDS and PCA, respectively, for a set of Spanish vowels uttered in various fricative contexts are compared. Acoustic cues to vowel quality were studied through examination of the weights given by PCA to specific spectral regions. The acoustic space formed by the first two formant frequencies was also evaluated. The results show that MDS has serious difficulties to be useful for the study of Spanish vowels. PCA offers a gross characterization compared to formant frequencies. Neither of them completely explains the perception of vowel quality.

The interaction of consonantal and vocalic segments in FV syllables regarding identification of place of articulation of fricatives has been studied. A probabilistic model for integration of acoustic information in both segments is proposed. The model weights each segment's contribution and integrates them in order to resemble listeners' perception. First, the perceptual validity of the model has been assessed. Overall correlations of the probabilistic model with listeners' responses to a set of natural and conflicting-cue (place of articulation) FV syllables were 0.88 and 0.74, respectively. The results showed that in order to increase correlations, the model should weight F and V segments differently for each fricative, and even for each vocalic context, since listeners gave more or less importance to the vocalic transition depending on the particular fricative and vocalic context. Acoustic analysis was also carried out computing the probabilistic model with a posteriori probabilities for F and V segments obtained with quadratic discriminant analysis. The results were disappointing, indicating that usual statistical methods fail to extract coarticulatory information. This prevents ASR systems from taking advantage of the enhancement of the characteristics of the consonant in the FV condition with respect to the F condition.

In this paper, we propose an error correction method using text corpora. In this method, recognition errors are corrected using phonetically similar examples in the text corpora. The reliability of the correction hypotheses are judged according to their semantic consistency and their phonetic similarity to the original input. We previously proposed an error correction method that uses a treebank [1]. However, the previous method was not flexible in its use of examples, because structural mismatches occurred between the input and examples due to recognition errors. In our new proposal, examples are treated as morpheme sequences. This enables us to use examples partially when there are no useful full-sentence-examples. We built our proposed method into a speech translation system and compared the translation quality for simple translation and translation with error correction. The rate of acceptable translation increased about 10% with our proposed method compared to simple translation.