Eduardo Castillo-Guerra

Reduction in the Number of Averages Required in BOTDA Sensors Using Wavelet Denoising Techniques

This paper reports on a new mechanism to decrease the number of averages and, consequently, the measurement time of Brillouin optical time-domain analysis (BOTDA) sensors using wavelet shrinkage techniques. Two different wavelet shrinkage techniques, VisuShrink and SureShrink, are applied to denoise signals acquired from measurements in BOTDA sensors. The conventional method to denoise signals in BOTDA sensors is ensemble averaging. Ensemble averaging is a time consuming technique, as it requires many acquisitions of signals to provide an acceptable SNR. To reduce the number of acquisitions, the setup of the BOTDA sensor is modified to denoise acquired signals using VisuShrink or SureShrink before applying ensemble averaging. Experimental results show a significant reduction in the number of averages required to provide an accurate measurement, and consequently, a substantial saving in the measurement time of the sensor. It has been shown that the combination of ensemble averaging with VisuShrink or SureShrink reduces the measurement time of the sensor up to 90%. This reduction in the measurement time enables the implementation of dynamic and fast measurements with BODTA sensors and opens opportunities to target a new range of applications.

Accurate estimation of Brillouin frequency shift in Brillouin optical time domain analysis sensors using cross correlation

Current methods of estimating the Brillouin frequency shift in Brillouin optical time domain analysis sensors are based on curve-fitting techniques. These techniques apply the same weight to all portions of the curve and dutifully fit into the peak and noisy ends of the curve. This makes them very sensitive to noise, initialization of fitting parameters, symmetry, and start and stop frequencies. We introduce a method based on the cross-correlation technique to estimate the central frequency of noisy Lorentzian curves, which is more robust to noise and free from initial settings of fitting parameters.

Automatic Modeling of Acoustic Perception of Breathiness in Pathological Voices

This paper revisits the modeling of acoustic perceptions of breathy voice (BV) quality for automatic assessment of perturbations in pathologic speech. Several acoustic measures related with the signal periodicity, harmonic components, and aspiration noise are studied to predict breathiness judgments performed on sustained vowel phonations. A novel comprehensive automatic measure is proposed that provides the highest correlation index (88.5%) with breathiness judgment performed by trained specialists on simulated and recorded utterances. The new measure reveals the most relevant aspects of BV quality and provides a vehicle to obtain reliable objectives judgments of such speech perturbation.

A Detailed Evaluation of the Correlation-Based Method Used for Estimation of the Brillouin Frequency Shift in BOTDA Sensors

This paper thoroughly describes and evaluates the method that was previously presented for estimating the central frequency of noisy Lorentzian curves (spectra) acquired from the measurements with Brillouin optical time domain analysis (BOTDA) sensors. The estimator is based on the cross-correlation technique and addresses the problem of sensitivity to noise and parameter initialization observed in other central frequency estimation methods employed with BOTDA sensors. Most of the current estimation methods rely on optimized rigorous least squares or maximum likelihood estimation (MLE) algorithms, which are sensitive to the parameter initialization and noise as they iteratively attempt to minimize the squared error or maximize the matching probability between the model and noisy curve. Alternatively, the estimation made with the cross-correlation based method is more accurate, noniterative, and insensitive to the parameter initialization. This statement is demonstrated and proved by comparing the correlation-based method with two commonly used iterative curve fitting methods based on the Levenberg-Marquardt algorithm and MLE.

Acceleration of Measurements in BOTDA Sensors Using Adaptive Linear Prediction

A conventional method to denoise signals in Brillouin optical time-domain analysis (BOTDA) sensors is ensemble averaging. This method necessitates the acquisition of thousands of signals to provide an acceptable signal-to-noise ratio (SNR). The signal acquisition is a time-consuming process that drastically increases the measurement time of BOTDA sensors. This paper presents a novel method on the basis of the adaptive linear prediction (ALP) technique to reduce the measurement time of such sensors. The conventional setup of BOTDA sensors is modified to denoise signals using the ALP technique before applying ensemble averaging. The application of the ALP technique removes a significant portion of noise while it preserves the abrupt changes and smooth pieces of signals. As a result, the number of signals required to obtain accurate measurements and, consequently, the measurement time of the sensor are reduced by up to 90%. This modification enables BODTA sensors to implement dynamic measurements of temperature and strain and opens opportunities to address a new range of applications.

Load aggregation from generation-follows-load to load-follows-generation

The growing interest in optimizing the generation, distribution, and delivery of electric power have motivated the implementation of several smart grid functions in many power systems around the globe. Among such functions are the peak-load management, demand response, direct load control, and integration of distributed power generation. Nowadays, smart grid functions are being implemented for industrial, residential, and/or commercial loads. One of the key requirements for implementing smart grid functions is the accurate and reliable load aggregation. The bottom-up, coordinated, and bus-split aggregation methods have been found applicable for different load types that are included in smart grid functions. This paper reviews the methods and approaches for performing the load aggregation, and provides a discussion for the critical role of load aggregation in power systems operating and smart grid functions. In addition, this paper discusses the applicability of the load aggregation methods in smart grid functions for residential loads.

Load Aggregation From Generation-Follows-Load to Load-Follows-Generation: Residential Loads

The growing interest in optimizing the generation, distribution, and delivery of electric power have motivated the implementation of several smart grid functions in many power systems around the globe. Among such functions are the peak-load management, demand response, direct load control, and integration of distributed power generation. Nowadays, smart grid functions are being implemented for industrial, residential, and/or commercial loads. One of the key requirements for implementing smart grid functions is the accurate and reliable load aggregation. The bottom-up, coordinated, and bus-split aggregation methods have been found applicable for different load types that are included in smart grid functions. This paper reviews the methods and approaches for performing the load aggregation, and provides a discussion for the critical role of load aggregation in power systems operating and smart grid functions. In addition, this paper discusses the applicability of the load aggregation methods in smart grid functions for residential loads.

A Neural Network Approach to Multi-step-ahead, Short-Term Wind Speed Forecasting

This paper presents a novel neural network-based approach to short-term, multi-step-ahead wind speed forecasting. The methodology combines predictions from a set of feed forward neural networks whose inputs comprehend a set of 11 explanatory variables related to past averages of wind speed, direction, temperature and time of the day, and their outputs represent estimates of specific wind speed averages. Forecast horizons range from 30 minutes up to 6:30 hours ahead with 30 minutes time steps. Final forecasts at specific horizons are combinations of corresponding neural network predictions. Data used in the experiments are telemetric measurements of weather variables from five wind farms in eastern Canada, covering the period from November 2011 to April 2013. Results show that the methodology is effective and outperforms established reference models particularly at longer horizons. The method performed consistently across sites leading up to more than 60% improvement over persistence and 50 % over a more realistic MA-based reference.

Real-time testing of Newton-phaselet method for calculating the power factor of single phase loads

A combination of Newton iterations and phaselet tight frames allows calculating the power factor of a single phase load. In this paper, the real-time implementation and experimental testing of the Newton-phaselet method are presented. The tested method is structured to employ the Newton iterations in order to estimate values for the apparent power S, and to utilize phaselet tight frames to calculate an angle v for the estimated S at each iteration. The estimated S and calculated v at each iteration provide a numerical value for the active power P. This calculated value of P is compared to the measured one in order to determine the required adjustment in S for the next iteration. The Newton-phaselet method is implemented in real time by using a digital signal processing board, where the measured active power is fed as the input. Experimental performances of the Newton-phaselet method are investigated for single phase linear, non-linear, and inverter-fed loads supplied at different frequencies. Test results demonstrate high accuracy, simple implementation, low memory requirements, fast convergence, and negligible sensitivities to harmonic components and supply frequencies.

Practical considerations for real-time implementation of speech-based gender detection

This paper describes a detailed analysis and implementation of a robust gender detector for audio stream applications. The implementation, based on melcepstral features and a Gaussian mixture model classifier, is designed to maximize gender classification performance in continuous speech. The described detector outperforms other reported systems based on statistically significant numbers of gender verifications (2136 unique speakers) obtained from the FISHER speech corpus. The system yields high accuracies for long and short utterances while a confidence figure of merit score for the decision ensures reliability in continuous audio streams.