TY - JOUR
T1 - Algoritmos Wavenet con aplicaciones en la aproximación de señales
T2 - Un estudio comparativo
AU - Domínguez Mayorga, C. R.
AU - Espejel Rivera, M. A.
AU - Ramos Velasco, L. E.
AU - Ramos Fernández, J. C.
AU - Escamilla Hernández, E.
N1 - Funding Information:
En (Chen and Hui-Qiang, 2007) se propone un método para implementar el convertidor analógico digital (ADC) de alta pre-cisión usando una red neuronal wavelet para aproximar y com-pensar la no linealidad del ADC. La red propuesta es un red de 3 capas, en donde la capa de salida implementa la función sig-moide, mientras que la wavelet madre Morlet es implementada en el resto de la red. Este tipo de red requiere de un número pequeño de iteraciones y parámetros en comparación con los perceptrones multicapa. El algoritmo propuesto en (Chen and Hui-Qiang, 2007) es muy similar al de este artículo, solo que la salida en este caso es la función identidad y en el presentado en (Chen and Hui-Qiang, 2007) es la función sigmoide, la cual acota la salida en el intervalo (01), que muestra ser útil para el caso que se estudia. Otra diferencia con el algoritmo aquí pro-puesto es que no se implementa el filtro IIR a la salida de la red.
PY - 2012
Y1 - 2012
N2 - In this paper adaptable methods for computational algorithms are presented. These algorithms use neural networks and wavelet series to build neuro wavenets approximators. The algorithms obtained are applied to the approximation of signals that represent algebraic functions and random functions, as well as a medical EKG signal. It shows how wavenets can be combined with auto-tuning methods for tracking complex signals that are a function of time. Results are shown in numerical simulation of two architectures of neural approximators wavenets: the first is based on a wavenet with which they approach the signals under study where the parameters of the neural network are adjusted online, the other neuro approximator scheme uses an IIR filter to the output of wavenet, which serves to filter the output, in this way discriminate contributions of neurons that are less important in the approximation of the signal, which helps reduce the convergence time to a desired minimum error.
AB - In this paper adaptable methods for computational algorithms are presented. These algorithms use neural networks and wavelet series to build neuro wavenets approximators. The algorithms obtained are applied to the approximation of signals that represent algebraic functions and random functions, as well as a medical EKG signal. It shows how wavenets can be combined with auto-tuning methods for tracking complex signals that are a function of time. Results are shown in numerical simulation of two architectures of neural approximators wavenets: the first is based on a wavenet with which they approach the signals under study where the parameters of the neural network are adjusted online, the other neuro approximator scheme uses an IIR filter to the output of wavenet, which serves to filter the output, in this way discriminate contributions of neurons that are less important in the approximation of the signal, which helps reduce the convergence time to a desired minimum error.
KW - Approximation algorithms
KW - Gradient methods
KW - Neural networks
KW - Self-adapting algorithms
KW - Signal processing
UR - http://www.scopus.com/inward/record.url?scp=84874466414&partnerID=8YFLogxK
U2 - 10.1016/j.riai.2012.09.001
DO - 10.1016/j.riai.2012.09.001
M3 - Artículo
AN - SCOPUS:84874466414
SN - 1697-7912
VL - 9
SP - 347
EP - 358
JO - RIAI - Revista Iberoamericana de Automatica e Informatica Industrial
JF - RIAI - Revista Iberoamericana de Automatica e Informatica Industrial
IS - 4
ER -