P. Muret et N. Omidi, ANALYSIS OF STEEPENED MAGNETOSONIC WAVES USING WAVELET TRANSFORMS ANDNEURAL NETWORKS, J GEO R-S P, 100(A12), 1995, pp. 23465-23479
In this study, wavelet transforms and neural networks are used to anal
yze the formation and evolution of steepened magnetosonic waves (shock
lets) generated in hybrid (fluid electrons, particle ions) simulations
, These waves model the shocklets observed upstream of planetary bow s
hocks and at comets, Specifically, the usefulness of wavelet transform
s and neural networks for understanding the nature of the steepening p
rocess, and the further evolution of shocklets into less coherent stru
ctures is investigated, Previous studies had suggested that the nonlin
ear steepening process is associated with the excitation of higher fre
quency waves within the original wave, This hypothesis, however, could
not be directly substantiated using Fourier transforms, In order to c
ontinue with the analysis of shocklets, it has become necessary to imp
lement new techniques using wavelet transforms and neural networks, Wa
velet transforms are tailored to the analysis of localized structures
such as shocklets, while neural networks have the ability to model non
linear dynamical systems, Application of wavelet transforms has verifi
ed the presence of higher frequency waves within the steepening shockl
et and has identified them as the forward propagating and the backward
propagating magnetosonic waves as well as the backward propagating Al
fven ion-cyclotron mode, The wavelet transform has also located the so
urce of the whistler wave packet attached to the shocklet to be the re
gion of steep magnetic field gradient, In order to understand the furt
her evolution of shocklets, neural networks have been applied in the a
nalysis, Used in conjunction with a translation invariant transform, n
eural networks have been successfully trained to identify shocklets, T
his allows for the scanning of large datasets as well as the developme
nt of a classification system for shocklets, A multinetwork classifica
tion system using various techniques, including wavelet preprocessing,
has been developed to analyze the further evolution of shocklets and
their components, Identification and classification of neural networks
have increased our understanding of shocklet evolution, The technique
s involving wavelet transforms and neural networks that have been empl
oyed in this study show considerable potential for the study of not on
ly shocklets, but also other wave phenomena in space plasmas.