With the development of machine-learning algorithms, many attempts have been
made to use Artificial Neural Networks (ANN) for complicated tasks related to
data classification, pattern recognition, and predictive modeling. Among such
applications include Binary Black Hole (BBH) and Binary Neutron Star (BNS)
merger Gravitational Wave (GW) signal detection and forecasting. Image neural
networks that use time-frequency spectrograms as inputs remain one of the most
prominent methods due to their relevance to highly efficient and robust ANN
architectures. Earlier studies used traditional Fourier transform-based
time-frequency decomposition methods for spectrogram generation, which have had
difficulties identifying rapid frequency changes in merger signals with heavy
background noise. The primary objective of this study is to develop a signal
decomposition technique for improved GW signal classification and detection
performance using ANN. We introduce the Joint-Chirp-rate-Time-Frequency
transform (JCTFT), in which complex-valued window functions are used to
modulate the amplitude, frequency, and phase of the input signal. In addition,
we outline general techniques for generating chirp rate enhanced time-frequency
spectrograms from the results of a JCTFT. We found improved signal localization
performance of the JCTFT in comparison to the short-time-Fourier-transform
method with a moderate-to-high amount of background noise. The JCTFT can be
applied to existing and next-generation GW detector signals. The inclusion of
the chirp rate makes the JCTFT computation more time-consuming. Further studies
will aim to improve the efficiency and performance of JCTFT numerical
computations.
