Trainable Wavelet Neural Network for Non-Stationary Signals

TL;DR

This paper presents a wavelet neural network that learns specialized filter-banks for non-stationary signals, enhancing interpretability and performance in digital signal processing tasks.

Contribution

It introduces a wavelet transform-based neural network with parameterized Morlet wavelets, demonstrating improved convergence, generalization, and performance over standard architectures.

Findings

Quick convergence on training data

Robust generalization to noisy signals

Outperforms standard neural network architectures

Abstract

This work introduces a wavelet neural network to learn a filter-bank specialized to fit non-stationary signals and improve interpretability and performance for digital signal processing. The network uses a wavelet transform as the first layer of a neural network where the convolution is a parameterized function of the complex Morlet wavelet. Experimental results, on both simplified data and atmospheric gravity waves, show the network is quick to converge, generalizes well on noisy data, and outperforms standard network architectures.