Data-driven Estimation of Sinusoid Frequencies

TL;DR

This paper introduces a new neural network architecture for frequency estimation in noisy signals, significantly improving accuracy and automation over existing methods, especially at higher noise levels.

Contribution

A novel neural network design that enhances frequency representation accuracy and includes a module for automatic frequency count detection, advancing state-of-the-art performance.

Findings

Outperforms existing methods at medium-to-high noise levels

Provides a fast and fully-automatic frequency estimation process

Achieves state-of-the-art accuracy in frequency estimation

Abstract

Frequency estimation is a fundamental problem in signal processing, with applications in radar imaging, underwater acoustics, seismic imaging, and spectroscopy. The goal is to estimate the frequency of each component in a multisinusoidal signal from a finite number of noisy samples. A recent machine-learning approach uses a neural network to output a learned representation with local maxima at the position of the frequency estimates. In this work, we propose a novel neural-network architecture that produces a significantly more accurate representation, and combine it with an additional neural-network module trained to detect the number of frequencies. This yields a fast, fully-automatic method for frequency estimation that achieves state-of-the-art results. In particular, it outperforms existing techniques by a substantial margin at medium-to-high noise levels.