Data-driven Estimation of Sinusoid Frequencies
This addresses frequency estimation for applications like radar and acoustics, but it is incremental as it builds on existing machine-learning approaches.
The paper tackles the problem of estimating frequencies in multisinusoidal signals from noisy samples, proposing a novel neural-network architecture that significantly improves accuracy and includes a module to detect the number of frequencies, achieving state-of-the-art results with substantial gains at medium-to-high noise levels.
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.