Integral Transforms from Finite Data: An Application of Gaussian Process\n Regression to Fourier Analysis

Luca Ambrogioni, Eric Maris

Abstract

Computing accurate estimates of the Fourier transform of analog signals from discrete data points is important in many fields of science and engineering. The conventional approach of performing the discrete Fourier transform (DFT) of the data implicitly assumes periodicity and bandlimitedness of the signal. In this paper, we use Gaussian process regression to estimate the Fourier transform (or any other integral transform) without making these assumptions. This is possible because the posterior expectation of Gaussian process regression maps a finite set of samples to a function defined on the whole real line, expressed as a linear combination of covariance functions. We estimate the covariance function from the data using an appropriately designed gradient ascent method that constrains the solution to a linear combination of tractable kernel functions. This procedure results in a posterior expectation of the analog signal whose Fourier transform can be obtained analytically by exploiting linearity. Our simulations show that the new method leads to sharper and more precise estimation of the spectral density both in noise-free and noise-corrupted signals. We further validate the method in two real-world applications: the analysis of the yearly fluctuation in atmospheric CO2 level and the analysis of the spectral content of brain signals.

Introduction

Analyzing analog signals—such as acoustic waves or neurophysiological recordings—requires sampling them at discrete intervals. The standard tool for frequency analysis, the Discrete Fourier Transform (DFT), relies on the assumptions that the signal is both periodic and bandlimited. However, real-world signals rarely satisfy these conditions perfectly. When these assumptions are violated, the DFT suffers from "spectral leakage," where energy from one frequency smears into others, creating artificial sidelobes that obscure low-energy components and reduce resolution.

Methodology: The Bayes-Gauss-Fourier (BGF) Transform

We propose the Bayes-Gauss-Fourier (BGF) transform, which leverages Gaussian Process (GP) regression to bridge the gap between discrete samples and continuous integral transforms.

Continuous Modeling and Analytical Transforms

Instead of treating data as a sequence of isolated points, we model the underlying signal as a continuous function $m_f(t)$ using GP regression. This function is expressed as a linear combination of covariance functions (kernels). Because the Fourier transform is a linear operator, the transform of the estimated signal is simply the sum of the transforms of the individual kernels. By selecting tractable kernels, such as the Squared Exponential kernel, we can derive the Fourier transform analytically in closed form.

Hierarchical Covariance Learning

The model does not rely on a fixed kernel. Instead, it employs a hierarchical learning step where the spectral density is estimated directly from the data using a gradient ascent method (Maximum A Posteriori estimation). This allows the BGF to "learn" the signal's frequency structure before performing the transform, effectively acting as an integrated denoiser.

Results and Performance

The BGF method was validated against synthetic signals, atmospheric $CO_2$ data, and human magnetoencephalography (MEG) recordings.

Sidelobe Suppression and Spectral Fidelity

In synthetic tests involving windowed oscillations, the BGF achieved sidelobe suppression up to $10^6$ times higher than a Hann-tapered DFT.

As illustrated in Figure 1, the BGF spectrum (green) closely follows the true spectrum (dashed black), whereas the standard DFT (blue) and Hann-tapered DFT (red) exhibit significant leakage and higher noise floors.

Comparison with Multitaper Methods

When compared to state-of-the-art multitaper methods using Discrete Prolate Spheroidal Sequences (DPSS), the BGF demonstrated superior performance in stopband suppression.

Quantitative analysis (Figure 6) confirms that BFG maintains high passband performance while significantly outperforming DPSS methods in stopband rejection. This leads to cleaner spectral estimates in MEG data, where it successfully resolved the 11 Hz alpha rhythm and its 22 Hz harmonic with minimal background noise.

Discussion

The BGF framework offers a flexible alternative to the DFT by treating the latter as a special, degenerate case of GP regression. By relaxing the rigid assumptions of periodicity and bandlimitedness, BGF provides sharper peaks and better noise handling. Future work will focus on extending this approach to non-stationary signals and optimizing the $O(N^3)$ computational complexity inherent in standard GP regression to accommodate larger datasets.

Summary updated

Abstract

Computing accurate estimates of the Fourier transform of analog signals from\ndiscrete data points is important in many fields of science and engineering.\nThe conventional approach of performing the discrete Fourier transform of the\ndata implicitly assumes periodicity and bandlimitedness of the signal. In this\npaper, we use Gaussian process regression to estimate the Fourier transform (or\nany other integral transform) without making these assumptions. This is\npossible because the posterior expectation of Gaussian process regression maps\na finite set of samples to a function defined on the whole real line, expressed\nas a linear combination of covariance functions. We estimate the covariance\nfunction from the data using an appropriately designed gradient ascent method\nthat constrains the solution to a linear combination of tractable kernel\nfunctions. This procedure results in a posterior expectation of the analog\nsignal whose Fourier transform can be obtained analytically by exploiting\nlinearity. Our simulations show that the new method leads to sharper and more\nprecise estimation of the spectral density both in noise-free and\nnoise-corrupted signals. We further validate the method in two real-world\napplications: the analysis of the yearly fluctuation in atmospheric CO2 level\nand the analysis of the spectral content of brain signals.\n
BibTeX
@inproceedings{ambrogioni2018integral,
  author = {Luca Ambrogioni and Eric Maris},
  title = {{Integral Transforms from Finite Data: An Application of Gaussian Process\textbackslash\{\}n Regression to Fourier Analysis}},
  booktitle = {International Conference on Artificial Intelligence and Statistics},
  year = {2018},
}
Figure