Edit on GitHub

Build Status codecov

SpectralDistances

This package facilitates the calculation of distances between signals, primarily in the frequency domain. The main functionality revolves around rational spectra, i.e., the spectrum of a rational function, known as a transfer function, an autoregressive model or an autoregressive moving-average model.

This repository implements all optimal-transport based distances between spectra detailed in the following pre-pre print

"New Metrics Between Rational Spectra and their Connection to Optimal Transport", Bagge Carlson and Chitre 2020

Installation

The package is registered and you may install it by

using Pkg
pkg"add SpectralDistances"

High-level overview

The main workflow is as follows

  1. Define a distance
  2. Evaluate the distance between two points (signals, histograms, periodograms, etc.)

You can also calculate barycenters, interpolations, and barycentric coordinates of signals under the chosen distance.

This package extends Distances.jl and all distance types are subtypes of Distances.PreMetric, even though some technically are true metrics and some are not even pre-metrics.

Many distances are differentiable and can thus be used for gradient-based learning. The rest of this manual is divided into the following sections

Contents

All Exported functions and types