Nonlinear Blind Source Separation Using Sensor-Independent Signal Representations

TL;DR

This paper introduces a method to identify and separate signals that are nonlinear mixtures of independent sources by analyzing local velocity distributions, enabling recovery of original signals in complex scenarios.

Contribution

It presents a novel approach to nonlinear blind source separation using local velocity analysis to determine separability and recover independent source signals.

Findings

Successfully separates nonlinear mixtures of signals.

Recovers original signals from complex combinations.

Applicable to speech and other real-world data.

Abstract

Consider a time series of signal measurements $x(t)$, having components $x_k \mbox{ for } k = 1,2, \ldots ,N$. This paper shows how to determine if these signals are equal to linear or nonlinear mixtures of the state variables of two or more statistically-independent subsystems. First, the local distribution of measurement velocities ($\dot{x}$) is processed in order to derive $N$ local vectors at each $x$. If the data are separable, each of these vectors is directed along a subspace traversed by varying the state variable of one subsystem, while all other subsystems are kept constant. Because of this property, these vectors can be used to determine if the data are separable, and, if they are, $x(t)$ can be transformed into a separable coordinate system in order to recover the time series of the independent subsystems. The method is illustrated by using it to blindly recover the separate utterances of two speakers from nonlinear combinations of their waveforms.