Blind Source Separation

Blind Source Separation (BSS) recovering source signals from their mixtures without knowing the mixing process has still been one of the most interesting topics in the field of signal processing. This page shows a BSS demo.

Algorithms

The algorithms used in the demo are as follows:

Conventional Methods

1. Original based Independent Vector Analysis (AuxIVA)

2. Independent Low-Rank Matrix Analysis (ILRMA)

3. ILRMA based on Multivariate complex Exponential Power Distribtion (ILRMA-MEPD)

Proposed Method

1. AuxIVA with Inter-Clique Dependence (AuxIVA-ICD)

2. ILRMA with Inter-Clique Dependence (ILRMA-ICD)

3. AuxIVA using Time-Varying-Variance source pdf models with Inter-Clique Dependence (AuxIVA-TVV-ICD)

All the NMF-based algorithms set ranks to 2. NMF variables were randomly initialized. As frequency dependence source priors, overlapped subband and harmonic clique models were considered in addition to a full-band radially symmetric joint pdf.