Call Today 716.688.4675

Compressed sensing approach to blind source separation of far field sources

Compressed sensing framework for blind source separation have been considered in recent years because of the inherent sparsity in the assumptions used for blind source separation. Consider a noise free instantaneous mixture signals described below:

$\begin{bmatrix} x_1(t) \\ x_2(t) \\ \vdots \\ x_N(t)\end{bmatrix} = \begin{bmatrix} a_{11} & a_{12} & \cdots & a_{1M}\\a_{21} & a_{22} & \cdots & a_{2M}\\\vdots & \vdots & \vdots & \vdots\\a_{N1} & a_{N2} & \cdots & a_{NM}\end{bmatrix} \begin{bmatrix} s_1(t) \\ s_2(t) \\ \vdots \\ s_M(t)\end{bmatrix}$

where there are $M$ sources and $N$ observations. Under the assumption that only one source is active, which implies the signal space is sparse, for each time, we define the following vectors and matrix.define the vector $b$ as:

$b = \begin{bmatrix} x_1(t) & x_1(t+1) & \cdots & x_1(t+T-1) & x_2(t) & \cdots & x_N(t+T-1)\end{bmatrix} ^T$

where $^T$ denotes the transpose. Also denote the vector $s$ as

$s = \begin{bmatrix} s_1(t) & s_1(t+1) & \cdots & s_1(t+T-1) & s_2(t) & \cdots & s_N(t+T-1)\end{bmatrix} ^T$

Finally, denote the matrix $A$ as

$\begin{bmatrix} A_{11} & A_{12} & \cdots & A_{1M}\\a_{21} & a_{22} & \cdots & A_{2M}\\\vdots & \vdots & \vdots & \vdots\\A_{N1} & A_{N2} & \cdots & A_{NM}\end{bmatrix}$

where

$A_{ij} = \begin{bmatrix} a_{11} &0 & \cdots &0\\0 & a_{11} & \cdots & 0\\\vdots & & \ddots & \vdots\\0 &\cdots &0& a_{11}\end{bmatrix}$

The problem can then be posed as solving the equation

$b = A s$

With $s$ being sparse by assumption, this is the classical compressed sensing signal recovery problem using

$min||s||_0 ~~ s.t. ~~ b = A s$

The signal space can be further given a sparser representation by using a dictionary D such that $s = \hat{A} \hat{x}$ with the $l_0$ minimization done on $\hat{x}$.

VOCAL Technologies offers custom designed solutions for beamforming with a robust voice activity detector, acoustic echo cancellation and noise suppression. Our custom implementations of such systems are meant to deliver optimum performance for your specific beamforming task. Contact us today to discuss your solution!

Solutions
Resources
VOCAL Technologies, Ltd.
520 Lee Entrance, Suite 202
Amherst New York 14228
Phone: +1-716-688-4675
Fax: +1-716-639-0713
Email: sales@vocal.com