Blind signal processing system employing information maximization to recover unknown signals through unsupervised minimization of output redundancy
First Claim
1. A method performed in a neural network having input means for receiving a plurality J of input signals (Xj) and output means for producing a plurality I of output signals (Ui) each said output signal Ui representing a combination of said input signals (Xj) weighted by a plurality I of bias weights (Wi0) and a plurality I2 of scaling weights (Wij) such that (Ui)=(Wij)(Xj)+(Wi0), said method minimizing the information redundancy among said output signals (Uj), wherein 0<
- i≦
I>
1 and 0<
j≦
J>
1 are integers, said method comprising;
(a) selecting initial values for said bias weights (Wi0) and said scaling weights (Wij);
(b) producing a plurality I of training signals (Yi) responsive to a transformation of said input signals (Xj) such that Yi =g(Ui), wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj) when J=I; and
(c) adjusting said bias weights (Wi0) and said scaling weights (Wij) responsive to one or more samples of said training signals (Yi) such that each said bias weight Wii0 is changed proportionately to a corresponding bias measure Δ
Wi0 accumulated over said one or more samples and each said scaling weight Wij is changed proportionately to a corresponding scaling measure Δ
Wij =ε
·
∂
(ln|J|)/∂
Wij accumulated over said one or more samples, wherein ε
>
0 is a learning rate.
2 Assignments
0 Petitions
Accused Products
Abstract
A neural network system and unsupervised learning process for separating unknown source signals from their received mixtures by solving the Independent Components Analysis (ICA) problem. The unsupervised learning procedure solves the general blind signal processing problem by maximizing joint output entropy through gradient ascent to minimize mutual information in the outputs. The neural network system can separate a multiplicity of unknown source signals from measured mixture signals where the mixture characteristics and the original source signals are both unknown. The system can be easily adapted to solve the related blind deconvolution problem that extracts an unknown source signal from the output of an unknown reverberating channel.
215 Citations
15 Claims
-
1. A method performed in a neural network having input means for receiving a plurality J of input signals (Xj) and output means for producing a plurality I of output signals (Ui) each said output signal Ui representing a combination of said input signals (Xj) weighted by a plurality I of bias weights (Wi0) and a plurality I2 of scaling weights (Wij) such that (Ui)=(Wij)(Xj)+(Wi0), said method minimizing the information redundancy among said output signals (Uj), wherein 0<
- i≦
I>
1 and 0<
j≦
J>
1 are integers, said method comprising;(a) selecting initial values for said bias weights (Wi0) and said scaling weights (Wij); (b) producing a plurality I of training signals (Yi) responsive to a transformation of said input signals (Xj) such that Yi =g(Ui), wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj) when J=I; and(c) adjusting said bias weights (Wi0) and said scaling weights (Wij) responsive to one or more samples of said training signals (Yi) such that each said bias weight Wii0 is changed proportionately to a corresponding bias measure Δ
Wi0 accumulated over said one or more samples and each said scaling weight Wij is changed proportionately to a corresponding scaling measure Δ
Wij =ε
·
∂
(ln|J|)/∂
Wij accumulated over said one or more samples, wherein ε
>
0 is a learning rate. - View Dependent Claims (2, 3)
- i≦
-
4. A neural-network implemented method for recovering one or more of a plurality I of independent source signals (Si) from a plurality J>
- I of sensor signals (Xj) each including a combination of at least some of said source signals (Si) wherein 0<
i<
I>
1 and 0<
j≦
J>
I are integers, said method comprising;(a) selecting a plurality I of bias weights (Wi0) and a plurality I2 of scaling weights (Wij); (b) adjusting said bias weights (Wi0) and said scaling weights (Wij) by repeatedly performing the steps of; (b.1) producing a plurality I of estimation signals (Ui) responsive to said sensor signals (Xj) such that (Ui)=(Wij)(Xj)+(Wi0), (b.2) producing a plurality I of training signals (Yi) responsive to a transformation of said sensor signals (Xj) such that Yi =g(Ui), wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj) when J=I, and(b.3) adjusting each said bias weight Wi0 and each said scaling weight Wij responsive to one or more samples of said training signals (Yi) such that said each bias weight Wi0 is changed proportionately to a bias measure Δ
Wi0 accumulated over said one or more samples and said each scaling weight Wij is changed proportionately to a corresponding scaling measure Δ
Wij =ε
·
∂
(ln|J|)/∂
Wij accumulated over said one or more samples, wherein ε
>
0 is a learning rate; and(c) producing said estimation signals (Ui) to represent said one or more recovered source signals (Si). - View Dependent Claims (5, 6)
- I of sensor signals (Xj) each including a combination of at least some of said source signals (Si) wherein 0<
-
7. A method implemented in a transversal filter having an input for receiving a sensor signal X that includes a combination of multipath reverberations of a source signal S and having a plurality I of delay line tap output signals (Ti) distributed at intervals of one or more time delays τ
- , said source signal S and said sensor signal X varying with time over a plurality J≧
I of said time delay intervals τ
such that said sensor signal X has a value Xj at time τ
(j-1) and each said delay line tap output signal Ti has a value Xj+1-i representing said sensor signal value Xj delayed by a time interval τ
(i-1), wherein τ
>
0 is a predetermined constant and 0<
i≦
I>
1 and 0<
j≦
J≧
I are integers, said method recovering said source signal S from said sensor signal X and comprising;(a) selecting a plurality I of filter weights (Wi); (b) adjusting said filter weights (Wi) by repeatedly performing the steps of (b.1) producing a plurality K=I of weighted tap output signals (Vk) by combining said delay line tap output signals (Ti) such that (Vk)=(Fki) (Ti), wherein 0<
k≦
K=I>
1 are integers, and wherein Fki =Wk+1-i when 1≦
k+1-i≦
I and Fki =0 otherwise,(b.2) summing a plurality K=I of said weighted tap signals (Vk) to produce an estimation signal ##EQU24## wherein said estimation signal U has a value Uj at time τ
(j-1), (b.3) producing a plurality J of training signals (Yj) responsive to a transformation of said sensor signal values (Xj) such that Yj =g(Uj) wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj) when J=I, and(b.4) adjusting each said filter weight Wi responsive to one or more samples of said training signals (Yj) such that said each filter weight Wi is changed proportionately to a corresponding leading measure Δ
W1 accumulated over said one or more samples when i=1 and a corresponding scaling measure Δ
Wi =ε
·
∂
(ln|J|)/∂
Wi accumulated over said one or more samples otherwise; and(c) producing said estimation signal U to represent said recovered source signal S. - View Dependent Claims (8, 9)
- , said source signal S and said sensor signal X varying with time over a plurality J≧
-
10. A neural network for recovering a plurality of source signals from a plurality of mixtures of said source signals, said neural network comprising:
-
input means for receiving a plurality J of input signals (Xj) each including a combination of at least some of a plurality I of independent source signals (Si), wherein 0<
i≦
I>
1 and 0<
j≦
J≧
I are integers;weight means coupled to said input means for storing a plurality I of bias weights (Wi0) and a plurality I2 of scaling weights (Wij); output means coupled to said weight means for producing a plurality I of output signals (Ui) responsive to said input signals (Xj) such that (Ui)=(Wij) (Xj)+(Wi0); training means coupled to said output means for producing a plurality I of training signals (Yi) responsive to a transformation of said input signals (Xj) such that Yi =g(Ui), wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj) when J=I;adjusting means coupled to said training means and said weight means for adjusting said bias weights (Wi0) and said scaling weights (Wij) responsive to one or more samples of said training signals (Yi) such that each said bias weight Wi0 is changed proportionately to a corresponding bias measure Δ
Wi0 accumulated over said one or more samples and each said scaling weight Wij is changed proportionately to a corresponding scaling measure Δ
Wij =ε
·
∂
(ln|J|)/∂
Wij accumulated over said one or more samples, wherein ε
>
0 is a learning rate. - View Dependent Claims (11, 12)
-
-
13. A system for adaptively cancelling one or more interferer signals (Sn) comprising:
-
input means for receiving a plurality J of input signals (Xj) each including a combination of at least some of a plurality I of independent source signals (Si) that includes said one or more interferer signals (Sn), wherein 0<
i≦
I>
1, 0<
j≦
J≧
I and 0<
n≦
N≧
1 are integers;weight means coupled to said input means for storing a plurality I of bias weights (Wi0) and a plurality I2 of scaling weights (Wij); output means coupled to said weight means for producing a plurality I of output signals (Ui) responsive to said input signals (Xj) such that (Ui)=(Wij) (Xj)+(Wi0); training means coupled to said output means for producing a plurality I of training signals (Yi) responsive to a transformation of said input signals (Xj) such that Yi =g(Ui), wherein g(x) is a nonlinear function and the Jacobian of said transformation is J=det(∂
Yi /∂
Xj);adjusting means coupled to said training means and said weight means for adjusting said bias weights (Wi0) and said scaling weights (Wij) responsive to one or more samples of said training signals (Yi) such that each said bias weight Wi0 is changed proportionately to a corresponding bias measure Δ
Wi0 accumulated over said one or more samples and each said scaling weight Wij is changed proportionately to a corresponding scaling measure Δ
Wij =ε
·
∂
(ln|J|)/∂
Wij accumulated over said one or more samples, wherein ε
>
0 is a learning rate; andfeedback means coupled to said output means and said input means for selecting one or more said output signals (Un) representing said one or more interferer signals (Sn) for combination with said input signals (Xj), thereby cancelling said interferer signals (Sn). - View Dependent Claims (14, 15)
-
Specification