Blind signal separation

• US 6,625,587 B1
• Filed: 03/10/2000
• Issued: 09/23/2003
• Est. Priority Date: 06/18/1997
• Status: Expired due to Term

• Current Assignee: CSR Technology Incorporated (Qualcomm, Inc.)
• Original Assignee: Clarity LLC
• Inventors: Salam, Fathi M., Erten, Gamze
• Primary Examiner(s): Starks, Jr., Wilbert L.
• Assistant Examiner(s): Hirl, Joseph P.
• Application Number:
  US09/445,778
• Time in Patent Office: 1,292 Days
• Field of Search: 706/22, 706/21, 706/24, 367/134
• US Class Current: 706/22
• CPC Class Codes: G06F 18/2134 based on separation criteria, e.g. independent component analysis
• CPC Class Codes: G10L 21/0272 Voice signal separating
• CPC Class Codes: H03H 21/0012 Digital adaptive filters

Abstract

Signal separation processing includes a signal separation architecture defining a relationship between at least one input signal and at least one output signal. The signal separation architecture includes a state space representation that establishes a relationship between the input and output signals. At least one output signal is based on the signal separation architecture.

Claims

1. A signal processing system, comprising:

at least one sensor for detecting a plurality of input signals comprising at least one source signal;

a storage device coupled to the at least one sensor for storing the input signals;

an architecture processor coupled to the storage device for defining and determining a signal separation architecture, the signal separation architecture defining a relationship between the plurality of input signals and at least one output signal, the relationship having constant parameters and time-varying parameters;

an update processor coupled to the architecture processor for determining a rate of change for the time-varying parameters in response to an associated rate of change; and

an output processor coupled to the architecture processor for providing the at least one output signal based on the signal separation architecture, the constant parameters, and the time-varying parameters, the at least one output signal estimating the at least one source signal;

wherein the signal separation architecture models a mixing environment and a separation environment as a linear dynamic system, wherein a mixing model is of the form

2. The signal processing system of claim 1, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-D_ 



-1

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{D}}^{\text{\hspace{1em}<br><br>}<br><br>-1}$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{\mathrm{BD}}}^{\text{\hspace{1em}<br><br>}<br><br>-1}<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{\mathrm{BD}}}^{\text{\hspace{1em}<br><br>}<br><br>-1}.$

3. The signal processing system of claim 1, wherein a number of measurements is greater than a number of source signals, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-[D_T

D_]-1

D_T

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>-{\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T.$

4. The signal processing system of claim 1, wherein the mixing model is represented by a time domain relation $m$

(t)=D_

s

(t)+∑

Lk=1

C_k

s(k)

(t)-∑

Nk=1

A_k

m(k)

(t),

wherein k represents the kth derivative in the continuous time case and k delay samples in the discrete time case.

5. The signal processing system of claim 4, wherein the signal separation architecture is of a form $u$

(t)=Dm

(t)+∑

L′

k=1

Ck

m(k)

(t)-∑

N′

k=1

Ak

u(k)

(t),

wherein dimensions of C and D are n×

m, and dimensions of A are n×

n, wherein

6. The signal processing system of claim 1, wherein computing parameter matrix C comprises using at least one weight update rule selected from a group comprising

7. The signal processing system of claim 6, wherein computing parameter matrix D comprises using at least one weight update rule selected from a group comprising

8. The signal processing system of claim 1, wherein at least one signal separation process is overlapped in time with at least one other signal separation process.

9. The signal processing system of claim 1, wherein at least one signal separation process uses adaptive parameter estimation, wherein the adaptive parameter estimation is performed using at least one select data set of a data stack, wherein the at least one select data set is controlled by at least one result of the at least one signal separation process.

10. The signal processing system of claim 1, wherein the at least one sensor is arranged in a sensor array, the sensor array having a directional response pattern, wherein the directional response pattern is capable of being modified by performing signal processing on the plurality of input signals.

11. The signal processing system of claim 1, wherein a quantity of the plurality of input signals and a quantity of the at least one output signal are not equal.

12. The signal processing system of claim 11, wherein at least one output signal is a function of at least two source signals.

13. The signal processing system of claim 11, wherein at least two output signals are functions of a same source signal.

14. The signal processing system of claim 1, wherein the provision of the at least one output signal is based on a plurality of internal states of the signal processing system.

15. A signal processing system, comprising:

at least one sensor for detecting a plurality of input signals comprising at least one source signal;

an architecture processor in communication with the at least one sensor for determining a signal separation architecture, the signal separation architecture defining a relationship between the plurality of input signals and at least one output signal, wherein the signal separation architecture comprises a state space representation that establishes the relationship between the plurality of input signals and the at least one output signal of the form

16. The signal processing system of claim 15, wherein the provision of the at least one output signal is further based on at least one previously computed state of the state space representation.

17. The signal processing system of claim 15, wherein the provision of the at least one output signal is further based on at least one current state and previously computed state of the state space representation.

18. The signal processing system of claim 15, wherein the state space representation is mapped onto a finite impulse response (FIR) filter.

19. The signal processing system of claim 15, wherein the state space representation is mapped onto an infinite impulse response (IIR) filter.

20. The signal processing system of claim 15, wherein the state space representation is generalized to a nonlinear time variant function.

21. A signal separation processing method, comprising:

receiving and storing a plurality of input signals comprising at least one source signal;

defining a signal separation architecture, the signal separation architecture defining a relationship between the plurality of input signals and at least one output signal;

initializing constant and time varying parameters of the relationship;

determining the time varying parameters in response to the rate of change associated with each time varying parameter; and

providing the at least one output signal based on the signal separation architecture and the constant and time varying parameters, the at least one output signal estimating the at least one source signal;

wherein the signal separation architecture models a mixing environment and a separation environment as a linear dynamic system, wherein a mixing model is of the form

22. The signal separation processing method of claim 21, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-D_ 



-1

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{D}}^{\text{\hspace{1em}<br><br>}<br><br>-1}$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{\mathrm{BD}}}^{\text{\hspace{1em}<br><br>}<br><br>-1}<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{\mathrm{BD}}}^{\text{\hspace{1em}<br><br>}<br><br>-1}.$

23. The signal separation processing method of claim 21, wherein a number of measurements is greater than a number of source signals, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-[D_T

D_]-1

D_T

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>-{\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T.$

24. The signal separation processing method of claim 21, wherein the mixing model is represented by a time domain relation $m$

(t)=D_

s

(t)+∑

Lk=1

C_k

s(k)

(t)-∑

Nk=1

A_k

m(k)

(t),

wherein k represents the kth derivative in the continuous time case and k delay samples in the discrete time case, s(t) is a source signal, m(t) is a mixture signal, and u(t) is a separated output signal, wherein the signal separation architecture is of a form $u<br><br>\left(t\right)=\mathrm{Dm}<br><br>\left(t\right)+\underset{k=1}{\stackrel{L^{\mathrm{\prime <br><br>}}}{\sum <br><br>}}<br><br>C_k<br><br>m^{\left(k\right)}<br><br>\left(t\right)-\underset{k=1}{\stackrel{N^{\mathrm{\prime <br><br>}}}{\sum <br><br>}}<br><br>A_k<br><br>u^{\left(k\right)}<br><br>\left(t\right),$

wherein dimensions of C and D are n×

m, and dimensions of A are n×

n, wherein

25. The signal separation processing method of claim 21, wherein the rate of change of at least one set of time varying parameters is contained in at least one array having at least two dimensions.

26. The signal separation processing method of claim 25, wherein the at least one array comprises a function of an outer product of a linear expansive, or compressive function of a set of the output signals arranged in a first one dimensional array and a linear, expansive, or compressive function of a set of the output signals arranged in a second one dimensional array.

27. A signal separation processing method, comprising:

receiving a plurality of input signals resulting from at least one source signal;

defining a signal separation architecture, the signal separation architecture defining a relationship between the plurality of input signals and at least one output signal, wherein the signal separation architecture comprises a state space representation that establishes a relationship between the plurality of input signals and the at least one output signal of the form

28. The signal separation processing method of claim 27, wherein a rate of change of at least one set of time varying parameters is contained in at least one array having at least two dimensions.

29. The signal separation processing method of claim 28, wherein the at least one array comprises a function of an outer product of a linear, expansive, or compressive function of a set of the output signals arranged in a first one dimensional array and a linear, expansive, or compressive function of a set of internal states of the state space architecture arranged in a second one dimensional array.

30. The signal separation processing method of claim 28, wherein the at least one array comprises a function of an outer product of at least one linear, expansive, or compressive function of a set of the output signals arranged in a first array having at least two dimensions and at least one linear, expansive, or compressive function of a set of the output signals arranged in a second array having at least two dimensions.

31. The signal separation processing method of claim 27, wherein a plurality of signal separation process operations are overlapped in time.

32. The signal separation processing method of claim 31, wherein at least one of the plurality of signal separation processes uses a predetermined set of constant values or a random set of numbers for initializing parameters.

33. The signal separation processing method of claim 31, wherein at least one of the plurality of signal separation processes uses parameters computed previously by another method overlapping in time.

34. The signal separation processing method of claim 31, wherein at least one of the plurality of signal separation processes is terminated.

35. The signal separation processing method of claim 34, wherein at least one of the plurality of signal separation processes uses parameters computed by at least one previously terminated signal separation process.

36. An acoustic signal discrimination system, comprising:

a plurality of acoustic sensors for detecting a plurality of input signals, wherein each input signal comprises different mixtures of a plurality of source signals;

a storage device coupled to the plurality of acoustic sensors for storing the plurality of input signals;

an architecture processor coupled to the storage device for determining an acoustic signal discrimination architecture that defines a relationship between the plurality of input signals and at least one output signal, the relationship comprising constant parameters and time varying parameters;

an update processor coupled to the architecture processor for determining a rate of change for the time varying parameters in response to an associated rate of change; and

an output processor coupled to the architecture processor for providing the at least one output signal based on the acoustic signal discrimination architecture, the constant parameters, and the time varying parameters, the at least one output signal estimating the at least one source signal;

wherein the acoustic signal discrimination architecture models a mixing environment and a separation environment as a linear dynamic system, wherein a mixing model is of the form

37. The acoustic signal discrimination system of claim 36, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-D_-1

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{D}}^{-1}$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{B<br><br>\text{\hspace{1em}<br><br>}<br><br>D}}^{-1}<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{B<br><br>\text{\hspace{1em}<br><br>}<br><br>D}}^{-1}.$

38. The acoustic signal discrimination system of claim 36, wherein a number of measurements is greater than a number of source signals, wherein inversion relationships between the parameter matrices are of the form $C$

=Δ



-[D_T

D_]-1

D_T

C_$D<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>-{\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T$$A<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>\stackrel{\_}{A}-{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{C},\text{\hspace{1em}<br><br>}<br><br>\mathrm{and}$$B<br><br>\stackrel{\mathrm{\Delta <br><br>}}{=}<br><br>{\stackrel{\_}{B}<br><br>\left[{\stackrel{\_}{D}}^T<br><br>\stackrel{\_}{D}\right]}^{-1}<br><br>{\stackrel{\_}{D}}^T.$

39. The acoustic signal discrimination system of claim 36, wherein the mixing model is represented by a time domain relation $m$

(t)=D_

s

(t)+∑

k=1L

 



C_k

s(k)

(t)-∑

k=1N

 



A_k

m(k)

(t),

wherein k represents the kth derivative in the continuous time case and k delay samples in the discrete time case, s(t) is a source signal, m(t) is a mixture signal, and u(t) is a separated output signal, wherein the signal separation architecture is of a form $u<br><br>\left(t\right)=D<br><br>\text{\hspace{1em}<br><br>}<br><br>m<br><br>\left(t\right)+\underset{k=1}{\overset{L^{\mathrm{\prime <br><br>}}}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>C_k<br><br>m^{\left(k\right)}<br><br>\left(t\right)-\underset{k=1}{\overset{N^{\mathrm{\prime <br><br>}}}{\sum <br><br>}}<br><br>\text{\hspace{1em}<br><br>}<br><br>A_k<br><br>u^{\left(k\right)}<br><br>\left(t\right),$

wherein dimensions of C and D are n×

m, and dimensions of A are n×

n, wherein

40. The acoustic signal discrimination system of claim 36, wherein the plurality of acoustic sensors comprises an acoustic sensor having a directional response pattern, wherein the directional response pattern is capable of being modified.

41. The acoustic signal discrimination system of claim 36, wherein a quantity of the plurality of input signals and a quantity of the at least one output signal are not equal.

42. The acoustic signal discrimination system of claim 41, wherein at least one output signal is a function of at least two source signals.

43. The acoustic signal discrimination system of claim 41, wherein at least two output signals are functions of the same source signal.

44. The acoustic signal discrimination system of claim 36, wherein provision of the at least one output signal is further based on internal states computed on at least one previous time instant.

45. An acoustic signal discrimination system, comprising:

at least one acoustic sensor for detecting a plurality of input signals, wherein each input signal comprises different mixtures of a plurality of source signals;

a storage device coupled to the at least one acoustic sensor for storing the plurality of input signals;

an architecture processor coupled to the storage device for determining an acoustic signal discrimination architecture that defines a relationship between the plurality of input signals and at least one output signal, the relationship comprising constant parameters and time varying parameters, the acoustic signal discrimination architecture is defined by at least one state space representation that establishes the relationship between input signals and output signals, each state space representation of the form

46. The acoustic signal discrimination system of claim 45, wherein provision of the at least one output signal is further based on at least one current state of the acoustic signal discrimination architecture.

47. The acoustic signal discrimination system of claim 45, wherein provision of the at least one output signal is further based on at least one state of the acoustic signal discrimination architecture computed on at least one previous time instant.

48. The acoustic signal discrimination system of claim 45, wherein provision of the at least one output signal is further based on at least one current state and at least one state of the acoustic signal discrimination architecture computed on at least one previous time instant.

49. The acoustic signal discrimination system of claim 45, wherein the at least one state space representation is mapped onto a finite impulse response (FIR) filter.

50. The acoustic signal discrimination system of claim 45, wherein the at least one state space representation is mapped onto an infinite impulse response (IIR) filter.

51. The acoustic signal discrimination system of claim 45, wherein the at least one state space representation is generalized to a nonlinear time variant function.