Signal separation method, signal separation device, signal separation program and recording medium

US 20060058983A1
Filed: 09/01/2004
Published: 03/16/2006
Est. Priority Date: 09/02/2003
Status: Active Grant

First Claim

Patent Images

1. :

A signal separation method that separates and extracts signals under conditions where N (N≧

2) signals are mixed together and observed with M sensors, comprising;

a procedure that transforms the observed signal values observed by said sensors into frequency-domain signal values, a procedure that uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), a procedure that clusters said relative values into N clusters, a procedure that calculates a representative value for each of said clusters, a procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦

V≦

M) signal sources, a procedure that uses said mask to extract said mixed signal values from said frequency-domain signal values, and a procedure that separates and extracts the values of V signals from said mixed signal values.

View all claims

1 Assignment

Timeline View

Assignment View

0 Petitions

Accused Products

Abstract

This invention achieves high-quality separation of mixed signals in situations where the relationship between the number of signal sources N and the number of sensors M is such that N>M. First, the values of the observed signal observed by M sensors are transformed into frequency domain values, and these frequency domain values are used to calculate the relative values of the observed values between the sensors at each frequency. These relative values are clustered into N clusters, and the representative value of each cluster is calculated. Then, using these representative values, a mask is produced to extract the values of the signals emitted by V (1≦V≦M) signal sources from the frequency-domain signal values, and this mask is used to extract the signal values emitted by V signal sources from these frequency-domain signal values. After that, if V=1 then the limited signal is output directly as a separated signal, while if V≧2 then the separated values are obtained from this limited signal by subjecting it to separation techniques such as ICA.

Citations

31 Claims

1. :
- A signal separation method that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a procedure that transforms the observed signal values observed by said sensors into frequency-domain signal values, a procedure that uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), a procedure that clusters said relative values into N clusters, a procedure that calculates a representative value for each of said clusters, a procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
  
  V≦
  
  M) signal sources, a procedure that uses said mask to extract said mixed signal values from said frequency-domain signal values, and a procedure that separates and extracts the values of V signals from said mixed signal values.
- View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
- - 2. :
    - A signal separation method according to claim 1, wherein said mask is a function that takes a high level value for said relative values that are within a prescribed range that includes V said representative values, and takes a low level value for said representative values that are not inside said prescribed range, and wherein the procedure that uses said mask to extract said mixed signal values from said frequency-domain signal values is a procedure in which said frequency-domain signal values are multiplied by said mask.
  - 3. :
    - A signal separation method according to claim 1, wherein said mask is a function that takes a low level value for said relative values that are within a prescribed range that includes V said representative values, and takes a high level value for said representative values that are not inside said prescribed range, and wherein the procedure that uses said mask to extract said mixed signal values from said frequency-domain signal values is a procedure in which the values obtained by multiplying said frequency-domain signal values by said mask are subtracted from said frequency-domain signal values.
  - 4. :
    - A signal separation method according to claim 2, wherein said mask is a function that the transitions from said high level value to said low level value that accompany changes of said relative value occur in a continuous fashion.
  - 5. :
    - A signal separation method according to claim 1, wherein the procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
      
      V≦
      
      M) signal sources is a procedure whereby said mask is generated by using the directional characteristics of a null beamformer (NBF).
  - 6. :
    - A signal separation method according to claim 1, wherein the procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
      
      V≦
      
      M) signal sources includes;
      
      a procedure that generates an (N−
      
      V+1)×
      
      (N−
      
      V+1) delay matrix H_NBF(f) in which the element at (j,i) is equal to exp(j2π
      
      fτ
      
      _ji), where τ
      
      _ji=(d_j/v_e)cosθ
      
      _i, v_eis the velocity of the signals, d_jis the distance between sensor 1 and sensor j (j=1, . . . ,N−
      
      V+1), θ
      
      ₁is any one of the estimated directions of the signal sources corresponding to the V said representative values, θ
      
      _i(i=2, . . . ,N−
      
      V+1) are the estimated directions of the signal sources corresponding to the other said representative values of the V said representative values, and f is a frequency variable, a procedure that calculates the inverse matrix W(f)=H_NBF^−
      
      1(f) of delay matrix H_NBF(f) as a NBF matrix W(f), a procedure that generates a directional characteristics function $FORMULA 54$ $F (f, θ) = \sum_{k = 1}^{N - V - 1} W_{1 k} (f) \exp (j2π {fd}_{k} \cos θ / v_{\underline{e}})$ where θ
      
      is a signal arrival direction variable, and the first row element of said NBF matrix W(f) is W_1k(f), and a procedure that uses said directional characteristics function F(f,θ
      
      ) to generate said mask.
  - 7. :
    - A signal separation method according to claim 1, wherein the procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
      
      V≦
      
      M) signal sources includes;
      
      a procedure that generates a function consisting of a single-peak function convolved with a binary mask, which is a function that takes a high level value for said relative values that are within a prescribed range including V said representative values and takes a low level value for said representative values that are not inside said prescribed range and where changes of the relative value are accompanied by discontinuous transitions from said high level value to said low level value, and a procedure that generates said mask as a function in which said relative values are substituted into said function consisting of a single-peak function convolved with a binary mask.
  - 8. :
    - A signal separation method according to claim 1, wherein the procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
      
      =V≦
      
      =M) signal sources is a procedure that generates said mask as a single-peak function obtained by mapping the differences between a first odd function that takes a value of zero when said relative value is the lower limit value a_minin a prescribed range including V said representative values and a second odd function that takes a value of zero when said representative value is the upper limit value a_maxin said prescribed range.
  - 9. :
    - A signal separation method according to claim 2 or claim 3, wherein said mask is a function that transitions from said high level value to said low level value occur in a discontinuous fashion.

10. :
- A signal separation method that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a procedure that transforms the observed signal values observed by said sensors into frequency-domain signal values, a procedure that uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), a procedure that clusters said relative values into N clusters, a procedure that calculates a representative value for each of said clusters, a procedure that generates a mask function that takes a high level value for said relative values that are within a prescribed range that includes one of the said representative values, and takes a low level value for said representative values that are not inside said prescribed range, wherein the transitions from said high level value to said low level value that accompany changes of said relative value occur in a continuous fashion, and a procedure that multiplies said frequency-domain signal values by said mask to extract the signal emitted from one signal source.

11. :
- A signal separation method that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a procedure that transforms the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors a_i(f) to represent each said cluster C_i(f), a procedure that extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a₁(f), a procedure that generates a mask M(f,m) represented by the formula $FORMULA 55$ $M (f, m) = {\begin{matrix} 1 & \max_{a_{p} (f) \in G_{k}} D (X (f, m), a_{p} (f)) < \min_{a_{q} (f) \in G_{k}^{c}} D (X (f, m), a_{q} (f)) \\ 0 & otherwise \end{matrix}$ where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of G_k, and D(α
  
  ,β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  , and a procedure that extracts the signal values emitted from V of said signal sources by calculating the product of said mask M(f,m) and said first vectors X(f,m).
- View Dependent Claims (13, 14)
- - 13. :
    - A signal separation method according to claim 11 or claim 12, wherein said clustering procedure is performed after performing the calculation $FORMULA 57$ $sign (X_{j} (f, m)) \leftarrow {\begin{matrix} X_{j} (f, m) / \langle X_{j} (f, m) \rangle & (\langle X_{j} (f, m) \rangle \neq 0) \\ 0 & (\langle X_{j} (f, m) \rangle = 0) \end{matrix} and X (f, m) \leftarrow {\begin{matrix} X (f, m) / sign (X_{j} (f, m)) & (\langle X_{j} (f, m) \rangle \neq 0) \\ X (f, m) & (\langle X_{j} (f, m) \rangle = 0) \end{matrix}$
  - 14. :
    - A signal separation method according to claim 13, wherein said clustering procedure is performed after performing the calculation $FORMULA 58$ $X (f, m) \leftarrow {\begin{matrix} X (f, m) /  X (f, m)  & ( X (f, m)  \neq 0) \\ X (f, m) & ( X (f, m)  = 0) \end{matrix}$ (where the notation ∥
      
      X(f,m)∥
      
      denotes the norm of X(f,m)). After said formula $X (f, m) \leftarrow {\begin{matrix} X (f, m) / sign (X_{j} (f, m)) & (\langle X_{j} (f, m) \rangle \neq 0) \\ X (f, m) & (\langle X_{j} (f, m) \rangle = 0) \end{matrix}$

12. :
- A signal separation method that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, wherein a procedure that transforms the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors a₁(f) to represent each said cluster C_i(f), a procedure that extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a_i(f), and a procedure that judges whether or not said first vectors X(f,m) satisfy the relationship
  max_a_p_(f)¥
  
  G_kD(X(f,m),a_p(f))<
  
  min_a_q_(f)∈
  
  G_k_cD(X(f,m),a_q(f))
  
  Formula 56 where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of G_k, and D(α
  
  ,β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  , and, if so, extracts said first vectors X(f,m) as the signal values emitted from V of the said signal sources.

15. :
- A signal separation method that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising a procedure that transforms the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)]^Tcomprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors ai(f) to represent each said cluster Ci(f), a procedure that calculates an N-row×
  
  M-column separation matrix W(f,m) that is the Moore-Penrose pseudo-inverse matrix of an M-row×
  
  N-column matrix in which 0 or more of the N said second vectors a_i(f) are substituted with zero vectors, and a procedure that calculates a separated signal vector Y(f,m)=[Y₁(f,m), . . . ,Y_N(f,m)]^Tby performing the calculation Y(f,m)=W(f,m)X(f,m).
- View Dependent Claims (16, 17, 18, 19, 20)
- - 16. :
    - A signal separation method according to claim 15, wherein;
      
      the procedure that calculates said separation matrix W(f,m) is a procedure that selects min(M,N) said second vectors a_i(f), generates a matrix A′
      
      (f,m) whose columns are the selected min(M,N) said second vectors a_i(f) and max(N−
      
      M,0) zero vectors, and calculates said separation matrix W(f,m) as the Moore-Penrose pseudo-inverse matrix of said matrix A′
      
      (f,m).
  - 17. :
    - A signal separation method according to claim 15, wherein;
      
      the procedure used to calculate said separation matrix W(f,m) when N>
      
      M is a procedure that selects M said second vectors a_i(f) in each discrete time interval m, generates a matrix A′
      
      (f,m) whose columns are the selected M said second vectors a_i(f) and N−
      
      M zero vectors, and calculates said (time-dependent) separation matrix W(f,m) as the Moore-Penrose pseudo-inverse matrix of said matrix A′
      
      (f,m), and the procedure used to calculate said separation matrix W(f,m) when N≦
      
      M is a procedure that calculates the Moore-Penrose pseudo-inverse matrix of a matrix comprising N said second vectors in each said cluster C_i(f) to yield said (time-invariant) separation matrix W(f,m).
  - 18. :
    - A signal separation method according to claim 15, wherein;
      
      said clustering procedure is performed after performing the calculation $FORMULA 59$ $sign (X_{j} (f, m)) \leftarrow {\begin{matrix} X_{j} (f, m) / \langle X_{j} (f, m) \rangle & (\langle X_{j} (f, m) \rangle \neq 0) \\ 0 & (\langle X_{j} (f, m) \rangle = 0) \end{matrix} and X (f, m) \leftarrow {\begin{matrix} X (f, m) / sign (X_{j} (f, m)) & (\langle X_{j} (f, m) \rangle \neq 0) \\ X (f, m) & (\langle X_{j} (f, m) \rangle = 0) \end{matrix}$
  - 19. :
    - A signal separation method according to claim 18, wherein;
      
      said clustering procedure is performed after performing the calculation $FORMULA 60$ $X (f, m) \leftarrow {\begin{matrix} X (f, m) /  X (f, m)  & ( X (f, m)  \neq 0) \\ X (f, m) & ( X (f, m)  = 0) \end{matrix}$ (where the notation ∥
      
      X(f,m)∥
      
      denotes the norm of X(f,m)). in addition to said formula $X (f, m) \leftarrow {\begin{matrix} X (f, m) / sign (X_{j} (f, m)) & (\langle X_{j} (f, m) \rangle \neq 0) \\ X (f, m) & (\langle X_{j} (f, m) \rangle = 0) \end{matrix}$
  - 20. :
    - A signal separation method according to claim 16, wherein said procedure that selects min(M,N) said second vectors a_i(f) is a procedure that initializes fourth vectors e with said first vectors X(f,m), and then repeats a process min(M,N) times in which it selects said second vectors a_q(u)(f) that maximize the absolute value of the dot product of a_q(u)(f)/∥
      
      a_q(u)(f)∥ and
      
      said fourth vectors, sets up a matrix Q=[a_q(1)(f), . . . ,a_q(k)(f)] representing the subspace subtended by all said second vectors a_q(u)(u=1, . . . ,k) selected so far, performs the calculation P=Q(Q^HQ)^−
      
      1Q^H, and updates the fourth vectors e with e=X(f,m)−
      
      P·
      
      X(f,m).

21. :
- A signal separation device that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a memory unit that stores the observed signal values observed by said sensors;
  
  and a processor which is connected to said memory unit and performs processing whereby it;
  
  transforms said observed signal values into frequency-domain signal values, uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), clusters said relative values into N clusters, calculates a representative value for each of said clusters, uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
  
  V≦
  
  M) signal sources, uses said mask to extract said mixed signal values from said frequency-domain signal values, and separates and extracts the values of V signals from said mixed signal values.

22. :
- A signal separation device that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a memory unit that stores the observed signal values observed by said sensors;
  
  and a processor which is connected to said memory unit and performs processing whereby it;
  
  transforms said observed signal values into frequency-domain signal values, uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), clusters said relative values into N clusters, calculates a representative value for each of said clusters, generates a mask, which is a function that takes a high level value for said relative values that are within a prescribed range that includes one said representative value, and takes a low level value for said representative values that are not inside said prescribed range, and where the transitions from said high level value to said low level value that accompany changes of said relative value occur in a continuous fashion, and extracts the values of a signal emitted from one signal source by multiplying said frequency-domain values by said mask.

23. :
- A signal separation device that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a memory unit that stores the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors;
  
  and a processor which is connected to said memory unit and performs processing whereby it;
  
  transforms said observed signal values x₁(t), . . . ,x_M(t) into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, calculates second vectors a₁(f) to represent each said cluster C_i(f), and extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a_i(f), generates a mask M(f,m) represented by the formula $FORMULA 61$ $M (f, m) = {\begin{matrix} 1 & \max_{a_{p} (f) \in G_{k}} D (X (f, m), a_{p} (f)) < \min_{a_{q} (f) \in G_{k}^{c}} D (X (f, m), a_{q} (f)) \\ 0 & otherwise \end{matrix}$ where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of G_k, and D(α
  
  ,β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  and extracts the signal values emitted from V of the said signal sources by calculating the product of said mask M(f,m) and said first vectors X(f,m).

24. :
- A signal separation device that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a memory unit that stores the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors;
  
  and a processor which is connected to said memory unit and performs processing whereby it;
  
  transforms said observed signal values x₁(t), . . . ,x_M(t) into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, calculates second vectors ai(f) to represent each said cluster C_i(f), extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a_i(f), judges whether or not said first vectors satisfy the relationship
  max_a_p_(f)∈
  
  G_kD(X(f,m),a_p(f))<
  
  min_a_q_(f)∈
  
  G_k_cD(X(f,m),a_q(f))
  
  Formula 62 where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of Gk, and D(α
  
  , β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  , and, if so, extracts said first vectors X(f,m) as the signal values emitted from V of the said signal sources.

25. :
- A signal separation device that separates and extracts signals under conditions where N (N≧
  
  2) signals are mixed together and observed with M sensors, comprising;
  
  a memory unit that stores the observed signal values x₁(t), . . . ,x_M(t) observed by said sensors;
  
  and a processor which is connected to said memory unit and performs processing whereby it;
  
  transforms said observed signal values x₁(t), . . . ,x_M(t) into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)]^Tcomprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, calculates second vectors a_i(f) to represent each said cluster C_i(f), calculates an N-row×
  
  M-column separation matrix W(f,m) that is the Moore-Penrose pseudo-inverse matrix of an M-row×
  
  N-column matrix in which 0 or more of the N said second vectors a_i(f) are substituted with zero vectors, and calculates a separated signal vector Y(f,m)=[Y₁(f,m), . . . ,Y_N(f,m)]^Tby performing the calculation Y(f,m)=W(f,m)X(f,m).

26. :
- A signal separation program that causes a computer to perform;
  
  a procedure that transforms observed signal values, which are mixtures of N (N≧
  
  2) signals observed with M sensors, into frequency-domain values, a procedure that uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), a procedure that clusters said relative values into N clusters, a procedure that calculates a representative value for each of said clusters, a procedure that uses said representative values to generate a mask for the purpose of extracting, from said frequency-domain signal values, mixed signal values comprising the signals emitted from V (2≦
  
  V≦
  
  M) signal sources, a procedure that uses said mask to extract said mixed signal values from said frequency-domain signal values, and a procedure that separates and extracts the values of V signals from said mixed signal values.
- View Dependent Claims (31)
- - 31. :
    - A computer-readable recording medium that stores a signal separation program according to any one of claims 26 through 30.

27. :
- A signal separation program that causes a computer to perform;
  
  a procedure that transforms observed signal values, which are mixtures of N (N≧
  
  2) signals observed with M sensors, into frequency-domain values, a procedure that uses said frequency-domain signal values to calculate at each frequency the relative values of the observed values between said sensors (including mapping these relative values), a procedure that clusters said relative values into N clusters, a procedure that calculates a representative value for each of said clusters, a procedure that generates a mask, which is a function that takes a high level value for said relative values that are within a prescribed range that includes one of said representative values, and takes a low level value for said representative values that are not inside said prescribed range, wherein the transitions from said high level value to said low level value that accompany changes of said relative value occur in a continuous fashion, and a procedure that extracts the signal values emitted from one signal source by multiplying said frequency-domain signal values by said mask.

28. :
- A signal separation program that causes a computer to perform;
  
  a procedure that transforms observed signal values x₁(t), . . . ,x_M(t), which are mixtures of N (N≧
  
  2) signals observed by M sensors, into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors a_i(f) to represent each said cluster C_i(f), a procedure that extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a_i(f), a procedure that generates a mask M(f,m) represented by the formula $FORMULA 63$ $M (f, m) = {\begin{matrix} 1 & \max_{a_{p} (f) \in G_{k}} D (X (f, m), a_{p} (f)) < \min_{a_{q} (f) \in G_{k}^{c}} D (X (f, m), a_{q} (f)) \\ 0 & otherwise \end{matrix}$ where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of G_k, and D(α
  
  ,β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  , and a procedure that extracts the signal values emitted from V of said signal sources by calculating the product of said mask M(f,m) and said first vectors X(f,m).

29. :
- A signal separation program that causes a computer to perform;
  
  a procedure that transforms observed signal values x₁(t), . . . ,x_M(t), which are mixtures of N (N≧
  
  2) signals observed by M sensors, into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)] comprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors a_i(f) to represent each said cluster C_i(f), a procedure that extracts V (1≦
  
  V≦
  
  M) third vectors a_p(f) (p=1, . . . ,V) from said second vectors a_i(f), and a procedure that judges whether or not said first vectors X(f,m) satisfy the relationship
  max_a_p_(f)∈
  
  G_kD(X(f,m),a_p(f))<
  
  min_a_q_(f)∈
  
  G_k_cD(X(f,m),a_q(f))
  
  Formula 64 where G_kis the set of said third vectors a_p(f), G_k^cis the complementary set of G_k, and D(α
  
  ,β
  
  ) is the Mahanalobis square distance between the vectors α and
  
  β
  
  , and, if so, extracts said first vectors X(f,m) as the signal values emitted from V of the said signal sources.

30. :
- A signal separation program that causes a computer to perform;
  
  a procedure that transforms observed signal values x₁(t), . . . ,x_M(t), which are mixtures of N (N≧
  
  2) signals observed by M sensors, into frequency-domain signal values X₁(f,m), . . . ,X_M(f,m), a procedure that clusters first vectors X(f,m)=[X₁(f,m), . . . ,X_M(f,m)]^Tcomprising said frequency-domain signal values X₁(f,m), . . . ,X_M(f,m) into N clusters C_i(f) (i=1, . . . ,N) at each frequency f, a procedure that calculates second vectors a_i(f) to represent each said cluster C_i(f), a procedure that calculates an N-row×
  
  M-column separation matrix W(f,m) that is the Moore-Penrose pseudo-inverse matrix of an M-row×
  
  N-column matrix in which 0 or more of the N said second vectors a_i(f) are substituted with zero vectors, and a procedure that calculates a separated signal vector Y(f,m)=[Y₁(f,m), . . . ,Y_N(f,m)]^Tby performing the calculation Y(f,m)=W(f,m)X(f,m).

Specification

Resources

Litigation Campaign Assessment

Current Assignee
Nippon Telegraph and Telephone Corporation
Original Assignee
Nippon Telegraph and Telephone Corporation
Inventors
Sawada, Hiroshi, Mukai, Ryo, Makino, Shoji, Araki, Shoko

Granted Patent

US 7,496,482 B2
Time in Patent Office

Days
Field of Search
US Class Current

702/190
CPC Class Codes

G06F 18/21347 using domain transformations

G10L 21/0272 Voice signal separating

Signal separation method, signal separation device, signal separation program and recording medium

First Claim

1 Assignment

0 Petitions

Accused Products

Abstract

Citations

31 Claims

Specification

Solutions

Use Cases

Quick Links

Signal separation method, signal separation device, signal separation program and recording medium

First Claim

1 Assignment

Subscription Required

Subscription Required

0 Petitions

Subscription Required

Accused Products

Subscription Required

Abstract

Citations

31 Claims

Specification

Subscription Required

Solutions

Use Cases

Quick Links