Adaptive noise cancellation device

• US 5,568,558 A
• Filed: 12/03/1993
• Issued: 10/22/1996
• Est. Priority Date: 12/02/1992
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Ramm, Dov, Chazan, Dan
• Primary Examiner(s): Coles, Sr., Edward L.
• Assistant Examiner(s): Grant, II, Jerome
• Application Number:
  US08/164,912
• Time in Patent Office: 1,054 Days
• Field of Search: 381/71, 381/94, 364/574
• US Class Current: 381/94.4
• CPC Class Codes: H03H 21/0027 filtering in the frequency domain
• CPC Class Codes: H04B 1/123 using adaptive balancing or compensation means adaptive filter circuits and algorithms H03H

Abstract

An adaptive noise cancellation device comprises: convolution logic (10) for convolving the signal from a reference input (x) with a discretized L-tap filter to form a filtered reference signal; and logic (20) for subtracting the filtered reference signal from a signal input to form an output signal; logic for generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and logic for repeatedly determining the coefficients C_k which minimize the power in the output signal (z), characterized in that N is less than the number of filter taps L and the basis functions have respective values over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains, in an embodiment they are gaussian. A full-duplex speakerphone is disclosed including such a noise cancellation device.

Claims

1. An adaptive noise cancellation device having a signal input (y), a reference input (x) and a signal output (z), the device comprising:

convolution logic (10) for convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

logic (20) for subtracting the filtered reference signal from the signal input to form the output signal;

logic for generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

logic for repeatedly determining the coefficients C_k which minimize the power in the output signal (z),

said N is less than the number of filter taps L and said N basis functions have respective values over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains

wherein said N basis functions have the form;

space="preserve" listing-type="equation">W.sub.k (t)=w(t)exp {jk(Δ

ω

)t}

for N values of an integer k, where w(t) is a window function having a portion of finite width, outside of which portion the function is substantially zero, both in the frequency and time domains.

2. An adaptive noise cancellation device as claimed in claim 1 wherein the window function w(t) is a gaussian function.

3. An adaptive noise cancellation device as claimed in claim 1 wherein the window function has the form:

space="preserve" listing-type="equation">w(t)=exp{-(tf.sub.s /γ

).sup.2 }

with Δ

ω

=2π

f_S /N and γ

in the range 0.25N to 0.75N.

4.

4. Adaptive noise cancellation device as claimed in claim 1 wherein N is a number one half of said filter length L.

5. An adaptive noise cancellation device having a signal input (y), a reference input (x) and a signal output (z), the device comprising:

convolution logic (10) for convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

logic (20) for subtracting the filtered reference signal from the signal input to form the output signal;

logic for generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

logic for repeatedly determining the coefficients C_k which minimize the power in the output signal (z),

said N is less than the number of said filter taps L and said N basis functions have respective values over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains,

wherein the logic for generating the filter response function comprises;

correlation logic (30) for generating the cross-correlation between said signal input (y) and said reference input (x) and for generating the autocorrelation of the reference input (x);

logic (40) for calculating the right hand side of a system of linear equations for the coefficients C_k using the equation

space="preserve" listing-type="equation">r.sub.i =G.sub.xy (iΔ

ω

)

where G_xy (ω

) is the Fourier transform of product of the window function w(t) and the cross-correlation between said signal input (y) and said reference input (x);

said logic (40) for calculating the coefficients of the system of linear equations using the equation, ##EQU12## where G_xx (ω

,μ

) is the Fourier transform of the product of the autocorrelation of the said reference input (x) and an auxiliary function

space="preserve" listing-type="equation">k(t,μ

)=F.sup.-1 [W*(ω

-μ

Δ

ω

)W(ω

+μ

Δ

ω

)]

where W(ω

) is the Fourier transform of the window function w(t); and

equation solving logic (50) for solving the equations a_ik C_k =r_i for C_k.

6. An adaptive noise cancellation device as claimed in claim 5 wherein said correlation logic (30) comprises:

logic for segmenting said signal input (y) and said reference input (x) into frames of a length equal to the filter length L;

logic for generating a Fourier transform of the frames;

logic for combining the Fourier transforms of the frames with a current segment of the auto-correlation function and cross-correlation functions weighted by a factor α

, where α

is between 0 and 1, to form an updated segment of the cross-correlation and auto-correlation functions.

7. An adaptive noise cancellation device as claimed in claim 6 comprising control logic (60) for calculating the factor α

as an increasing function of the ratio of the power or amplitude of said reference input (x) and said signal output (z).

8. A speakerphone capable of full-duplex operation having a signal input and a signal output (z) comprising:

an adaptive noise cancellation device (90);

a microphone (100) arranged to provide the signal input (y) to said adaptive noise cancellation device;

and a loudspeaker (110) connected to said signal input of the speakerphone, wherein said signal input of said speakerphone provides said reference input (x) for said adaptive noise cancellation device and said signal output (z) of said adaptive noise cancellation device provides said signal output (z) of said speakerphone;

said adaptive noise cancellation device including;

convolution logic (10) for convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

logic (20) for subtracting the filtered reference signal from the signal input to form the output signal;

logic for generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

logic for repeatedly determining the coefficients C_k which minimize the power in the output signal (z),

Said N is less than the number of said filter taps L and said N basis functions have respective values over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains,

wherein said N basis functions have the form;

space="preserve" listing-type="equation">Wk(t)=w(t)exp{jk(Δ

ω

)t}

for N values of an integer k, where w(t) is a window function having a portion of finite width, outside of which portion the function is substantially zero, both in the frequency and time domains.

9. A method for adaptive noise cancellation of noise in a signal input (y) by using a reference input (x) to provide a signal output (z), comprising the steps of:

convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

subtracting the filtered reference signal from the signal input to form said output signal (z);

generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

repeatedly determining the coefficients C_k which minimize the power in said output signal (z),

wherein said N is less than the number of said filter taps L and said N basis functions have respective value over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains, wherein said N basis functions have the form;

space="preserve" listing-type="equation">Wk(t)=w(t)exp{jk(Δ

ω

) t}

for N values of an integer k, where w(t) is a window function having a portion of finite width, outside of which portion the function is substantially zero, both in the frequency and time domains.

10. The method of claim 9 wherein said window function w(t) is a gaussian function.

11. The method of claim 10 wherein said window function has the form:

space="preserve" listing-type="equation">w(t)=exp{-(tf.sub.s /γ

).sup.2 }

with Δ

ω

=2π

f_S /N and γ

in the range 0.25N to 0.75.

12. A method for adaptive noise cancellation of noise in a signal input (y) by using a reference input (x) to provide a signal output (z), comprising the steps of:

convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

subtracting the filtered reference signal from the signal input to form said output signal (z);

generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

repeatedly determining the coefficients C_k which minimize the power in said output signal (z),

wherein said N is less than the number of said filter taps L and said N basis functions have respective value over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains,

wherein said step of generating the filter taps includes the steps of;

generating the cross-correlation between said signal input (y) and said reference input (x) and generating the autocorrelation of the reference input (x);

calculating the right hand side of a system of linear equations for the coefficients C_k using the equation

space="preserve" listing-type="equation">r.sub.i =G.sub.xy (iΔ

ω

),

where G_xy (ω

) is the Fourier transform of product of the window function w(t) and the cross-correlation between said signal input (y) and said reference input (x);

said step of calculating the coefficients of the system of linear equations including using the equation, ##EQU13## where G_xx (ω

,μ

) is the Fourier transform of the product of the autocorrelation of the said reference input (x) and an auxiliary function

space="preserve" listing-type="equation">k(t, μ

)=F.sup.-1 [W*(ω

-μ

Δ

ω

)W(ω

+μ

Δ

ω

)]

where W(ω

) is the Fourier transform of the window function w(t); and

solving the equations a_ik C_k =r_i for C_k.

13. The method of claim 12 wherein said step of generating the cross-correlation includes the step of:

segmenting said signal input (y) and said reference input (x) into frames of a length equal to the filter length L;

generating a Fourier transform of the frames;

logic for combining the Fourier transforms of the frames with a current segment of the auto-correlation function and cross-correlation functions weighted by a factor α

, where α

is between 0 and 1, to form an updated segment of the cross-correlation and auto-correlation functions.

14. The method of in claim 13 further including the step of calculating the factor α

as an increasing function of the ratio of the power or amplitude of said reference input (x) and said signal output (z).

15. A method for adaptive noise cancellation of noise in a signal input (y) by using a reference input (x) to provide a signal output (z), comprising the steps of:

convolving the signal from the reference input (x) with a discretised L-tap filter to form a filtered reference signal; and

subtracting the filtered reference signal from the signal input to form said output signal (z);

generating the filter taps as a linear combination of N basis functions each having a corresponding coefficient C_k ; and

repeatedly determining the coefficients C_k which minimize the power in said output signal (z),

wherein said N is less than the number of said filter taps L 14 and said N basis functions have respective value over a portion of finite width, outside of which portion the functions are substantially zero, both in the frequency and time domains,

wherein N is selected to be a number one half of said filter length L.