Frequency domain adaptation with dynamic step size adjustment based on analysis of statistic of adaptive filter coefficient movement

Patent metadata

Field	Value
Publication number	US-10984778-B2
Application number	US-201916516644-A
Country	US
Kind code	B2
Filing date	Jul 19, 2019
Priority date	Jul 19, 2019
Publication date	Apr 20, 2021
Grant date	Apr 20, 2021

Abstract

An adaptive filter includes a frequency domain adaptation block that analyzes a statistic of coefficient movement in the frequency domain. The adaption block adjusts, in the frequency domain, a parameter (step size or leakage factor) that affects speed of convergence of the adaptive filter based on the analyzed statistic of filter coefficient movement. The filter includes an associated coefficient, statistic of coefficient movement, and parameter for each frequency bin. The coefficients may be complex numbers, and separate real and imaginary statistics and parameters are maintained. The statistic may be direction counts of the filter coefficient movement. The step size may be adjusted to a predetermined minimum value when the current direction of movement of the filter coefficient is different than the predominant direction and otherwise the step size is adjusted approximately proportionally to an amount of predominance by a value based on a direction count of the filter coefficient movement.

First claim

The invention claimed is:

1. A method, comprising:
analyzing, in a frequency domain, a statistic of coefficient movement of an adaptive filter; and
adjusting, in the frequency domain, a parameter that affects speed of convergence of the adaptive filter based on the analyzed statistic of filter coefficient movement; and
wherein the statistic of coefficient movement comprises a measure of an attribute of a sample of a change of adaptive filter coefficients between a current and next value of the coefficients.

2. The method of claim 1 ,
wherein the adaptive filter includes, for each frequency bin of a plurality of frequency bins, an associated:
coefficient;
statistic of coefficient movement; and
parameter that affects speed of convergence of the adaptive filter;

wherein said analyzing the statistic in the frequency domain comprises, for each frequency bin of the plurality of frequency bins, analyzing the statistic of coefficient movement associated with the frequency bin; and
wherein said adjusting the parameter in the frequency domain comprises, for each frequency bin of the plurality of frequency bins, adjusting the parameter associated with the frequency bin based on the analyzed statistic associated with the frequency bin.

3. The method of claim 2 ,
wherein said adjusting the parameter comprises decreasing the step size when the frequency bin is above a predetermined upper frequency bin.

4. The method of claim 2 ,
determining a current maximum distance of the filter coefficient movement among the frequency bins and maintaining a smoothed version of the maximum distance;
wherein said analyzing comprises determining a condition that is whether the current maximum distance exceeds a product of the smoothed version and a scaling factor; and
wherein said adjusting the parameter comprises setting the parameter to a value that causes the speed of convergence to be approximately zero when the condition is satisfied.

5. The method of claim 1 ,
wherein coefficients comprise complex numbers having a real part and an imaginary part;
wherein the statistic of coefficient movement comprises separate statistics for the real part of the coefficients and the imaginary part of the coefficients; and
wherein the parameter comprises separate parameters for the real part of the coefficients and the imaginary part of the coefficients.

6. The method of claim 1 ,
wherein the statistic comprises direction counts of the filter coefficient movement.

7. The method of claim 1 ,
wherein the parameter that affects speed of convergence is a step size.

8. The method of claim 7 ,
wherein said adjusting the parameter based on the analyzed statistic comprises adjusting the step size approximately proportionally to an amount of a predominance of a direction of movement of the filter coefficient.

9. The method of claim 8 ,
wherein said adjusting the parameter based on the analyzed statistic further comprises adjusting the step size by a predetermined minimum value when a current direction of coefficient movement does not conform to the predominance of the direction of movement of the filter coefficient.

10. The method of claim 8 ,
wherein said adjusting the step size approximately proportionally to an amount of a predominance of a direction of movement of the filter coefficient comprises adjusting the step size by a value based on a direction count of the movement of the filter coefficient.

11. The method of claim 1 ,
wherein the parameter that affects speed of convergence is a leakage parameter.

12. The method of claim 11 ,
wherein said adjusting the parameter based on the analyzed statistic comprises adjusting the leakage parameter approximately inversely proportionally to an amount of a predominance of a direction of movement of the filter coefficient.

13. The method of claim 1 ,
wherein the adaptive filter is used in an active noise cancellation system.

14. The method of claim 13 ,
wherein the adaptive filter is used in the active noise cancellation system to generate an anti-noise signal or to model an electro-acoustic path of an audio device.

15. An adaptive filter, comprising:
a digital signal processor programmed to:
analyze, in a frequency domain, a statistic of coefficient movement of the adaptive filter; and
adjust, in the frequency domain, a parameter that affects speed of convergence of the adaptive filter based on the analyzed statistic of filter coefficient movement; and

wherein the statistic of coefficient movement comprises a measure of an attribute of a sample of a change of adaptive filter coefficients between a current and next value of the coefficients.

16. The adaptive filter of claim 15 ,
wherein the digital signal processor maintains, for each frequency bin of a plurality of frequency bins, an associated:
coefficient;
statistic of coefficient movement; and
parameter that affects speed of convergence of the adaptive filter;

wherein to analyze the statistic of coefficient movement in the frequency domain, the digital signal processor analyzes, for each frequency bin of the plurality of frequency bins, the statistic of coefficient movement associated with the frequency bin; and
wherein to adjust the parameter in the frequency domain, the digital signal processor adjusts, for each frequency bin of the plurality of frequency bins, the parameter associated with the frequency bin based on the analyzed statistic associated with the frequency bin.

17. The adaptive filter of claim 16 ,
wherein to adjust the parameter, the digital signal processor decreases the step size when the frequency bin is above a predetermined upper frequency bin.

18. The adaptive filter of claim 16 ,
wherein the digital signal processor determines a current maximum distance of the filter coefficient movement among the frequency bins and maintains a smoothed version of the maximum distance;
wherein to analyze the statistic of coefficient movement in the frequency domain, the digital signal processor determines a condition that is whether the current maximum distance exceeds a product of the smoothed version and a scaling factor; and
wherein to adjust the parameter in the frequency domain, the digital signal processor sets the parameter to a value that causes the speed of convergence to be approximately zero when the condition is satisfied.

19. The adaptive filter of claim 15 ,
wherein coefficients comprise complex numbers having a real part and an imaginary part;
wherein the statistic of coefficient movement comprises separate statistics for the real part of the coefficients and the imaginary part of the coefficients; and
wherein the parameter comprises separate parameters for the real part of the coefficients and the imaginary part of the coefficients.

20. The adaptive filter of claim 15 ,
wherein the statistic comprises direction counts of the filter coefficient movement.

21. The adaptive filter of claim 15 ,
wherein the parameter that affects speed of convergence is a step size.

22. The adaptive filter of claim 21 ,
wherein to adjust the parameter based on the analyzed statistic, the digital signal processor adjusts the step size approximately proportionally to an amount of a predominance of a direction of movement of the filter coefficient.

23. The adaptive filter of claim 22 ,
wherein further to adjust the parameter based on the analyzed statistic, the digital signal processor adjusts the step size by a predetermin

Assignees

• Cirrus Logic Int Semiconductor Ltd
• Cirrus Logic Inc

Inventors

• Zhou Dayong
• Yong Chin Huang

Classifications

• G10K11/17879

  using both a reference signal and an error signal · CPC title

• G10K11/17854 Primary

  the filter being an adaptive filter · CPC title

• H04B3/237

  using two adaptive filters, e.g. for near end and for end echo cancelling · CPC title

• H03H2021/0058

  Block LMS, i.e. in frequency domain · CPC title

• G10K11/17881

  the reference signal being an acoustic signal, e.g. recorded with a microphone · CPC title