NONLINEAR REFERENCE SIGNAL PROCESSING FOR ECHO SUPPRESSION

• US 20120237047A1
• Filed: 03/16/2012
• Published: 09/20/2012
• Est. Priority Date: 03/18/2011
• Status: Active Grant

• Current Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Original Assignee: Dolby Laboratories Licensing Corporation (Dolby Laboratories Incorporated)
• Inventors: Dickins, Glenn N., Neal, Timothy J.
• Granted Patent: US 8,804,977 B2
• Time in Patent Office: Days
• US Class Current: 381/66
• CPC Class Codes: H04B 3/234 using double talk detection
• CPC Class Codes: H04M 9/082 using echo cancellers echo cancellers per se H04B3/23

Abstract

An echo suppression system and method, and a computer-readable storage medium that is configured with instructions that when executed carry out echo suppression. Each of the system and the method includes the elements of a linear echo suppressor having a reference signal path, with a nonlinearity introduced in the reference signal path to introduce energy in spectral bands. Unlike an echo canceller, the echo suppression system and method are relatively robust to errors in the introduced nonlinearity.

Claims

1. A signal processing apparatus comprising:

an input transformer and a spectral bander to accept samples of one or more input signals to form a banded frequency-domain amplitude metric of the one or more input signals for a plurality of spectral bands;

a reference signal path to accept one or more reference signals and to form a banded frequency-domain amplitude metric of an echo reference signal representing the one or more reference signals, wherein the reference signal path includes a nonlinear element, such that the banded frequency-domain amplitude metric is of a nonlinearly distorted version of the one or more reference signals;

an adaptive linear filter to accept the banded frequency-domain amplitude metric of the echo reference signal and to determine a frequency-domain amplitude metric of a predicted echo, the adaptive filter using a set of adaptively determined filter coefficients;

a filter coefficient updater coupled to the output of the input transformer and spectral bander and to the output of the adaptive linear filter, the filter coefficient updater to update the filter coefficients;

an echo suppressor coupled to the output of the input transformer and spectral bander and to the output of the adaptive linear filter, the echo suppressor to suppress echoes and to determine an echo-suppressed output signal.

2. A signal processing apparatus as recited in claim 1, wherein the echo suppressor comprises a gain calculator coupled to the output of the input transformer and spectral bander and to the output of the adaptive linear filter, the gain calculator to calculate a set of gains;

and a gain applicator coupled to the output of the input transformer and to the gain calculator to apply the set of gains to the input signal and determine echo-suppressed frequency bins of an output signal.

3. A signal processing apparatus as recited in claim 1, wherein the updating of the filter coefficients includes double talk detection to gate updating of the filter coefficients.

4. A signal processing apparatus as recited in claim 1, wherein the apparatus includes an estimator of a banded spectral estimate of a noise amplitude metric, and wherein the echo suppressor uses the estimate of the noise amplitude metric to achieve simultaneous echo suppression and noise suppression.

5. A signal processing apparatus as recited in claim 1, wherein the spacing of the spectral bands is log-like.

6. A signal processing apparatus as recited in claim 1, wherein the spectral bands are perceptually spaced.

7. A signal processing apparatus as recited in claim 1, wherein the reference signal path includes a transformer and a spectral bander, and wherein the nonlinearity precedes the transformer such that the transforming is of a nonlinearly distorted echo reference signal representing the one or more reference signals.

8. A signal processing apparatus as recited in claim 7, wherein the nonlinearity includes a polynomial function.

9. A signal processing apparatus as recited in claim 7, wherein the nonlinearity includes a clipping function.

10. A signal processing apparatus as recited in claim 7, wherein the nonlinearity is determined by measurements.

11. A signal processing apparatus as recited in claim 7, wherein the reference signal path includes a time-to-frequency transformer, and wherein the nonlinearity is after the transforming by the transformer, such that the nonlinearity applies nonlinear distortion in the frequency domain.

12. A signal processing apparatus as recited in claim 11, wherein the nonlinearity distributes power from one or more fundamental components to higher order components in the frequency bin domain, the distribution from a first component to the next higher component being according to an attenuation factor.

13. A signal processing apparatus as recited in claim 12, wherein the amount of power distributed is decayed over time.

14. A signal processing apparatus as recited in claim 1, wherein the frequency domain amplitude metric is frequency domain power.

15. A signal processing apparatus comprising a processing system and configured to suppress undesired signals including echoes, the processing apparatus configured to:

accept an input signal and an echo reference signal, the echo reference representative of one or more reference signals;

determine a transformed frequency bin representation of the input signal for a plurality of frequency bins;

determine a transformed banded representation of an amplitude metric of the input signal for a plurality of spectral bands by spectral banding;

determine a transformed banded representation of an amplitude metric of the reference signal via a reference signal path, wherein the reference signal path includes a nonlinear element, such that the transformed banded representation is of the amplitude metric of a nonlinearly distorted version of the echo reference signal;

linearly filter the transformed banded representation by a filter that uses an adaptively determined set of filter coefficients, the filtering to determine a transformed banded representation of an amplitude metric of an estimate of an echo for suppression;

adaptively update the filter coefficients using the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal; and

suppress echoes by processing the input signal according to the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal in order to determine an output signal.

16. A signal processing apparatus as recited in claim 15, wherein the suppressing includes determining a set of gains to apply to the input signal to determine the output signal.

17. A method comprising:

accepting an input signal and an echo reference signal;

determining a transformed frequency bin representation of the input signal for a plurality of frequency bins;

determining a transformed banded representation of an amplitude metric of the input signal for a plurality of spectral bands;

determining a transformed banded representation of an amplitude metric of the reference signal via a reference signal path, wherein the reference signal path includes a nonlinear element, such that the transformed banded representation is of the amplitude metric of a nonlinearly distorted version of the echo reference signal;

linearly filtering the transformed banded representation by a filter that uses an adaptively determined set of filter coefficients, the filtering to determine a transformed banded representation of an amplitude metric of an estimate of an echo for suppression;

adaptively determining the filter coefficients using the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal; and

suppressing echoes by processing the input signal according to the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal in order to determine an output signal.

18. A method as recited in claim 17, wherein the suppressing includes determining a set of gains to apply to the input signal to determine the output signal.

19. A method as recited in claim 17, wherein the adaptively determining the filter coefficients is gated by a double talk detector.

20. A method as recited in claim 17, further comprising determining an estimate of the amplitude metric spectrum of the echo to use for the suppressing.

21. A method as recited in claim 17, wherein the method includes determining a banded spectral estimate of a noise amplitude metric, and wherein the processing the input signal is further according to the banded spectral estimate of a noise amplitude metric to achieve simultaneous echo suppression and noise suppression.

22. A method as recited in claim 17, wherein the nonlinearity in the reference signal path is in the time-sample domain before the reference signal is transformed.

23. A method as recited in claim 22, wherein the nonlinearity includes a polynomial function.

24. A method as recited in claim 22, wherein the nonlinearity includes a clipping function.

25. A method as recited in claim 17, wherein the nonlinearity in the reference signal path is in the frequency domain.

26. A method as recited in claim 25, wherein the nonlinearity distributes power from one or more fundamental components to higher order components in the frequency bin domain, the distribution from a first component to the next higher component being according to an attenuation factor.

27. A method as recited in claim 26, wherein the amount of power distributed is decayed over time.

28. A method as recited in claim 17, wherein the spacing of the spectral bands is log-like.

29. A non-transitory computer-readable medium configured with instructions that when executed by at least one processor of a processing system, cause processing hardware to carry out a method comprising:

accepting an input signal and an echo reference signal;

determining a transformed frequency bin representation of the input signal for a plurality of frequency bins;

determining a transformed banded representation of an amplitude metric of the input signal for a plurality of spectral bands;

determining a transformed banded representation of an amplitude metric of the reference signal via a reference signal path, wherein the reference signal path includes a nonlinear element, such that the transformed banded representation is of the amplitude metric of a nonlinearly distorted version of the echo reference signal;

linearly filtering the transformed banded representation by a filter that uses an adaptively determined set of filter coefficients, the filtering to determine a transformed banded representation of an amplitude metric of an estimate of an echo for suppression;

adaptively determining the filter coefficients using the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal; and

suppressing echoes by processing the input signal according to the banded representation of the amplitude metric of an estimate of an echo and the banded representation of the amplitude metric of the input signal in order to determine an output signal.