Digital adaptive filter and acoustic echo canceller using the same

A frequency domain adaptive filter for modeling a signal path between a loudspeaker and a microphone of a hands-free communication device, comprising:

• an input terminal adapted to receive an input signal;
• and
• filter coefficient update means to successively update filter coefficients in accordance with the input signal, an estimated power of the input signal, and an error signal between the input signal filtered in the frequency domain adaptive filter and an input signal propagated along an external path being modeled by the frequency domain adaptive filter, the filter coefficient update means adapted to calculate a step size for at least each frequency band individually in dependence on a background noise level of each frequency band, wherein the filter coefficient update means updates the filter coefficients according to
• H
• ⁡
• (
• k
• )
• 
• s
• ·
• t
• =
• H
• ⁡
• (
• k
• )
• 
• s
• +
• μ
• ⁡
• (
• k
• )
• ∘
• E
• ⁡
• (
• k
• )
• PX
• ⁡
• (
• k
• )
• 
• s
• ∘
• X
• ≠
• (
• k
• )
• ∘
• G
• ⁡
• (
• k
• )
• k
• =
• 0
• ,
• …
• ⁢
• ,
• N
• 2
• +
• 1
• ,
• where
• μ(k) is the individual step size for each frequency band,
• E(k) is a transformed error signal,
• X*(k) is a complex conjugate of a transformed loudspeaker signal,
• PX(k) is an estimated power of the transformed loudspeaker signal, and
• G(k) is a gradient constraint operation.

A subband adaptive filter for modeling a signal path between a loudspeaker and a microphone of a hands-free communication device, comprising:

• an analysis filter bank adapted to filter an input signal into at least two frequency bands;
• a corresponding subband filter for each of the at least two frequency bands to filter the corresponding frequency band supplied thereto;
• and
• a synthesis filter bank adapted to generate a time domain output signal from the output of the subband filters, wherein each subband filter comprises filter coefficient update means to successively update filter coefficients in accordance with the corresponding frequency band supplied thereto, an estimated power thereof, and a subband error signal between the corresponding frequency band and a corresponding frequency band input signal propagated along an external path being modeled through the subband adaptive filter, and wherein the filter coefficient update means is adapted to calculate a step size for each corresponding frequency band individually in dependence on a background noise level for each corresponding frequency band, and wherein the filter coefficient update means of each subband filter is adapted to perform recursive smoothing for an increasing input power and a decreasing input power asymmetrically, wherein an input signal power estimation means performs the recursive smoothing of the input power according to:
• PI
• ⁢
• ⁢
• 
• =
• β
• ∘
• Pin
• ⁢
• |
• r
• ⁢
• +
• (
• J
• -
• β
• )
• ∘
• PJ
• 
• i
• ·
• j
• β
• =
• {
• β
• up
• PI
• ⁢
• ⁢
• increasing
• β
• down
• PI
• ⁢
• ⁢
• decreasing
• .

A digital adaptive filtering method for modeling a signal path between a loudspeaker and a microphone of a hands-free communication device, comprising the steps of:

• successively updating filter coefficients in accordance with an input signal, an estimated power of the input signal, and an error signal between the input signal for the digital adaptive filtering method and die input signal propagated along an external path being modeled through the digital adaptive filtering method;
• and
• performing recursive smoothing for an increasing input power and a decreasing input power asymmetrically with a different weighting factor for en increasing input signal power level and a decreasing input signal power level, wherein the input power is smoothed recursively according to:
• PI
• ⁢
• ⁢
• 
• =
• β
• ∘
• Pin
• ⁢
• |
• r
• ⁢
• +
• (
• I
• -
• β
• )
• ∘
• PI
• 
• i
• ·
• j
• β
• =
• {
• β
• up
• PI
• ⁢
• ⁢
• increasing
• β
• down
• PI
• ⁢
• ⁢
• decreasing
• .

A frequency domain adaptive filtering method for modeling a signal path between a loudspeaker and a microphone of a hands-free communication device, comprising the steps of:

• successively updating filter coefficients in accordance with an input signal, an estimated power of the input signal, and an error signal between the input signal filtered according to the frequency domain adaptive filtering method and an input signal propagated along an external path being modeled through the frequency domain adaptive filtering method;
• and
• calculating a step size for at least two frequency bands individually in dependence on a background noise level of each frequency band, wherein filter coefficients are updated according to
• H
• ⁡
• (
• k
• )
• 
• t
• ·
• s
• =
• H
• ⁡
• (
• k
• )
• 
• t
• +
• μ
• ⁡
• (
• k
• )
• ∘
• E
• ⁡
• (
• k
• )
• PX
• ⁡
• (
• k
• )
• 
• t
• ∘
• X
• *
• ⁡
• (
• k
• )
• ∘
• G
• ⁡
• (
• k
• )
• k
• =
• 0
• ,
• …
• ⁢
• ,
• N
• 2
• +
• 1.
• where
• μ(k) is the individual step size for each frequency band,
• E(k) is a transformed error signal,
• X*(k) is a complex conjugate of a transformed loudspeaker signal,
• PX(k) is an estimated power of the transformed loudspeaker signal, and
• G(k) is a gradient constraint operation.

A digital adaptive filtering method of the subband type for modeling a signal path between a loudspeaker and a microphone for a hands-free communication device, comprising the steps of:

• filtering an input signal into at least two frequency bands in an analysis filter bank;
• filtering each of the at least two frequency bands in a corresponding subband filter;
• generating a time domain output signal from the output of the subband filters in a synthesis filter bank, wherein in each subband filter coefficients are successively updated in accordance with the frequency band supplied thereto, an estimated power thereof, and a subband error signal between the frequency band and a corresponding frequency band input signal propagated along an external path being modeled through the digital adaptive filtering method of the subband type, and wherein a step size for each frequency band is calculated individually in dependence on a background noise level for each frequency band;
• and
• performing recursive smoothing for an increasing input pouter and a decreasing input power asymmetrically, wherein the recursive smoothing of the input power is performed according to:
• PI
• ⁢
• ⁢
• 
• =
• β
• ∘
• Pin
• ⁢
• |
• r
• ⁢
• +
• (
• I
• -
• β
• )
• ∘
• PI
• 
• i
• ·
• j
• β
• =
• {
• β
• up
• PI
• ⁢
• ⁢
• increasing
• β
• down
• PI
• ⁢
• ⁢
• decreasing
• .