Method and device for cancelling stereophonic echo with frequency domain filtering

DETAILED DESCRIPTION Accordingly, the invention proposes a method as outlined above, further comprising computing a spectral energy P.
sub.
11 (f) of the first input signal, a spectral energy P.
sub.
22 (f) of the second input signal, an inter-spectral energy P.
sub.
12 (f) of the first and second input signals and a coherence value .
GAMMA.
(f) for each of the frequencies f in the set of 2N frequencies.
According to that method, the updating of the vector H.
sub.
1 for the processing the next block is performed on the basis of a modified gradient in the form: .
gradient.
.
sub.
M1 =A.
sub.
11 X.
gradient.
.
sub.
1 +A.
sub.
12 X.
gradient.
.
sub.
2 =(A.
sub.
11 XX*.
sub.
1 +A.
sub.
12 XX*.
sub.
2)XE (1) where A.
sub.
11 denotes a vector of size 2N whose component relating to a frequency f is ##EQU1## and A.
sub.
12 denotes a vector of size 2N whose component relating to a frequency f is ##EQU2## G(f) being an increasing real function of the coherence value .
GAMMA.
(f) such that 0.
ltoreq.
G(f)<1.
On the other hand, the updating of the vector H.
sub.
2 for the processing the next block is performed on the basis of a modified gradient in the form: .
gradient.
.
sub.
M2 =A.
sub.
22 X.
gradient.
.
sub.
2 +A*.
sub.
12 X.
gradient.
.
sub.
1 =(A.
sub.
22 XX*.
sub.
2 +A*.
sub.
12 XX*.
sub.
1)XE (2) where A.
sub.
22 denotes a vector of size 2N whose component relating to a frequency f is ##EQU3## Another aspect of the present invention relates to a stereophonic echo canceller arranged to implement a method as defined above.
Normalisation of the gradient takes account of crossed terms between the two channels, with a weighting determined by the coherence between the two input signals.
This approach to adapting the filter coefficients represented by the components of H.
sub.
1 and H.
sub.
2 may be derived from a "block Newton" type of algorithm, expressed in the frequency domain, with limited hypotheses on the form of the signal correlation matrices.
In a preferred embodiment of the method, the spectral energies P.
sub.
11 (f) and P.
sub.
22 (f) are respectively averages of