05-06-2012, 03:11 PM
ROBUST VARIABLE STEP-SIZE AFFINE PROJECTION ALGORITHM SUITABLE FOR ACOUSTIC ECHO CANCELLATION
ROBUST VARIABLE STEP-SIZE AFFINE PROJECTION ALGORITHM.pdf (Size: 549.79 KB / Downloads: 0)
ABSTRACT
The affine projection algorithm (APA) and different versions
of it have proved to be very attractive choices for acoustic
echo cancellation (AEC). In this context, a classical APA
with a constant step-size has to compromise between two
performance criteria, i.e., 1) high convergence rates and
good tracking capabilities, and 2) low misadjustment and
robustness against background noise variations and doubletalk.
INTRODUCTION
In acoustic echo cancellation (AEC) applications, an adaptive
filter identifies the acoustic echo path between the terminal’s
loudspeaker and microphone, i.e., the room impulse response
[1]. In this context, several specific problems have to be addressed.
First, the echo path can be extremely long and it
may rapidly change at any time during the connection. Secondly,
the background noise that corrupts the microphone
signal can be strong and non-stationary in nature. Third, the
involved signals (i.e., mainly speech) are non-stationary and
highly correlated. Finally, the behaviour during double-talk
(i.e., the talkers on both sides speak simultaneously) has to be
considered.
SIMULATION RESULTS
The simulations were performed in an AEC context. The
measured acoustic impulse response has L = 512 coefficients;
the same length is used for the adaptive filter. The far-end
signal x(n) is a speech sequence. An independent white
Gaussian noise signal w(n) is added to the echo signal y(n),
with 20 dB signal-to-noise ratio (SNR). The performance is
evaluated in terms of the normalized misalignment (in dB),
defined as 10
20log (|| h − hˆ (n) || / || h ||) , where ||•|| denotes
the l2 norm. Besides the classical APA, another two members
of the APA family are chosen for comparisons, i.e., the variable
regularized APA (VR-APA) recently proposed in [6] and
the robust proportionate APA (R-PAPA) [9].
CONCLUSIONS
The basic idea of the classical APA, i.e., to cancel p a posteriori
errors, was modified within the proposed VSS-APA in
order to take into account the existence of the near-end signal.
The variable step-size formula of the proposed algorithm
does not require any additional parameters from the acoustic
environment. Consequently, the proposed VSS-APA is very
suitable in practice. As compared to other APAs, it was found
to be more robust to near-end signal variations like the increase
of the background noise or double-talk.