Nonlinear Adaptive Filters for Acoustic Echo Cancellation in Mobile Terminals - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 247

30

VOLT 2

25

20

15

10

5

MMD

LIN

0

0 5 10 15 20 25 30

L/N

FIGURE 7.12
Echo cancellation level (dB) for a set of echo cancelers vs. second-order nonlinear distortion level
(dB).

model was obtained by the identification of the true acoustic echo path of a
commercial handset terminal.
To evaluate the relevance and the benefits of a nonlinear approach to acous-
tic echo cancellation, we consider the results of simulations done in the pres-
ence of different levels of quadratic nonlinearities. To this purpose, we define
as L N the ratio between the average power of the signal at the output of the

linear part of the system modeling the echo path, and the average power of the
signal at the output of the second-order part of the model. As a figure of merit
of the echo cancellation achieved, we consider the ratio between the average
echo signal power and the average residual echo signal power OP

/

MSE. This
figure of merit is plotted in Figure 7.12 vs. the nonlinear distortion level L

/

/

N
for different truncated Volterra filters and bilinear filters. The first set of filters
always includes a linear part having a memory of 100 samples, whereas the
bilinear filter, according to its recursive structure, permits a shorter memory
with respect to the input signal, here fixed to N 1 =

15 samples. The following

filters have been considered:

LIN: Linear FIR filter with 100 coefficients
MLNLF: Cascade of a memoryless polynomial preprocessor with order
7 and a linear FIR filter with 100 coefficients