Speech Enhancement - Digital Speech: Coding for Low Bit Rate Communication Systems - page 412

Table 11.1 ITU-T recommendations and ANLMS system performance
results

ITU-T recommendation G.165 ANLMS

Tests Input levels Recommendation Results

Steady state − 30 dBm0 − 48 dBm0 − 83 dBm0

residual echo level − 20 dBm0 − 42 dBm0 − 72 dBm0

− 10 dBm0 − 36 dBm0 − 60 dBm0

Convergence − 30 dBm0 attenuation ≥ 27 dB 30 dB

− 20 dBm0 attenuation ≥ 27 dB 30 dB

− 10 dBm0 attenuation ≥ 27 dB 30 dB

Leak rate − 30 dBm0 (For all input levels, (Echo level

(i.e. slow divergence − 20 dBm0 residual echo level increase of 6 dB

when no signal) − 10 dBm0 should not increase was evident for

more than 10 dB) all input levels)

Infinite return
loss convergence − 30 dBm0 ≤− 37 dBm0 − 78 dBm0

(i.e. rapid return to − 20 dBm0 ≤− 37 dBm0 − 68 dBm0

− ≤− −

convergence after 10 dBm0 37 dBm0 57 dBm0

an interrupt to
echo path)

far-end and near-end ports. A test is devised (see [22]) for each of the require-
ments in Table 11.1, listed with the results obtained for the various tests.
The requirements for echo canceller performance for double-talk situations
is subdivided into two tests. The first is related to the double-talk detection
part of the echo canceller. As there is no such double-talk detector used in
the ANLMS system this test is not performed. The second part of the test is
aimed at ensuring that, in double-talk situations, the divergence is low. The
requirement for this part is that only a 10 dB increase in residual echo level
of the results listed in the steady state test (Test No. 1 of [22]) are permitted.
The ANLMS is well within this requirement.
Note that this does not mean that systems based on either LMS or NLMS
do not satisfy the ITU-T requirements. On the contrary, they do satisfy them,
but the advantages of ANLMS are the continued filter coefficient adaptation
even during cross-talk scenarios and that there is no need for switching or
VADs, which results in more consistency.
In order to improve the overall system performance, a noise suppressor
and an echo canceller can be used jointly. The noise suppressor may be
integrated either prior to the echo canceller or after it. Integrating prior to
the echo canceller in order to remove the noise from the near-end signal