CS6422
CS6422
4.3.2.2 Convergence Time
4.3.2.3 Half-Duplex Switching
Convergence time is a measure of how quickly the
adaptive filter can model the echo path. From
cleared coefficients, the training signal is injected
Although the CS6422 transitions from half-duplex
to full-duplex operation from reset after only a few
utterances are passed through the system, the per-
into the echo canceller and the time for the ERLE formance of half-duplex is critical to the end-user
to reach a given threshold value is the convergence
time. Different customers will have different
in cases where the echo canceller is not adequate.
The half-duplex switching characteristics can be
threshold levels, so Crystal does not specify con- subjectively tested with the following procedure:
vergence time.
Set the CS6422 Microcontroller Interface to the
The following example will measure convergence
time for the acoustic echo canceller:
nominal register values for the system, but clear the
acoustic and network echo canceller coefficients.
This will force the CS6422 to remain in half-duplex
mode.
Set up the system as for the ERLE test. Clear the
acoustic canceller coefficients through the Micro-
controller Interface. Apply the training signal to
The most useful test of practical performance
NI, set the coefficients to normal, and simulta- found at Crystal has been the “alternating counting
neously start a timer. Once the measured ERLE test.” In this test the person at the near-end counts
reaches the threshold the system designer desires,
stop the timer. The elapsed time is the convergence
time. A good value for the threshold would be the
AErle value from Register 3, since this would be
all the odd numbers and the person at the far-end
counts all the even numbers. This tests the inter-
ruptibility of the half-duplexer. During testing, sys-
tem parameters for the half-duplex may need to be
the time for the CS6422 to go from half-duplex changed to accommodate the level of performance
mode to full-duplex mode.
expected for the product. See Section 4.1.2, “Half-
Duplex Mode” and Section 3.2.2, “Register Defini-
tions”for more details.
A good tool for this measurement is a digital stor-
age oscilloscope set to a slow sweep so that about
five seconds of signal is shown on the screen. One
channel of the oscilloscope should monitor the
ADC input (for an uncancelled reference), and an-
other channel should monitor the echo cancelled
output. This technique is especially effective when
speech is the training signal.
We see about 2-5 seconds of training time using
known good equipment. This time assumes contin-
uous speech as the training signal. Pauses will ex-
tend the convergence time.
41
41
DS295F1
CIRRUS [ CIRRUS LOGIC ]
