Rockwell RFX144V24-S23, Designer'S Manual

Page: 175 / 198

RFX144V24-S23 and RFX96V24-S23 Modem Designer’s Guide
1070 12-5
The default speaker volume control is -10 dB. It is advised to check the dial-tone loudness at this setting. Adjust the volume
control, if necessary, to make sure its loudness is at the right level. Disabling the speaker AGC function may be considered
during the dial-tone reception. At the end of dial-tone, selecting Speakerphone Mode (CONF = 91h) will reinitiate the AGC
parameters. The microphone channel should be muted during the dial-tone reception.
12.3.7 Microphone and Speaker Muting
The muting function is available in both the FDX Speakerphone Mode and the Handset Mode. Muting does not affect the use
of tone transmit function. AGC functions can be left either enabled or disabled.
The muting function overwrites the output of the corresponding voice channel with zero, totally silencing the voice output.
The rest of the muted voice channel still functions on the channel's voice input, keeping the echo canceller active, speech
detector active, loop gain calculation active, and so on. The AGC update, if enabled, is active only if the other channel is not
muted.
The channel not being muted is fully open with a fixed AGC (if enabled) gain determined prior to the muting of the other
channel. Disabling the AGC function forces the AGC gain to unity. The channel's volume control can still be used to adjust
the voice output level.
The host sets or releases the muting function through the MUTEI and MUTEX bits.
MUTEI MUTEX Action
0 0
No muting
0 1
Microphone muted (default)
1 0
Speaker muted
1 1
Both channels muted
12.3.8 Microphone and Speaker AGC
The microphone and speaker voice AGC algorithms are designed to raise or lower the channel's voice energy to a desired
level. The AGCs can be enabled (set to a 1) or disabled (reset to a 0) through the AGCIE and AGCXE control bits. When
enabled, the AGC involvement is controlled by the speech detector. When the talker is silent, the AGC is turned off to avoid
amplification of noise. When speech is detected, the AGC process is activated in a fast-attack and slow-decay fashion, so
that the AGC gain is applied to the signal appropriately.
The AGC gain adaptation, on the other hand, is a slow-rise and fast-drop process. The gain's rise time is controlled by the
AGC slew rate. The long-term sample by sample, averaged signal energy after the AGC is calculated and compared with the
AGC's reference level. When the signal energy is below its AGC reference level, the AGC gain will be increased till the signal
energy reaches the reference level. When the signal energy is greater than its AGC reference level, the AGC gain will be
decreased until the signal energy drops to the reference level. The maximum AGC gain is 18 dB in each voice channel.
There is no low limit for the AGC compression loss.
Initially the AGC gain is set at its maximum 18 dB in both the microphone and speaker channels. Due to the fast-drop nature
of the algorithm, the convergence process of the AGC gain quickly settles down to reach the reference level. Very little
clipping sound can be heard during this process. Each time the microphone AGC function is disabled, the AGC gain is reset
back to unity. It then takes about four seconds to raise the AGC gain from unity to its maximum with the default slew rate,
depending on the input signal level. The speaker AGC gain after reset is set to +6 dB to compensate the 6 dB loss in the
receive interface circuitry to the line.
The AGC slew rate and the reference level are totally separate parameters. But to tune up the slew rate, one should set the
reference level first so that the slew rate can be adjusted accurately. Increasing the slew rate value linearly increases the
AGC convergence speed.
12.3.9 Speech Detectors
Two speech detectors work together in the speakerphone. They are fully adaptive with room and line noise floors, and can
distinguish true talkers from noise and echo feedback. The room speech detector detects the talker activities in the local
room while the line speech detector detects the talker activities at the remote site.
Speech detectors activate or deactivate the AGCs depending on talker activity. When the talker is silent, the AGC is turned
off to avoid noise amplifications. When speech is detected, the AGC process is activated in the fashion of fast-attack and
slow-decay. This is to avoid initial clipping at the beginning of a sentence and tail-end chopping at the end of the sentence. A