MT92220 Zarlink Semiconductor, MT92220 Datasheet - Page 91

• Manufacturer Part Number: MT92220
• Description: 1023 Channel Voice Over IP/AAL2 Processor
• Manufacturer: Zarlink Semiconductor

Data Sheet
MT92220
8.7
Silence Suppression
The other operation that the packet assembly module can perform is calculating silence suppression information on
data contained in circular buffers. The objective is simple: after performing calculations on a block of PCM data
(using the PCM bytes), the calculator must decide if the channel currently being treated is silent or not. It then
communicates this information to the packet assembler that will discard all RTP packets containing that channel
until the channel re-enables.
Silence suppression calculations are performed on the block of data that is about to be sent in the xxPCM packet by
the assembly process. By synchronizing the two events, the assembly process will have the most up-to-date
information concerning the suppression of its packets and voice quality will be maintained to a maximum.
The silence suppression process reads its data from the same circular buffer as the packet assembly; the minimum
size for any TX circular buffer is 512 bytes, because silence suppression information is only contained in the lower
byte of each word in the buffer (thus the minimum amount of useful data in a circular buffer is 256 bytes).
There are 2 sequential steps to be performed in the silence suppression calculation. The first is the filtering of the
input signal. In some cases, PCM signals are not centered around 0 and have a constant offset that is added to
them: performing energy calculations with these offset values would lead to erroneous results and poor
performance. Thus, to eliminate this error, the silence suppression module uses a first-order high-pass butterworth
filter with a cutoff frequency of 10 Hz to eliminate DC and extremely low-frequency signals. The butterworth filter
keeps the entire context it needs in the Local Butterworth HP 10 Hz Context field. This filter can be enabled or
disabled by using the Local Butterworth enable bit.
To filter the signal, it must first be expanded into linear form. The Local Law bit indicates if the input signal is coded
using u-law or A-law. Using the correct law, each PCM input sample is expanded into its linear value.
Once the filtering of the input signal has been calculated, silence suppression can be correctly performed on the
data. The silence suppression process uses an adaptive algorithm to determine whether the input signal represents
silence or voice. The basic approach used in the silence suppression algorithm is to smooth out the level of the
signal to remove any frequency-dependent components, then to establish the minimum level at which the signal
maintains itself over a reasonable period of time. Then, the current level of the signal is compared to the floor level,
and if it is measured to be larger, then the signal is deemed to be voice; otherwise it is considered to be silence.
If the packet is silent, then a CN (for RTP) or SID (for AAL2) packet may potentially be generated. If the last packet
was not suppressed, then a CN or SID packet will definitely be transmitted. A CN or SID packet will also be sent at
other points in time when the padding energy at the remote end needs to be updated. The Last Suppress bit
indicates the state of the last packet, which allows a CN or SID packet transmission decision to be taken.
The CN or SID energy is calculated by summing the energies of the most recent samples received. Depending on
whether the current state is voice or silence, the energy will be summed on a different number of samples, using the
First Energy Period and Subsequent Energy Period respectively. This is done because energy on background
noise and silence can be calculated on a much longer period to obtain greater accuracy. The sum of samples is
kept in the Energy Sum and the sample count is kept in the Energy Counter. When the Energy Counter reaches the
terminal count, a linear-to-dB conversion is performed to take the Energy sum divided by the Energy Counter to
obtain the average linear energy, then converted to a logarithmic scale to obtain an energy in dBov. The TX Silence
Suppression Structure contains a dB Correction that allows the dB energy to be converted to any scale, like dBm0
instead of dBov, for example. There is also a Maximum dB Value and a Minimum dB Value that can be set, making
sure that the value in CN or SID packets stays within a certain range.
Zarlink Semiconductor Inc.
91