Page 64 - From GMS to LTE

50 From GSM to LTE-Advanced Pro and 5G

TCH

t
Normal voice DTX period
transmission Transmission of information
for the comfort noise simulation
during DTX periods
Figure 1.40 Discontinuous transmission (DTX).

downlink direction, this is managed by the voice activity detection (VAD) algorithm in
the TRAU, while in the uplink direction the VAD is implemented in the mobile device.
Simply deactivating a speech channel, however, creates a very undesirable side effect.
As no speech signal is transmitted anymore, the receiver no longer hears the back-
ground noise on the other side. This can be very irritating, especially for high‐volume
background noise levels such as when a person is driving a car or sitting in a train.
Therefore, it is necessary to generate artificial noise, called comfort noise, which simulates
the background noise of the other party for the listener. As the background noise can
change over time, the mobile device or the network, respectively, analyzes the back-
ground noise of the channel and calculates an approximation for the current situation.
This approximation is then exchanged between the mobile device and the TRAU every
480 milliseconds. Additional benefits for the network and mobile device are the ability
to perform periodic signal quality measurements of the channel and the ability to use
these frames to get an estimation on the current signal timing to adapt the timing
advance for the call if necessary. How well this method performs is clear from the audibility
as this procedure is used in all mobile device calls today and the simulation of the back-
ground noise in most cases cannot be differentiated from the original signal.
Despite the use of sophisticated methods for error correction, it is still possible that
parts of a frame are destroyed beyond repair during transmission on the air interface. In
these cases, the complete 20‐millisecond voice frame is discarded by the receiver and
the previous data block is used instead to generate an output signal. Most errors that are
repaired this way remain undetected by the listener. This trick, however, cannot be used
indefinitely. If after 320 milliseconds a valid data block has still not been received, the
channel is muted and the decoder keeps trying to decode the subsequent frames. If,
during the following few seconds no valid data frame is received, the connection is ter-
minated and the call drops.
Many of the previously mentioned procedures have specifically been developed for
the transmission of voice frames. For example, for circuit‐switched data connections
that are used for fax transmissions or end‐to‐end encrypted voice calls, a number of
modifications are necessary. While it is possible to tolerate a number of faulty bits for
voice frames or to discard frames if a CRC error is detected, this is not possible for data
calls. If even a single bit is faulty, a retransmission of at least a single frame has to be
performed as most applications cannot tolerate a faulty data stream. To increase the
likelihood of correctly reconstructing the initial data stream, the interleaver spreads the
bits of a frame over a much larger number of bursts than the eight bursts used for voice
frames. Furthermore, the channel coder, which separates the bits of a frame into differ-
ent classes based on their importance, had to be adapted for data calls as well, as all bits
are equally important. Thus, the convolutional decoder has to be used for all bits of a