Digital Signal Processing Reference
In-Depth Information
decrease of the compression ratio for two main reasons: the slice headers would
reduce the available bandwidth, and the context-based entropy coding would
become less efficient. Long slices, on the other hand, are more likely to contain
transmission errors, which leads to reduced transmission efficiency and higher
packet losses. In this chapter the performance trade-offs involved in the packet
creation process will be investigated.
4. RESULTS
4.1 The simulation scenarios
Simulations have been carried out using the
ns
[15] network simulator fed
with the well known
Foreman
video sequence. The video sequence is coded us-
ing the H.264 test model software [5], enabling most of the new characteristics
of the H.264 standard, in particular multiple reference frames and Lagrangian
optimized motion search for macroblocks down to 4×4 pixels size. The se-
quence size is CIF at 15 fps, and is encoded using a fixed quantization param-
eter, set to achieve a bit rate of about 256 kbit/s. The sequence length is 149
frames, and one B frame is introduced after each P frame. The transmitted se-
quence is obtained concatenating the base video sequence 80 times, reaching
a length of 794.6 s at 15 fps. In order to improve error resilience, an I frame
is interposed at the beginning of each repetition of the sequence (i.e., every
148 frames.) The video sequences are packetized according to the IP Network
Adaptation Layer (NAL) specification of the H.264 standard and transmitted
using the RTP/UDP/IP protocol stack. At the receiver, a playout buffer mecha-
nism has been implemented to compensate the delay jitter of the packets. If not
specified otherwise, the playout buffer size has been set to 1 s.
When present, the interfering data traffic is carried by greedy TCP connec-
tions; the NewReno version of TCP is used.
At the MAC layer, the duration of the time slot and of the DIPS time interval
has been set to and respectively. When not specified otherwise, the
maximum number of transmission attempts is set to 2.
The 802.11b radio channel is modeled as a Gilbert channel. Two states,
good
and
bad,
represent the state of the channel during an 802.11b time slot: an
MPDU is received correctly if the channel is in state
good
for the whole dura-
tion of the MPDU transmission; it is received in error otherwise. We denote the
transition probability from state
good
to state
bad
by and the transition proba-
bility from state
bad
to state
good
by
q.
The average error probability, denoted
by is given by the average length of a burst of consecutive errors is
equal to 1/
q
time slots. In the simulated scenarios the value of
q
is set to 0.9 so
