Digital Signal Processing Reference
In-Depth Information
Fig. 21.5 Packet interleaver
with block size
n ¼
4 and
interleaving depth
m ¼
3.
Packets are transmitted by
columns following the
sequence numbers in brackets
interleaving provides the advantages of (1) being computationally simple and (2)
not requiring any increase in bit rate. Furthermore, packet interleaving can easily be
coupled with FEC techniques.
A potential drawback of packet interleaving is that it requires additional delay.
Interleaving delay is of particular concern in high interactive applications, such as
Internet telephony, that cannot tolerate a delay above 400 ms [
7
]. However, the
required delay, which depends on channel burst length characteristics, can gener-
ally be bound to relatively short values, so even in this kind of applications, the end-
to-end delay introduced by this technique is usually acceptable. Since many
approaches for interleaving exist, we introduce the specific packet interleaving
strategy used in this study.
A simple packet interleaver that permutes the packet transmission order is
represented in Fig.
21.5
. At the sender, packets are first written into the interleaver
in rows, with each row corresponding to a block of
n
packets; among them,
k
1
are data packets, and the last one is a XOR-based FEC packet. Then the packets are
transmitted by columns as soon as
m
rows of packets fill up. At the receiver, when
packets are reordered using their timestamp and sequence number, loss bursts are
converted into separated losses. Let us consider, for example, the case of a transmis-
sion channel afflicted by a burst loss of length three occurring during the transmission
of the first three packets. Using the
(n, m)
interleaver shown in Fig.
21.5
, the burst loss
affects separated packets 1, 4, and 7 instead of successive packets 1, 2, and 3.
The effectiveness of the interleaver depends on the block size and the
interleaving depth as well as the loss characteristics of the channel. With an
interleaving depth of
m
, a burst loss of length
B
can be converted into a shorter
burst with a maximum length of
¼
n
d
B/m
e
, where
d
x
e
denotes the smallest integer not
smaller than
x
. In an ideal case, when
m
B
, the burst loss can be converted into
isolated losses.
In this case, the separation between any two losses is either
n
or
n
1. A larger
interleaver is more effective in that it can convert a longer burst loss into isolated
losses or increase the separation of the converted isolated losses. However, this is at
the cost of higher latency. At the client, an interleaved packet received cannot be
used until all the packets it depends on are received. For an
(n, m)
interleaver, the
nth packet in the original order suffers from the highest delay, as it has to be
transmitted in the ((
n
1)
m
) + 1th place. Hence,
the decoding delay
corresponding to an
(n, m)
interleaver is
ð
n
1
Þ
m
(21.1)
Search WWH ::
Custom Search