Digital Signal Processing Reference
In-Depth Information
Fig. 21.3 Two basic sample schemes using generic FEC as defined in RFC 5109
FEC data that takes a set of packet payloads and applies the binary exclusive or
(XOR) operation across the payloads. This scheme allows the recovery of missing
data in the case where
one
of the original packets is lost, but the FEC packet is
received correctly. The RTP payload format for using generic FEC based on XOR
operations has been published in RFC 5109 [
4
].
In recent years, several proposals have been made to use well-known error-
correcting codes, such as Reed-Solomon [
5
] codes, for packet loss recovery as well.
However, the weakness of the more complex schemes is the computational com-
plexity, which may cause performance problems with long packets and a large
number of parity packets. This is why we limit the scope of this chapter to XOR-
based FEC codes only. Nevertheless, the basic principles discussed here can be
easily extended to other kinds of linear codes.
Figure
21.3
shows two basic schemes using the generic FEC defined in RFC
5109. In this chapter, we adopt the definition of function
f(x, y,
)
to denote the
resulting FEC packet when the XOR operation is applied to the packets
x, y,
...
.
In example (a), a single packet loss every three packets (in the original media
stream) can be recovered, and in example (b), every packet loss can be recovered,
assuming that the FEC stream is received correctly in both cases.
Clearly, both schemes require more network bandwidth because of the redun-
dancy overhead. Example (a), that is denoted FEC 3:1, introduces an overhead of
33% since an FEC packet is sent every three data packets, while example (b), that is
denoted FEC 1:1, introduces an overhead of 100%. In general, an FEC i:1
introduces an FEC packet for every i data packets, causing an overhead of
(100/i)
%.
In practice, the media stream and the FEC stream are usually transmitted using
the same transport medium. This is why we cannot expect packet losses to occur
only in the media stream as both streams are likely to suffer from similar error
characteristics. In the network perspective, it is realistic to assume the media stream
and the FEC stream to form a single stream containing both media and FEC
packets. Given a sequence of media and FEC packets, we can easily see the
variation in error recovery rates when we examine the residual media data loss
rate after applying different kinds of FEC patterns to the sequence. In Fig.
21.4
,we
plot the packet loss rate at the network level for a real wireless inter-vehicle
...
Search WWH ::
Custom Search