Digital Signal Processing Reference
In-Depth Information
21.5.1 FEC and Interleaving Real-Time Protection
Technique (FIR)
The proposed XOR-based adaptive FEC scheme uses the averaged loss rate
p
reported periodically by RTCP to adjust the amount of redundancy (FEC) to be
transmitted. XOR-based FEC protocol produces an additional redundancy packet
from
k
media packets, and it has the capacity to overcome a single packet loss over
the
n
k+1
consecutive packets. This provides resiliency against a maximum
packet loss rate of
p
¼
1/n when considering that even FEC packets may be
affected by loss. Thus, based on the averaged packet loss rate measurements such
as that provided by the RTCP feedback, it is possible to constantly adjust the
redundancy amount by changing the number of media packets (
k
) covered by the
FEC packet as follows:
¼
1
p
k
¼
1
(21.2)
The maximum acceptable loss rate threshold beyond which the streaming server
triggers FEC adaptation may differ depending on the nature of the audiovisual
content and its loss resiliency characteristics (e.g., according to Eq.
21.2
, if the
maximum threshold is set to 10%, the maximum value of
k
has to be set to 9).
The other dimension of the interleaving matrix, i.e., the number of rows (
m
),
depends on the overall delay that can be tolerated by the real-time application. The
total end-to-end delay consists of three components: the codec delay, the network
delay, and the playout delay. The latter is set according to the jitter introduced by
the network transmission, and when interleaving is used, it should be increased so
that it can accommodate also the interleaving delay. Playout buffer size is set by the
receiver at the beginning of the transmission, before media decoding and, in the
simplest scenarios, it is usually kept constant. So, if we denote it by
d
po
¼
d
j
+d
i
,
where
d
j
corresponds to the jitter component and
d
i
to the interleaving component,
the value of
m
can be dynamically calculated from Eq. 21.1 as a function of
d
i
and
n
as
d
i
m
¼
(21.3)
n
1
An additional issue that must be considered is that the FEC adaptation model
poses a problem when dealing with channels that exhibit varying packet loss rates
over time. The frequency of the receiver reports, which give to the sender an estimate
about the network loss rate and other parameters, may reduce the responsiveness of
the FEC scheme, leading to suboptimal FEC efficiency. A high frequency would
enhance the responsiveness at the sender while causing high variations between
successive measurements and possibly leading to instability, not
to mention
Search WWH ::
Custom Search