Scalable H.264 Wireless Video Transmission over MIMO-OFDM Channels - Advanced Concepts for Intelligent Vision Systems - page 37

Fig. 1. Hierarchical prediction structure for SVC for a GOP size of 8

value for video encoding, and the physical layer parameters, RCPC coding rate
and the symbol constellation choice for the STBC block code. The optimization
is considered on a GOP-by-GOP basis and is constrained by the total available
bandwidth (symbol rate) B budget . We assume that the SVC codec produces L
layers μ 1 ,μ 2 ,...,μ L via a combination of temporal and FGS scalability. Then,
the bandwidth allocation problem can be described as:
{ QP ∗ , R c , M ∗ }

= gmin
{ QP , R c , M }

E { D s + c } s.t. B s + c ≤ B budget (4)

where B s + c is the transmitted symbol rate, B budget is the total available symbol
rate and E is the total expected end-to-end distortion due to source

and channel coding, which needs to be estimated as discussed in Section 5.
QP , R c and M are the admissible set of values for QP, RCPC coding rates
and symbol constellations, respectively. For all the layers of each GOP, QP ∗ =
{

{ D s + c }

, R c = and M ∗ = define

the QP values, the RCPC coding rates and the symbol constellations for each
scalable layer, respectively, obtained after optimization. The transmitted symbol
rate B s + c can be obtained as

QP μ 1 ,...,QP μ L } { R c,μ 1 ,...,R c,μ L } { M μ 1 ,...,M μ L }

L

R s,μ l
R c,μ l ×

B s + c = (5)

log 2 ( M μ l )

l =1

where R s,μ l is the source coding rate for layer μ l in bits/s and depends on the
quantization parameter value used for that layer; R c,μ l is the channel coding
rate for layer μ l and is dimensionless; M μ l is the constellation used by layer μ l
and log 2 ( M μ l ) is the number of bits per symbol.
The problem of Eq. (4) is a constrained optimization problem and is solved
using the Lagrangian method.

5 Decoder Distortion Estimation

In order to perform the optimization of Eq. (4), it is necessary to estimate the
expected video distortion at the receiver E .Inthispaper,weusethe

distortion estimation technique of [11] and we also propose a new technique.

{ D s + c }