WU et al.: ENERGY-MINIMIZED MULTIPATH VIDEO TRANSPORT TO MOBILE DEVICES 1163
Fig. 2. Overall design of the proposed EVIS (Energy-Video aware multIpath tranSport) protocol.
Sender Components: The working components at the
sender side include the prioritized frame scheduler and
unequal loss protector. In particular, the scheduling algo-
rithms for these decision processes are described in Section IV
(Algorithm 1–2). The main control parameters in the Raptor
coding process involve the code rate and encoded symbol size.
For each decision epoch, the video frames are converted into
encoded symbols and encapsulated in different data packets.
The data distributor will partition the coded video traffic into
multiple sub-flows based on allocation vector over the available
paths. These sub-flows are transmitted to the destination using
the UDP sockets. The data packets allocated for each path will
be temporarily stored and queueing at the send buffers before
being delivered. To guarantee the fairness with other TCP
flows, a TCP-friendly rate control (TFRC) [32] unit is adopted
to regulate the transmission rate of each sub-flow. However,
the TFRC may unnecessarily reduce the transmission rate if
wireless losses occurs. Therefore, the ZigZag scheme [33] is
employed for the packet loss differentiation (i.e., distinguish
congestion losses from wireless losses).
Receiver Components:
At the receiver side, EVIS reorders
the received packets from multiple paths and checks whether
they have passed the decoding deadline. The overdue packets
will be dropped since they cannot contribute to the video dis-
play process. If any data packet is lost during transmission,
the Raptor decoding is performed to recover the lost packets.
It is a interesting solution to leverage the overdue frames as
the decoding reference of subsequent frames to mitigate drift-
ing distortion. The information feedback unit is implemented
at the receiver side to periodically provide the network status
information (available bandwidth, round trip time, packet loss
rate) through a most reliable communication path to the sender
side. In this research, we employ the multipath status estimation
model developed in [28] to periodically capture the physical
characteristics of each communication path. These information
are involved in the decision process of FEC coding and data
distribution at the sender side. Each data packet is associated
with a transmission sequence number for the traffic reassemble
at the receiver side.
The decision process for the energy minimization with qual-
ity and delay constraints includes the frame scheduling and
loss protection. This joint optimization problem involves the
TABLE II
M
ATHEMATICAL NOTATIONS
mathematical models of heterogeneous wireless network [4],
video quality [7], systematic Raptor coding [34], and energy
consumption of mobile device [35]. For the sake of complete-
ness and integration with the system framework, these models
are briefly introduced in this section. The basic notations used
in this paper are summarized in Table II.
B. Model Description
Heterogeneous Wireless Network
Without loss of generosity, we consider a heterogeneous
wireless network environment integrating P access networks.
Each access network represents an independent end-to-end
communication path and includes wired-cum-wireless domains
between end devices. In this research, each communication
path is characterized by the physical properties of round trip
time RT T
p
, packet loss rate π
B
p
and available bandwidth
μ
p
. The detailed descriptions of these physical properties are
presented in [4].