Digital Signal Processing Reference
In-Depth Information
Anticipative Energy and QoS Management
in which
i
denotes the number of combinations to select
i
fragments out of
m
.
The resulting probability to deliver a frame (
JFR
)is:
p
JFR
p
(
K
)
D
j
(P
e
,
K
).
1
−
=
(6.19)
j
=
0
Time and energy requirements are respectively given by:
TXOP
p
(
K
)
=[
mT
good
(
K
)
]+[
pT
bad
(
K
)
]
,
(6.20)
p
E
p
(
K
)
D
j
(
K
)
=
×{
mE
good
(
K
)
+
jE
bad
(
K
)
}
j
=
0
H
p
(
K
).
(6.21)
The expected energy for a given configuration is the sum of the probabilities that
the transmission will succeed after
m
good and
j
bad transmissions multiplied by
the energy needed for good and bad transmissions. A second term
Z
p
(
K
)
should be
added to denote the energy consumption for a failed job, hence when there are less
than
m
good transmissions, and (
p
Z
p
(
K
)
+
+
+
1) bad ones:
E
bad
(
K
)
m
−
1
D
j
p
+
Z
p
(
K
)
JFR
p
(
K
)
=
×
+
1
(
K
)
j
=
0
×
(jE
good
(
K
)
+
(P
+
1
)E
bad
(
K
))
.
(6.22)
The third term
H(
K
)
denotes the cost that has to be added once every scheduling
period. We will show later that this cost corresponds to a wake-up cost only and no
reconfiguration cost should be taken into account, where we assume that the cost for
each configuration
K
is constant.
We determine the Energy-Time-JFR trade-off as a function of the system state
and number of retransmissions for each
K
. This specifies the full profile for the
system, and is determined only once during design or calibration time. These can
then be combined into the 3D profiles of the system (Fig.
6.9
(a)). This cost, resource
and quality profile information is stored in each node's driver. We further prune the
3D profile to obtain per-flow Energy-TXOP curves. Configuration points that do
not meet the target
JFR
∗
are pruned. Next, the convex minorant is computed in
both Energy and TXOP dimensions. The resulting two-dimensional Pareto-optimal
trade-off curves are shown in Fig.
6.12
for the considered channel and frame size
scenarios.
6.4.2 Anticipative Control in the 802.11 MAC Protocol
Based on the Energy and TXOP curves for each node, the scheduler in the AP can
efficiently derive a near-optimal resource allocation at run-time using the greedy
