Digital Signal Processing Reference
In-Depth Information
be extended to a broader range of quality targets. We note again that, next to this
explicit QoS constraint that can be adapted or configured at run-time, application
bitrate constraints are implicit constraints given through by the use of scenarios.
3.
Shared Resource
(
R
i
): In this dissertation, we consider the particular case where
access to the channel is only divided in time. The fraction of resource consumed
by node
i
is denoted by
R
i
.
To summarize, each flow
F
i
is associated with a set of possible system states
S
i,m
,
which affects the mapping of the control dimensions
K
i
to the Cost (
K
i
,S
i,m
→
C
i
),
Resource (
K
i
,S
i,m
→
Q
i
) that will be specified in
the next section. It is essential to note that for each user, depending on the current
state, the relative energy gains possible by rate scaling or sleeping are different and
should hence be exploited differently. Each user experiences different channel and
application dynamics, resulting in different system states over time, which may or
may not be correlated with other users. This is a very important characteristic which
makes it possible to exploit multi-user diversity for energy efficiency.
As the system performance requirements are specified at the application layer and
the energy consumption is at the lower (hardware) layers, it is essential to: (a) trans-
late the application layer requirements to relevant metrics at each intermediate layer
and (b) to define clean interfaces between layers for an Energy-Performance feed-
back mechanism (see Fig.
6.7
). This is to allow for a local calibration of the hard-
ware which makes implementation more feasible, still enabling translation of to
application specific quality metrics. For delay-sensitive traffic, the QoS metric of
interest is the target
JFR
∗
.
At the link-layer, each application
frame
is fragmented into one or more fixed-
sized
fragments
. An application frame size or rate requirements are hence translated
into a local Queue Size. Next, as shown in Fig.
6.7
,the
JFR
∗
is translated at the
link layer to a PER constraint, which corresponds to a maximum Block Error Rate
(BlER) as a function of the physical layer low-level knobs. The BlER is a result of
the receive SINAD for given PHY parameters. Each target JFR may be satisfied by
one or more control dimension configurations,
K
i
, each associated with the energy
consumed (cost) and time required (resource) to complete the frame transmission in
the current system state. The state is defined by a discrete channel state and traffic
requirement (i.e. current frame size), which can easily be monitored as the Queue
Size. Channel classification and monitoring is typically a more difficult problem.
At run-time, based on a node's current system state and
JFR
∗
, the corresponding
Cost-Resource set of points are fetched from memory. From each curve, configu-
ration settings are then chosen using the fast greedy algorithm such that the total
transmission time for all nodes is less than the deadline. We first discuss the control
dimension mapping to obtain those Pareto-optimal trade-off sets.
R
i
) and Quality (
K
i
,S
i,m
→
6.2.4 Anticipating the Performance
The key aspects of the method for design of an energy efficient cognitive radio
are the mapping of the control dimensions to cost, resource and quality profiles
