Image Processing Reference
In-Depth Information
4.4.5 Cross-Layer Optimization
Traditional system architectures are commonly known to adhere to the layered protocol stack design
where every layer operates in a completely independent manner. he Internet browser on a PC, for
example, does not require any knowledge about the kind of network connectivity available. It works
regardless of whether the user is connected via Ethernet or .b. Such a layered approach however,
is not optimal from the energy efficiency point of view especially when considering WSNs. This is
because unlike the network being used in an office LAN, WSNs are typically application-specific
networks. hus it only makes sense to try and make the various components of the WSN architecture
more application-aware by performing certain cross-layer optimizations thereby helping to improve
network lifetime. Note however, that while the term “cross-layer optimization” in the field of sensor
networks might refer to the optimization between the application and routing layers for instance,
in this case we refer to the more radical approach where optimization is performed between the
application and MAC layer.
Asusageofthetransceiverisanenergy-consumingtask,itisimperativethatmaximumbeneitis
derived during the time it is operational. Thus rather than encountering energy-wasting collisions
during data transmission or actively waiting for messages that do not arrive, current cross-layer opti-
mization techniques use a variety of methods to try and schedule tasks in an energy-efficient manner.
Wenowreviewanumberofthesetechniques.
TinyDB [] uses an interval-based communication scheduling protocol to collect data where par-
ent and child nodes receive and send data (respectively) in the tree-based communication protocol.
The cross-layer optimization in TinyDB involves (i) reducing the burden on the MAC using specifi-
cations from the injected query and (ii) routing data from the source nodes to the root. Each node is
assumed to produce exactly one result per epoch (time between consecutive samples), which must
be forwarded all the way to the base station. As shown in Figure ., every epoch is divided into a
number of fixed-length intervals which is dependent on the depth of the tree. he intervals are num-
bered in reverse order such that interval is the last interval in the epoch. Every node in the network
is assigned to a specific interval which correlates to its depth in the routing tree. Thus for instance
if a particular node is two hops away from the root node, it is assigned the second interval. During
its own interval, a node performs the necessary computation, transmits its result and goes back to
sleep. In the interval preceding its own, a node sets its radio to “listen” mode collecting results from
its child nodes. hus data flows up the tree in a staggered manner eventually reaching the root node
during interval .
While TinyDB's slotted scheduling protocol does help conserve energy by keeping nodes asleep
a significant proportion of time, it is primarily designed for servicing a single query posed to the
entire network. he scheme is unusable if there are multiple concurrent queries with different epoch
requirements.
Chatterjea et al. [] describe an Adaptive, Information-centric, and Lightweight MAC (AI-
LMAC) protocol for WSNs that adapts its operation based on the requirements of the application.
The amount of bandwidth that is allocated to a node is proportional to the amount of data that is
expected to flow through it in response to the query it is servicing. Bandwidth allocation is done in a
distributed manner and is not static but changes depending on the injected query. Information about
the expected data traffic through a node is obtained using a completely localized data management
framework that helps capture information about traffic patterns in the network. A major advantage
of this approach is that the MAC protocol reduces its duty cycle on nodes that are not taking part
in servicing the query, thus improving energy efficiency and limiting communication activity only
to areas of the network where it is actually required. he data management framework is also used
for efficient query dissemination (i.e., directing queries to only the relevant parts of the network)
and query optimization. hus cross-layer optimization in AI-LMAC addresses the entire spectrum
of data management issues, i.e., operation of the MAC, routing, and query optimization.
