Image Processing Reference
In-Depth Information
supply. For such heavily energy constrained systems, power consumption needs to be meticulously
minimized. Since power consumption is such a pivotal factor for a WSN system, the integration
of tests especially targeted for power consumption is required throughout the whole design and
development process. Code size is another critical factor as program memory is limited. Compile
time checks integrated into a automated build framework address these issues. Another vital aspect
of the test architecture is visualizing the trends of program memory usage as individual software
components are added to the design.
11.1.1.2 Communication Intricacies
Radio communication is a characterizing trait of WSNs. It introduces unreliability in message trans-
missions and incurs significant power consumption. Unreliable communication requires each node
to incorporate countermeasures for failed transmissions due to interferences and collisions. The
broadcast nature of the medium necessitates arbitration of the broadcasting nodes to avoid collisions.
Noise, anisotropies, fading, and multi-path effects deteriorate packet reception rates.
A particular aspect of typical WSNs is that computation is cheaper than communication, i.e.,
E
comp
E
comm
. hus processing of data fundamentally consumes less power than communication
of data [,]: While computation is not for free, sending kb of information across m consumes
the amount of energy required for executing millions of instructions on a general-purpose micro-
processor. In general, energy for transmitting and receiving are comparable in power consumption.
Even idle listening is comparable in its current draw, necessitating coordination in transmission to
avoid wasting energy. As communication is expensive in terms of energy costs, a general goal is to
minimize the amount of messages to be transmitted or received. A customized design must consider
trade-offs concerning the accuracy, periodicity, and latency requirements of the sensor data.
Considering the small amount of payload the sensor nodes typically transmit, the control overhead
concerning the maintenance and protocols are significant. Additionally, debug or health information
as well as design for testing mechanisms have to be carefully selected as they can have a significant
impact on these low data transport applications. However, the protocol stack itself can benefit from
additional visibility []. There is an inherent trade-off between the observability of a system and
its efficiency in the use of resources. A system spending the majority of the time in a sleep state or
accessible only by a highly bandwidth-limited and unreliable wireless link is challenging to observe,
which is a requirement for the successful analysis and identification of causes of error.
Execution information typically does not allow for standard analysis of the communication as
the interpretation heavily depends on customized, nonstandardized protocols and the instrumenta-
tion of the system []. As an example, individual message transactions may be observed by logging
n
send
≪
k
sending attempts. Due to lost messages, missed acknowledges and retransmission attempts,
the number of according reception successes
n
succ
=
. However, instrumentation on the appli-
cation level may return just a single notification of transmission success on each the sender and the
receiver for multiple send and receive events on the medium access control (MAC) layer. Although
the same action has occurred, the event traces and their interpretation differ considerably.
∈[
,
k
]
11.1.1.3 Distributed System State
In WSNs, system state is distributed over the sensor nodes and messages. Sensor nodes are highly
concurrentandonlylooselycoupledthroughunreliable,wirelesscommunication.Testingofa
comprehensive WSN system includes the checking of predicates on the distributed, global state.
Observing global state via local instrumentation or passive inspection raises issues of consistency
of snapshots and causality []. Concurrent and thus partial ordered actions in between synchro-
nization events, produce a substantial number of linear extensions, each a valid representation of
system execution. Probe effects are aggravated as the probed device and its context in the system are
perturbed. hus, the same test inputs may result in different outputs for individual runs.
