Image Processing Reference
In-Depth Information
optimized for its operational use in order to minimize costs. For sensor nodes, the inter-
faces are limited to the radio, a serial interface, individual pins, and some LEDs. Radio
and LEDs are expensive in their power consumption, limiting their use in real deploy-
ments with batteries. In order to avoid probe effects [], using the radio for inspection
shouldbeavoidedduringtestingthesystem.Passiveinspection[]allowingformoni-
toring a system without perturbation is limited in its applicability in testing as fails cannot
be attributed to the system as the platform itself cannot provide any guarantees on a reli-
ability. Off-line gathering of data collected in unused memory during the test is a viable
option. However, memory access can interfere radio communication when they share
a bus, e.g., on the Tmote Sky Platform. LEDs merely allow for primitive instantaneous
checking.
Simulation tools allow for detailed inspection of the system. The monitoring is reliable
and does not perturb the system under test. Platforms offering high visibility allow for
extensive logging to find correlation between log events. Substantial data gathered with
TOSSIM [] allows for diagnostic simulation using data mining techniques []. For
testing on real sensor nodes with lower visibility, monitoring data is focussed to detect
merely a specific error or failure. Monitoring data volume is highly test-specific: Test-
ing for reboots under environmental conditions merely requires a single bit, while tests
concerning distributed protocols such as for routing incur a substantial logging amount.
.
Abstraction
Testing the system in its final deployment location and its ultimate scale is typically pro-
hibitive. hus, test platforms have to abstract from the final system and build models for
the testing (cf. Figure .). The models include device, network, communication, and
environmental abstractions.
(a) Device
While execution on a sensor node and tests using an instruction-level simulator use
the comprehensive compiled application, simulation tools typically abstract from
the device. TOSSIM uses actual NesC code, but provides simulation libraries for
hardware-dependent code. Other discrete event simulators such as GloMoSim []
or Castalia [] are platform and operating system agnostic, nevertheless allowing for
modeling comprehensive communication models.
(b) Network and communication
Network topology and radio communication have a substantial impact on a sensor
network. Wireless communication over low-power radios is varying in time and dis-
tance. he result is transitional, asymmetric, and irregular links. his is exacerbated
when the topologies become complex and obstructions create multipath and fading
effects. This fact has a considerable impact on the protocol stack. When simulating
sensor networks great care must be taken to accustom for this considerable impact.
Models must be finely tuned to represent an actual deployment. Overly simplis-
tic models and assumptions [] hide the complexity and nondeterministic behavior
found in actual WSN deployments.
(c) Environment and energy
Sensor nodes are deeply embedded with its environment. As such a deployment is
heavily dependent on its environment: Temperature and humidity affect node oper-
ation and communication. Solar energy to be harvested and battery drainage heavily
affect longevity and sustainability. Simulators typically do not consider these effects.
Indoor testbeds may incorporate special instruments to allow for controlling bat-
tery profiles of a power supply. Temperature effects for alpine environments can be
simulated by using climate chambers.
