Image Processing Reference
In-Depth Information
11.4.5.2 Emulating the Environment
An apparent deficiency of current WSN testbeds is that they are typically set up in office environ-
ments and thus can only offer some realism with respect to the devices and an actual radio channel
used. hey still abstract from the real deployment as depicted in Figure .. But they fail to capture
the exact effects of a target deployment region, e.g., temperature variation, or changing operating
conditions, e.g., depleting power resources. However these effects are important factors that have
lead to a number of reported failures for real WSN deployments [,]. [,].The effect of temperature on
electronic devices is well known, often basically considered at design time but not elaborated in detail
in a system context toward the end of a development phase. Since software running on WSN devices
is rigorously optimized to the hardware, small changes in the operating environment can lead to a
malfunction of the system. In a similar way the capacity of any battery is influenced both by the dis-
charge behavior and the environmental conditions, e.g., the temperature. Today, this can neither be
emulated nor estimated and requires testing on a live object.
We therefore suggest to further augment a testbed/CI architecture with means to emulate and
control the environment in which nodes are deployed on a testbed. In practice this means that nodes
under test are to be placed in a temperature and humidity cycling [temperature cycling test (TCT)]
chamber and powered from different power sources, e.g., real batteries or preferably a programmable
power source to emulate the behavior of a battery or a solar cell. his allows for testing the response
of a system under test in both cyclic and boundary conditions, e.g., depletion of the battery. In com-
bination with the ability of certain testbeds to insert asynchronous events [], e.g., errors like a node
failureoradelayedstartofasubsetofnodes,worst-casescenarioscanbeemulatedonthesystem
under test. Mixed scenarios are easily feasible, with a portion of nodes in a TCT chamber and a por-
tion located outside subject only to the regular ambient conditions. Since in a mixed scenario the TCT
chamber affect the propagation of the radio signals, we propose to situate all antennas of the devices
under test outside of the TCT chamber using extension cabling. In cases where this is not feasible and
yet weak signal conditions need to be emulated, (programmable) attenuators could be incorporated
in the infrastructure. We have so far refrained from the increase in complexity and simply performed
test runs alternating antennas both inside the TCT chamber and outside.
11.4.6 Physical Parameter Extraction
For the physical characterization of motes, we employ two different approaches: On the one hand
we observe long-term trends to determine the development process effects on the characteristics.
With detailed snapshots of individual software builds, we perform an in-depth analysis allowing for
regression testing of physical parameters.
11.4.6.1 Long-Term Trending
Current testbeds only has means of profiling power consumption on select nodes []. High-
precision distributed energy and power monitoring will be possible when tools such as SPOT []
are available. Up to this point, due to the lack of inexpensive tools, widespread adoption has not
been possible. A supervision of all nodes under test is mandatory for assuring reliable operation and
sufficient test coverage. We are using a combination of a DSN [] node pair with a custom power
monitoring board (see Figure .) that uses the internal analog-to-digital convertor (ADC) on the
ATmegal and a current sense amplifier in combination with the network logging tool Cacti. he
DUT can be powered from different sources (battery, line power) and the power status is sampled
on request once every min. Although limited in precision and temporal resolution, basic long-term
trending with coarse granularity is very helpful for testbed supervision on long test sequences, giving
