Image Processing Reference
In-Depth Information
performance and resource usage. Abstracting layers, such as HAL or OS, are often omitted due to
resource constraints and low-power aspects.
Microcontrollers are often developed and programmed for a specific, well-defined task. his lim-
itation of the application domain leads to high-performance embedded systems even with strict
resource constraints. Development and programming of such systems are too much effort. Further-
more, an application developed for one microcontroller is in most cases not portable to any other one,
so that it has to be reimplemented from scratch. Microcontroller and application form an insepara-
ble unit. If the application domain of an embedded system changes often, the whole microcontroller
must be replaced instead of writing and downloading a new program.
For sensor nodes, application-specific microcontrollers are preferred instead of general-purpose
microprocessors. This is because of the small size and the low energy consumption of those con-
trollers. However, requirements concerning a sensor node exceed the main characteristics of a
conventional microcontroller and its software. The main reason for this is the dynamic character
of a sensor node's task. Sensor nodes can adopt different tasks, such as sensor data acquisition, data
forwarding, or information processing. he task assigned to a node with its deployment is not fixed
until the end of its life cycle. Depending on, for instance location, energy level, or neighborhood of a
sensor node, a task change can become advantageous or even necessary.
Additionally, software for sensor nodes should be reusable. An application running on a certain
sensor node should not be tied to a specific microcontroller but to some extent be portable onto
different platforms to enhance interoperability of sensor nodes with different hardware platforms.
Not limited to software development for wireless sensor networks is the general requirement for a
straightforward programmability and, as a consequence, a short development time.
It is quite hard or even impossible to meet the requirements mentioned above with a monolithic
application. Hence, at present there is much research effort in the areas of middleware and service
architectures for wireless sensor networks. A middleware for wireless sensor networks should encap-
sulate required functionality in a layer between OS and application. Incorporating a middleware layer
has the advantage that applications get smaller and are not tied to a specific microcontroller. At the
same time, development effort for sensor node applications (SNA) reduces since a significant part of
the functionality moves from application to middleware. Another research domain tends to service
architectures for wireless sensor networks. A service layer is based on mechanisms of a middleware
layer and makes its functionality more usable.
12.2.1 Architectural Layer Model
Like in other networking systems, the architecture of a sensor network can be divided into different
layers (see Figure .). .).The lower layers are the hardware and HAL. .).The OS layer and protocols are
above the hardware-related layers. he OS provides basic primitives, such as multithreading, resource
management, and resource allocation that are needed by higher layers. Also access to radio interface
and input/output operations to sensing devices are supported by basic OS primitives. Usually in node-
level OSs these primitives are rudimentary and there is no separation between user and kernel mode.
On top of the OS layer reside middleware, service, and application layer.
In recent years, much work has been done to develop sensor network node devices (e.g., Berkeley
Motes []), OSs, and algorithms, for example, for location awareness, power reduction, data aggre-
gation, and routing. Today researchers are working on extended software solutions including mid-
dleware and service issues for sensor networks. The main focus of these activities is to simplify
application development process and to support dynamic programming of sensor networks. The
overall development process of sensor node software usually ends with a manual download of an
executable image over direct wired connections or over-the-air (OTA) interface to the target node.
After deployment of nodes, it is nearly impossible to improve or adapt new programs to the tar-
get nodes. But this feature is necessary in future wireless sensor networks to adapt the behavior of
