WSN - one platform, two approaches

Post date: Oct 20, 2014 8:16:54 AM

Wireless sensor network (WSN), or sensornet, is a promising sensing and computing platform for all spectrum of electrical and computer engineering. From an electrical engineering perspective, WSN represents a significant twist on the traditional, mature field of statistical inference: distributed inference. On the other hand, computer engineers often perceive WSN as opportunity for new, innovative architectures which are energy-efficient. It should not be surprising to learn that each of these approaches has its own pros and cons, simply because they do not look at the platform as a whole (Think the classical story: blind men and an elephant). 

The architecture-oriented approach often focuses too much on the architecture and overlook the importance of the data itself. Therefore works pursuing this approach often make simplifying assumptions on the sensor readings, such as they are simply imperfect representation of the physical process. The plus side is that they can truly create energy-efficient architecture for the simple data type that was initial assumed!

On the other school, inference-focus approach often fails to create truly energy-efficient system (or any system at all), due to the fact that important bottlenecks in practical systems are often neglected, or unforeseeable from theory. However, this school of thought does not make artificially simplifying assumption on the nature of the data and can technically deal with any type of data, including unstructured data such as audio and video.

Therefore a holistic approach is necessary to design the optimal WSN, in the energy/performance trade-off sense.