| dc.description.abstract | Small size of wireless sensor nodes limit the size of the battery supported on it which directly limits the capacity of the battery and lifetime of the sensor nodes. Though energy scavenging is one of the solutions, it is not always reliable due to the unavailability of energy to be scavenged (solar, wind, vibrations etc) all the time. Wireless Sensor Networks (WSN) are required to work over longer period as they have been deployed to detect or monitor some rare events or objects. It creates the need for stringent power optimization at all the layers of sensor network. In order to have a WSN alive sufficient number of sensor node needs to be alive covering the sensing field and providing a route to the sink.
Aim of this research is to increase the life time of battery powered wireless sensor nodes along with the Quality of Service (QoS) ensured. QoS parameter considered is data loss due to buffer overflow. Each sensor node can individually and independently make decisions about its operating state depending on the current workload. QoS parameter is kept at utmost priority along with the power optimization. Increasing the lifetime of sensor nodes means decreasing the time averaged power consumption of processing and transmission units. This research creates the need for a dynamically reconfigurable sensor node in which computational as well as communication tasks can be carried out with different speeds and power as per the requirement. DVFS (Dynamic Voltage Frequency Scaling) and DMS (Dynamic Modulation Scaling) techniques help to increase the power efficiency of communication unit and radio unit respectively. In this thesis we are focusing not only on the simultaneous use of DVFS and DMS but also on coordinated integration of the two technologies what we have termed as Coordinated Adaptive Power Management (CAPM) technique.
We have modeled a wireless sensor node as a tandem queue. Microcontroller has been considered as first server and radio transmitter is the second one. In case of event detection applications, data traffic is not uniform all the time. In the absence of an event, data arrived in a node is very small, while data traffic increases suddenly when an event occurs. We have tried to vary the service rate of processor using DVFS and transmitter using DMS techniques as per the actual workload. Operating with lower service rate consumes less power and with higher service rate, it consumes more power. Though increasing the lifetime of a sensor node is our main objective, equally we are concerned with the data loss due to buffer overflows. So we are suggesting a power optimization technique that trades off between power consumption and buffer overflow probability. During the period of heavy traffic, reducing the data loss due to buffer overflow is of higher priority and during low traffic period, power saving is of higher priority.
Comparison is made for a sensor node with various capabilities- fixed service rate, only DVFS, only DMS, both DVFS and DMS together and coordinated with each other. After analyzing the tandem queue model of sensor node, we have considered a sensor node as a single server having DVFS and DMS coordinated inside it. By changing the service rate of sensor node (μ1, μ2,...), specific processing rate and transmission rate are selected internally. M[x]/M/1/N model has been used to capture bulk data arrival and represented with a Markov chain. As it has become very complex to solve Markov chain state equations for performance analysis, we have developed GSPN models of sensor node with various capabilities. Using SHARPE software analytical tool, we have analyzed all these models for various performance parameters. MATLAB simulations carried out and compared with the analytical results. For WSN applications where non uniform data traffic exists, CAPM has shown the lifetime improvement along with reduced buffer overflow probability.
We have analyzed the performance of a wireless sensor node under various data arrival rates and results obtained from analytical tool SHARPE and MATLAB simulations for various models of sensor node have shown that CAPM technique reduces the idle time probability and idle power wastage during normal periods and reduces data loss due to buffer overflow during catastrophic period and increases the lifetime of wireless sensor network node. As compared to fixed service rate sensor node lifetime increase of 15% was seen when only DVFS was implemented on a sensor node while implementing only DMS it was 17.5% but DVFS and DMS together (CAPM) applied on a sensor node resulted in 27.22% lifetime increase. All these results support the concept of CAPM technique. | |
