Interplay of Communication and Computation Energy Consumption for Low Power Sensor Network Design

International Journal of Ad hoc, Sensor & Ubiquitous Computing (IJASUC) Vol.3, No.4, August 2012
DOI : 10.5121/ijasuc.2012.3407 65
INTERPLAY OF COMMUNICATION AND
COMPUTATION ENERGY CONSUMPTION FOR LOW
POWER SENSOR NETWORK DESIGN
Zeeshan Ali Khan, Mustafa Shakir
Department of Electrical Engineering,
COMSATS Institute of Information Technology,
Islamabad, Pakistan.
{zakhan,mustafa.shakir}@comsats.edu.pk
ABSTRACT
The sensor network design approach normally considers the communication energy consumption for
evaluating a communication protocol. This is true for the low power devices such as MICAz/MICA2
which do not consume a lot of energy for the data treatment. However, recently developed sensor devices
for multimedia applications such as iMote2 do consume considerable amount of energy for data
processing. In this article, we consider various scenarios for routing the data in wireless multimedia
sensor networks by considering the local design parameters of devices such as PXA27x and beagleboard.
The proposed routing solution considers node level optimizations such as data compression, dynamic
voltage and frequency scaling (DVFS) for making a routing decision. The proposed approaches have
been simulated to prove the effectiveness of the approach.
KEYWORDS
Wireless Sensor Networks, Low Power Design, Communication versus Computation Energy
Consumption, Real-time data delivery.
1. INTRODUCTION
The wireless sensor networks are employed in various monitoring applications. The main
optimization aspect of the sensor network is to minimize the energy consumption by choosing
an energy efficient data dissemination strategy. These routing strategies normally consider the
communication energy consumption and ignore the computation energy consumption. This is
true for sensor devices such as MICA2/MICAz/TelosB as they normally have a micro controller
that operates at the lower frequency which minimizes the computation energy consumption.
However, the multimedia sensor network applications have to do a lot of computation before
sending the data on the wireless networks. Thus, in this case we cannot ignore this energy
consumption while making a decision at the routing layer. The strategies that significantly effect
the computation energy consumption include the video compression in the multimedia
transmission that consumes a considerable amount of energy.
The article is organized as follows; section 2 describes the context of the presented work with
respect to the related work in the field of energy efficient routing and node level optimizations
in wireless sensor networks. Then, the chosen platform and proposed routing solution is
explained in sections 3 and 4. Finally, the experimental results are explained in section 5.

66
2. CONTEXT
A large volume of sensor network research is focussed on applications where the processor is
only capable of dealing with low data rate applications such as temperature and pressure
monitoring. However, there is a new paradigm in research where there are sensor network
applications which have to process a large amount of data. Here, the existing techniques can not
be directly applied [1]. In general, the sensor networks are deployed in regions where it is
difficult to replace the battery. Thus, it is necessary to have power efficient algorithms at the
system as well as network layer. In this case, we need to consider the local optimizations such
as data compression, dynamic voltage and frequency scaling (DVFS) and dynamic power
management (DPM) while making a routing decision. Thus, we need to find an interrelation
between node level and network level design techniques in order to optimize the overall design.
There are some techniques where authors have developed the power consumption model of a
specific device and used it to make the energy efficient algorithms [2].
There is a lot of work that has been done using DVFS in the wireless sensor networks. These
include [3] which propose an energy efficient task scheduling for structural health monitoring
by considering dynamic voltage and frequency scaling. [4] explains the design of a low power
video monitoring application on wireless sensor network using 802.15.4 protocol. They also
explain the power consumption strategy by using the DVFS strategy at the system level.
Now we describe how this type of routing can be used inside a wireless sensor network
deployed using imote2 or TelosB. The energy consumption profile for these nodes is listed in
table 1 for different data compression techniques such as lossless, lossy and raw technique. The
raw technique sends the data without any compression, lossless compression ratio reduces the
size of data by 3 and lossy compression reduces the size of data 10 times. For raw data
transmission, the energy consumption on imote2 platform is 1242J while this energy
consumption is 374.4J on the TelosB platform. The energy consumption for the lossless
compression is 835.5J for imote2 platform while it is 4618J for the TelosB platform. This shows
that local optimizations such as compression can significantly reduce the device’s overall
energy consumption by reducing the size of transmitted data. This energy consumption is
computed for the network diagram shown in figure 1. The results are listed in table 2 and it
shows that iMote2 has higher energy consumption for the transmission, but we can efficiently
compress the data using this platform to decrease the global energy consumption.
In this article, we shall discuss first the local optimizations which are possible and then
we shall explain their relation with the network layer design.
Figure 1. Example Network

67
Table 1. Compression Energy Consumption for TelosB versus iMote2 [1]
.
Imote2 (J) Telos (J)
Computation Transmission Total Computation Transmission Total
Raw 0 1242 1242 0 374.4 374.4
Lossless 7.7 827.7 835.5 4369 250 4618
Lossy 12.3 124.1 136.4 6480 37.4 65.7
Table 2. Analytical Evaluation on Example Network (Figure 1)
.
Imote2 (J) Telos (J)
Max. Energy
Consumed
Min. Energy
Consumed
Max. Energy
Consumed
Min. Energy
Consumed
Raw 6210 1242 1872 374.4
Lossless 4146.2 835.5 5619 4618
Lossy 632.8 136.4 6667 6517
3. PLATFORM
We have chosen two platforms for evaluating the proposed idea. The first chosen platform is
PXA27x processor is an integrated microprocessor which has dynamic voltage and frequency
scaling. The available frequencies are in the range from 13 to 416 MHz. The tables III and IV
shows the operating points and corresponding power consumption that is possible in the
PXA27x processors [2]. We take these operating points in order to find out an optimal
frequency for the processor that is involved in mild or higher processing activity. This will help
in finding out the optimal operating point for a processor in an offline manner.
The second platform chosen for this purpose is OMAP3530 (beagleboard) and table V provides
measures of the OMAP 3530 processor power consumption at different operating frequencies. It
is taken from [5]. Also [4] gives the following observation about the beagleboard. Typically,
• If the bit rate is less than 2KB/sec, CPU should run at maximum 125MHz.
4. ROUTING PROPOSAL
The network architecture consists of different nodes that can operate at different frequency
levels. If a node operates at the higher frequency then it will be able to process the task fastly,
consume more energy and this will be helpful in determining the optimal operating frequency.

68
The problem is modelled in subsection 4.1 and the routing solution is proposed in the
subsections 4.2 and 4.3 for PXA27x and Beagleboard respectively.
4.1. Model
The network consists of N nodes n=1,2,...,N and each node can be operated on M discrete
frequency levels m = 1,2,...,M. The processor activity level is shown with the help of a variable
a=1,2,...,A. We have to map this processor activity level to a certain frequency level m so that
the node n can operate on that frequency. This will help in determining the optimal processing
frequency for a certain processor that meets the data reporting deadline and also helps in
optimizing the overall network energy consumption. The processor activity level depends
whether this node is located close to the sink or not. In that case, it will have a large processor
activity otherwise it will have a low processor activity level. This will help in determining the
voltage on which we want the device to operate.
4.2. Frequency Selection for PXA27x
The mapping function will map the node to a certain frequency in an offline manner. The
processor activity level is determined with the help of the position of node with respect to the
sink. The node which is located near the sink has the higher activity level as it has to gather the
packets that have been generated from the far-off node. Thus, we choose:
a α (1 / Distance of node from the sink)
Where, the sink is the node in the wireless sensor network which is responsible for collecting all
the data. Thus, we choose:
h = Number of hops from the sink
In order to choose a frequency for the processor, we map the variable a to the frequency such
that,
f α (1/h)
And we try to choose the minimal frequency as well to minimize the energy consumption. As
we have selected various operating points and each point has specific power consumption. We
choose the frequency of a processor based on its activity level, a.
4.3. Frequency Selection for Beagleboard
The mapping function for beagle board depends on the processor activity which depends on the
data rate that can be supported. For a certain processing activity, we can use the supported data
rate listed in the previous section. This gives the maximum frequency that can be assigned to a
node w.r.t its distance from the sink (in number of hops) from the sink. If distance from sink is
larger, a lower frequency will be selected and higher frequency will be chosen for a node
located near to the sink.

69
5. EXPERIMENTAL ANALYSIS
We choose the network shown in figure 2 for evaluation and the simulation parameters are
described in table 3. The ideal MAC is considered in this case and we use a C++ based
simulator to perform the experiments. The proposed DVFS approaches are implemented on the
chosen multimedia wireless sensor nodes. The experiments and the results are described in the
following sub-sections.
Figure 2. Simulated Network
Table 3. Simulation Parameters.
Parameter Value
Network size 17
Routing Technique Geographical
Traffic Generation Model Constant Bit Rate

70
5.1. PXA27x
The properties for PXA27x are shown in tables 4 and 5. We use these properties to select the
frequency based on the frequency selection function described in section 4.2. If we consider
three processing frequencies such as 125, 250 and 500 MHz for sending the data from source to
sink, then the processing results are described in table 6 considering that the transmission rate is
not a bottleneck to the transmission rate. These results show that for processing an application,
we can use the proposed design approach for data reporting. As its end-to-end delay lies
between the approaches where either all the nodes are operating at the minimum or the
maximum frequencies. Choosing lower frequency will decrease the processor power
consumption but it will increase the end-to-end packet reporting delay where each application
will take 5 micro seconds to process on the processor for the lowest frequency. Choosing
maximum frequency for all the nodes increases the power consumption but decreases the end-
to-end delay.
Table 4. Clock Frequencies for the PXA27X processor [2].
OP Core (f) Bus (f) Memory (f) SDRAM (f) Core(f)
1 13 13 13 13 850 mV
2 104 104 104 104 900 mV
3 208 208 208 104 1125 mV
4 312 208 208 104 1200 mV
5 416 208 208 104 1250 mV
Table 5. Operating Point Characterization for the PXA27X processor [2].
cpu-bench mem-bench
OP Power (mW) T(sec) Power(mW) T(sec)
5 568 2.21 616 0.96
4 432 2.95 472 1.18
3 304 4.41 336 1.65
2 172 8.83 188 3.18
1 108 74 112 25.37
Table 6. Experimental Results for PXA 27x.
Max. Power
Consumed
Min. Power
Consumed
End-to-end Delay
Proposed Approach 303 mW 228 mw 13.75 micro sec.
Min. Energy 228 mW 228 mW 20 micro sec.
Max. Energy 303 mW 303 mW 5 micro sec.

71
5.2. Beagleboard
The power consumption versus frequency for a particular application executing on Beagleboard
is listed in table 7. The frequency for this board is selected based on the processor activity level
described in section 4.3. Here, we know the data rate that can be achieved for a particular
frequency and we use this value to calculate the bottleneck data rate supported. This is shown in
table 8 where we can select the processor frequency to minimize the processor power
consumption by increasing the data rate. The proposed approach for frequency selection and the
maximum frequency selection approach have the best data rate at the network bottleneck and
the energy consumption of the proposed approach is comparable to the minimum frequency
selection approach.
Table 7. Power Consumption versus Frequency for Beagleboard [5].
Frequency (MHz) P (mW)
125 57
250 130
500 303
550 371
600 445
Table 8. Experimental Results for Beagleboard.
Max. Power
Consumed
Min. Power
Consumed
Data Rate at
Network Bottleneck
Proposed Approach 303 mW 228 mw 8 KB/sec
Min. Energy 228 mW 228 mW 2 KB/sec
Max. Energy 303 mW 303 mW 8 KB/sec
6. CONCLUSIONS
In this article, we have coupled various node level optimizations with the global optimizations
to help in further optimizing the global power consumption in the wireless sensor networks. The
local node level optimizations that have been considered include data compression and DVFS
and these techniques were coupled and compared with the routing energy consumption for
different network scenarios. The results show that we can achieve better end-to-end data
reporting delay and minimize the overall network energy consumption by considering the node
level optimizations while making a routing decision.
REFERENCES
[1] L. Nachman, J. Huang, J. Shahabdeen, R. Adler & R. Kling, (2008) “IMOTE2: Serious
Computation at the Edge”, Proceedings of International Wireless Communications and Mobile
Computing Conference, Crete Island.
[2] A. Castagnetti, C. Belleudy, S. Bilavarn & M. Auguin, (2010) “Power Consumption Modeling
for DVFS Exploitation”, Proceedings of 13th Euromicro Conference on Digital System Design:
Architectures, Methods and Tools, Lille.

72
[3] A. Ravinagarajan, D. Dondi & T.S. Rosing, (2010) “DVFS based task scheduling in a harvesting
WSN for Structural Health Monitoring”, Proceedings of Design, Automation and Test in Europe
Conference and Exhibition (DATE), Dresden.
[4] S. Bilavarn, A. Castagnetti & L. Rodrigue, (2011) “A Video Monitoring Application for
Wireless Sensor Networks using IEEE 802.15.4”, Proceedings of ARCS, Como.
[5] J. Kriegel, F. Broekaert, A. Pegatoquet & M. Auguin, (2010) “Power optimization technique
applied to real-time video application”, 13th Sophia Antipolis Microelectronics Forum (SAME),
University Booth, Valbonne, France.

Interplay of Communication and Computation Energy Consumption for Low Power Sensor Network Design

More Related Content

What's hot (19)

Similar to Interplay of Communication and Computation Energy Consumption for Low Power Sensor Network Design (20)

Recently uploaded (20)

Interplay of Communication and Computation Energy Consumption for Low Power Sensor Network Design