The second problem arises whenever the consumption of the sensing subsystem is not negligible Data driven techniques presented in the following are designed to reduce the amount of sampl
Trang 1Energy Efficiency for Wireless Sensor Networks Based
on Internet of Things
Thu Ngo Quynh', Chung Nguyen Due, Anh Nguyen Quynh
Hanoi University ofScience and Technology, No 1 Dai Co Viet Sir, Ha Noi, Viet Nam Received: March 04, 2014: accepted' Aprd 22 2014
Abstract
Recently, internet of Things (loT) enables the convergence of Wireless Sensor Networt<s with the IP wortd and the connectivity of smart objects to the Internet by using RPL routing protocol for transmitting IPv6 over small sensor nodes of WSN One most important disadvantage of RPL is - many control messages of RPL such as DIS, DIO and DAO can lead to the increase of energy consumption of sensors That is why it
IS necessary to propose different solutions for energy savings in loT, In this paper, we propose three energy efficient data processing schemes that are intergrated with RPL routing protocol and evaluate them
to 9% energy consumed
Keywords, Energy Efficiency, Routing protocol, Internet of Things
1 Introduction
Recently, Internet of Things (loT) becomes a
potential future scenario of the applicability and
impact of technology In human life loT can extend
the concept of Internet fi-om a network of rather
homogeneous devices such as computers to network
of heterogeneous devices (home appliances,
consumer electronics or sensors nodes of Wireless
Sensor Networks) Since loT systems consist of
sensor nodes that have weak processing power, to
suit with new type of network, techniques used
Internet are required adjustment or even new
techniques, protocols or mechanisms are also
suggested
For enabling the implementation of loT over
WSN or making IPv6 packets to be earned over
IEEE 802.4 feasible, IETF Working Group Routing
over Wireless Sensor Networks (WSN) investigated a
routing protocol named RPL [1] RPL was proposed
protocols such as AODV, OLSR or OSPF could meet
the specification of Low power and Lossy Networks
(LLN), Networks running RPL are connected in such
a way that no cycles are present In order to have no
cycles, a Destination Oriented Directed Acyclic
Graph (DODAG), which is routed at a root, is built
For establishing this DODAG, 4 types of control
messages are defined in RPL specification: DIO,
DAO, DIS and DAO-ACK
* Conespongdmg Author; Tel (+84)912.528.824
Expenmental measurements have shown that data transmission in general costs expensive m terms
of energy con sumption while data processing consumes significantly less energy [3], When approximately the same is thit needed for processing
a thousand operations in a t^pital sensor node [4]
ot loT and the design ot RPL significantly because
descnbed above (DIO, DAO, DIS,,,), Smce most of crucial to limit the amount of sent control messages over the network That is why it is necessary to propose different solutions for energy savmg of RPL
m loT One possibility is to adapt the sending rate of DIO messages by extending the Trickle algorithm [5]
or using energy associated with battery index as object fimcnon of RPL [6,7] Another possibility is to
implement data driven techniques that are designed
to reduce the amount of sampled data by keeping the sensing accuracy within an acceptable level for the appU cation
In this paper, we utilize the data set of Intel Berkeley research lab and present three methods (intergrated with RPL) for reducing the amount of this set This reduced sampled data will be sent to the root by using RPL routing protocol implemented in Contiki operating system The simulation results show that our schemes can help to save from 4 to 9% energy consumed for RPL
This paper is organized as follows In Section
I, we present the background of data driven
Trang 2techniques for energy efficiency in WSN, Section 3
describes the proposed data processing technologies
that are intergrated with RPL routing protocol
Section 4 descnbes our simulation scenarios using
Contiki operating system and evaluate the
performance of these three methods Section 5
concludes the paper and discuss our ongoing work
2 Background
2.1 Energy conservation for WSN
In WSN, different data-driven techniques can
be used to improve the energy efficiency In fact, data
.sensing impacts on sensor' energy consumption in
two ways:
• Unneeded samples Normally, sampled data
have strong spatial and/or temporal
correlations [7] That is why there is no need
to communicate the redundant data to the root,
• Power consumption of the sensing system
Reducing communication is not enough when
the sensor Itself is power hungry
In the first case, redundant samples result in
useless energy consumption, even if the cost of
sampling is negligible because they result in
unneeded communications The second problem
arises whenever the consumption of the sensing
subsystem is not negligible Data driven techniques
presented in the following are designed to reduce the
amount of sampled data by keeping the sensing
accuracy within an acceptable level depending on the
requirement of applications
Data-driven techniques can be divided to: data
processing and energy efficient data acquisition
case of unneeded samples, while energy-efficient
data acquisition schemes often aimed at reducing the
energy consummed by the sensing subsystem
However, some of them can reduce the energy spent
for communication as well
In this paper, it is important to discuss here
one more classification level related to
data-processing algonthms All these approaches aim at
reducing the amount of data to be delivered to the
root However the design principles of these
approaches are rather different In-network
processing performs data aggregation at intermediate
sensors between the sources and the root By doing
this, the amount of data is reduced while traversing
in-network processing technique depends on the
specific application and must be tailored to it In [8]
an up-to-date survey about in-network processing
applied to reduce the amount of data sent by sources This scheme includes encodmg information at sensors which generate data, and decoding it at the root There are different methods for compressing data [9-14], As compression techniques are general (i.e, not necessanly related to WSNs), we will omit a detailed discussion of them to focus on other
approaches specifically tailored to WSNs Data
sensed phenomenon, i,e, a model descnbing the evolution of data The model can predict the values sensed by sensors, and resides both at the sensors and
at the root On the other side, explicit communication
the model is not accurate enough, i.e the actual sample has to be retrieved and/or the model has to be number of data sent by sensor nodes and the energy
spent for communication as well Data reduction are
schemes that reduce the amount of data sent by source by removing unneeded samples
In this paper, we are concentrated on developing data processing schemes intergrated with RPL routmg protocol for the data set of Intel Berkeley Research, This reduced sampled data will
be sent to the root using RPL routing protocol implemented in Contiki operating system In the next section, the characteristics of this data set are examined
2.2 Data set of Intel Berkeley lab
The data set used in this paper is collected from Intel Berkeley Research lab It consists of temperture, humidity and light intensity that are sampled each 31
in the above table From this table, we realize that temperature and humidity vary slowly while light intensity stays unchanged during a long period From
schemes: data reduction using difference, linear data
three methods are presented in the following section
Table 1 Variation of three metrics
Tetnperatiire
19.9884
19.3024
19.1652
19.1456
Humidity
37.0933
37.0933
38.4629
38.8039
Light Intensity 45.08
45 08
45.08
45.08
Trang 33 Data driven techniques
3.1 Linear data prediction
The most important characteristic of this data
set is that all three metncs varies slowly according
the time That is why it is possible to implement a
data linear predicted function that can predict Ihe
intensity) based on n data samples collected
difference between real data value and its predicted
value By implementing an appropriate linear data
predicted fimction, this difference is smaller than the
original sample and transmitting only this difference
of this scheme are described as follows:
Step 1 Sensors collect values of n samples 5 s ( 0
with i=l n, save these values and transmit towards
root Root receives this data and saves for later
process
Step 2 The next data value at (n+1)"' sample is
collected by sensors At this time, sensors predict also
the value of this data sample by using following
linear prediction function:
gin + 1) = CnflOO + c„_i5Gi - l ) -I- - + Cjg
4-(2)
where flsCO is data value at i"' sample with i=l n,
g,Oi-l- l ) f t ( « -I- l)is predicted value, q with i=l n
is predicted coefficient that satisies following
condition: c„ > £-„_! > ••• > c^ > 0 and
I?=lCi = 1
Step 3 Sensors evaluate difference between original
and predicted value according to the following
equation:
^9s = SsCn + l ) - a t ( n - l - l )
This difference is transmitted to the root using
RPL routing protocol of loT
Step 4, Root receives this difference and recalculates
the linear predicted fiinction:
SB (« + ! ) = c^Bs W + %-iSfi in-i} + - + c^g^ (1)
It reevaluates the value of data at (n+1)'''
sample as
follows-5HCn-|-l)=fl.Cn-|-l)-|-Afl«
Step 5 Sensors and root clear data at die / " sample
and save data value of f>i-i-7/* sample for later use
This strategy can reduce the amount of sampled data
memory for saving n samples and the calculation of
linear predicted function consumes also much energy That is why we present in the next section another data reduction algorithm usmg difference between two consecutive data samples
3.2 Data reduction using difference between consecutive samples:
As described in previous sections, three metncs of this data set (temperture, humidity and light intensity) vary slowly according the time That
IS why the difference between two consecutive samples stays small compared to original samples By transmitting only this difference, we can achieve energy efficiency for RPL routing protocol The detail of this data reduction method is described as follows:
Step 1 Sensors collect the data value of /"' sample
root, this value g^ (l) is received
Step 2 Sensors collect next data sample and evaluate
the difference between the P' and 2'"' values:
i & = S < 2 ) - S ( l ) Step 3 Sensors transmit this difference towards root Step 4 Root receives this difference and evaluate the value of 2"^ sample by implementing this equation 5fl(23=5fl(l)+Afl„
Step 5 Root and sensor clear the value of / " sample while keeping 2"'' sample
Obviously, the difference-based algorithm is more simple than linear predicted scheme because it
of only the difference between two consecutive samples consumes less energy than the calculation of
present the data collected at sensors when implementing these two strategies applying to 5 samples and the original
data-From this figure, we find that the difference between original data and data after precessing of the first strategy (linear prediction) is higher than the second strategy That is why the second method can
be reexamined in different simulation scenarios described in section 4
Trang 43.3 Entropy-based Data Reduction
In this section, we present a data reduction
algorithm that utilizes an entropy-based threshold in
order to decide when sensors need to transmit data
towards root In order to receive this threshold, it is
necessary to calculate the entropy value of data
according to the following function:
HiV^.V, Vi) = - 2 ] Pin-^2-Vn)l0S2pin'-^2 ^)
where V^ with i=l n is sensor node, v^ is data
collected at this node (, pCi'i-i'i-.t^} is the joint
probability distnbution fimction ofVi_,i?2•••!'„, After
calculating this entropy value, we select an
entropy-threshold based that examines the increment of the
increased amount entropy AHii) The value of dW(i)
can be calculated as follows:
d//CO rHCFi ^) -WC^i V^_t)withi> 1
^ 1 with i = 1
Please refer to [15] for more information
relating the calculation of entropy-based threshold
Step 1 Sensors collect the value of / " data sample
fljCl), save it and transmit to root by using RPL
routing protocol Root receives the value gg ( l )
Step 2 Nodes collect 2"'' data value and evaluate the
difference: Aflj = g, il) — fljCl)
Step 3 If this difference is higher than the
enfropy-based threshold Ags > Hyjj.gai,oi(j, nodes will transmit
data
Step 4 Root receives this difference Ag^ and
evaluate the value of 2"'' sample by applying the
equation 5^ (2) =g (l) + Ag^ ,giiC2) = g t l } -I- AgR
Step 5 Nodes and root clear the value of 1" sample
and while keeping the value of 2'"' sample
ig 2 Data collected at sensors of 3"* scheme
In the following figure, we present the value
of temperature collected at sensors when applying entropy-based threshold of 0.02 and 0.01 We realize that this difference is negligible compared to the origmal data value For humidity, the characteristic of data is similar to temperature and we can also apply
humidity, light intensity stay unchanged during a long period
4 Performance evaluation
In this section, we implement three above techniques intergrated with RPL routing protocol by using Contiki operating system The simulated topology consists of 15 nodes Nodes that have rank I are (2,3,6-8,12)andnodes that have rank 2 are (1,5,7, 9-11,13-16), Input data is the data set received at Intel Berkeley research once each 31s Root is situated in the middle of the topology
In this topology, we evaluate the performance
of 4 following scenarios:
- In the 1" scenario, all three meh-ics
temperature, humidity and light intensity are transmitted in a RPL IPv6 packet without data driven techniques
- In the 2"'' scenario, we apply the linear
predicted data reduction algonthm for M=5 samples (in case of temperature and hunidity only)
- In the J"^ scenario, data reduction algorithm using difference between consecutive samples are implemented
- In the 4'* scenario, we implement entropy-based data reduction algorithm with the
for humidity and 0 for light intensity
In the following figure, energy consumed of
2"'' scenario is compared to the P' one
From this figure, we realize that linear predicted data redimction algorithm achieves better
data driven techniques (4.07%) However, the
methods is not much because of the complexity of
this data reduction (large memory for saving n
samples and complex calculations for linear predicted
function) For 3"^ scenario, energy consumed is
presented in the following figure:
Fig 3, Simulated Topology
Trang 5Fig 5 Energy of 3'*' and 1 ^' scenarios
Fig 7 Energy of all 4 scenarios
We realize that the data reduction method
based on difference between two consecutive samples
achieves also better energy efficiency than the P'
scenario (5%) It can be explained simply because all
three metrics change slowly and the energy
consumed for transmitting difference between
consecutive samples stays small
Next, we examine energy efficiency of 4"'
scenario Obviously, in the entropy-based method
scenario
More concretelly, the energy efficiency of all
4 scenarios is presented together in the following
The last scenario with entropy-based data reduction
method achieves the best energy efficiency — 9%
5 Conclusion
In this paper, we present three data processing
methods, linear prediction, data reduction based on
difference and data reduction based on
entropy-threshold These three methods are also intergrated
with RPL routing protocol for transmitting data to the
root by using IPv6 protocol Simulation results by
Contiki show that entropy-based data reduction
method can save to 9% energy consummed, while
method using difference between consecutive
samples and linear predicted function achieve only
V]
G Montenegro, N Kushalnagar, J Hui, and D 802.15 4 networks," IETF, RFC 4944, 2007, A- Brandt, J Hui, R Kelsey, P Levis, K Pister, R Slmik, JP Vasseur, R.Alexander, RPL, IPv6 Routing Protocol for Low-Power and Lossy Networks IETF RFC 6550, May 2012
V Raghunathan, C Schurghers, S Park, M Snvastava, Energy-aware wireless microsensor
40-50
G Pottie, W Kaiser, Wireless integrated network seusors, Communication of ACM 43 (2000) 51-58
P, Levis, N Patel D, Culler, and S Shenker, Trickle:
A self-regulatmg algorithm for code maintenance and propagation m wireless sensor networks In Proceedmgs of the USENIX NSDI Conference, San Francisco, CA, USA (2004) 15-28,
A, Barbato, A Capone, M Barrano, N Figiani Protocol for IPv6 Wireless Sensor Networks, IEEE Online Conference on Green Communications (2013) 163-168
Cheng-Yen Liao, Lin-Huang Chang, Tsung-Han Lee, Shu-Jan Chen, An Energy-Efficiency-Oriented Routing Algorithm over RPL AIT Conference 2013, [8] M,C Vuran, OB, Akan, l.F, Akyildiz, Spatio-temporal correlation- theory and applications for wireless sensor networks, Computer Networks Joumal 45 (3) (2004) 245-261
[9] E Fasolo, M Rossi, J Widmer, M Zorzi, hi-network aggregation techniques for wireless sensor networks:
70-87-[10] S.S Pradhan, K Ramchandran, DisQributed source coding using syndromes (DISCUS) design and constraction, IEEE Transactions on Information Theory 49 (2003) 626-643
[II] C Tang, C S Raghavendra, Compression Techniques for Wireless Sensor Networks, Book Wireless Sensor
231 (Chapter 10)
[12] M Wu, CW Chen, Multiple Bh Stream Image Transmission over Wireless Sensor Networks, Book Sensor Network Operations, IEEE & Wiley Interscience (20O6) 677-687 (Chapter 13) [13] Z Xiong; A.D Livens, S Cheng, Distributed source codmg for sensor networks, IEEE Signal Processing Magazine 21 (2004)80-94,
[14] Pietro Gomzzi, Gianluigi Fenari, Paolo Medagliani, Jeremie Leguay Data Storage and Retrieval with RPL routing 9* Wireless Communications and Mobile Computing Conference IWCMC (2013) 1400-1404 [15] Hiroaki Taka, Hideyuki Uehara, Takashi Ohira, Intermittent Transmission Method based on Aggregation Model for Cluslenng Scheme, IEEE