Energy Efficient Smart Routing Based on Link Correlation Mining

for Wireless Edge Computing in IoT

Bahir Dar University
Bahir Dar Institute of Technology (BiT)
Master of Science in Communication Systems Engineering
Dessie Fikir (BDU1500410)
Email:[email protected]
February 6, 2024

Abstract Key words: Smart Routing, Network Coding, Link Correla-

tion Mining, IoT, Edge Computing
In today’s IoT applications, edge computing is introduced to
reduce data latency and communication bandwidth between
cloud servers and IoT edge devices. However, Inefficient 1 Introduction
routing can cause transmission failures and unnecessary data
retransmissions, reducing reliability and increasing energy IoT (Internet of Things) technologies have spread through ev-
consumption. Therefore, designing efficient routing proto- ery corner of our modern society, from wearable devices to
cols is essential for high-quality data mining with reliability mobile applications. However, in this technology, there are
and energy efficiency. This paper critically reviews the use challenges such as data security, communication reliability,
of network coding and opportunistic routing based on link cor- energy efficiency, and smart design [1][2][3]. These chal-
relation to improve energy efficiency and reliability in wireless lenges affect the dependability of Internet of Things applica-
IoT infrastructure. We have made a comparison analysis with tions, data mining quality, and communication infrastructure
the related works and analyzed the proposed algorithm based stability. One of the key challenges to achieving high-quality
on the baseline of the previous work. Finally, we have crit- data mining is is to ensure that every individual end and edge
ically examined strong side and weak side of the papaer. device must be able to communicate with each other reliably
We also discussed potential future developments for this pa- and in the most energy efficient manner. Thus, the paper
per. Some studies disregarded link correlation and some of has aimed to develop a novel routing mechanism for modern
them are without Network Coding, leading to estimation er- wireless IoT infrastructure to support data stream mining. As
rors in the expected number of transmissions by forwarders. a solution to improve energy efficiency and reliability across
This impacts the selection of the forwarder set and ultimately wireless sensor networks [4][5], efficient design in routing
affects the protocol’s performance. In this context, this paper protocols has become an attractive topic.
proposed an algorithm for choosing a more optimal forwarder In traditional routing mechanism, the forwarding path is fixed,
set and a smart routing method to estimate the number of in this regard due to the broadcast nature of the wireless link,
transmissions required by forwarders accurately. The network the devices or nodes which do not need to participate in packet
coding mechanism is based on the mining of link correlation. transmission may listen to the data packet (which is an unnec-
Simulation results show that the suggested method can achieve essary forwarding of a message). However, these nodes must
fewer transmissions and provide more energy-efficient com- discard the received packets even if they may be ”closer” to the
munications for wireless edge Internet of Things applications. destination node. This leads to inefficient use of network re-

1
sources, and additionally, traditional routing chooses the next of transmissions required for a packet to be successfully re-
hop before transmitting a packet; but, when the link quality ceived at its destination) to select candidates. However, using
is poor, the probability that the chosen next hop receives the single-path ETX as a metric for candidate selection is an ap-
packet is low. proximation because it cannot capture the opportunistic paths.
Although the ExOR design uses a mechanism that selects a
forwarding node based on which is closest to the destination,
the paradigm is still more or less similar to fixed rout-
ing because nodes that receive each packet must choose one
forwarder; there is no option to select more than one for-
warder to increase flexibility or boost throughput. Even the
Figure 1: Routing, packet forwarding simulation result shows that throughput is measured in Kilo
Bytes, indicating an improvement over traditional routing by
Traditional routing predetermines the path before transmis- one step. And generally, ExOR requires node coordination,
sion. It sends traffic along the path “source→R→destination”, which is more difficult in larger networks. Moreover, this
which has the highest delivery probability. However, wireless study didn’t consider the network coding and link correlation.
is a broadcast medium. When a node transmits, there is al-
In [7] Biswas and Morris . proposed Opportunistic routing in
ways a chance that a node closer than the chosen next hop to
wireless multi-hop networks which is a modification of [6].
the destination overhears the packet. For example, assume the
In this study, to select the forwarder that is closest to the des-
source sends 2 packets, p1 and p2. The next hop, R, receives
tination node from multiple candidate nodes, opportunistic
both, and the destination happens to overhear p1. It would be
routing implemented a Strict Scheduling mechanism at the
a waste to have node R forward p1 again to the destination.
MAC layer. This approach resulted in increased transmission
In contrast, opportunistic routing allows any node that over- reliability and throughput overall compared to conventional
hears the transmission and is closer to the destination to par- fixed-path wireless routing. The ExOR architecture uses a
ticipate in forwarding the packet. To conserve energy and distributed algorithm to solve the problem of selecting a for-
ensure a longer network life, an intra-session network coding warding node after transmission. Initially, a basic schedule
combined with a routing mechanism is implemented, which outlining the priority order in which the packet should be for-
can reduce transmission times to a reasonable degree without warded by possible recipients is included in each packet that a
degrading the network performance. node transmits. The schedule is calculated by the node using
measurements of inter-node delivery rates that are exchanged.

To confirm that the receivers agree on who the highest pri-

2 Literature Review
ority receiver was, ExOR then employs a distributed slotted
In this regard, there are works of literature that tried to design MAC technique for acknowledgments. The paper chooses a
a routing protocol to improve energy efficiency. candidate set based on the minimum number of hops (priori-
tized by delivery rates) to determine the candidate forwarder
Biswas and Morris proposed the extremely opportunistic rout-
set. First, identify the shortest path to the destination. The
ing ExOR: opportunistic multi-hop routing for wireless
first node in this path is the highest-priority candidate. Then
networks [6]. In this study, the intermediate nodes that listen
ExOR deletes that node from the loss rate matrix again finds
to the packets and ”closer” to the destination can participate
the shortest route, and uses the first hop on that route as the
in the forwarding of the packet. The packet path is chosen
candidate with the second priority. It repeats this process to
using the ExOR algorithm, which can also indicate which
find the remaining candidates. Including the ID of the sender
intermediate nodes received the transmission. Each packet’s
of the highest-priority ACK heard so far helps suppress dupli-
recipient nodes need to decide on a single forwarder and agree
cate forwarding.
on their identities. Until the packet is received at the desti-
nation, this procedure keeps on. The algorithm uses single However, this ExOR implementation will fail in the cases
path ETX (Expected Transmission Count, expected number of dense networks, because ExOR only has a relatively small

2
number of candidate selections. In such cases, a heuristic destination node can decode all packets.
that chooses members from dynamic and dense networks can
One challenge of this mechanism is how to avoid duplicate
be used. The strict scheduling made the design and imple-
transmissions and guarantee reliable transmissions. MORE
mentation of the protocol complex and difficult to extend to
proposes a credit mechanism in which each node calculates
multicast (hard to extend to alternate traffic types) networks.
how effective it would be in forwarding coded packets to
Due to this strict schedule, nodes farther away from the desti-
downstream nodes. MORE also supports multicast trans-
nation, cannot transmit packets at the same time as nodes close
mission, which has three important modifications compared
to the destination, since they have to wait for the nodes close
to unicast transmission. First, the candidate nodes are the
to the destination to finish transmitting. Moreover, this paper
union of candidates of different unicast traffic. Second, the
didn’t consider the link correlation and network coding.
transmission credit value is calculated from the maximum
In [8] Chachulski et al. Investigated a combined random of transmission rates of different unicast traffic. Third, the
linear network coding with opportunistic routing, and pro- source transmits the next batch until the packets in the pre-
posed a routing mechanism which did not require the strict vious batch reach all destinations. The main drawback of
scheduling between nodes. This paper presents MORE, a MORE is that the source only forwards a group of coded pack-
MAC-independent opportunistic routing protocol, that aimed ets from one batch until the destination can decode all packets
to reduce unnecessary packet transmission and improve the of that batch. The transmission of the next batch will be sus-
bandwidth utilization performance. MORE randomly mixes pended while awaiting the acknowledgment. Moreover, this
packets before forwarding them. This randomness ensures approach was designed based on the assumption that wireless
that routers that hear the same transmission do not forward links are independent. Without considering link correlation,
identical packets. Thus, MORE needs no special scheduler the opportunistic routing schemes do not fully capture the
to coordinate routers. In this paper Network coding offers an diversity benefit of the wireless broadcast medium.
elegant solution to overcome the above problem mentioned in
H. Dubois et al. propose least-cost opportunistic routing
paper [7].
(LCOR) [9], which takes EAX as the metric to select the
In the previous example Figure 1, the destination has overheard candidate sets. Similar to the well-known Bellman-Ford al-
one of the transmitted packets, p1, but node R is unaware of gorithm, LCOR exhaustively searches all possible candidates’
this fortunate reception. With network coding, node R nat- sets to find the paths with minimum transmissions. However,
urally forwards linear combinations of the received packets. its computational cost increases dramatically in dense net-
For example, R can send the sum 𝑝1 + 𝑝2. The destination works because of the exponential explosion in the exhaustive
retrieves the packet p2 it misses by subtracting from the sum search. Therefore this study is not more applicable for dense
and acknowledges the whole transfer. Thus, R need not know networks. Moreover, it assumed that the links are independent,
which packet the destination has overheared. Generally, the which causes an error in the determinations of the number of
source broadcasts its packets. Routers create random linear forwards and forward set results in degraded performance.
combinations of the packets they hear. The paper validated
Some research has been conducted to exploit link correlation
that network coding has reduced the need for retransmissions,
to improve the performance of network protocols.
improving the overall throughput.
In this regard, Wang et al.[10]. It introduced the link cor-
It selects potential candidates at the source before data trans-
relation into the opportunistic routing protocol. To select
missions; these candidates are “closer” to the destination than
the candidate forwarder set with good link quality, prioritize
the source. Then data packets will be grouped into batches
them according to the ETX value. First, initialize the candi-
and coded packets will be generated according to the data
date set by adding nodes with smaller ETX values and create
packets in that batch. Coded packets will be forwarded with
a directed acyclic graph from the source to the destination
the order of batch index, in which the transmission of the next
eliminate candidates that have higher ETX values, and again
batch of coded packets begins after all coded packets from the
loop this process. Then add the node to the candidate set and
previous batch reach the destination. The source node keeps
remove it from the initial set. Loop the above procedures until
transmitting the current batch via candidate nodes until the

3
we find enough candidates to meet the predesigned set size. tions among candidates, ensuring successful packet delivery
Testbed evaluation and extensive simulation show that higher even if one candidate fails [11].
link correlation leads to fewer diversity benefits and that, with
the link-correlation-aware design, the number of transmis-
sions is reduced by 38 percent. However, this study does not 3.2 Link Correlation
include smart routing based on Network coding which helps
to increase the node coordination in achieving accuracy of Link correlation indicates the correlation of packet loss be-
packet delivery to the destination and significantly reduces tween various links. A positive correlation can occur due
the required number of forwarding. Moreover, this mech- to cross-network interference, such as high-power wireless
anism only considered the link correlation in the routing networks like Wi-Fi causing destructive noise in low-power
mechanism or only calculated how many coded packets for- networks like ZigBee, leading to synchronized packet loss
warders need to transmit without suggesting how to select across multiple links. Negative correlation, due to correlation
the forwarder set. fading, where the receiving terminals receive packets from
the signal source separately depending on the location of the
Having identified the limitations in existing literature, this
moving obstacle. As a metric the packet receive/lost proba-
paper concentrates on Network Coding Combined with Link
bility of multiple links can be used [11].
Correlation (NCCLC) to enhance energy efficiency in wireless
edge IoT applications. The paper introduces an intra-session Link Correlation has an impact on both packet forwarding
network coding mechanism that accounts for link correlation quantity and packet forwarder selection The paper validated
in determining the necessary forwarding steps and selecting the impact of Link Correlation.
the forwarder set. Particularly this paper has contributed as
follows:
3.3 Network coding
• Analyzed the impact of link correlation on smart routing
in the number of required forwarding and the selection Network coding combines multiple missed packets into a sin-
of forwarder set. gle transmission, enhancing transmission efficiency. This
• Energy efficient forwarding strategy is designed based on technique enables intermediate nodes to merge packets before
link correlation analysis to obtain a minimum forwarding forwarding, offering significant potential to boost efficiency
number and to guarantee the network packets are routed for both broadcast and unicast communication. Illustrated in
to the destination correctly. Figure 2 .

• A smart routing mechanism is developed to accurately

calculate the number of transmissions for a forwarder
and forwarder set selection.

3 State of the art

3.1 Opportunistic Routing (OR)
In multi-hop wireless networks, opportunistic routing (OR) Figure 2: Network Coding [11]
selects a group of potential forwarders for packet delivery
rather than designating a single next-hop forwarder. To deter-
mine which candidates have received the packet, OR waits to In traditional designs, to make sure that both receivers get the
choose the subsequent hop. Candidate selection is critical for two broadcast packets, the source node u needs to send packets
performance since the variety of received packets from candi- p1 and p2 using two transmissions. With the help of network
dates is essential to OR’s effectiveness. The effectiveness of coding, node u broadcasts an XOR packet 𝑝1 ⊕ 2 using one
opportunistic routing relies on the diversity of packet recep- transmission, thus saving one transmission.

4
4 Smart Routing Based On Link Cor- From the above equation the average number of required for-
warding for node j to receive a linear independent packet from
relation Mining
a node with a higher ETX value can be determined as:
Methods to calculate the expected number of forwarding, and
select forwarder set based on link correlation are discussed.
Strictly considered the link correlation metrics in terms of the
probability of packet loss rate and packet received. From this,
If the estimated value of 𝑇 𝑋 𝑐𝑟𝑒𝑑𝑖𝑡 is too large, there may be
the ETX (Expected Transmission Count), can be calculated
many redundant packet transmissions in the network. Such
starting from the destination through the intermediate node
redundancy may flood the network, cause more collisions, and
to the source The ETX, calculates the average number of
eventually result in unnecessary energy consumption. If the
transmissions that a packet requires to be received by the
value of 𝑇 𝑋 𝑐𝑟𝑒𝑑𝑖𝑡 is too small, the destination node cannot
destination. The ETX value of a corresponding link is the
decode the coded packet since there are not enough linearly
inverse of the packet delivery probability of that link. The
independent packets.
ETX value on a path can be derived by adding up all ETX
values of each link on this path.
4.2 Packet Forwarder Set Selection
The idea for packet forwarder set selection is to first pick a
4.1 Packet Forwarding Estimation
single forwarder with the fewest number of transmissions, and
In this how the average number of required packet forward- then cyclically add forwarders with fewer forwarding to the
ing is calculated based on link correlation is presented. To forwarder set. Finally, the forwarder set with the least number
calculate/determine this, first the number of packets and the of transmissions can be obtained, or the size of the forwarder
total number of transmission that a given/certain node needs set can reach a predetermined threshold.
to forward and needs to make is calculated respectively.

Let’s given N nodes in a network, the source node is s, and 5 Implementations of the proposed al-
the destination node is d. For two nodes 𝑖 and 𝑗 assume 𝑗 < 𝑖
mean node 𝑗 is closer to the destination node than node𝑖 or
gorithms
ETX (Expected Transmission Count) value of the node 𝑗 is A novel smart routing method is proposed to accurately es-
smaller than that of node 𝑖. 𝑃𝑖, 𝑗 indicates the packet loss rate
timate the number of transmissions required by forwarders,
between node i and node. 𝑍𝑖 Represents the expected number together with an algorithm for selecting a forwarder set with
of forwarding required for the intermediate node 𝑖 between more optimal number of transmissions.
node s and node d to successfully send a packet from s to
d. The number of packets that node 𝑗 needs to forward is Packet Forwarding Estimation Mechanism: The source
calculated as follows. node sends a flood request before sending data packets. Each
forwarder that receives the request sends the packet loss rate
of its link back to the source node, and the ETX value from
each node to the destination node is calculated. The selected
forwarders are recorded in the packet header and arranged
Where, 𝐸 𝑖, 𝑗 = the event that node j successfully received the
in descending order of ETX values. The reception reports
packet sent by node 𝑖, 𝜙 𝑗 = the event that node 𝑗 successfully
are carried in the data packet and then sent. Each forwarder
received the packet sent by node 𝑖, and 𝐸¯ 𝑖, 𝑗 is the event that
checks the forwarder list, finds its upstream nodes, and re-
transmission from 𝑖 is lost at node 𝑘. Therefore, the total
quests link correlation metric from them when receiving a
number of transmission node 𝑗 needs to make is calculated as
new batch of encoded packets. After receiving them, it cal-
follows.
culates its own 𝑇 𝑋 𝑐𝑟𝑒𝑑𝑖𝑡 value and 𝑍 value by all the above
equations. Nodes also maintains and updates values of 𝑍
of all upstream nodes. When an acknowledgement packet is

5
received by nodes, the current 𝑍 value record, the link corre- the packet loss rate, when link transmissions are affected by
lation metric, and the packet loss rate of each link are cleared. an interference source, the calculation based on link loss rate
Then the upstream set of the new batch is found. Finally, the probability can be expressed as follows.
𝑇 𝑋 𝑐𝑟𝑒𝑑𝑖𝑡 value is recalculated.

Forwarder Set Selection Algorithm: The core idea of the

algorithm is to select a forwarder set with the smallest EAX
value. Since the calculation of the EAX value takes into ac-
count the quality and correlation of the link, the forwarder
set selected using this algorithm can achieve fewer required
transmissions. The forwarder set selection algorithm is shown
in the Algorithm Table 1.

Table 1: Forward set selection algorithm

6.1 Performance evaluation

The paper uses network throughput (as a metric). As shown in

Figure 3, the throughput distribution of the 100 streams with
one interference source. Observed that NCCLC has a large
probability distribution towards up to 3Mbps, whereas the
MORE and LCAOR are mostly at 1-2Mbps, demonstrates the
effectiveness of the NCCLC in improving network throughput.

6 Result and Analysis

In the simulation result the paper considers multiple sources
of interference at different locations in reality, different sce-
narios of interfering sources are designed in the experiment
to evaluate performances of the proposed NCCLC in terms
of the throughput and distribution of streams under different
degrees of link correlations. To demonstrate the effectiveness
of the proposed method, the paper uses two previous baseline
methods as the benchmark protocols that we have seen in the
Figure 3: Throughput distribution of 100 streams
literature, which are:

MORE (MAC-independent Opportunistic Routing and En-

coding) [14] and LCAOR (Link-Correlation-Aware Oppor-
tunistic Routing) [24].
In Figure 4 NCCLC has a much better probability distribution
To simulate the link correlation, an interference model needs at higher throughput (i.e. up to 3 Mbps), with 41 streams that
to be added to the simulator (which accounts a positive cor- have more than 4 hops between the source and destination
relation). Considering the impact of interference sources on hop.

6
 LCAOR and NCCLC become better due to the utilization of
link correlation.

Figure 4: Throughput distribution of 41streams

Figure 6: Throughput distribution of 45 streams over 4-hops with 3
interference sources
With new network topologies and increase the number of in-
terference sources to three as shown in Figure 5. This is
to evaluate whether the NCCLC can perform consistently and 7 Critical Review (Criticism) of the pa-
improve the network throughput under the scenario with multi-
per
interference sources. Demonstrate that the NCCLC outper-
forms the other two methods by achieving larger portion of 7.1 Strong sides of the paper
higher throughputs.
• Effectively improve the network throughput for multi-
hop wireless IoT mesh networks with one or more in-
terference sources, where most of the previous work de-
clined their throughput in dense networks due to lack of
Network coding combined with Link correlation (NC-
CLC) Mechanism.

• The newly designed network coding (intra-session) can

effectively reduce the unnecessary transmissions of the
forwarders, which introduces effective node coordina-
tion, where the downstream nodes can notify the up-
stream nodes to stop sending packets when they receive
enough coded packets, thus these upstream nodes can
send more packets of other streams. Additionally, multi-
Figure 5: Throughput distribution of 100 streams with 3 interference ple missed packets are encoded together and then broad-
sources
cast in a single transmission, thus improving the trans-
mission efficiency.
Further to evaluate the 45 streams that have more than 4 hops
• The proposed algorithm keep its consistency (stability)
with three interference sources is plotted in Figure 6, in which
to improve the network throughput with variable number
the source and destination hop will not interfere with each
of interference as we have seen in the simulation result.
other. NCCLC consistently outperforms the other two meth-
ods. Moreover, results in Table VI and Table III indicate • A metric to calculate the link correlation was newly de-
that the performance of MORE degrades with the presence signed, which significantly improve the reliability of the
of multiple sources of interference, while the performance of network.

7
 • A novel link correlation-based smart routing method to- which several wireless devices with disparate technologies
gether with an algorithm for selecting a forwarder set can coordinate their communication.
with more optimal number of transmissions, can achieve
fewer transmissions and offer more energy efficient com-
munications for wireless edge IoT applications.
8 Conclusion
This paper examined the importance and effects of link cor-
7.2 Weak sides of the paper relation in intelligent routing. The investigation research led
• The paper didn’t consider the scalability of the for- to the proposal of a unique link correlation-based routing
warding set. When the size (number) of the forward technique that improves energy-efficient communication in
set increases, the nodes in the network are closely dis- wireless edge Internet of Things environments. The tech-
tributed, the time complexity of the algorithm will rise nique includes mechanisms to precisely estimate the amount
rapidly. This explicitly indicates the algorithm designed of packet forwarding required and choose an ideal forwarder
in this paper will no more applicable for large dense set. In detail, a metric to calculate the link correlation was
networks. newly designed, which could be applied in estimating the ex-
pected number of forwarding packets for a node, to ensure a
• The effect of correlation fading (negative correlation) packet would arrive at its destination successfully. The for-
has not been conducted, which is real in multi-hop warder set with the lowest transmission quantity was chosen,
wireless IoT mesh networks. It is introduced by highly and by effectively reducing the number of forwarders, it may
dynamic environments in which networks are located. In be possible to lower the chance of collisions and increase net-
reality, wireless signals suffer shadow fading caused by work throughput. The simulation results demonstrated that
the presence of obstacles in the propagation path of the the proposed mechanism could effectively improve network
radio waves, leading to correlated reception performance throughput compared with some existing related works.
among receivers that are closely located. Correlated
Finally, we have identified the strong side and weak side of
shadowing, a channel propagation phenomenon in which
the proposed algorithm by cross-evaluation from the related
nearby links are affected by the same shadowed, may also
works and also based on the state of the art in edge comput-
introduce correlated packet losses to wireless networks.
ing of IoT applications. and also discussed potential future
• The algorithm needs some new modifications to be appli- developments.
cable in dense networks with acceptable computational
costs.
References
• Metrics in the simulation result is only throughput, it is
better to add a variety of metrics, for example, energy [1] Giancarlo Fortino, Wilma Russo, Claudio Savaglio,
efficiency, complexity, etc. Weiming Shen, and Mengchu Zhou. Agent-oriented
cooperative smart objects: From iot system design to
• The paper needs more evaluations in different situations,
implementation. IEEE Transactions on Systems, Man,
for instance, to improve the algorithms accuracy for
and Cybernetics: Systems, 48(11):1939–1956, 2018.
multi hope networks and large and dynamic interference
sources. [2] Zhipeng Cai and Xu Zheng. A private and efficient
mechanism for data uploading in smart cyber-physical
7.3 Future work systems. IEEE Transactions on Network Science and
Engineering, 7(2):766–775, 2020.
One unique approach to solving the cross-technology interfer-
ence issue, in my opinion, is to develop cross-technology com- [3] Zhikun Deng, Po Yang, Youbing Zhao, Xia Zhao, and
munication. In order to address the issue of communication Feng Dong. Life-logging data aggregation solution for
failure under cross-technology interference, cross-technology interdisciplinary healthcare research and collaboration.
communication offers a channel of communication through pages 2315–2320, 2015.

8
 [4] Bin Cao, Jianwei Zhao, Po Yang, Peng Yang, Xin Liu,
and Yuan Zhang. 3-d deployment optimization for
heterogeneous wireless directional sensor networks on
smart city. IEEE Transactions on Industrial Informatics,
15(3):1798–1808, 2019.

[5] Zhipeng Cai and Tuo Shi. Distributed query processing

in the edge-assisted iot data monitoring system. IEEE
Internet of Things Journal, 8(16):12679–12693, 2021.

[6] Sanjit Biswas and Robert Morris. Opportunistic routing

in multi-hop wireless networks. 34(1), 2004.

[7] Sanjit Biswas and Robert Morris. Exor: opportunistic

multi-hop routing for wireless networks. 2005.

[8] Szymon Chachulski, Michael Jennings, Sachin Katti,

and Dina Katabi. Trading structure for randomness in
wireless opportunistic routing. 37(4), 2007.

[9] Henri Dubois-Ferriere, Matthias Grossglauser, and Mar-

tin Vetterli. Least-cost opportunistic routing. 12 2010.

[10] Anas Basalamah, Song Min Kim, Shuo Guo, Tian He,
and Yoshito Tobe. Link correlation aware opportunistic
routing. pages 3036–3040, 2012.

[11] Ting Zhu, Ziguo Zhong, Tian He, and Zhi-Li Zhang. Ex-
ploring link correlation for efficient flooding in wireless
sensor networks. Proc. NSDI, pages 49–64, 04 2010.