Improving QoS-based Routing by Limiting Interference in Lossy Wireless Sensor Networks

International Journal of Wireless & Mobile Networks (IJWMN) Vol.2, No.4, November 2010
DOI : 10.5121/ijwmn.2010.2404 44
Improving QoS-based Routing by Limiting
Interference in Lossy Wireless Sensor Networks
Shuang Li1
, Alvin Lim2
and Cong Liu3
1
Department of Computer Science and Engineering, The Ohio State University,
Columbus, Ohio, USA
li.908@osu.edu
2
Department of Computer Science and Software Engineering, Auburn, Alabama, USA
lim@eng.auburn.edu
3
Department of Computer Science and Engineering, The University of North Carolina
at Chapel Hill, Chapel Hill, NC, USA
cong@email.unc.edu
ABSTRACT
Many new routing protocols have been proposed for wireless sensor networks to maximize throughput,
minimize delay or improve other QoS metrics in order to solve the problems of resource-constrained
sensor nodes in large networks. However, many of them are based on flooding or its variants. Many
routing messages are propagated unnecessarily and may cause different interference characteristics
during route discovery phase and in the actual application data transmission phase. As a result, incorrect
routes may be selected. Epidemic algorithms have been used to limit flooding in the field of wireless
sensor networks. Directed diffusion has been commonly used in wireless sensor networks because it is
designed to improve energy efficiency and scalability. However, the intrinsic flooding scheme for interest
subscriptions prevents it from achieving the maximal potential of these two goals. We propose a routing
protocol that uses ID-free epidemic flooding to limit interference in conjunction with metrics for
increasing throughput and reducing delay. Simulation results in ns2 show that there is an optimal number
of neighbors to achieve the best throughput and delay performance. For a fixed topology of a certain size,
there exists an optimal percentage of neighbors that forward the flooding message to achieve the best
throughput and delay performance.
KEYWORDS
Delay, epidemic flooding, interference, throughput, wireless sensor networks
1. INTRODUCTION
Recent advances in low-power electronics design have enabled the rapid development of tiny,
wireless unattended sensors. The field of wireless sensor networks has been under intensive
research in the past ten years. Routing is one of the most important issues in protocol design.
Most routing protocols proposed for wireless sensor networks, such as SPIN[1], directed
diffusion[2], LEACH[3], GPSR[4], GAF[5] and other variants, target at maximizing
throughput, minimizing delay or improving other QoS (Quality of Service) metrics since
resource-constrained sensor nodes are usually deployed in large numbers.
Routing protocols that require location information, such as LAR [6], GPSR [4], and DREAM
[7], do not need to flood routing requests. Others, such as DSR [8], AODV [9], ZRP [10], and
TORA [11], suffer from the effects of flooding, even with some optimizations, since nodes do
not know their locations. Flooding causes many routing messages to be propagated
unnecessarily.

45
To reduce the number of routing messages sent and to guarantee reliable data dissemination,
epidemic algorithms, which was first used in replicated databases [12], has begun to be used in
the context of wireless sensor networks, such as GOSSIP [13] and Fireworks[14]. These
protocols require the sender to have knowledge of the potential receivers, which is achieved by
exchanging beacons. Anonymous Gossip (AG) [15] overcomes this problem by attempting to
send a gossip message and waiting until the other node sends back a gossip reply, incurring
higher overhead.
Directed diffusion, though regarded as an epidemic algorithm in [16] since it avoids broadcast
storm, does not perform well with interest flooding. No matter what metrics are used in
selecting a route (basic directed diffusion uses delay), the route which performs best during
route discovery phase may not perform well during the actual data transmission phase due to
differences in interference levels caused by the different traffic patterns. Interest flooding
increases traffic in the network and causes maximum level of interference. Exploratory data are
flooded to determine the best path, which follows gradients established in the interest
propagation phase. Actual application traffic only flows through the reinforced path, which is
not affected by inter-path traffic at all, assuming there is no other data transmission at that time.
Every node has an interference range. Interference set [18] and conflict graph [19] are used to
schedule network traffic or theoretically analyze the impact of interference on wireless
networks. However, no routing protocol could have the prior knowledge about which path the
actual data traffic will go through and what the traffic pattern will be like before the route is
determined.
Our goal is to design solutions which make more accurate routing decisions by reducing the
interference level during the route discovery phase and making it more similar to that during the
actual data transmission phase. We limit the interest flooding in a modified directed diffusion,
which uses a hybrid metric described in [17]. We measure the throughput and delay
performance in different topologies and identify the optimal percentage of neighbors that
propagate interests.
2. RELATED WORK
Most of the GPS-free routing protocols for ad hoc or sensor networks are based on flooding,
such as DSR [8], AODV [9], and TORA [11]. In DSR, a source uses the complete hop-by-hop
route to the destination determined by flooding the network with route request (RREQ) packets
during route discovery. If the source is notified with a route error (RERR) packet, the source
removes any route using this link from its cache and a new round of route discovery will be
initiated. AODV and TORA, which share DSR’s on-demand characteristics, also discover
routes on an as needed basis through a similar route discovery phase.
Message redundancy caused by excessive flooding has been realized in recent years. The
taxonomy of the major proposed solutions is described in [14]: Probabilistic-based schemes,
area-based methods and neighbor knowledge methods. GOSSIP [13], Anonymous gossip [15]
and Fireworks [14] fall under the first category. In GOSSIP, when a node first receives a route
request, it broadcasts the request to its neighbors with probability p and it discards the request
with probability 1-p. In Fireworks, a node re-broadcasts the message to all its neighbors with
probability p and it sends it to only c randomly selected neighbors with probability (1-p). The
Fireworks protocols results in higher reliability given the same number of links over which the
broadcast packet is transmitted. Anonymous gossip (AG) does not require any member to know
the other members of the multicast group because the sender sends a gossip message and it will
gossip until the other node sends back a gossip reply. The other two categories require location
awareness or two hop neighborhood knowledge, both of which incur more overhead than the
first category.

46
The broadcast nature of wireless medium leads to interference, which is one of the most
significant features of wireless networks. Network performance, such as throughput and delay,
is affected by node interferences. Jain et. al. [19] theoretically analyzed the impact of
interference on multi-hop wireless network performance and determined the maximum
throughput that can be supported by the resulting network given a specific placement of
wireless nodes in physical space and a specific traffic workload. Chaporkar et. al. [18] proposed
maximal scheduling to maximize the network throughput under arbitrary topologies and
interference models. Both assume prior knowledge of a specific topology.
3. PROBLEM STATEMENT
In this section, we first introduce the error and interference model we use to capture the lossy
and interference natures of WSNs. We then discuss how to design an efficient routing algorithm
for the problem.
3.1. Error and Interference Model
Wireless links are affected by propagation loss, shadow fading, and multipath Rayleigh fading.
SNR (Signal-to-Noise ratio) is a good indicator of link quality and can be determined from the
hardware. Different SNRs cause different BERs (Bit error rate) [20]. Lee et. al. [21] derived the
mathematical formula for calculating BER.
)(5.0
fN
WP
erfcBER r
×
×
×=
Pr is the received power, W the channel bandwidth, N the noise power, f the transmission bit
rate, and erfc the complementary error function. Most wireless cards typically measure:
N
P
SNR r
log10=
The above formula is the case when only one frame is received by the receiver. If other frames
arrive at the receiver when it is receiving one frame, the SNR (or SINR to be more accurate,
where I means interference) is,
∑=
+
= n
i
ri
r
PN
P
SNR
1
log10
∑=
n
i
riP
1
is the interference component and riP is the signal strength of the other frames at the
receiver. n is the number of other frames that arrive at the receiver simultaneously. Given the
MAC frame size, FER (Frame Error Rate) could be calculated from BER.
3.2. Overview of Directed Diffusion and Variants
Directed diffusion uses a publish/subscribe communication model in which a sink node floods
interests as requests for a named data. As the interest is propagated through the network, each
intermediate node sets up a gradient with its neighbors and enables data that match the interest
to be pulled towards the sink. Sensor nodes with data that match the interest will forward
exploratory data propagated by intermediate nodes through established gradients to the sink.
The sink initiates a reinforcement message to the node that first forwarded the new data to it.

47
Other nodes use the same rule to reinforce the upstream neighbor. The source node continues to
send data through the reinforced path after it received the reinforcement.
Based on the above rule, basic diffusion generally selects route with the lowest delay. In the
past few years, researchers have proposed a variety of single or hybrid metrics with the purpose
of improving the performance, including throughput, delay, jitter, and deadline-hit ratio, of
wireless networks. Single metrics include RTT [22], PktPair [22], ETX [23], WCETT [24], and
EDR [25]. Most of them can be implemented in directed diffusion. In order to consider both
throughput and delay and take into account intra-path interference, a hybrid metric was
proposed in [17].
3.3. Route Selection Problems
Directed diffusion and its variants select paths which perform best during the exploratory data
phase. However, during this phase, the network has very high traffic since the exploratory data
follow all the gradients set up during the interest propagation phase, which effectively results in
flooding. Different topologies and traffic types lead to different interference levels for each node
in the network. As a result, the performance of a candidate path is determined by how each link
performs under high interference level due to exploratory data flooding. As shown in Figure 1,
the interference range usually differs from the transmission range, typically by a factor of 2.
Node 3 can send to Nodes 2, 4, 6, and 7 but not Nodes 1, 5, 8, and 9. However, Node 3’s
transmission will interfere with that of Nodes 1, 5, 8, and 9.
Figure 1. Node 3’s transmission and interference ranges. The solid circle is the transmission
range and the dashed circle is the interference range. R = 2r. We assume homogeneous nodes.
Figure 2. A simple topology of triangular tessellation with 19 nodes. Each node has the same
distance to its nearest neighbors.
source
1
2
sink
1 2 3 4 5
6
7
8
9 R
r

48
After reinforcement, one path is selected for application data transmission and it rarely suffers
from interference from other paths used in route discovery since no data will be sent over them
once a route is selected and reinforced. Different interference level in the two phases causes
incorrect routes to be selected since a path that performs best under high interference
environment may perform worst when transmitting data without other flooded paths. Border
links are good examples (Figure 2). The inner circle is the transmission range and the outer
circle is the interference range. Consider Nodes 1 and 2. Almost every node in the topology is
in Node 1’s interference range; thus all the links starting from or ending at Node 1 should have
very bad performance when exploratory data are propagated. In contrast, Node 2 is on the
border and only the upper half of the nodes is in its interference range. This is also true of other
border nodes. However, when there is only one data path in the network, a central path (through
Node 1) will perform better than a detour path composed of border links.
Interference causes route selection problems. Unpredictable traffic pattern and unknown
geographic information aggravate this problem.
4. EPIDEMIC ALGORITHM IN DIFFUSION
In this section, we present a set of localized epidemic algorithms (without using neighbor
information) for a modified directed diffusion. The main purpose is to reduce interference
during the exploratory data phase, make the interference level more similar to that during the
actual data transmission phase and improve route selection. The key challenge for designing
such a protocol is that probability-based schemes usually makes use of some basic
understanding of the network topology to assign to a node a probability p to broadcast, which
means the sender knows the IDs of potential receivers. However, directed diffusion is a
localized ID-free routing protocol, where no node knows its neighbors before interest flooding.
(We assume this original property of directed diffusion although in practice it may be possible
for the sender to determine the MAC addresses of the receivers.) So we have two choices here:
receivers probabilistically drop interest or senders probabilistically send exploratory data since
neighbor information is known by the sender. In both ways, we are taking the risk of losing
candidates. There is a trade-off between the degree of interference level match and the
completeness of the candidate pool. (Our results below show the optimal probability that will
maximize throughput or minimize delay for each network configuration.)
4.1. Epidemic Interest Flooding
Interest flooding is the first phase of directed diffusion. To adapt it to directed diffusion, when a
node receives an interest packet, it updates the gradient table with a probability of p, i.e. with a
probably of 1-p, the node will not update its gradient table (Table 1). (We assume the sender
has no knowledge about its neighbors.) Strictly speaking, interests are still flooded through the
network. However, exploratory data are propagated selectively, which satisfies our goal, since
only parts of the gradients, with the percentage of p, are established in the earlier phase.
TABLE 1. Pseudocode segment for epidemic interest flooding
Receiver side:
If it is a new interest packet
Set r = random ()
If r > 1-p
Update gradient table
Else
Drop interest

49
In Figure 3, interests from Nodes 2 and 4 are probabilistically dropped by Node 1; thus only the
interest from Node 3 is updated in Node 1’s gradient table. In the next phase, exploratory data
flows from the other side to Node 1 and only Node 3 will be chosen as the downstream node.
However, if Node 1 does not drop interests from Nodes 2 and 4, the exploratory data flowing
from Node 1 to Node 3 will suffer from the interference coming from Node 2 and 4 since Node
1 is also sending exploratory data to them.
TABLE 2. Pseudocode segment for epidemic exploratory data propagation
Figure 3. Probabilistic interest dropping reduces interference level during exploratory data
phase. The dotted circle is the interference range of Node 1.
4.2. Epidemic Exploratory Data Propagation
Another way to limit exploratory data flooding is to drop exploratory data probabilistically
during the exploratory data phase. The sender selects gradients from the table probabilistically
as the downstream links. This is the direct implementation to reduce flooding during the
exploratory data phase. Both GOSSIP and Fireworks algorithms can be used.
Obviously, GOSSIP algorithm eliminates a variety of candidates, where the change in
interference level may be very abrupt. In Figure 4, Node 1 discards the exploratory data
completely while Node 1’ randomly selects two neighbors (c = 2). Although GOSSIP reduces
interference in this case, it also eliminates all candidate links from Node 1. Nodes 2 and 2’ send
to all nodes in the gradient table. Links in the transmission range of Node 2 has the same
interference range as those in the transmission range of Node 2’. The transmission of Node 1’
rarely affects that of Node 2’ if Node 1’ and 2’ are beyond the interference range. The two
transmissions in the transmission range of 1’ cause limited interference with each other
Sender side:
1) Gossip version
If it is exploratory data
Set r = random ()
If r <= p
Send it to all nodes in gradient table
Else
Discard it
Else
Execute original code
2) Fireworks version
If it is exploratory data
Set r = random ()
If r <= p
Send it to all nodes in gradient table
Exploratory
DataInterests
1
2
4
3

50
compared to full flooding. However, the advantage is that more candidate links are being
considered.
The combination of (A) and (B) above is worth considering. Intuitively, there will be less
interference and fewer candidates than either scheme. We will not discuss the details here.
(a) GOSSIP algorithm (b) Fireworks algorithms
Figure 4. Comparison of two gossip-based algorithms.
5. IMPLEMENTATION
We conducted the simulation of epidemic interest flooding with a hybrid metric Costp [17] (we
use delay * ETX3
in the simulation) for a modified directed diffusion in ns2. We measure the
throughput and delay performance in different topologies with different network sizes and
identify the optimal percentage of neighbors that drop interests.
5.1. Simulation Methodology
There is a source and a sink in each of the four topologies: grid, triangular tessellation, hexagon
tessellation (Figure 5) and random. For the first three topologies, the source is at the bottom left
corner and the sink is at the top right corner. Both are one hop inside the border. For the random
topology, we manually select source-sink pairs whose distances are almost the same. We
simulate the algorithm using a modification of directed diffusion release 3.2.0 in ns2.29 with the
error and interference model introduced in the earlier section.
We use the IEEE 802.11 protocol for the MAC layer with a channel bandwidth of 2 Mbps. The
transmission range is 250m and the interference range is 550m. In the first two topologies, the
distance between the closest pair of nodes is 250m. We fix the size of random topology to
2000m by 2000m with different numbers of nodes. The simplest CBR traffic is generated at the
rate of 10 packets per second.
Figure 5. Topologies of triangular, grid and hexagon tessellation, from left to right. Every node
has 6, 4, and 3 neighbors, respectively.
1
1’
2 2’

51
We plug in the BER formula (we use a much smaller constant 0.0000005 instead of 0.5 because
it matches the experimental data more accurately) into the interference model and calculate FER
from BER. To make it simple, we set the packet size the same as frame size so that packet loss
rate is FER. Although ETX can be computed by broadcasting probe packets before interest
flooding, to simplify the simulation, we put the FER into the common header of ns2 and let the
protocol read FER and calculate ETX. The noise levels we use in the error model are from the
specification of the Orinoco cards. Ns2 has a global scheduler to record all the traffic so that the
error model is able to capture the dynamic interference.
5.2. Performance Evaluation
We run each setting, with different topology and percentage of nodes that do not drop interests,
for 50 times and take the average of the metrics. In our discussions below, let p be the fraction
of nodes that do not drop interests. We compare the throughput, end-to-end delay and directed
diffusion related metrics, such as, percentage of cases where no interest reaching the source,
percentage of cases where no path was built (which includes the previous cases) and percentage
of cases with zero throughput (which includes the previous two cases). Each run lasts for 300
seconds.
Figure 6. Throughput of triangular tessellation topology with different sizes.
Figure 7. End-to-end delay of triangular tessellation topology with different sizes.
We first run the application for the triangular tessellation with different network sizes as shown
in Figure 5 and measure the throughput (Figure 6) and end-to-end delay (Figure 7). The
throughput increases rapidly when p increase from 0.1 to 0.5. It achieves the highest throughput
when 60% of the nodes do not drop interests. Beyond 60%, the throughput drops because of the
extra interference during exploratory data phase when more or all candidate gradients are
considered (fewer or no node drop interests). Note that when p=1.0, it becomes the original
directed diffusion algorithm and the throughput is lower than the peak, when p=0.6, which is
20% higher than the original directed diffusion. When many nodes drop interests, i.e. p is small,
the throughput is lower because the number of candidates is very small. Network with smaller

52
sizes achieves higher throughput since it suffers less from cumulative packet loss. Figure 7
shows that end-to-end delay goes up to 80 ms when p=0.7 or p=0.8 and drops down after that.
The higher delay is due to high traffic when the throughput peaks. Larger networks have longer
delay since there are more hops between the source and the sink. Trade-off between throughput
and delay is also shown in these two figures. Although we do not have the highest delay when
p=0.7 where throughput is maximum, we believe they reflect the trend that there should be a
magic percentage, between 60% and 70% (or 80%), for this topology.
Figure 8. Percentage of zero throughput, no path built and no interest reaching source of
triangular tessellation topology with 77 nodes.
One of the shortcomings of probabilistic forwarding is that in some cases no forwarding path for
interest, exploratory data or application data, may be found. We analyze the above-mentioned
probability phase by phase. In the first phase, interests may fail to reach the source due to the
epidemic flooding. We measure the percentage of such instances in Figure 8. If more nodes are
updating the gradient table, the percentage of interests reaching the source is monotonically
higher. Also, we compare the percentage of cases where no path was built, which includes the
cases where no interest reaches the source. Another reason for this might be that interference
causes exploratory data to be dropped half way even if interests can reach the source
successfully. As a result, it is no surprise to find that when p=0.6, more paths are built than
when p=1 (original directed diffusion). Percentage of cases with zero throughput (plotted in the
same figure) includes the cases where no path is built. When p=0.6 we not only achieves the
highest throughput, but also achieve the highest percentage of successful delivery of application
data.
Figure 9. Throughput of grid topology with different sizes.

53
Figure 10. End-to-end delay of grid topology with different sizes.
Figure 9 and Figure 10 show the throughput and delay in grid topology with different network
sizes. Delay follows almost the same trend as in the previous topology although the magic
percentage is 0.6 here. Throughput looks quite different in grid topology and it keeps increasing
till all nodes update their gradient tables. The reason is that every node has only 4 neighbors,
which is small and hard to guarantee connectivity. Compared to the previous topology with 6
neighbors, the product of 6 and 0.6 is approximately 4, which is evidence that a magic number,
instead of magic percentage, does exist for the number of neighbors with different topologies.
Figure 11. Throughput of hexagon tessellation with 32, 50, 72 nodes.
Figure 12. End-to-end delay of hexagon tessellation with 32, 50, 72 nodes.
Hexagon tessellations with different sizes are compared in Figure 11 and Figure 12 in terms of
throughput and end-to-end delay. Throughput keeps increasing with higher percentages of
nodes that do not drop interests since the topology has only 3 neighbors, even smaller than grid

54
topology. The peak of end-to-end delay is p=0.7 or p=0.8, which is higher than grid topology at
0.6. The highest delay is due to high traffic level. For throughput, p=0.6 is a breakpoint, beyond
which smaller network sizes have larger delay than larger network sizes. We explain this
phenomenon as: the probability of larger networks having connectivity increases much more
rapidly than that of smaller networks. Figure 12 shows that in all network sizes, the highest
end-to-end delay occurs at p=0.7 or p=0.8 due to the high traffic level.
Figure 13. Throughput of random topology with 60, 80, 90, 100 nodes.
Figure 14. End-to-end delay of random topology with 60, 80, 90, 100 nodes.
We also test random topology with different network sizes (Figure 13 and Figure 14).
Throughput always follows the “up and down” trend and the network with 60 nodes performs
the best. The magic percentage varies between 0.3 and 0.8. Network size of 60 nodes achieves
the highest throughput at a higher percentage than that of 80, 90, or 100 nodes because every
node has fewer neighbors. The delay graph (Figure 14) also indicates the magic number. The
delay of 80-node topology keeping increasing till p=1. Smaller network achieve lower end-to-
end delay than larger ones most of the time. Since nodes are uniformly distributed in this
random topology, given the transmission range, area of the network, and number of nodes, we
calculate the average number of neighbors is 3.9 (almost 4) in the 80-node network and 2.9
(almost 3) in the 60-node network. We may give the same explanation as in Figure 9 or Figure
10 although that is for throughput.
To show the magic number for different topologies, we put all four of them in the same figure
(Figure 14 and Figure 15). We keep the hop count between the source and the sink the same for
them. In Figure 14, hexagon tessellation performs the best and random topology the worst.
Fewer neighbors cause less interference. Although the average number of neighbors of random
topology is 4, the uneven deployment aggravates the side-effects of interference. An interesting
thing is that hexagon tessellation throughput exceeds that of the 7 by 7 grid at p=0.6. The reason
is that it is hard for the hexagonal tessellation topology to maintain connectivity with two few
neighbors. Hexagon tessellation also performs best in end-to-end delay.

55
Figure 14. Throughput comparison of 4 topologies.
Figure 15. End-to-end delay comparison of 4 topologies.
Figure 16. Throughput comparison of 3 topologies with full flooding.
Figure 17. End-to-end delay comparison of 3 topologies with full flooding.

56
To further test the detrimental effect of full flooding, we put the throughput (end-to-end delay)
of the 3 regular topologies with different sizes in the same figure (Figure 16 and Figure 17). We
still make the hop count between the source and the sink the same for each point in the same
column. Hexagon tessellation is affected less than others in both throughput and end-to-end
delay since it has the smallest number of neighbors. The grid topology has the best overall
throughput and delay performance since it has the magic number of neighbors.
6. CONCLUSION
In this paper, we propose a set of epidemic flooding algorithms for QoS-based directed
diffusion, which makes the routing decision more accurate by reducing interference level during
route discovery, as shown in our simulation results. We observe that different topologies have
different magic percentages of neighbors to achieve optimal routing decision (which is different
from the connectivity problem). We also identify the magic number of neighbors that can be
applied to different topologies for optimizing throughput and delay.
For all the topologies we have tested, the optimal number of neighbors for achieving the best
throughput is four. When the number of neighbors is greater than four, maximum throughput is
achieved when some nodes drop interests, i.e. the percentage of nodes not dropping interest is
less than 100%. In these cases, when more nodes drop interests, the level of interference is
reduced and higher throughput is achieved in the resulting routes. As the number of neighbors
decreases, the optimal percentage of nodes that do not drop interests for achieving maximum
throughput moves toward 100%. In most cases, delay is highest when the percentage of nodes
not dropping interest is optimal. Random topologies perform worse than topologies with
regular tessellation no matter what density we choose because of the uneven distribution of
nodes resulting in hotspots with high interference level and difficulty for the routing algorithm
to estimate interference characteristics.
REFERENCES
[1] W. R. Heinzelman, J. Kulik, and H. Balakrishnan, “Adaptive Protocols for Information
Dissemination in Wireless Sensor Networks,” in Proc. ACM MobiCom '99 Seattle, WA, 1999,
pp. 174-185.
[2] C. Intanagonwiwat, R. Govindan, and D. Estrin, “Directed Diffusion: A Scalable and Robust
Communication Paradigm for Sensor Networks,” in Proc. ACM MobiCom '00 Boston, MA,
2000, pp. 56-67.
[3] W. Heinzelman, A. Chandrakasan, and H. Balakrishman. “Energy-Efficient Communication
Protocols for Wireless Microsensor Networks”. In Proceedings of Hawaiian International
Conference on System Science, January 2000.
[4] B. Karp and H. T. Kung, “GPSR: Greedy Perimeter Stateless Routing for Wireless Networks,”
in Proceedings of MOBICOM 2000, Boston, MA, USA, 2000, pp. 243-254.
[5] Y. Xu, J. Heidemann, D. Estrin, “Geography-informed energy conservation for ad hoc routing,”
MobiCom 2001, Rome, Italy, pp. 70-84, July 2001.
[6] Y. B. Ko and N. H. Vaidya. Location-aided routing (LAR) in mobile ad hoc networks. In Proc.
Fourth Annual ACM/IEEE International Conference on Mobile Computing and Networking
(MobiCom), pages 66-75, 1998.
[7] S. Basagni, I. Chlamtac, V. R. Syrotiuk, and B. A. Woodward. A distance routing effect
algorithm for mobility (DREAM). In Proc. Fourth Annual ACM/IEEE International Conference
on Mobile Computing and Networking (MobiCom), pages 76-84, 1998

57
[8] D. B. Johnson and D. A. Maltz. Dynamic Source Routing in Ad Hoc Wireless Networks. Kluwer
Academic Publishers, 1996.
[9] C. E. Perkins and E. M. Royer. Ad-hoc on-demand distance vector routing. In Proc. 2nd IEEE
Workshop on Mobile Computing Systems and Applications, pages 90-100, February 1999.
[10] Z. Haas and M. Pearlman. The performance of query control schemes for the zone routing
protocol. In Proc. ACM SIGCOMM, pages 167-177, August 1998.
[11] V. Park and M. S. Corson. A highly adaptive distributed routing algorithm for mobile wireless
networks. In Proc. INFOCOM, pages 1405-1413, April 1997.
[12] A. Demers, D. Greene, C. Hauser, W. Irish, J. Larson, S. Shenker, H. Sturgis, D. Swinchar, D.
Terry, epidemic algorithms for replicated database maintenance, in: Proceedings of the sixth
ACM Symposium on Principles of Distributed Computing, Vancouver, Canada, August 1987.
[13] Zygmunt Haas, Joseph Halpern, and Li Li. Gossip based ad hoc routing. In INFOCOM, New
York, USA, June 2002.
[14] L. Orecchia, A. Panconesi, C. Petrioli, A. Vitaletti, Localized techniques for broadcasting in
wireless sensor networks, Proceedings of the 2004 joint workshop on Foundations of mobile
computing, October 01-01, 2004, Philadelphia, PA, USA.
[15] Anonymous Gossip: Improving Multicast Reliability in Mobile Ad-Hoc Networks, Proceedings
of the The 21st International Conference on Distributed Computing Systems, p.275, April 16-19,
2001.
[16] D. Ganesan, B. Krishnamachari, A. Woo, D. Culler, D. Estrin, S. Wicker, An empirical study of
epidemic algorithms in large scale multihop wireless networks, Tech. Rep. IRB-TR-02-003, Intel
Research, March 2002.
[17] S. Li, A. Lim, S. Kulkarni, C. Liu, “EDGE: A Routing Algorithm for Maximizing Throughput
and Minimizing Delay in Wireless Sensor Networks”, Proceedings of the 26th Military
Communications Conference (MILCOM’07), October 2007.
[18] P. Chaporkar, K. Kar, and S. Sarkar. Throughput guarantees through maximal scheduling in
wireless networks. In Proc. 43rd Allerton Conf. on Commun., Control, and Comp., September
2005.
[19] Kamal Jain , Jitendra Padhye , Venkata N. Padmanabhan , Lili Qiu, Impact of interference on
multi-hop wireless network performance, Proceedings of the 9th annual international
conference on Mobile computing and networking, September 14-19, 2003, San Diego, CA, USA.
[20] J. J. Lemmon, “Wireless Link Statistical Bit Error Model,” NTIA Report 02-394, June 2002.
[21] S. Lee, B. Bhattacharjee, and S. Banerjee. Efficient Geographic Routing in Multihop Wireless
Networks. In ACM Mobihoc’05, May 2005.
[22] R. Draves, J. Padhye, and B. Zill, Comparison of Routing Metrics for Static Multi-Hop Wireless
Networks. Proc. of the Conference on Applications, Technologies, Architectures, and Protocols
for Computer Communications (SIGCOMM), pages 133–144, ACM Press, 2004.
[23] D. D. Couto, D. Aguayo, J. Bicket, and R. Morris. A high-throughput path metric for multi-hop
wireless routing. Proc. of MobiCom, 2003.
[24] R. Draves, J. Padhye, and B. Zill, Routing in multi-radio, multi-hop wireless mesh networks.
Proc. of the 10th Annual International Conference on Mobile Computing and Networking, 2004.
[25] J. Park and S. Kasera, Expected data rate: an accurate high-throughput path metric for multi-hop
wireless routing. Proc. of second Annual IEEE Communications Society Conference on Sensor
and Ad Hoc Communications and Networks, 2005.

58
Authors
Shuang Li is a PhD student in the department of Computer Science and Engineering at
the Ohio State University. She received her M.S. degree from Auburn University,
Auburn, Alabama in 2008. Her research interests are routing metrics in wireless sensor
networks, power control and resource allocation in MANETs and VANETs.
Alvin Lim is currently an associate professor of computer science and software
engineering at Auburn University. He received his Ph.D. degree in computer science
from University of Wisconsin at Madison in 1993. His research interests include self-
organizing sensor networks, mobile and pervasive computing, network security,
wireless networks, reliable and dynamically reconfigurable distributed systems,
complex distributed systems, mobile and distributed databases, distributed operating
systems, and performance measurement and analysis. He has published widely in
journals and conferences in these networking and distributed systems areas. He is a subject area editor of
the International Journal of Distributed Sensor Networks. His work had been supported by the National
Science Foundation, the DARPA SensIT program, U.S. Air Force Research Lab and the U.S. Army.
Cong Liu is a PhD student in the department of Computer Science at the University of
North Carolina – Chapel Hill. He received his Master’s degree from the department of
Computer Science and Software Engineering at Auburn University in Dec. 2007. His
current research interests are real-time systems and scheduling theory on real-time
multicore platforms.

Improving QoS-based Routing by Limiting Interference in Lossy Wireless Sensor Networks

More Related Content

What's hot (18)

Similar to Improving QoS-based Routing by Limiting Interference in Lossy Wireless Sensor Networks (20)

Recently uploaded (20)

Improving QoS-based Routing by Limiting Interference in Lossy Wireless Sensor Networks