A unified framework for topology management multi-rate ad hoc networks

Abstract

Finite battery energy and limited bandwidth resources are the two major constraints in ad hoc networks. Therefore, researchers have persistently sought for optimization algorithms to reduce the control overhead and to increase bandwidth utilization efficiency, using mechanisms such as topology control and management, multi-rate adaptation etc. to cut down the energy expenditure in ad hoc networks. However, the energy conservation and network throughput improvement are handled separately in many cases. In this paper, we propose a unified framework that combines the multi-rate adaptation and clustering mechanisms so as to provide the optimal network throughput under very low control overhead. The framework incorporates two control mechanisms into the networking stack. The first mechanism is based on a novel clustering algorithm, called PATM (priority-based adaptive topology management), that constructs the backbone of the network topology for routing protocols. The second mechanism is a medium access control protocol, called RMAC (relay-based MAC), that provides high data transmission rates between adjacent backbone nodes. The efficiency of the unified framework is evaluated using theoretical analysis and extensive simulations with DSR (dynamic source routing) as the underlying routing protocol. We demonstrate that our solution not only significantly reduces the routing control overhead, but also substantially improves the network throughput.

Full text

Page 1

1
A Unified Framework for Topology Management in
Multi-Rate Ad Hoc Networks
Haixia Tan, Weilin Zeng, Lichun Bao and Tatsuya Suda
Abstract—Finite battery energy and limited bandwidth resources are the
two major constraints in ad hoc networks. Therefore, researchers have per-
sistently sought for optimization algorithms to reduce the control overhead
and to increase bandwidth utilization efficiency, using mechanisms such as
topology control and management, multi-rate adaptation etc. to cut down
the energy expenditure in ad hoc networks. However, the energy conserva-
tion and network throughput improvement are handled separately in many
cases. In this paper, we propose a unified framework that combines the
multi-rate adaptation and clustering mechanisms so as to provide the opti-
mal network throughput under very low control overhead. The framework
incorporates two control mechanisms into the networking stack. The first
mechanism is based on a novel clustering algorithm, called PATM (Priority-
based Adaptive Topology Management), that constructs the backbone of
the network topology for routing protocols. The second mechanism is a
medium access control protocol, called RMAC (Relay-based MAC), that
provides high data transmission rates between adjacent backbone nodes.
The efficiency of the unified framework is evaluated using theoretical anal-
ysis and extensive simulations with DSR (Dynamic Source Routing) as the
underlying routing protocol. We demonstrate that our solution not only
significantly reduces the routing control overhead, but also substantially
improves the network throughput.
Index terms — topology management, relay MAC, cross-
layer design, ad hoc networks.
I. INTRODUCTION
In comparison to cellular networks which are supported by
a fixed, wired infrastructure, and scheduled by the central
base stations, ad hoc networks [14] are self-organizing, self-
configuring wireless networks.
that autonomously establish connectivity via multi-hop wireless
communications without relying on any existing, pre-configured
network infrastructure or centralized control. Because of the au-
tonomy, ad hoc networks are gaining popularity in emergency
situations such as battlefield operations, disaster recovery, and
commercial applications such as group conferencing, home net-
working and vehicular communication. Furthermore, the grow-
ing interest in sensor network applications has created a need for
protocols and algorithms for large-scale ad hoc networks.
Two major constraints in ad hoc networks are the finite bat-
tery energy and limited bandwidth resources. Therefore, most
research on ad hoc networks has focused on optimization algo-
rithms aimed at reducing control overhead and increasing the ef-
ficiency of bandwidth utilizations. For instance, topology man-
agement techniques try to provide the minimal network connec-
tivity information to routing protocols so as to reduce the control
overhead by cutting back the amount of topological information
maintenance and the routing updates [3]. In other occasions, au-
tomatic data transmission rate selection protocols allow wireless
devices to operate at high data rate when the channel conditions
are sufficiently clear so as to improve the network throughput
and to increase the bandwidth efficiency [7].
It consists of mobile nodes
Donald Bren School of Information and Computer Sciences, Univer-
sity of California, Irvine, CA 92697, Email:
suda}@ics.uci.edu
{htan, wzeng, lbao,
Topology control mechanisms are currently one of the most
active research areas studied for energy conservations [10] [11]
[17]. A topology control mechanism, ABTC, [20] introduced a
distributed algorithm where each node determines the minimal
power, by which it can either directly receive packets sent by its
original neighbors, or it can receive the packets through relays in
its original neighbor set. However, ATC [12] observes that min-
imally connected topology does not always provide the optimal
performance. Therefore ATC estimates the traffic load within a
separate control channel, and adapts the transmission power of
each node accordingly to achieve the maximum throughput per
unit energy.
Topology management is different from topology control in
that topology control usually refers to power control, and tries
to physically manipulate the hop-by-hop links of the network
topology by adjusting the transmission power levels. In com-
parisons, topology management keeps the original network con-
nectivities intact, and tries to construct a connected dominat-
ing set or backbone in the network graph to carry out the con-
trol functionalities [5] [6] [21]. Backbone nodes usually remain
in active state, and can originate, receive and forward packets.
Other nodes are usually in passive state, and only send or re-
ceive packets, or even sleep in a power saving mode. This way,
we can reduce the interference and collision in the network, and
save energy consumption.
Theadvantagesoftopologycontrolarethatthepower-savings
and co-channel interference reduction are immediate although it
introduces unidirectional links to the network [15], while the
topology management provides a foundation for more advanced
managementfunctionalitiessuchassleepingmodearrangement,
channel access scheduling. These two schemes are not contra-
dictory, and compliments each other.
The backbone construction starts out with maintaining a clus-
ter hierarchy of the network nodes, and are completed by means
of gateways connecting the cluster-heads. SPAN [5] adaptively
elects coordinators according to the remaining energy and the
number of pairs of neighbors a node can connect. Geographic
Adaptive Fidelity (GAF) for sensor networks [24] subdivides
a sensor network into small grids, such that only one node in
each grid is active at each point of time. In Cluster-based En-
ergyConservation(CEC)[23], thecluster-headelectionisbased
on the battery lifetime and the degree of each node. Topology
Management by Priority Ordering (TMPO) [3] proposed to con-
struct and maintain a network backbone based on MDS (Mini-
mal Dominating Set) and CDS (Connected Dominating Set) us-
ing only two-hop neighbor information. On-Demand Cluster
Formation (ODCF) [25] is an on-demand traffic-driven cluster-
ing algorithm, which additionally optimized by piggybacking
topology management information onto outgoing data packets.
On the other hand, the advance of wireless communications

Page 2

2
technology enable wireless devices to operate at different data
rates under different channel conditions by changing the mod-
ulation/coding schemes. This potentially brings the higher data
throughput than the fixed data rate communication. The routing
layer control mechanisms have to take such multi-rate transmis-
sion capabilities into account in order to improve the network
throughput. Otherwise, ignorance of such facts can lead unde-
sirable consequences for the higher network layer [9].
A fair amount of research interest has focused on how to ex-
ploit this multi-rate capability in wireless ad hoc networks at
both MAC layer alone and the cross-layer coordinations be-
tween MAC and the routing protocols [18] [19]. Of the ap-
proaches based on coordinated MAC and routing protocols,
Yuen et al. considered using route selection criteria such as in-
terference, throughput or delay [26] in the route selection pro-
cess. On the other hand, Awerbuch et al. introduced a new met-
ric called Medium Time Metric, which considers how different
network components affect each other in a multi-rate environ-
ment [2].
The rest of the paper is organized as follows. Section II de-
scribes PATM (Priority-based Adaptive Topology Management)
and discusses other topology management algorithms. Section
III presents RMAC (Relay-based MAC protocol), and compares
RMAC with another similar protocol, RBAR [7]. Section IV
describes the unified framework combining PATM with RMAC.
Performance evaluations based on theoretic analysis and exten-
sive simulations are presented in Section V. Section VI summa-
rizes the paper.
II. TOPOLOGY MANAGEMENT
We propose a Priority-based Adaptive Topology Management
(PATM) for topology management purposes.
In PATM, we assume that each mobile node in the ad hoc net-
work has a unique identifier and an omni-directional transceiver.
The transmission powers are set to the same value so that bidi-
rectional links are maintained between adjacent neighbors. If
two nodes u and v are within the packet-reception range of each
other, they are called one-hop neighbors of each other. The set
of one-hop neighbors of a node i is denoted by N1
that are not connected but share at least one common one-hop
neighbor are called two-hop neighbors of each other.
The dominating set of a graph is a subset of nodes with the
following property: each node is either in the dominating set, or
is adjacent to a node in the dominating set. Similar to TMPO
[3], the dominating set election is based on the priority values
computed for each node. If the priority of a node is the highest
amongitsone-hopneighborsoramongtheone-hopneighborsof
one of its one-hop neighbors, the node becomes a cluster-head.
In addition, the priority computation follows the same heuristics
as in TMPO, and is a function of the node’s ID, current time,
and the willingness of the node to serve as a cluster-head.
PATM improves the performance of ad hoc network topology
management by using the following mechanisms:
1. While some clustering algorithms require such assumptions
as node position information [24], knowledge of IDs of neigh-
bors [3] and distance to neighbors [20], or synchronization
among nodes [3], PATM uses simple and readily available in-
formation such as neighbor IDs, and its own remaining energy
i. Two nodes
level of each node to elect cluster-heads, instead of the more
expensive approaches that require synchronization or distance
information etc.
2. In ad hoc networks, collecting accurate and up-to-date topol-
ogy information incurs a heavy traffic overhead. Similar to
ODCF [25], PATM reduces the control over-head through pig-
gybacking the control information onto the data traffic as much
as possible, while sufficiently keeping nodes informed of the
topology updates for topology management purposes.
3. Different from TMPO, where node IDs, current node types
and willingness values are exchanged to derive node priorities at
each node, PATM requires nodes directly exchange priority and
node type information. This way, the synchronization overhead
is eliminated, and nodes change their priorities only when it is
necessary.
4. PATM is a proactive topology management scheme. Nodes
adapt the interval of their priority computation and control infor-
mation exchange according to the network traffic and mobility
conditions. For example, when a region of the network carries
verylighttraffic, PATMprolongstheintervalofpriorityupdates,
causing less control overhead, and providing more energy sav-
ings.
(B) Cluster−head Election
bd
e
f
g
h
i
j
(A) A Sample Topology
aa
c
bd
e
f
g
h
i
j
c
Host
Clusterhead
Host
Fig. 1. Cluster-head Election in A Sample Network Topology
As an example, Fig. 1 (A) shows the topology of a sample ad
hoc network, and Fig. 1 (B) shows one of the MDS (Minimal
Dominating Set) of the network topology, {c,g,j}, elected by
PATM. In fact, the minimum dominating set could be {b,f}.
After the MDS is formed, the CDS (Connected Dominating
Set) is constructed by adding doorways and gateways in the
MDS. Doorways are necessary when two cluster-heads in the
MDS are separated by three hops and there are no other cluster-
heads between them. Gateways are added for simply connecting
two clusterheads or one clusterhead and another doorway when
there are no other cluster-heads between them.
The construction of the backbone topology is completed by
adding links between the elected cluster-heads, doorways and
gateways to the CDS. Routing protocols can now exchange rout-
ing information and control messages among the nodes in the
backbone, which greatly reduces the nodes and links participat-
ing the route maintenance. Data communication is carried out
by the source node sending data packets to its corresponding
cluster-head first, then the data packets are forwarded along the
backbone nodes and links toward the destination.
The topology management using PATM is transparent to the
routing protocols. Therefore, any routing protocols could be
used on top of PATM, including the commonly referred AODV
[13] and DSR [8].
Fig. 2 (C) shows one possible result of the doorway election

Page 3

3
(C) Adding Doorway
e
f
g
h
i
j
(D) Adding Gateways And Connections
a
c
b
a
c
bd
e
f
g
h
i
j
d
Host
Clusterhead
Doorway
Host
Clusterhead
Doorway
Gateways
Fig. 2. Doorway and Gateway Elections
because node d may have the maximum priority value on the
three-hop path between node c and node g. Afterward in Fig.
2 (D), nodes b and f become gateways between two cluster-
heads or between a cluster-head and a doorway. The backbone
topology is formed after adding all the possible links in blue
and red between gateways, cluster-heads, doorways that are in
the adjacency in Fig. 2 (D).
III. MULTI-RATE ADAPTATION
The physical layer of IEEE 802.11b can provide data rates at
1Mbps, 2Mbps, 5.5Mbps and 11Mbps at different transmission
ranges in typical deployments as shown in Fig. 3 [1].
Distance (m)
1
2
3
4
5
6
7
8
9
10
11
30m 60m 100m
Data Rate (Mbps)
Fig. 3. The Relationship between Transmission Ranges And Achievable Data
Rates
In ad hoc networks based on IEEE 802.11b and the mini-
mum hop routing scheme, the lack of communication between
the multi-rate adaptive MAC layer and the routing protocol
can result in worse performance than what the ad hoc net-
work could support because the minimum hop routing protocol
tendstochoosetheshortest-hopdistancebutlowdata-ratepaths,
whereas the longer-hop paths with higher data-rate could have
provided better network throughput under certain conditions.
In the MAC-alone approaches, a rate adaptive MAC, called
Receiver-Based AutoRate (RBAR), is presented in [7] to use
RTS-CTS packet exchange to select the appropriate rate for the
following data packet transmission. In RBAR, RTS-CTS carries
the information such as packet size and the data rate, instead of
the duration of the reservation. With this information, the re-
ceiver can exchange the data rate information with the sender
and the neighboring nodes are also able to calculate to the dura-
tion of the requested reservation.
However, when high data-rate links that form two-hop path
between a pair of nodes are available, RBAR would have to send
a data packet through the intermediate node using the RTS-CTS-
DATA-ACK sequence multiple times [7]. This is the problem
that we try to resolve.
We propose a Relay-based MAC, called (RMAC), to support
multi-rate communication with minimum-hop routing protocols
in ad hoc networks. In RMAC, the data transmission can be
performed in one of two modes: direct transmission or relay-
based transmission, whichever provides a higher data-rate for
a certain data packet size. RMAC discovers an intermediate
node residing between the transmitter and the receiver, and em-
ploys the intermediate node as a high-speed relay to send the
data packet using a single RTS-CTS-DATA-DATA-ACK mes-
sage sequence. In addition, the RTS and CTS control messages
have been adapted in RMAC for data-rate and relay selections.
Different from the approach of modifying routing metrics [2]
[26], our scheme affects the MAC layer alone while inherently
assisting routing decisions.
(a)
2 Mbps
5.5 Mbps
11Mbps
A
B
C
Regular Node
Relay Node
(b)
t
ABC
RTS
CTS
DATA
ACK
11Mbps
5.5 Mbps
2 Mbps
2 Mbps
2 Mbps
Relay
tt
Fig. 4. Multiple Data-Rate Links between Two Nodes
Fig. 4 illustrates a sample scenario of multi-rate adaptation,
where the direction data transmission between nodes A and C
can only achieve 2 Mbps (mega-bit per second) as shown in
Fig. 4 (a). RMAC opportunistically replaces the long low-rate
data link (A,C) with two-hop high date-rate links at 5.5 Mbps
and 11Mbps through node B using a single round of RTS-CTS-
DATA-DATA-ACK handshake sequence in Fig. 4 (b).
RMAC achieves multi-rate adaptation in two steps: relay
neighbor discovery and data communication based on the RTS-
CTS-DATA-ACK sequence.
Duringtherelayneighbordiscovery, RMACmaintainsarelay
table at each node to keep track of neighbors to find out the pos-
sible data rates with and without relays. Possible relay nodes’
ID and the corresponding data rates are recorded for future data
transmissions.
In IEEE 802.11 standard, to guarantee correct operation in the
MAC protocols, the initial RTS-CTS exchange and ACK mes-
sage are set at the base rate even though the channel conditions
are sufficiently good for high data rate transmission. In RMAC,
a node overhears signal strength of RTS, CTS and DATA mes-
sages exchanged between its neighbors, and calculates the data
rate it can provide between itself and the transmitter and re-
ceiver, respectively. Then the node determines whether its par-

Page 4

4
ticipation in the communication could provide higher through-
put than the original communicating pair. If so, it sends out an
RTR (Request To Relay) packet to the transmitter in a separate
process. The RTR packet contains the data rate that could be
used in transmitting packets from the transmitter to the relay,
and the data rate from the relay to the receiver.
To determining the optimality of a relay node between a pair
of other nodes, RMAC computes the transmission durations of a
given packet for both direct and relay-based transmissions under
the available data rate. For such purpose, symbols Sdata, Srts,
Scts, Sackand Sphyare introduced to represent the respective
sizes of data frame, RTS, CTS and ACK control frames, and
the physical layer header. Symbols Rbase, Rdataare introduced
to denote the base rate and the data transmission rate, respec-
tively. The transmission cycle for a data packet includes the
time for transmitting RTS, CTS, DATA and ACK. For simplic-
ity, we have ignored the propagation delay, SIFS (short inter-
frame space), and DIFS (distributed inter-frame space).
Then, the transmission time in direct transmission mode is
expressed as
Tdirect=Sdata
Rdata
+Srts+ Scts+ Sack+ 4Sphy
Rbase
.
Similarly, denote Rdata1, Rdata2as the data rates of the first
hop and the second hop, respectively in relay-based transmis-
sions. Then, the transmission time in the relay transmission
mode is:
Sdata
Rdata1
Rdata2
Trelay=
+
Sdata
+Srts+ Scts+ Sack+ 5Sphy
Rbase
.
Therefore, the condition to use the relay node between two
nodes is if Eq. (1) holds.
Trelay−Tdirect=
where Rdata1, Rdata2are the data rates between the relay node
and the source and destination nodes, respectively, whereas
Rdatais the data rate between the source and destination nodes.
To illustrate the differences, we use the data rates provide by
IEEE 802.11b to compute the threshold packet sizes in direct
and relay-based transmissions under several scenarios. It ap-
pearsthatwhenthedatarateofdirecttransmissionis5.5Mbpsor
11Mbps, direct transmission time is always smaller than relay-
based transmission time regardless of the data packet size. That
is, the threshold data frame size is infinite when direct trans-
mission rate is 5.5 Mpbs or 11 Mbps. However, when direct
transmission operates at 1 Mbps or 2 Mbps, the thresholds of
data frame sizes are given in Table I for a few scenarios. When
the packet size is greater than the threshold, relay-based trans-
missions takes less time than direct transmissions.
In RMAC, a single RTS-CTS-DATA-DATA-ACK sequence
saves the overhead of additional one RTS, one CTS, one ACK
and one backoff latency for renewed channel access as done in
the previous approaches [7]. RMAC incurs less control over-
heads than routing algorithms supporting multi-rate features be-
cause routing protocols involve multi-hop coordination. Using
RMAC, a simple one-hop coordination is used while still uti-
lizing the multi-rate feature of the wireless ad hoc networks,
therefore hiding the complexity of multi-rate capabilities in the
physical layer.
Sdata
Rdata1+Sdata
Rdata2+Sphy
Rbase−Sdata
Rdata
< 0. (1)
TABLE I
MINIMUM DATA FRAME SIZES FOR RELAY-BASED TRANSMISSIONS
Relay-Based \ Direct
2 Mbps + 5.5 Mbps
2 Mbps + 11 Mbps
5.5 Mbps + 5.5 Mbps
5.5 Mbps + 11 Mbps
11 Mbps + 11 Mbps
1 Mbps
75 B
63 B
40 B
35 B
31 B
2 Mbps
∞
∞
177 B
106 B
76 B
IV. A UNIFIED FRAMEWORK
The main purpose of topology management is to reduce the
routing complexity and energy consumption, while keeping the
network connectivity at the same time. Using multi-rate aware
MAC protocols, such as RMAC, we can additionally mitigate
the reduced throughput due to the longer data links in ad hoc
networks, and the communication between the cluster heads
can still be achieved at the high data rate.
edge, the topology management and multi-rate adaptation al-
gorithms were optimized independently so far. We provide a
unifiedframeworkforefficientnetworkoperationsinadhocnet-
works by combining multi-rate adaptation algorithm with topol-
ogy management to achieve high network throughput with low
overhead.
To our knowl-
(b)
Network Layer (Routing)
Layer
Link
Physical Layer
Management
Topology
MAC (RMAC)
Network Layer (Routing)
Layer
Link
MAC (802.11 DCF)
Physical Layer
(a)
Fig. 5. A Unified Framework for Topology Management and RMAC
Most computer network architecture use the layered approach
[22]. Fig. 5 (a) illustrates a networking architecture based on
IEEE 802.11 [1]. The upward arrows represent routing informa-
tion flows, and the downward arrows denote routing decisions
for data forwarding.
Our unified framework seamlessly combines topology man-
agement and multi-rate adaptation into the networking architec-
ture as shown in Fig. 5 (b). The topology management com-
ponent is inserted between the networking layer and the MAC
layer so as to provide a succinct presentation of the network
topology to the network layer, and to control the routing update
overhead. A second component, the Relay-based MAC proto-
col, replaces IEEE 802.11 DCF [1] to enhance the routing de-
cisions of the network layer by increase the data transmission
rate. The end-results are efficient network routing protocol, and
enhanced data forwarding capability.
Fig. 6 gives an example of how topology management and
routing protocols work together in a multi-rate ad hoc network.
Suppose link (g,i) has data rate 2 Mbps, and links (g,h), (g,h)

Page 5

5
(B) Relay−based Minimum−hop Path
c
bd
e
f
g
h
i
j
(A) Minimum−hop Path
a
c
bd
e
f
g
h
i
j
a
Host
Clusterhead
Doorway
Gateways
Gateways
Relay
Host
Clusterhead
Doorway
Fig. 6. Topology Management And Rate-Adaption
have data rate 11 Mbps. Without RMAC, the topology manage-
ment and routing protocol based on minimum-hop routing result
in a low throughput path a − g − i between nodes a and i, as
shown Fig. 6 (a). With RMAC support, a high throughput path
a − g − h − i is chosen in Fig. 6 (b), instead. However, the
routing protocol still sees the path a − g − h − i as a − g − i
because RMAC has hidden the multi-rate adaptation within the
MAC layer.
V. PERFORMANCE EVALUATION
A. Theoretic Analysis
A.1 Probabilities of Node States
In PATM, the role of a node can be either one of the four:
cluster-head, gateway, doorway, and host. The probabilities of a
node becoming a cluster-head, a gateway, a doorway and a host
are denoted by pch, pg, pdand ph, respectively. Thus
pch+ pg+ pd+ ph= 1
Bao et al. computed pchand derived the expected size of the
MDS in an area with N nodes in the uniform distribution [3]:
|MDSN| = N · pch.
(2)
Similarly, the expected size of the CDS is:
|CDSN| = N · (pch+ pg+ pd) = N · (1 − ph).
Although pchwas easily derived assuming the Poisson distri-
bution for the number of nodes located in unit area, the theoreti-
cal computations of pg, pd, and phare very complicated because
of the relative geographic location estimation between two-hop
neighbors. Instead, we use extensive simulations to estimate the
probabilities.
Assumingauniformdistributionofnodesonaninfiniteplane,
the various probabilities of nodes in cluster-head, gateway, door-
way and host states in topology management is directly related
with the average number of one-hop neighbors, which is the
node distribution density times the area covered under the trans-
mission range. To find out the relationship between the average
number of one-hop neighbors and ph, we did a set of simula-
tions by randomly placing 500 nodes in a 4000 × 4000 square
feet area, and varying the antenna transmission range from 100
feet to 1100 feet. In each simulation scenario, the numbers of
nodes in various states are periodically collected, and the aver-
age number of one-hop neighbors and the average probability of
a node in the host state are printed out after long period of time.
Fig. 7 shows the relationship between the average number of
(3)
one-hop neighbors of each node and the probability of nodes in
the host state, ph.
02040 60 80100120
0
0.2
0.4
0.6
0.8
1
Avg Number of One−Hop Neighbors
Probability of Being A Host
Fig. 7. Probability of Being a Host
A.2 Collision Probabilities
In order to compute the network throughput under flat topol-
ogy scenarios and topology-managed scenarios, we follow the
analytical methods used in [4].
throughput of an ad hoc network, which is defined as the max-
imum the system throughput achieved as the offered load in-
creases. It assumed ideal channel conditions without hidden
terminals, and a fixed number n of contending stations to each
node, which always has backlogged packets for transmission.
Our computations are based on some conclusions of [4].
In CSMA/CA channel access scheme, suppose that the min-
imum contention window size is W = CWmin, and the max-
imum contention window size is 2mW = CWmax, where m
is a predefined parameter of the CSMA/CA protocol. In addi-
tion, denote p as the constant and independent probability that a
transmitted packet encounters collision, regardless of the num-
ber of retransmissions that the packet has already suffered. Then
the probability τ that a station transmits in a randomly chosen
slot time is expressed as [4]:
[4] evaluated the saturation
τ =
2(1 − 2p)
(1 − 2p)(W + 1) + pW(1 − (2p)m)
Because hidden terminal problems are not considered, under
flat topology organizations, the aforementioned collision proba-
bility p of a node is equal to the probability that at least one of
the node’s one-hop neighbors transmits [4], which is:
(4)
p = 1 − (1 − τ)n−1
(5)
where n is the average number of one-hop neighbors because all
nodes are in active states under the flat topology organizations.
Under the network topology management by PATM, the num-
ber of contending nodes dramatically decreases because more
nodes are in dormant host state. Therefore, the average number
of contending nodes, denoted by n?, becomes n?= n(1 − ph),
and the collision probability p?of a node is:
p?= 1 − (1 − τ?)n?−1= 1 − (1 − τ?)n(1−ph)−1
where τ?is derived by replacing the p with p?in Eq. (4).
As an example for comparison purposes, when the average
number of one-hop neighbors is n = 40, n?= n(1−ph) = 40×
(6)

Page 6

6
(1−0.65) = 14. Accordingly, Eq. (4) and Eq. (5) have a unique
solution: τ = 0.018, and p = 0.50. Similarly, Eq. (4) and Eq.
(6) have a unique solution: τ?= 0.032, and p?= 0.34. As we
can see, with topology management, the collision probability
decreases by p−p?= 0.50−0.34 = 16%, which demonstrates
that topology management can reduce collision probability.
A.3 Successful Transmission Probabilities
Wecomputetransmissionprobabilitiesunderflattopologyor-
ganization and under topology management. Denote Ptras the
probability that there is at least one transmission in a considered
slot time under flat topology, and denote P?
under topology management. Then,
tras the probability
Ptr= 1 − (1 − τ)n
(7)
P?
tr= 1 − (1 − τ?)n?
(8)
Furthermore, denote Psas the probability that a transmission
over the channel is successful under flat topology organization,
and denote P?
then we have [4]:
sas the probability under topology management,
Ps=nτ(1 − τ)n
Ptr
=
nτ(1 − τ)n
1 − (1 − τ)n
=n?τ?(1 − τ?)n?
1 − (1 − τ?)n?
(9)
P?
s=n?τ?(1 − τ?)n?
P?
tr
(10)
As an example, taking the same parameters as in Section V-
A.2 into Eq. (9) and Eq. (10), we have Ps= 0.69, and P?
0.80. Itiseasytoseethatthesuccessfultransmissionprobability
under topology management can increase by P?
0.69 = 11% when the average number of one-hop neighbors is
40.
s=
s−Ps= 0.80−
B. Simulations Results
The unified framework for topology management in multi-
rate ad hoc networks is simulated using NS-2 simulator [16].
In the simulations, we run the Dynamic Source Routing proto-
col (DSR) [8] as the basic routing protocol with combinations
of PATM and multi-rate MAC algorithms. DSR is a reactive
routing protocol, consisting of route discovery and maintenance
mechanisms. Route discovery mechanism has two phases: route
request and route reply. Under topology management scheme,
we modify the route request phase such that every node re-
broadcasts an RREQ packet only if the node is not a host. As
a result, hosts are excluded from serving as intermediate nodes
on a routing path. However, since the nodal type is initialized to
host, we allow hosts to forward RREQ packets when the hosts
do not have a cluster-head in its one-hop neighborhood. After
the initialization phase, only non-host nodes can broadcast and
forward RREQ packets.
To see the differences between various optimizations in ad
hoc networks, we compare the performance of the following dif-
ferent combinations of routing protocols and MAC protocols.
1. DSR: DSR combined with the plain IEEE 802.11 MAC.
2. DSR-PATM: DSR with PATM and the plain IEEE 802.11.
3. DSR-SPAN: DSR with SPAN [5] and the plain IEEE 802.11.
4. DSR-PATM-RBAR: DSR with unified PATM and RBAR
(Receiver-Based AutoRate).
5. DSR-PATM-RMAC: DSR with unified PATM and RMAC
(Relay-Based MAC).
The simulations are carried out in ad hoc networks generated
over a 1000 × 400 square meter area with 70 nodes moving in
random directions at random speeds. The transmission range is
fixed at 250 meters. Each simulation runs for 890 seconds.
Both TCP and Constant Bit Rate (CBR) data traffic are sim-
ulated. Each source starts a session randomly with data rate 4
packets/second and 1460 bytes payload size. Accordingly, the
following metrics are used to show the performance of each pro-
tocol.
1. Normalized Control Overhead: the total number of control
packets divided by the total number of data packets delivered to
destinations.
2. Delivery Ratio: the total number of data packets delivered
to destinations divided by the total number of data packets sent
from sources.
3. Average Delay: the average delay of all the data packets de-
livered to destinations.
4. Throughput: theamountofdatadeliveredtodestinationsdur-
ing a simulation divided by the time span of the simulation.
Fig. 8 shows the performance comparisons under different
metrics between the five protocols when 20 TCP flows are gen-
erated between randomly chosen pairs of nodes. Different de-
grees of mobility are simulated where the maximum speeds of
the nodes are spaced by 5 meters/second in speed range from
0 to 30 meters/second. It is apparent that our unified solution
PATM+RMAC improves the throughput, and reduces the rout-
ing overhead and average delay dramatically, as shown by the
“*” curves in Fig. 8.
In Fig. 9 shows the performance of the five protocols under
different amount of traffic loads by changing the number of TCP
flows. In these simulations, we fixed the maximum speed of the
nodes to 20 meters/second, and vary the number of TCP flows
from 5 to 40 by add 5 flows at a time. As shown in the Fig-
ure, we got the similar conclusion that the unified framework
makes great contribution to throughput improvement, and over-
head and delay reduction. Especially, almost one order of mag-
nitude difference is observed between the best and worse proto-
cols in terms of the normalized overhead (see Fig. 8(a) and Fig.
9(a)).
Fig. 10 compares the performance of the five protocols under
various CBR flows. The maximum speed of the nodes is 20
meters/second, and the number of flows varies from 10 to 50
by 5 flows difference. As mentioned earlier, four packets are
sent per second in the CBR flows, and each packet has 1460
bytes of payload. Our unified PATM+RMAC solution provides
the best delivery ratio, and reduces the routing overhead and
average delay, and save more energy in most of the simulated
cases.
In order demonstrate the overall performance differences be-
tween the protocols by accounting for all the previous measure-
ments, a combined metric is computed to further quantify the
general performance of each protocol. The combined metric
takestheproductofthenormalizedoverheadandtheaveragede-
laydividedbythethroughputforalltheTCPflowmeasurements

Page 7

7
0.01
0.1
1
10
05 10 15202530
Maximum Speed (m/s)
Normalized Overhead (log)
DSR
DSR-SPAN
DSR-PATM
DSR-PATM-RBAR
DSR-PATM-RMAC
0.5
1
1.5
2
2.5
3
3.5
4
4.5
05 1015202530
Maximum Speed (m/s)
Throughput (Mbps)
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
0510 152025 30
Maximum Speed (m/s)
Average Delay (s)
(a) Normalized Overhead (log) (b) Throughput(c) End-to-End Delay
Fig. 8. TCP Performance Under Various Speeds
0.01
0.1
1
10
5 101520253035 40
# of TCP flows
Normalized Overhead (log)
DSR
DSR-SPAN
DSR-PATM
DSR-PATM-RBAR
DSR-PATM-RMAC
0
1
2
3
4
5
6
5 1015 2025 30 3540
# of TCP flows
Throughput (Mbps)
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
5 10152025303540
# of TCP flows
Average Delay (s)
(a) Normalized Overhead (log)(b) Throughput (c) End-to-End Delay
Fig. 9. TCP Performance Under Various Loads
0.1
1
10
100
10 15 20253035 40 4550
# of CBR flows
Normalized Overhead (log)
DSR
DSR-SPAN
DSR-PATM
DSR-PATM-RBAR
DSR-PATM-RMAC
0
0.2
0.4
0.6
0.8
1
1.2
10 15 202530 3540 4550
# of CBR flows
Delivery Ratio
(a) Normalized Overhead (log) (b) Delivery Ratio
45000
50000
55000
60000
65000
10152025303540 4550
# of CBR flows
Remaining Energy
DSR
DSR-SPAN
DSR-PATM
DSR-PATM-RBAR
DSR-PATM-RMAC
0
1
2
3
4
5
6
7
8
9
101520253035404550
# of CBR flows
Average Delay (s)
(c) Remaining Energy(d) End-to-End Delay
Fig. 10. CBR Performance Under Various Loads
as shown in Fig. 11 (a) and (b). Although it seems meaningless
to simply combine several independent metrics, the combination
fairly compares the overall performance if the individual metrics
are equally important. The lower the combined metric of a pro-
tocol, the better the protocol performs. The combined metrics in
the TCP flow scenarios demonstrate that our unified framework
always performs the best among all the protocols, in both low
mobility and high mobility scenarios. However, in Fig. 11(c)
for CBR traffics, a combined metric is defined slightly different
astheproductofnormalizedoverheadandaveragedelaydivided
by the product of delivery ratio and remaining energy. We can
see that our unified topology management PATM+RMAC yields

Page 8

8
0.001
0.01
0.1
1
10
05 101520 2530
Maximum Speed (m/s)
Combined Metrics (log)
0.001
0.01
0.1
1
10
51015 20 25 303540
# of TCP flows
Combined Metrics (log)
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
10 1520253035 404550
# of CBR flows
Combined Metrics (log)
DSR
DSR-SPAN
DSR-PATM
DSR-PATM-RBAR
DSR-PATM-RMAC
(a) CombinedMetricofTCPFlows with
Mobility
(b)CombinedMetricofTCPFlowswith
Various Loads
Fig. 11. Performance under Combined Metrics in Logarithmic Scales
(c) Combined Metric of CBR Flows
the best performance metrics in most simulated cases.
Over all, in a variety of mobility scenarios and different traffic
loads, the simulation results demonstrate that the combination
of topology management and multi-rate algorithms can reduce
the routing overhead dramatically while improving the routing
performance.
VI. CONCLUSIONS
We have proposed a novel unified framework that provides
efficient topology management and high network throughput si-
multaneously. Our unified framework consists of three compo-
nents: an adaptive topology management algorithm, a multi-rate
adaptive channel access mechanism, and the seamless combina-
tion of the two components in ad hoc networks. The Priority-
based Adaptive Topology Management (PATM) provides an op-
timal presentation of the network topology to routing protocols.
Relay-based MAC (RMAC) takes advantage of the multi-rate
data transmission capabilities in ad hoc networks to employ in-
termediate nodes to forward packets at high data rates. Perfor-
mance evaluations based on analysis and simulations demon-
strate that the unified framework is promising in reduction of
the routing overhead and improvement of the network capacity
and energy savings.
REFERENCES
[1]IEEE Std 802.11. Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Specifications. Technical report, IEEE, Jul. 1997.
B. Awerbuch, D. Holmer, and H. Rubens. High throughput route selection
in multi-rate ad hoc wireless networks. In First Working Conference on
Wireless On-demand Network Systems, Trento, Italy, Jan. 2004.
L. Bao and J.J. Garcia-Luna-Aceves. Topology Management in Ad Hoc
Networks. In Proc. of the 4th ACM Interational Symposium on Mobile
Ad Hoc Networking and Computing (MOBIHOC), Annapolis, Maryland,
USA, Jun. 2003.
G. Bianchi. Performance analysis of the IEEE 802. 11 Distributed Co-
ordination Function. IEEE journal on selected areas in communications,
18(3), Mar. 2000.
B. Chen, K. Jamieson, H. Balakrishnan, and R. Morris. Span: an Energy-
Efficient Coordination Algorithm for Topology Maintenance in Ad Hoc
Wireless Networks. In Proc. 7th ACM MOBICOM, Rome, Italy, Jul. 2001.
M. Gerla and J.T.C. Tsai. Multicluster, mobile, multimedia radio network.
Wireless Networks, 1(3):255–65, 1995.
G. Holland, N. H. Vaidya, and P. Bahl. A rate-adaptive MAC protocol
for multi-hop wireless networks. In Seventh Annual ACM International
Conference on Mobile Computing and Networking (MOBICOM), Rome,
Italy, Jul. 2001.
D.B. Johnson, D.A. Maltz, and Y.C. Hu. The Dynamic Source Routing
Protocol for Mobile Ad Hoc Networks (DSR). Technical report, IETF
MANET Working Group, Jul. 2004.
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]V. Kawadia and P.R. Kumar. A cautionary perspective on cross layer de-
sign. IEEE Wireless Communication Magazine, Jul. 9 2003.
[10] L. Li, J. Halpern, V. Bahl, Y.M. Wang, and R. Wattenhofer. Analysis of a
Cone-Based Distributed Topology Control Algorithms for Wireless Multi-
hop Networks. In ACM Symposium on Principle of Distributed Computing
(PODC), Aug. 2001.
[11] S. Narayanaswamy, V. Kawadia, R. S. Sreenivas, and P. R. Kumar. Power
Control in Ad-Hoc Networks: Theory, Architecture, Algorithm and Imple-
mentation of the COMPOW Protocol. In Proc. of the European Wireless
Conference – Next Generation Wireless Networks: Technologies, Proto-
cols, Services and Applications, pages 156–162, Florence, Italy, Feb. 25-
28 2002.
[12] S.J. Park and R. Sivakumar. Poster: Adaptive topology control for wireless
ad-hoc networks. In ACM SIGMOBILE Mobile Computing and Commu-
nications Review, Poster Session: Summaries of MobiHoc 2003 posters,
volume 7(3), pages 37 – 38, Jul. 2003.
[13] C.Perkins, E.Belding-Royer, andS. Das. RFC3561 -Ad hocOn-Demand
Distance Vector (AODV) Routing. Technical report, Internet Engineering
Task Force (IETF), Jul. 2003.
[14] C.E. Perkins, editor. Ad hoc networking. Addison-Wesley, 2000.
[15] R. Prakash. Unidirectional links prove costly in Wireless Ad-Hoc Net-
works. In Proc. of the Discrete Algorithms and Methods for Mobile Com-
puting and Communications - DialM, Seattle, WA, Aug. 20 1999.
[16] VINT Project.The
http://www.isi.edu/nsnam/ns/.
[17] R. Ramanathan and R. Rosales-Hain.
Wireless Networks using Transmit Power Adjustment. In Proc. of IEEE
Conference on Computer Communications (INFOCOM), page N.A. IEEE,
Mar. 26-30 2000.
[18] B. Sadeghi, V. Kanodia, A. Sabharwal, and E. Knightly. Opportunistic
Media Access for Multi-rate Ad Hoc Networks. In Proc. of ACM MOBI-
COM, Atlanta, GA, Sep. 2002.
[19] Y. Seok, J. Park, and Y. Choi. Multi-rate Aware Routing Protocol for
Mobile Ad Hoc Networks. In IEEE VTC, Jeju, Korea, Apr. 2003.
[20] C. Shen, Z. Huang, and C. Jaikaeo. Ant-based distributed topology control
algorithms for mobile ad hoc networks. Technical report, DEGAS Group,
Dept. of CIS, University of Delaware, Apr. 2003.
[21] P. Sinha, R. Sivakumar, and V. Bharghavan. Enhancing ad hoc routing
with dynamic virtual infrastructures. In Proceedings of IEEE Conference
on Computer Communications (INFOCOM), pages 1763–72, Anchorage,
AK, USA, Apr. 22-26 2001.
[22] A.S. Tanenbaum. Computer networks. Prentice Hall PTR, 4 edition, 2003.
[23] Y. Xu and S. Bien. Topology control protocols to conserve energy in wire-
less ad hoc networks. Technical report, CENS Technical Report 0006, Jan.
2003.
[24] Y. Xu, J. Heidemann, and D. Estrin. Geography-informed energy conser-
vation for ad hoc routing. In Proc. of MobiCom 2001, pages 70–84, Rome,
Italy, Jul. 2001.
[25] Y. Yi, M. Gerla, and T.J. Kwon. Efficient flooding in ad hoc networks
using on-demand (passive) cluster formation. In Proc. of the 3rd ACM
international symposium on mobile Ad Hoc networking and Computing,
Lausanne, Switzerland, Jun. 9-11 2002.
[26] W. H. Yuen, H. Lee, and T.D. Andersen. A simple and effective cross-layer
networking system for mobile ad hoc networks. In Proc. of IEEE PIMRC,
2002.
NetworkSimulator-ns-2.
Topology Control of Multihop