Mobile Netw Appl (2007) 12:43–56
DOI 10.1007/s11036-006-0005-x
A Cross-layer Approach to Channel Assignment
in Wireless Ad Hoc Networks
Michelle X. Gong · Scott F. Midkiff · Shiwen Mao
Published online: 4 December 2006
© Springer Science + Business Media, LLC 2006
Abstract To improve the capacity of wireless ad hoc
networks by exploiting multiple available channels, we
propose a distributed channel assignment protocol that
is based on a cross-layer approach. By combining chan-
nel assignment with routing protocols, the proposed
channel assignment protocol is shown to require fewer
channels and exhibit lower communication, computa-
tion, and storage complexity than existing channel as-
signment schemes. A multi-channel MAC (MC-MAC)
protocol that works with the proposed channel assign-
ment protocol is also presented. We prove the cor-
rectness of the proposed channel assignment protocol.
In addition, through a performance study, we show
that the proposed protocol can substantially increase
throughput and reduce delay in wireless ad hoc net-
works, compared to the IEEE 802.11 MAC protocol
and an existing multi-channel scheme.
Keywords ad hoc routing protocol ·
cross-layer design · distributed channel assignment ·
multi-channel medium access control ·
wireless ad hoc networks
M. X. Gong
Cisco Systems, Inc., San Jose, CA 95134, USA
S. F. Midkiff
The Bradley Department of Electrical and Computer
Engineering, Virginia Tech, Blacksburg, VA 24061, USA
S. Mao (
B
)
Department of Electrical and Computer Engineering,
Auburn University, Auburn, AL 36849, USA
e-mail: smao@ieee.org
1 Introduction
Despite recent advances in wireless local area network
(WLAN) technologies, today’s WLANs still cannot
offer the same data rates as their wired counterparts.
The throughput problem is further aggravated in multi-
hop wireless environments due to intra-flow interfer-
ence introduced by adjacent nodes on the same path
and inter-flow interference generated by nodes from
neighboring paths. For instance, it has been shown that
the maximum capacity that the IEEE 802.11 MAC can
achieve for a chained network could be as low as one
seventh of the link bandwidth [10].
All current IEEE 802.11 physical (PHY) standards
divide the available frequency into several orthogonal
channels, which can be used simultaneously within a
neighborhood. Therefore, increasing capacity by ex-
ploiting multiple channels becomes particularly appeal-
ing. However, IEEE 802.11 WLANs that operate in ad
hoc mode rarely use multiple channels simultaneously,
partly because the IEEE 802.11 MAC is not designed
to operate with multiple channels. For instance, an ad
hoc network that is based on IEEE 802.11a technology
uses only one out of 12 available orthogonal channels,
wasting more than 90% of the potentially available
spectrum.
Consequently, there has been substantial interest in
channel assignment schemes that can achieve higher
throughput by exploiting multiple available channels [8,
11, 14, 18]. The channel assignment problem has been
showntobeNP-complete and, thus, computationally
intractable [3, 7]. In this paper, we propose an effi-
cient distributed channel assignment protocol and an
accompanying multi-channel MAC protocol. We de-
sign this channel assignment algorithm based on three
44 Mobile Netw Appl (2007) 12:43–56
design principles. First, to reduce complexity, channel
assignment and routing should be jointly considered.
Most existing multi-channel MAC protocols have two
management entities: channel assignment and medium
access control. We propose to bring the channel as-
signment entity to the routing layer. This “cross-layer”
design approach is motivated by the fact that both the
channel assignment algorithm and the ad hoc routing
algorithm will be invoked when there is a change in the
network topology. Exploiting this design principle can
greatly reduce the complexity of channel assignment al-
gorithms. In addition, the resulting multichannel MAC
protocol can be simplified since it is relieved the chan-
nel assignment burden.
Second, channels should be assigned only to active
nodes. This “on-demand” channel assignment princi-
ple is motivated by the fact that only active nodes
need communication channels. Fewer channels may
be required if this on-demand assignment principle
is implemented. Finally, both collisions and interfer-
ence should be taken into consideration. Two or more
wireless nodes may generate primary collisions if they
are one hop away from each other, while secondary
collisions can be generated by nodes that are two
hops away from each other [4]. Most existing chan-
nel assignment protocols consider only secondary col-
lisions, since they are mainly designed to solve the
“hidden terminal” problem [3, 7]. Nevertheless, pri-
mary collisions and interference are also important
factors that adversely affect channel utilization and net-
work capacity. To improve network performance, dis-
tinct channels should be assigned in such a way that
collisions and interference can be avoided as much as
possible.
We propose to combine channel assignment with
a proactive ad hoc routing protocol called Optimized
Link State Routing (OLSR) [5]. The proposed Channel
Assignment OLSR (CA-OLSR) protocol assigns dis-
tinct channels to active nodes. Further, a multi-channel
MAC protocol (MC-MAC) is incorporated with CA-
OLSR to manage multi-channel medium access. A
proactive routing protocol allows each node to have the
complete up-to-date topology information of the net-
work based on periodic exchanges of control messages.
Therefore, proactive routing protocols can be closely
coupled with channel assignment, without causing sig-
nificant modifications to the protocols themselves.
As a result of the cross-layer approach, CA-OLSR
exhibits significantly lower complexity than exist-
ing channel assignment protocols. Following the on-
demand channel assignment principle, CA-OLSR is
invoked only when there is a channel conflict, thus
further reducing control overhead. Unlike some chan-
nel assignment schemes that ignore a node’s one-hop
neighbors [6, 7], CA-OLSR assigns distinct channels
to active nodes within a k-hop neighborhood to avoid
collisions and to mitigate interference.
The rest of the paper is organized as follows. We
formulate the channel assignment problem in Section 2.
We then describe the proposed CA-OLSR protocol
and the multi-channel MAC protocol in Section 3,
and provide a correctness proof in Section 4.Aper-
formance study with ns-2 simulations is presented in
Section 5.Section6 presents related work and Section 7
concludes this paper.
2 Problem formulation
A wireless ad hoc network can be modeled as a graph
G ={V, E},whereV is the set of nodes and E is
the set of edges that represent links. We assume that
nodes use omnidirectional antennas and radio links
are bidirectional. A link is assumed to exist between
two nodes if and only if the two nodes are within
each other’s radio range. Both primary and secondary
collisions can be eliminated if nodes within a two-hop
range of each other transmit on different orthogonal
channels. The interference range is defined to be the
k-hop neighborhood of a node.
Before we formulate the distributed channel assign-
ment problem, we define V
t
⊂ V to be the set of active
transmitters and V
r
⊂ V the set of active receivers.
Note that V
t
is determined by the underlying schedul-
ing algorithm (e.g., IEEE 802.11 MAC or a multi-
channel MAC). Let v
r,i
∈ V
r
be a particular receiver
and v
t, j
∈ V
t
be a particular transmitter. Let P(v
t, j
,v
r,i
)
denote the received power at node v
r,i
,whichisa
function of the distance between nodes v
t, j
and v
r,i
,
the transmit power, and the channel condition (e.g.,
path loss). C denotes the set of all available chan-
nels in the network. Table 1 summarizes the notation
used.
If the number of available channels, i.e. |C|,issuf-
ficiently large, distributed channel assignment algo-
rithms should assign distinct channels to any nodes
within a k-hop neighborhood. However, in many cases,
|C| may be less than the number of nodes in a k-
hop neighborhood. Therefore, the objective of distrib-
uted channel assignment is to minimize the maximum
number of nodes sharing the same channel with any
designated node v
t, j
∈ V
t
among this node’s k-hop
neighbors. We can formulate the channel assignment
problem as follows.
Mobile Netw Appl (2007) 12:43–56 45
Table 1 Notation
Symbol Definition
C Set of available channels
V Set of nodes in the network
E Set of edges that represent radio links
v ∈ V A node in the network
n
k
(v) Number of k-hop neighbors sharing
the same channel with node v
V
t
⊂ V Setofactivetransmitters
V
r
⊂ V Set of receivers
v
t, j
∈ V
t
A particular transmitter
v
r,i
∈ V
r
A particular receiver
v
t,T(i)
Desired transmitter
P(v
t, j
,v
r,i
) Power level of the received signal
from transmitter v
t, j
to receiver v
r,i
S(v
t,T(i)
,v
t, j
) Relation between the channels used by
v
t,T(i)
and v
t, j
β SINR threshold for successful reception
P
N
Power level of additive white Gaussian noise
Minimize:
max n
k
(v
t, j
), ∀ v
t, j
∈ V
t
(1)
subject to:
P(v
t,T(i)
,v
r,i
)
v
t, j
∈V
t
\v
t,T(i)
P(v
t, j
,v
r,i
)·S(v
t,T(i)
,v
t, j
)
+P
N
≥β,∀v
r,i
∈V
r
(2)
where P
N
is the power level of the additive white
Gaussian noise (AWGN) noise and β is the minimum
signal-to-interference-noise-ratio (SINR) required for
a successful packet reception. The term S(v
t,T(i)
,v
t, j
)
indicates the cross-correlation between the channels
used by the desired transmitter v
t,T(i)
and transmitter
v
t, j
.
The SINR constraint implies that the cumulative
interference generated by active transmitters sharing
the same or interfering data channels as the designated
transmitter should be less than a certain threshold to
ensure that the receiver can decode the data packet suc-
cessfully. Thus, a channel assignment protocol should
distribute available channels within any k-hop neigh-
borhood in such a way that the maximum number of
transmitters sharing the same data channel is mini-
mized. Meanwhile, the same set of channels should be
re-used in such a way that the cumulative interference
generated on any particular data channel is below a
certain threshold.
3 Protocol description
Because the distributed channel assignment problem is
showntobe
NP-complete [7], it is a great challenge
to design practical channel assignment algorithms for
general ad hoc networks. To this end, we explore ef-
ficient design of heuristic algorithms that can achieve
near-optimal performance with low complexity. In this
section, we describe the proposed CA-OLSR protocol
in detail and provide a brief overview of the multi-
channel MAC protocol. CA-OLSR seeks to use as
many channel as possible within a neighborhood and
to keep channel conflicts to a minimum. Meanwhile,
the multi-channel MAC protocol guarantees that even
when two neighboring nodes choose the same channel,
contention and collisions on the channel are controlled
and resolved.
3.1 Overview of Optimized Link State Routing
(OLSR)
As the name suggests, OLSR is essentially a link state
routing protocol with an optimized flooding method
that can effectively reduce routing control overhead [5].
Specifically, OLSR minimizes the overhead of control
packet flooding by using only selected neighbors, called
multipoint relays (MPRs), to retransmit control mes-
sages. Each node selects its MPR set among its one
hop neighbors in such a way that the set covers all
the two-hop neighbors. For instance, the black nodes
in Fig. 1 are MPRs selected by the node in the center.
Once a node is selected as a multipoint relay, it not only
retransmits routing control messages, but also serves as
an intermediate node on routing paths.
In OLSR, two types of control messages are trans-
mitted periodically: HELLO message and topology con-
trol (TC) message. HELLO messages permit each node
to learn the topology of its neighbors up to two hops
away. Based on this information, each node in the
neighbor node
multipoint Relays
current node
Figure 1 Illustration of multi-point relays
46 Mobile Netw Appl (2007) 12:43–56
network independently selects its own set of MPRs
that covers all the two-hop neighbors. TC messages
are sent periodically by each node to declare its MPR
selector set that consists of the list of neighbors who
have selected the sender node as an MPR. TC messages
are forwarded by nodes in the MPR set to the entire
network. A TC message is larger than a HELLO message
and is sent out less frequently than the HELLO message.
A routing table is constructed at each node based
on the information contained in received HELLO and
TC messages. The calculation of the routing table can
follow any standard link state algorithm. An example
procedure is given in RFC 3626 to explain how the
routing table is computed [5].
3.2 The Channel Assignment and Optimized Link
State Routing (CA-OLSR) protocol
CA-OLSR combines distributed channel assignment
with the OLSR routing protocol. We assume that there
is one dedicated control channel and up to N equivalent
data channels in the network. The dedicated control
channel is shared by all the nodes in the network to
exchange routing and MAC control messages. Each
transmitter is assigned one data channel. The assigned
data channel is then used by MC-MAC for data packet
transmissions.
The basic ideas of CA-OLSR are to use routing con-
trol messages to exchange channel information and to
assign distinct channels to active transmitters within a
k-hop neighborhood. CA-OLSR identifies active nodes
within a certain time period and then give them higher
priority when there is a channel conflict. Because
HELLO messages are sent more frequently and can
carry channel information up to two hops away, we
choose HELLO messages to carry channel information.
The neighborhood size in CA-OLSR is defined to be
k = 2, based on the assumption that interference range
is about twice the radio range [11, 15].
With CA-OLSR, data channels are assigned accord-
ing to the following procedure. During initialization,
each node in the network randomly chooses a Node-
Number and a channel from a set of all available chan-
nels, denoted by
C. The random NodeNumber is used
for resolving channel conflicts. The range of NodeNum-
ber should be large enough such that the probability of
two neighboring nodes choosing the same NodeNum-
ber at the same time is extremely low. For instance,
we choose the maximum range of a four-byte unsigned
integer as the range of NodeNumber. Each time a node
updates its data channel, it should re-generate a Node-
Number. When two neighboring nodes that choose the
same data channel happen to have the same NodeNum-
Procedure sendHello()
HELLO.myChannelIndex = myChannel;
HELLO.NeighborNumber[0] = myNumber;
for i =1 to k
HELLO.NeighborChannelIndex[i] = NeighborChannel[i];
HELLO.NeighborNumber[i] = NeighborNumber[i];
endfor
Broadcast HELLO;
endprocedure
Procedure recvHello(HELLO)
(neighborChannels,NodeNumber) = getActiveNeighborInfo(HELLO);
Update the available channel set A;
channelConflict = FALSE;
if (Detect a channel conflict)
if (activeNode == TRUE)
if (activeMe == FALSE)
channelConflict = TRUE;
elseif (NodeNumber ≤ myNumber)
channelConflict = TRUE;
endif
else
if (activeMe == FALSE && NodeNumber ≤ myNumber)
channelConflict = TRUE;
endif
endif
endif
if (channelConflict == TRUE)
myChannel = randomChannel(A);
myNumber = randomNumber();
endif
endprocedure
Figure 2 Procedures sendHello() and recvHello(HELLO)
ber, both of them should update their data channels and
randomly choose another NodeNumber, as shown in
Fig. 2. A node sends out a HELLO message that contains
a list of its one-hop active neighbors, their channels,
and their NodeNumbers. Nodes detect active neighbors
by listening on the common control channel. Neighbors
that exchange RTS and CTS control messages in one
hello interval are considered active neighbors, which
are then indicated in the HELLO messages sent in the
next hello interval. Upon receiving HELLO messages
from neighboring nodes, a node builds an available
channel list, A, by marking channels that are taken by
active neighbors as unavailable.
If there is a channel conflict between the current
node and an active neighbor, the current node should
choose another channel from the available channel set
A. This is to ensure that active nodes have higher
priority to obtain distinct channels than other nodes
when the number of available channels is fewer than
the number of nodes in the two-hop neighborhood.
However, if there is a channel conflict between two
active nodes, the node with the smaller NodeNumber
retains its channel while the other node should mark
the channel-in-conflict as unavailable and randomly
pick a new channel from its updated available channel
set A. The same procedure applies when two inactive
Mobile Netw Appl (2007) 12:43–56 47
nodes have a channel conflict. In general, active nodes
have higher priority over in-active nodes. Within the
same set of conflicting active or inactive nodes, the
nodes with smaller NodeNumber have higher priority.
Because NodeNumbers are randomly generated each
time there is a channel update, CA-OLSR does not
favor one node over another within the same set of
conflicting active or inactive nodes. It is worth noting
that the use of random NodeNumbers achieves fairness
among the contending nodes, as compared with prior
approaches that resolve channel conflicts based on sta-
tic node IDs [6].
CA-OLSR has two main procedures that relate
to channel assignment. These two procedures,
sendHello() and recvHello(HELLO), are summarized
in Fig. 2. In the recvHello(HELLO) procedure, the
getActiveNeighborInfo(HELLO) function retrieves
channel information from HELLO messages received
from neighboring nodes, while the randomChannel(A)
function returns a channel index that is randomly
chosen from the available channel set
A. Note that
the channel assignment procedure in CA-OLSR is
invoked “on-demand,” only when a channel conflict
occurs after a topology change. Further, CA-OLSR
seeks to assign distinct channels to active nodes in the
two-hop neighborhood to minimize collisions as well as
to mitigate interference.
3.3 Overview of the Multi-Channel MAC protocol
Most current multi-channel MAC protocols have two
functionalities: channel assignment and medium access
control [8, 11, 18]. Because the channel assignment
is performed by CA-OLSR, our Multi-Channel MAC
(MC-MAC) protocol needs to manage only medium ac-
cess control on multiple data channels. As a result, the
design of MAC protocol is significantly simplified. A
second advantage of our approach, i.e. the separation of
channel assignment and medium access control, is that
it enables optimization of different modules separately.
For instance, channel assignment can be combined with
different reactive or proactive routing protocols. The
MAC protocol can also be designed independently
without the knowledge of how channels are assigned
to individual nodes. In addition, the separation of func-
tions makes it possible to design backward compatible
and practical multi-channel MAC protocols.
MC-MAC is a transmitter-based protocol. It is as-
sumed that all nodes in the network share the same
common control channel and each node is equipped
with two half-duplex transceivers. One transceiver at
each node listens on the common control channel all
the time, whereas the other transceiver can switch
D B RTS
S CTS
S DATA
S ACK
Node A
Node B
Nodes that have a
channel conflict with
Node A
Common control channel Node A’s data channel
NAV(RTS) NAV(DATA) D
DNAV(CTS)
Defer Access
B: backoff D: DIFS
S: SIFS
Figure 3 Four-way handshake procedure of MC-MAC
from one channel to another. Nodes are assigned data
channels by the CA-OLSR routing protocol. When a
node is ready to transmit, it first informs the destination
node of its assigned data channel. As shown in Fig. 3,
when the sender Node A intends to transmit, it first
uses the control channel to broadcast a request-to-send
(RTS) message, carrying its own data channel index,
c
A
. Upon receiving the RTS message, the destination
Node B uses the control channel to return a clear-
to-send (CTS) message carrying c
A
and switches its
receiving channel to c
A
. After Node A receives the CTS
message on the control channel, the other transceiver
switches to the confirmed data channel c
A
and starts
data transmission. Neighboring nodes that overhear the
RTS/CTS exchange but do not share the same data
channel with Node A should defer only for the dura-
tion of the control message transmission. If a node is
assigned the same data channel c
A
, two situations may
happen, as illustrated in Fig. 3.
– If the node overhears a CTS message, it should
defer from using the data channel c
A
until the end
of the data transmission to avoid causing a collision
at the receiver.
– The node that overhears only an RTS message, but
not a CTS message, should first defer from using the
control channel only for the duration of the con-
trol packet transmission. Then, it performs carrier
sensing on its own data channel c
A
. If the carrier
is busy, which means that the transmitting node
has successfully acquired the medium, the node
should defer for the duration of the data packet
transmission. However, if the carrier is not busy, the
node can start to contend for the control channel
immediately.
After the data transmission, the sender listens on
the data channel until an ACK is received or a time-
out occurs. If a node receives an RTS while its data
transceiver is busy communicating with another node,
it replies with a Negative CTS (NCTS) on the control
