1

AbstractA cognitive radio-based wireless mesh network is

considered. In addition to forwarding the data packets, each mesh
node also senses the channels of a target primary system to
identify the spectrum opportunities, and uses them for its own
data transmission. Interference temperature model is used to
define the occupancy and availability of a channel. A cooperative
algorithm based on interference temperature model is proposed
for computation of available channels by mesh nodes. Cases for
mesh nodes with fixed transmission power and adaptive
transmission power are considered separately. Finally, link and
end-to-end routing metrics are proposed to select appropriate
channels from the computed set of available channels.

Index TermsChannel selection, Cognitive radio, Interference
temperature, Opportunistic Spectrum Access, Routing metric

I. INTRODUCTION
ince spectrum is scarce resource in wireless
communication, it needs to be utilized efficiently. Most
part of the spectrum is already licensed, except for ISM
band and U-NII band, which are unlicensed bands.
Recent studies, such as [1], have shown that a
significant part of the licensed spectrum is extremely
under-utilized. Current spectrum regulations do not
allow these under-utilized parts of the licensed spectrum
to be used by unlicensed devices. On the other hand, the
unlicensed bands have become overcrowded due to
heavy proliferation of wireless services and devices in
these bands. Recognizing the potential of using under-
utilized parts of licensed spectrum bands, the regulatory
bodies across the world are reorganizing the policies for
spectrum allocation and usage. One possibility being
explored is to allow unlicensed devices (also referred as
secondary devices or underlay devices) to use those
parts of the licensed spectrum that are not currently
utilized by the licensed devices (also referred as primary
devices), provided the unlicensed devices do not
obstruct the privileges of the licensed device, and
relinquish the spectrum band (also referred to as

*
Kanwal Rekhi School of Information Technology, IIT Bombay, Powai,
Mumbai, India, email: {manuj, sahoo}@it.iitb.ac.in
#
Advanced Numerical Research and Analysis Group, DRDO, Hyderabad,
India
frequency band or channel) as soon as the licensed
device intend to use it. In order to realize such sharing of
licensed spectrum, two approaches are being
investigated. One, the Opportunistic Spectrum Access
(OSA), in which an unlicensed device continuously
senses the licensed spectrum bands and
opportunistically use those bands for transmissions that
are currently not used by licensed spectrum (spectrum
white spaces or holes). Second, the Negotiated
Spectrum Sharing or Secondary Market approach, in
which unlicensed users negotiate with the licensed
users, either off-line or on-line, for the use of under-
utilized licensed bands. The licensed users usually lease
the under-utilized bands to the secondary users for a
fixed duration and cost. In Section II, we elaborate on
these two approaches for spectrum sharing between the
licensed and unlicensed users.
One of the criteria, proposed by the Interference
Protection Working Group of Spectrum Policy Task
Force (set up by FCC) [2], to opportunistically share the
licensed spectrum bands and to quantify and manage the
interference is interference temperature. Interference
temperature is a measure of the RF power available at a
receiving antenna to be delivered to a receiver power
that is generated by other emitters and noise sources.
FCC in its Notice for Proposed Rule Making (NPRM)
[3] has suggested that unlicensed devices can be allowed
to use licensed spectrum bands in a geographical region
provided the interference temperature at each licensed
receiver in the region does not exceed an interference
temperature threshold. The regulatory bodies need to
define these thresholds for all the frequencies in various
geographical regions. Under this model, which we refer
as Interference Temperature Model, an unlicensed
device can transmit on any of those licensed channels
for which its transmission does not increase the
interference temperature beyond the interference
temperature threshold at any licensed receiver within its
interference range.
In this paper, we describe algorithms for selecting the
most appropriate channel for transmission by a multi-
interface node in a cognitive mesh network. We consider
Channel Selection under Interference Temperature Model in
Multi-hop Cognitive Mesh Networks
Manuj Sharma
*
, Anirudha Sahoo
*
, and K. D. Nayak
#

S
2
the Interference Temperature Model in which a channel
is assumed to be available for transmission by a mesh
node if the nodes transmission does not increase the
interference temperature within its interference range
beyond a predefined threshold. We consider two cases:
one, in which each mesh node transmits with a fixed
power, and second, in which each mesh node employs
an adaptive transmission power control. Each mesh node
computes a set of channels available for transmission
(without increasing the interference temperature beyond
the threshold in its interference range). It then uses a
per-hop link cost metric and end-to-end routing metric
(described in section V) to select channels from the
available channels set for each hop on the end-to-end
path. We introduce the channel stability factor and the
consumed transmission power in per-hop link cost
calculation for each channel.
The rest of the paper is organized as follows. Section
II describes three approaches to spectrum sharing by
licensed and unlicensed devices: opportunistic spectrum
sharing, negotiated spectrum sharing, and static license
allocation of spectrum. Section III describes the
Interference Temperature Model, including its
formalization and possible options for measuring it. In
section IV, we describe the architecture of the cognitive
mesh network that we have considered in this paper. We
also describe a 2-state Markov chain model for
formalizing the notion of mean availability time of a
channel. Section V describes the notations and
algorithms for computing the available channel set for
both the fixed and adaptive transmission power cases. In
section VI, we propose a per-hop link metric to select
channel for a given hop. This metric provides the cost of
using a channel on that link (hop), taking into account
the channel stability factor. We also describe an end-to-
end routing metric that is used to select the channels on
end-to-end path hops. The channels are selected from
the set of available channels that are computed in
section V. Section VII briefly highlights the possible
future work.

II. SPECTRUM SHARING BY LICENCES AND
UNLICENSED DEVICES
A. Opportunistic Spectrum Sharing (OSA)
Opportunistic Spectrum Sharing approach enables an
unlicensed device to continuously monitor the spectrum
usage to identify the unused spectrum bands or
spectrum holes, and use these bands opportunistically
for its own data transmission. These spectrum holes
are also referred as spectrum opportunities. This
opportunistic use of unused spectrum bands by
unlicensed devices is completely transparent to licensed
devices. The unlicensed devices must ensure that they
vacate the occupied band within a very short interval if
it interferes with the operation of licensed devices
beyond an acceptable limit. The mechanism to
implement the OSA must ensure that the privileges of
licensed user are not compromised because of the
opportunistic use of the spectrum by unlicensed devices.
In this paper, we assume that a channel is the basic unit
of a spectrum band that can be used for communication
of data. Broadly, we can model channel availability
from a secondary device perspective in two ways: first, a
binary model, in which the secondary device considers a
channel to be occupied or available depending on
whether it does or does not detect the presence of any
primary device signal on that channel; and second,
interference temperature model, in which the secondary
device considers a channel to be unavailable for
transmission if its transmission on that channel would
result in increase of interference temperature beyond the
predefined threshold within its interference range
1
. In
the binary model, a secondary transmitter must
relinquish an occupied channel as soon as it detects the
presence of a primary device signal on that channel. In
case of interference temperature model, a secondary
transmitter must relinquish an occupied channel as soon
as it discovers that the interference temperature for the
channel has exceeded the threshold at any of the primary
receivers within its interference range. Typically in both
the models, a secondary transmitter must sense the
channels and detect the presence of primary signals.
Techniques such as cyclostationary signal processing
and matched filters are used by secondary devices for
primary signal detection if the modulation and
waveform details of primary signals are known a priori.
In absence of such details, conventional radiometric
detectors (or energy detectors) are used to compare the
aggregate signal energy against a predefined threshold.
If the aggregate energy exceeds the threshold, the
detector concludes the presence of primary device.
Please refer to [4] for a comparative study of spectrum
sensing techniques.
The protection of privileges of primary users is of
paramount importance in OSA approach. Since the
opportunistic spectrum access by secondary devices is
completely transparent to primary users, the primary

1
Strictly speaking, at any primary receiver within the interference range of
the secondary transmitter.
3
users experience a short but measurable interference as
soon as they start transmission on a channel that is
currently used by a secondary device for its data
transmission. This assumes that a primary device starts
transmitting on the channel without first checking
whether the channel is occupied by any secondary
transmitter. An alternative approach [5] is that a primary
user checks a channel before using it. If the channel is
occupied by a secondary transmitter, it searches for
another free channel in the spectrum pool. In case it
finds no free channel, it starts using an occupied
channel, which then must be vacated by the secondary
transmitter using it within a short time interval). This
interference to the primary transmitter persists as long as
the secondary transmitter detects the appearance of the
primary transmitter and relinquishes the occupied
channel. Therefore, not only that a secondary transmitter
should quickly detect the primary signals, it must also
immediately take corrective action. This action can be
the channel reversion and switching to a new channel, or
reducing the transmission power sufficiently so as to
reduce the interference temperature, or to completely
stop transmission temporarily if there is no other
appropriate free channel is available. The larger the
detection and correction times, the larger the
interference experienced by the primary receivers.
The opportunistic usage of licensed spectrum by
secondary devices face several legal issues that will
need to be resolved before wide-spread deployment of
such approaches. The legal issue arises because
licensees have paid enormous license fees for acquiring
the spectrum and are legally protected against any
interference in their transmissions due to unlicensed
operations.
B. Negotiated Spectrum Access (Secondary Market
approach)
Negotiated Spectrum Access approach enables a
secondary device (or a secondary network as a whole) to
negotiate, either off-line or on-line, the use of under-
utilized spectrum bands with the licensee and takes it on
lease for fixed duration and for a price [6]. The terms
and conditions, including tariffs, are voluntarily agreed-
upon between the licensee and lessee. The decision to
lease the unused/under-used channels of licensed
explicitly lies with the licensee. Since the licensed
devices are completely aware of the channels being used
by secondary devices, their transmissions usually do not
interfere with the transmissions of the secondary
devices. This approach is likely to be more acceptable to
the licensees as, unlike in OSA approach, they have
financial incentive in leasing their unused part of
spectrum [7]. Similarly, the secondary users (lessees)
temporarily get part of spectrum for their data
transmission without making huge investments in
acquiring licenses. Recognizing the potential benefits of
Secondary Market approach both for licensee and lessee,
FCC has also issued a NPRM exploring the use of
secondary markets [8].
C. Static license Allocation of Spectrum
This is the traditional approach to spectrum allocation
in which the licenses are allocated to the vendors usually
through a bidding process. Some of the bands are
reserved for specific purposes such as public safety and
for government operations, whereas some bands are
unlicensed (such as ISM and U-NII bands).

III. INTERFERENCE TEMPERATURE MODEL
A. Definition and Formalization
Interference generally limits the useable range and
effectiveness of communication signals. It also affects
the efficiency of spectrum use. Due to substantial
increase in radio communication services, and fixed and
mobile RF devices in recent times, it is expected that the
current approaches to interference management will no
longer be adequate. In order to meet the challenges of
future, the Interference Protection Working Group of
FCC Spectrum Policy Task Force recommended the use
of Interference Temperature metric as a means of
quantifying and managing interference. Interference
temperature is defined as a measure of the RF power
available at a receiving antenna to be delivered to a
receiver. This is the power generated by other emitters
and noise sources. More specifically, it is the
temperature equivalent of the RF power available at a
receiving antenna per unit of bandwidth, measured in
units of Kelvin [2]. There are several equivalent
definitions of interference temperature [2, 9], but we use
the following formulation in this paper:

( )
( )
c
c c
c c c
kB
B f Pi
B f T
,
, = (1)
where T
c
(f
c
, B
c
) is the interference temperature for
channel c, with central frequency f
c
and bandwidth B
c
.
Pi(f
c
, B
c
) is the average interference power in Watts (at
the antenna of a receiving or measuring device) centered
at frequency f
c
and covering the bandwidth B
c
(in
Hertz). k is Boltzmanns constant (1.38 10
-23
Joules
per Kelvin).
4
In its effort to use the spectrum more intensively and
efficiently, the FCC proposed an interference
temperature model for opportunistic use of licensed
spectrum bands by unlicensed devices [3]. In this model,
an unlicensed device can opportunistically transmit the
data on a channel provided the interference temperature
for that channel does not exceed a predefined threshold
at any primary receiver within the transmitters
interference range. It is proposed that the FCC (and
similarly, other regulatory bodies in different countries)
would define the interference temperature threshold for
all the frequency bands within a given region.
In interference temperature model, there are two
possible interpretations of what constitute the
interference at a primary receiver [10]. According to one
interpretation, the combined signals due to a primary
transmitter, the other secondary transmitters, and noise
are jointly considered to calculate the interference
temperature at a primary receiver. This interference
temperature due to aggregate signal strength received at
a primary receiver is compared against the threshold
value. Primary device signals are not detected
separately. In other interpretation, only the noise and
signals due to other secondary transmitters are
considered as interference, and they must be separated
from the received primary signals to calculate the
interference temperature. In this case, primary device
signals must be explicitly detected in the total signals
received at a primary receiver.
B. Interference Temperature Measurement Options
The FCC NPRM [3] suggests three possible methods
of measuring interference temperature at different
locations in a region. In this subsection, we briefly
summarize the three methods.
1) Measurement by Primary Receiver
Since, ideally the interference temperature needs to be
measured and controlled at primary receivers, the most
appropriate approach is that the primary receivers
themselves measure the interference temperature at their
antenna, and send the values back to the unlicensed
secondary devices in the region. Though this approach is
most accurate (as primary receivers know the exact
modulation type and waveform details of transmitted
primary signals), it requires major hardware and
software modifications in the primary receivers. This is
clearly infeasible, especially for devices (such as TVs,
Laptops, Mobile phones, PDAs, etc.) which are already
developed and deployed. Moreover, it requires a channel
for explicitly transmitting the measured values to the
secondary transmitters.
2) Grid of Monitoring Stations
Another approach to measure interference temperature
is to deploy a grid of monitoring stations in the target
region. These devices are dedicated for measuring the
interference temperature at the location of their
deployment, and send these measurements back usually
to a well-known sink node in the grid (from which the
secondary devices can obtain the interference
temperature in their nearby region). A major advantage
of this approach is that it does not require any
modification in primary system. Moreover, these
devices, being dedicated for interference temperature
measurement, can be fine tuned for high precision, and
are usually not power-starved. This is in contrast to
small devices (either secondary or primary), such as
mobile phones and PDAs, where incorporating
measurement capabilities are costly in terms of silicon
real-estate and battery power consumption.
On the other hand, this approach has several
disadvantages too. First, the interference temperature is
measured at locations (i.e. at grid nodes), which are
different from where it need to be controlled (i.e. at
primary receivers). The interference temperature at these
different locations may be different due to differing
terrain and path loss conditions. This approach also
suffers from hidden terminal problem. Grid nodes
usually cannot exactly know the path loss between
secondary transmitters and primary receivers, as well as
the shadowing and fading effects experienced at primary
receivers.
3) Measurement by Secondary Transmitters
The simplest but somewhat inaccurate method to
measure the interference temperature is to let a
secondary device itself measure the interference
temperature locally. This approach neither requires any
modification in primary system, nor any additional
deployment of measuring services. But the interference
temperature that the secondary device measures locally
and the one that is present at a primary receiver may
differ significantly (unless both the devices are very near
to each other), due to location and terrain-dependent
multipath interference and shadowing affects. Moreover,
secondary devices need to be equipped with interference
temperature measurement capabilities.

IV. COGNITIVE MESH NETWORKS
A. Architecture
A cognitive radio is a smart radio which has the
ability to sense the external environment, learn from the
5
history, and make intelligent decisions to adjust its
transmission parameters according to the current state of
environment [11]. In this paper, we consider a static,
multi-hop wireless mesh network in which each node is
equipped with multiple cognitive radios. The mesh
consists of several types of nodes, including pure
wireless routers, which only relay incoming data packets
from a downstream mesh node to the appropriate next
hop upstream mesh node. The other type of nodes is
called Traffic Access Points (TAPs) which interact with
the end-user devices. The third category is of that of
gateway nodes that connect the mesh network to a wired
network. The mesh nodes are assumed to have location
determination capabilities (using any location
determination technique or algorithm) so that they can
determine their own location. They sense the target
license bands and opportunistically use the available
channels to relay data packets over multiple hops. We
assume that the mesh nodes are homogeneous with
respect to the number of interfaces they support, and
their processing capabilities. The channels in the target
primary system are also homogeneous with respect to
their transmission ranges and peak bandwidth
2
. The
available channel set, however, dynamically changes
depending on the traffic imposed by the primary users.
We assume that the mesh nodes know the exact
waveform details and modulation type of primary
signals (i.e. the mesh network knows the primary system
a priori) and are able to detect the presence of primary
signals using techniques such as cyclostationary signal
processing and matched filter detection. In other words,
the mesh nodes are able to separate out the primary
signals from other interfering signals (due to other mesh
nodes and noise).
In order to measure the interference temperature
within a region, we suggest a collaborative approach
among the mesh nodes. We adopt a combination of
methods 2 and 3 described in Section III-B. Since a
mesh network is itself a grid of stations, we propose to
incorporate the interference temperature measurement
capabilities in each mesh node (which is also a
secondary device). Each mesh node measures the
interference temperature for each channel locally, and
exchanges it with the other nodes within its interference
range. Based on the collected statistics, it derives the
interference temperature for each channel at various
locations in that area. This collaborative approach to
determine the interference temperature forms the basis

2
The channels may differ if they come from widely spaced regions of
spectrum. But since we assume that the channel pool belongs to one specific
primary system, these channels can be taken as homogeneous.
of available channel set computation, as described in
Section V.
B. Channel Modeling
We consider interference temperature model to model
the channel availability. A channel c with central
frequency f
c
and bandwidth B
c
is said to be available for
transmission at a mesh node m if the transmission due to
m does not increase the interference temperature at any
other mesh node n (within the interference range of m)
beyond a predefined threshold. Following the notations
of [10], we formally represent the above constraint as

( )
( )
c
c
c c m
c
mn
c c c
TH
kB
B f Pt L
B f T < +
,
, (2)
Here, T
c
(f
c
, B
c
) is the interference temperature for
channel c with central frequency f
c
and bandwidth B
c
,
c
mn
L is the path loss in transmission from node m to n on
channel c, Pt
m
(f
c
, B
c
) is the transmission power of m,
TH
c
is the interference temperature threshold for channel
c, and k is the Boltzmanns constant. In this paper, we
consider two variants of transmission power of mesh
nodes. In the first case, each mesh node transmits with a
fixed transmission power; in the second case, a mesh
node adapts its transmission power according to the
noise and interference conditions at the receiver. The
adaptive transmission power control leads to higher
network throughput and reduced battery consumption.
Using constraint (2), as it is, may lead to the following
problem. In Figure 1, consider a mesh node x which is at
the boundary of interference range of two other mesh
nodes, say A and B. Both A and B independently
obtains the interference temperature value at node x for
channel c (which is lower than the predefined threshold)
and start transmitting on it. As a result of their
simultaneous transmission, the cumulative power
received at node x makes the interference temperature at
x overshoot the threshold value.



Fig. 1. Mesh node x is at the boundary of interference ranges of nodes A and
B. Dark circle represents the transmission range and dashed nodes represents
the interference range of a node.
x
A
B
6
We propose two possible solutions to this problem.
First, the nodes should coordinate in a distributed
manner to decide which one of the several potential
secondary transmitters should use the channel. Such
coordination will ensure that one and only one mesh
node will transmit on a channel at a time such that the
interference temperature at any other mesh node within
its interference range does not exceed the threshold. The
second approach is to conservatively scale down the
predefined threshold value by a fraction, so as to start
corrective action before the temperature exceeds the
actual threshold. In this paper we pursue the second
approach, deferring the first one for future work.
Following the second approach, we rewrite constraint
(2) as follows:

( )
( )
c
c
c c m
c
mn
c c c
TH
kB
B f Pt L
B f T < +
,
, (3)
Where is the scaling-down factor ( ). 1 0 < < If a mesh
node calculates that its transmission on a channel c
would not violate constraint (3), then the node can use
the channel for data transmission. On the other hand, if a
mesh node reports that interference temperature
measured by it for a channel c exceeds TH
c,
the other
nodes in the region using the channel performs channel
reversion and switch to some other appropriate channel.
Readers should note that a channel satisfying the
above constraint, though categorized as available
channel, may not be the most appropriate channel for
transmission of data of a target application. In section
VI, we formulate a routing metric to select the most
appropriate channel out of the set of available channels.
The metric incorporates a number of parameters of
available channels to select the most appropriate one.
We use a simple 2-state Markov Chain (shown in
Figure 2) to model a channels availability and
unavailability, based on its interference temperature.
States A and B represents a channel being available and
unavailable respectively at a mesh node. The conditions
for channel availability and unavailability are already
described above in this subsection. Symbol
c
represents
the probability with which the interference temperature
for a channel c overshoots the predefined threshold
(thereby making the channel c transit from the state of
availability to that of unavailability). Symbol
c
represents the probability with which the interference
temperature drops down from higher to lower values
across the threshold (thereby making the channel c
transit from the state of unavailability to that of
availability). The sojourn time of each state is
geometrically distributed. We are particularly interested
in the sojourn time of state A, which denotes the average
time for which channel c remains available. Let us
denote this time by
A
c
. Being geometrically distributed,
the mean availability time for channel c is given as
A
c
= 1 / ) 1 (
c

(4)
We propose to use
A
c
as one of the parameters in
routing metric to select the most appropriate channel
from the set of all the available channels. The challenge
in the above model lies in calculating the transition rates
(
c
and
c
) of interference temperature value (for
channel c) across the threshold.

V. AVAILABLE CHANNEL SET COMPUTATION
ALGORITHM
A. Notations
N: Set of all the mesh nodes, {1..N}.
C: Set of all the channels (frequency bands) in the
spectrum pool, {1..C}.
m: A mesh node, which is under consideration for
computing the available channel set for sending its
data.
TR
m
: Set of mesh nodes which are within the
transmission range of node m.
r: Receiver mesh node (within the transmission range
of node m) that has to receive data from sender node
m. (rTR
m
).
IR
m
: Set of mesh nodes which are within the interference
range of node m. Note that, jTR
m
jIR
m
.
n: A mesh node within the interference range of node
m, i.e. nIR
m
.
SIR_Th
j
: The threshold for signal-to-interference ratio at
a node j (where j N) for which node j can
successfully receive a packet j. It is assumed that
SIR_Th
j
is constant for all the channels c C .
LT
m
: Local interference temperature vector measured by
node m, c C .
LT
r
: Local interference temperature vector measured by
node r, c C .
LT
n
: Local interference temperature vector measured by
A B

c
1 -
c

c
1 -
c
A B

c
1 -
c

c
1 -
c

Fig. 2. A 2-state Markov chain to model a channel.
7
node nIR
m
, c C .
LT
m
[c], LT
r
[c], and LT
n
[c]: Local interference
temperature measured by node m, r, and n,
respectively for channel c C .
V
j
: Tuple containing a nodes identifier (j), the local
interference temperature vectored measured by node
j, and SIR_Th
j
;

that is, (n, LT
j
, SIR_Th
j
).
T
n
m
: Vector of aggregate interference temperatures for
all channels experienced at node n due to
transmissions by node m; it is calculated by node m
by adding its own potential contribution to the
existing interference temperature at node n (for all
channels).
T
n
m
[c]: Aggregate interference temperature for channel
c experienced at node n due to transmissions by
node m.
T
m
m
[c]: Aggregate interference temperature for channel
c experienced in the neighborhood of m due to its
transmissions; it is calculated by node m by adding
its own potential contribution to the existing
interference temperature that it has measured for
channel c.
TH
c
: Predefined interference temperature threshold for
channel c.
: Scale-down factor for interference temperature
threshold.
L
c
mn
: Path loss in transmission from node m to n on
channel c.
f
c
: Central frequency of channel c.
B
c
: Bandwidth of channel c.
Pt
j
(f
c
, B
c
): Transmission power of a node j over channel
c, (where j N).
Pr
j
(f
c
, B
c
): Received signal power at a node j over
channel c, (where j N).
Pi
j
(f
c
, B
c
): Aggregate interference power at a node j
over channel c, (where j N).
k: Boltzmann constant (1.38 10
-23
Joules per
Kelvin).
PC
m
: Vector of probable channels
AC
m
: Set of channels available at node m for
transmission without increasing interference
temperature beyond defined threshold at any node
n IR
m.

B. Initialization at each mesh node
1. Compute location coordinates (x
i
, y
i
, z
i
) iN
using any location determination technique.
2. For each node iN, exchange (x
i
, y
i
, z
i
) with all
the other nodes j (N IR
i
).
3. For each node iN, compute its distance from
every other node in its interference range:
d
ij
= ((x
i
x
j
)
2
+ (y
i
y
j
)
2
+ (z
i
z
j
)
2
)
1/2
,
j(N IR
i
).
4. For each node iN, calculate L
ij
c
using d
ij
,
j(N IR
i
) and cC.
After initialization, node m under consideration will
have L
c
mn
for all n IR
m
and all cC.
C. Computation of Available Channel Set: The Fixed
Transmission Power Case
In our proposed algorithm, the mesh nodes collaborate
to derive the set of available channels at each node.
Following are the main steps of the algorithm, assuming
that each mesh node transmits with a fixed transmission
power.

1) At each mesh node n IR
m
:
1.1 Compute LT
n
.
1.2 Transmit tuple V
n
= (n, LT
n
) to all nodes
within its interference range.
2) At mesh node m
2.1 Receive V
n
from each node n IR
m
.
2.2 For each c C :
2.2.1 T
m
m
[c] = {Pt
m
(f
c
, B
c
) / (k B
c
) } +
LT
m
[c]
2.2.2 If (T
m
m
[c] . TH
c
) then
PC
m
= PC
m
{c}
2.3 For each mesh node n IR
m
:
2.3.1. For each c PC
m
:
2.3.1.1 T
n
m
[c] = {(L
c
mn
Pt
m
(f
c
, B
c
))
/ (k B
c
)} + LT
n
[c]
2.3.1.2 If (T
n
m
[c] . TH
c
) then
AC
m
= AC
m
{c}

Every node n within the interference range of node m
measures locally the interference temperature for all the
channels and periodically sends the measured
information to all the nodes within its interference range
(statements 1.2 and 1.2 in above algorithm). This
dissemination of information can be done on a dedicated
802.11 channel of unlicensed band. For the nodes that
are outside the communication range but within the
interference range of the sending node, multihop relay
communication, or increased transmission power need to
be used. For example, consider node m in Figure 3,
8


Fig. 3. Nodes a, b, and r are within the transmission range of node m (thick
solid circle centred at m), whereas nodes d, e, and f are outside the
transmission range but within the interference range (dashed circle centred at
m) of node m. Concentric thin solid circle and dotted circle centred at r
denotes, respectively, the transmission range and interference range of node r
to illustrate the bi-symmetrical nature of the channel.

which computes the available channels set. The
transmission and interference ranges of m is shown as
concentric thick solid circle, and a dashed circle centred
at m. Nodes a, b, and r are within the transmission range
of m. The figure also shows (as thin solid circle and
dotted circle centred at r) the transmission and
interference range of node r for illustrating the bi-
symmetrical channel. Nodes d, e, and f are outside the
transmission range but within the interference range of
m. Each node (m, a, b, r, d, e, and f) measures the
interference temperature locally and disseminate this
value to all the nodes within its interference range.
Assuming bi-symmetrical channel, node m receives the
interference temperature vector for all channels from all
other nodes within its interference range (nodes a, b, r,
d, e, and f), and vice-versa. On receiving the interference
temperature vector (for all the channels) from these
nodes, node m first measures the interference
temperature locally for all the channels and calculates
the aggregate interference temperature (for all channels)
that would be experienced in its neighborhood due to its
transmissions (statement 2.2.1). It then discards those
channels for which interference temperature exceeds the
scaled-down threshold value, and keeps the remaining
channels in the set of probable channels, PC
m
(statement
2.2.2.). These channels are probable, because though
they do not increase the interference temperature beyond
the threshold in the neighborhood of node m, they may
increase the temperature beyond the threshold at other
mesh nodes which are within the interference range of
node m. In statement 2.3.1.1, node m computes the
aggregate interference temperature experienced by all
nodes n within its interference range due its
transmission on all the probable channels. Out of all the
probable channels, it then selects those channels as
available channels (AC
m
) for which interference
temperature would not exceed the threshold value at any
mesh node within its interference range when it (node
m) transmits on those channels (statement 2.3.1.2).
Therefore, available channel set at a node m is
essentially the set of those channels on which
transmission by m will not increase the aggregate
interference temperature beyond a predefined threshold.
D. Computation of Available Channel Set: The
Adaptive Transmission Power Case
In this subsection, we relax the fixed transmission
power assumption, and allow the mesh nodes to adapt
their transmission power according to the interference
conditions at the receiver. Following are the main steps
of the algorithm, assuming adaptive transmission power
control for each mesh node.

1) At each mesh node n IR
m
:
1.1 Compute LT
n
.
1.2 Transmit tuple V
n
= (n, LT
n
, SIR_Th
n
) to all
nodes within its interference range.
2) At mesh node m
2.1 Receive V
n
from each node nIR
m
, including
from the intended receiver node r.
2.2 For each c C :
2.2.1 Calculate the interference power level
experienced by receiver node r:
Pi
r
[c] = k B
c
LT
r
[c]
2.2.2 Calculate the minimum signal power
required (under given interference
temperature) at receiver r to receive
the packet correctly:
Pr
r
[c] = k B
c
LT
r
[c] SIR_Th
r

2.2.3 Node m must therefore transmit with
the following power:
Pt
m
[c] = Pr
r
[c] / L
c
mr

2.2.4 T
m
m
[c] = {Pt
m
[c] / (k B
c
) } + LT
m
[c]
2.2.5 If (T
m
m
[c] . TH
c
) then
PC
m
= PC
m
{c}
2.3 For each mesh node n IR
m
:
2.3.1. For each c PC
m
:
2.3.1.1 T
n
m
[c] = {(L
c
mn
Pt
m
[c]) /
m r
a
d
b
e
f
9
(k B
c
)} + LT
n
[c]
2.3.1.2 If (T
n
m
[c] . TH
c
) then
AC
m
= AC
m
{c}

In most of the adaptive transmission power control
algorithms, the basic idea is that the transmitters obtain
the interference condition and the Signal-to-Interference
threshold value at the intended receiver. With these
values in hand, the transmitter then computes the
appropriate transmission power so that the signal
received at the intended receiver exceed the SIR
threshold under the given interference conditions [12].
Consider Figure 4 as an illustration of intuitive benefits
of adaptive transmission power control in a scenario in
which we assume that node m is the sender node under



Fig. 4. Node m is sender and node r is the intended receiver (both shown as
black dots). Node m adaptively controls its transmission power based on the
interference conditions at the intended receiver r. The transmission power is
just sufficient to successfully transmit to node r. This adaptive power control
results in the reduced transmission and interference ranges of sender node m
(show as concentric thick solid and dashed circles centred at m), as compared
to fixed transmission case (see Figure 3). The ranges of the other nodes are
shown same as in Figure 3. The nodes from which m now needs to receive
interference temperature vector and SIR threshold values reduces to four (as
against six in Figure 3).

consideration, whereas node r is the intended receiver.
The sender-receiver pair remains within the transmission
range of each other and is shown as black dots in the
figure. Depictions of transmission and interference
ranges in Figure 4 remain same as in Figure 3 except
that sender node m now employs adaptive transmission
power control to transmit with a power that is just
sufficient to communicate to the intended receiver node
r. This results in the reduced transmission and
interference range. (We have assumed the interference
range to be approximately equal to twice of the
transmission range for pictorial depiction in the figure).
Nodes a and r remains in the new transmission range,
but node b now goes out of transmission range but
remain within the interference range of m. Node d
remains within the interference range of m, but nodes e
and f goes completely out of it. Each node (m, r, a, b, d,
e, and f) measures the interference temperature for all
the channels and periodically sends this vector (along
with its SIR threshold value) to every other node within
its interference range. Contrary to the fixed transmission
power case (Figure 3), node m now needs to receive and
process the interference temperature vector and SIR
threshold value only from four other node (nodes a, b, r,
and d), which are within its new interference range
(statement 2.1 of above algorithm). Using the received
interference temperature values, the sender node
calculates the interference signal strength at the receiver
for each channel. This calculation follows from the
definition of interference temperature (see equation (1)),
according to which, the interference temperature at r for
channel c is given as:
c
kB
[c]
r
Pi
[c]
r
LT =
Using above equation, the interference signal strength
can be calculated as (statement 2.2.1),
[c] LT kB [c]
r
Pi
r c
=
Once node m obtains the interference signal strength at
receiver r for channel c, it then computes the minimum
signal strength with which it must transmit so that the
signal strength received at r exceed its (node rs) SIR
threshold. Node m uses SIR threshold of node r
(SIR_Th
r
) to compute this value as follows (statement
2.2.2). By definition,
r
Pi
[c]
r
SIR_Th
r
Pr
=
r r r
Th SIR c Pi c _ ] [ ] [ Pr =
r r c r
Th [c]SIR LT kB c _ ] [ Pr =
The above expression for Pr
r
[c] gives the minimum
signal strength required at receiver r for correct
reception of packet at channel c, given that the
interference temperature at r for channel c is LT
r
[c].
Taking into account the path loss (L
c
mr
) for signal
between node m and r over channel c, the sender node
m, therefore, must transmit with a signal power (Pt
m
[c]),
which is equal to (Pr
r
[c] / L
c
mr
) (statement 2.2.3). Once
the transmission power has been calculated by node m,
remaining steps are similar to those of the fixed
m
d
r
a
f
e
b
10
transmission power case (Section V-D).
E. Updation of Interference Temperature
The algorithm mentioned in the above subsection is
executed periodically to compute the interference
temperature and the available channel set. Let us denote
this period by t
p
. To highlight the need for additionally
updating any changes in interference temperature, we
once again consider Figure 1. We assume that node x
sends its local interference temperature vector (LT
x
) at
time instant t
1
that is received by both nodes A and B.
Node A selects channel c for its transmission and starts
sending data over channel c, as a result of which the
interference temperature for channel c increases at node
x. But the next periodic update of the local interference
temperature vector by node x will be sent only at time
instant t
2
(= t
1
+ t
p
). Assume that at time t, where t
1
< t <
t
2
, node B decides to send a frame. Based on the
interference temperature value for channel c (LT
x
[c]),
sent by node x at time instant t
1
, node B also selects
channel c for its data transmission. As a result of this
transmission by node B, the cumulative interference
temperature at node x due to transmission by node A and
B may exceed the threshold. To prevent this unexpected
overshooting of interference temperature, node x must
immediately disseminate any changes in its locally
observed interference temperature to all the nodes that
are within its interference range. These are essentially
the steps 1.1 and 1.2 of the algorithm described in the
previous section. In general, in addition to periodically
disseminating its local interference temperature vector
to all the nodes within its interference range, a mesh
node must also disseminate information as soon as it
detects any changes in its locally measured interference
temperature within its periodic updation period. This
solution, in conjunction with the scaled-down threshold
method described in section IV-B, ensures that any two
nodes (such as A and B in the above example) do not
generally select the same channel for transmission that
will lead to overshooting of interference temperature
over threshold. Even if they select the same channel
simultaneously, the scaled-down threshold will try to
mitigate the chances of increasing the interference
temperature threshold.

VI. LINK AND ROUTING METRICS
A. Per-hop Link Metric for Channel Selection
Once a node computes the set of all the available
channels, it must then select one appropriate channel for
transmission to its next hop neighbor. First prerequisite
of channel selection for a link between two neighboring
nodes is that both the nodes must have a non-empty
intersection of their available channel sets, i.e. for
neighboring nodes m and n to communicate, AC
m

AC
n
. Assuming that node m needs to transmit
data to node n, and the above constraint holds for both
the nodes, node m must then select a channel out of the
common channel set for the link between itself and node
n. The channel selection for a link should be based on
following criteria: (i) The selected channel should take
minimum possible amount of time to transmit a link-
layer frame from node m to n; (ii) it should have
minimum possible switching cost, (iii) it should be
stable in its availability (i.e. its transition rate from
available to unavailable state should be as low as
possible), and (iv) it should lead to minimum
transmission power (in adaptive transmission control
case). For first two criteria, we adopt the parameters
proposed in [13] and [14]. For the third criteria, we
formulate a channel stability factor, and use it
conjunction with the remaining two parameters. The
transmission powers of all the available channels are
calculated as a part of available channel computation
algorithm (section V-D). Finally, we formulate the link
metric for channel selection as a weighted combination
of the above parameters.
There has been significant amount of work in recent
years in designing link and routing metrics for 802.11-
based multi-channel, multi-hop wireless mesh networks.
One such metric, called ETT, is based on the expected
frame transmission time over a link between two nodes
[14]. This metric captures the first criteria for channel
selection, but it does not take into account the channel
switching cost in a multi-channel network. The ETT
value in [14] is calculated as:
) / ( * B S ETX ETT = (5)
where ETX is the expected number of transmission
(including retransmissions) for successfully transmitting
a frame across the link, S is the frame size, and B is the
peak bandwidth of the channel (used for that link). This
limitation was rectified in MCR metric [13], which not
only consider ETT but also considers the channel
switching cost (SC) while selecting a channel. For
details of calculating the channel switching cost, please
refer to [13]. In addition to these parameters, we propose
two additional parameters while considering a channel
to select: the channel stability factor,
c
SF and the
transmission power (Pt
m
[c]) required for successful
transmission over the channel (only for adaptive power
11
controlled nodes). We define the channel stability factor
for a channel c as the average amount of time for which
the channel remain available for transmission (i.e. the
average amount of time for which its interference
temperature remained below the scaled-down threshold
( .TH
c
)). The average channel availability time,
'
c
,
can be calculated as weighted moving average of the
previously measured availability times, and the time
measured in the current measurement duration (
c
). In
analytical model of a channel, as described in Section
IV-B, the average channel availability time can be given
as
'
c
=
'
c
+
c
) 1 (

where is the smoothing factor for
'
c
. As per above
definition, we take
SF
c
=
'
c
(6)
With all the parameters defined, we formulate the cost
metric for channel selection for a given link (LC
c
) as
follows.

For fixed transmission power case:
LC
c
= w
1
ETT
c
+ w
2
SC
c
+ w
3
(1 / SF
c
) (7)

For adaptive transmission power case:
LC
c
= w
1
ETT
c
+ w
2
SC
c
+ w
3
(1 / SF
c
) + w
4
Pt
m
[c] (8)

where LC
c
is cost of a link when channel c is used, and
other symbols are as defined above for channel c. The
values w1, w2, w3, and w4 are the weights that can be
assigned to different parameters under the constraint that
w1 + w2 + w3 + w4 = 1. The appropriate value for these
weights and the basis for deciding them is a subject of
further study.
B. End-to-end Routing Metric
Assuming that the end-to-end route r consists of p
hops, we update the routing metric suggested in [13] to
use the link cost definition given in eq.(7). The routing
metric is formulized as:
RM(r) =

p
i
i
LCc
1
* ) 1 ( +
C j 1
max ( * )
j
X (9)
Here, RM(r) is the routing metric for route r,
i
LCc is
the cost of using channel c on hop i,
j
X is the total
number of times channel j is used in route r, and C is the
total number of channels in the spectrum pool.
i
LCc is
given by equations (7) or (8) depending upon whether
fixed power or adaptive power controlled nodes are
assumed in the network.

VII. FUTURE WORK
In the work reported in this paper, we have made two
crucial assumptions in our algorithm, which we would
relax in our future work. First, we have assumed that
available channels are homogeneous in nature in terms
of their transmission power, range, etc. This assumption
is valid if all the available channels come from a single
primary system, and the secondary devices completely
know the characteristics of the primary system. But in
absence of such knowledge about the primary system,
the secondary devices are required to consider a
heterogeneous channel set. The heterogeneous channel
set brings in unique challenges, such as described in
[15]. In our future work, we would study the effect of
heterogeneous channel set in conjunction with dynamic
channel set (which is already considered in this paper).
One important question in dealing with heterogeneous
channel set is to decide the protocol stack layer where
this channel heterogeneity is to be handled [15]. Second,
the ETT value used in equations (7) and (8) is calculated
using eq. (5), which takes a channels peak bandwidth B
into account. A more accurate estimate of ETT can be
obtained by using available bandwidth (instead of peak
bandwidth) in eq. (5). This requires estimating the
available bandwidth for each channel. It needs to be
investigated whether the existing proposals, such as [16]
and [17], for estimating available bandwidth in multi-
hop ad hoc networks require modifications when used in
cognitive mesh networks. We must also note that some
of these proposals, such as [17], are coupled with the
MAC layer used in the network. Another area of future
study in this work is to investigate the challenges in
design of higher layer protocols, such as transport layer,
for dynamic and heterogeneous channels set, and to
formulate appropriate solution for them. Finally, the
design of a virtual MAC layer abstraction that can work
with different heterogeneous channels remains an
important and interesting area of future work.
REFERENCES
[1] Shared Spectrum Spectrum Measuements
(www.sharedspectrum.com).
[2] FCC Spectrum Policy Task Force, Report of the
Interference Protection Working Group, November
15, 2002.
[3] FCC, Notice of Inquiry and Notice of Proposed
Rule Making, In the matter of establishment of an
interference temperature metric to quantify and
manage interference and to expand available
12
unlicensed operation in certain fixed, mobile and
satellite frequency bands, ET Docket No. 03-237,
November 13, 2003..
[4] Denijela Cabric, Shridhar Mubaraq Mishra, and
Robert W. Brodersen, Implementation issues in
spectrum sensing for cognitive radios, in Proc.
Asilomar Conf. Signals, Systems, and Computers,
2004, pp. 772-776.
[5] F. Capar, I. Martoyo, T. Weiss, and F. Jondral,
Comparison of bandwidth utilization for controlled
and uncontrolled channel assignment in a spectrum
pooling system, in Proc. IEEE VTC 2002, pp.
1069-1073.
[6] Jon M. Peha, and Sooksann Panichpapiboon, Real-
time secondary markets for spectrum, Elsevier J.
Telecommunications Policy, vol. 28, pp. 603-618,
2004.
[7] Comments of the Wireless Communications
Associations International, Inc. on FCC NOI/NPRM
ET Docket No. 03-237.
[8] FCC, Notice of Inquiry and Notice of Proposed Rule
Making, In the matter of promoting efficient use of
spectrum through elimination of barriers to the
development of secondary markets, WT Docket
No. 00-230, November 9, 2000.
[9] Paul J. Kolodzy, Interference temperature: a metric
for dynamic spectrum utilization, Wiley Int. J.
Network Management, vol. 16, pp. 103-113, 2006.
[10] T. Clancy, "Formalizing the Interference
Temperature Model," to appear in Wiley Wireless
Communications and Mobile Computing 2006.
(Available online at:
http://www.cs.umd.edu/~clancy/docs/itma-
wcmc06.pdf).
[11] J. Mitola III, Cognitive Radio: An Integrated
Agent Architecture for Software Defined Radio,
Doctor of Technology Dissertation, Royal Institute
of Technology (KTH), Sweden, May 2000.
[12] Jeffrey P. Monks, V. Bharghavan, and Wen-mei W.
Hwu, A Power Controlled Multiple Access
Protocol for Wireless Packet Networks, in Proc.
IEEE INFOCOM, 2001.
[13] Pradeep Kyasanur, and Nitin H. Vaidya,
Routing and link-layer protocols for multi-channel
multi-interface ad hoc wireless networks, in ACM
Mobile Computing and Communications Review,
vol. 1, no. 2, January 2006.
[14] Richard Draves, Jitendra Padhye, and Brian Zill,
Routing in multi-radio, multi-hop wireless mesh
networks, in Proc. ACM MobiCom, 2004.
[15] Pradeep Kyasanur, and Nitin H. Vaidya, Protocol
Design Challenges for Multi-hop Dynamic
Spectrum Access Networks, Proc. of IEEE
DySPAN, November 2005.
[16] Yaling Yang, and Robin Kravets, Contention-
Aware Admission Control for Ad Hoc Networks,
in IEEE Transactions on Mobile Computing, Vol. 4,
No. 4, July/August 2005.
[17] Kimaya Sanzgiri, Ian D. Chakeres, and Elizabeth
M. Belding-Royer, Pre-Reply Probe and Route
Request Tail: Approaches for Calculation of Intra-
Flow Contention in Multihop Wireless Networks,
in MONET Journal, 11(1), Feb 2006.