1958
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
Spectrum-Aware Opportunistic Routing in
Multi-Hop Cognitive Radio Networks
Yongkang Liu, Lin X. Cai, Member, IEEE, and Xuemin (Sherman) Shen, Fellow, IEEE
Abstract—In this paper, cognitive routing coupled with spectrum sensing and sharing in a multi-channel multi-hop cognitive
radio network (CRN) is investigated. Recognizing the spectrum
dynamics in CRN, we propose an opportunistic cognitive routing
(OCR) protocol that allows users to exploit the geographic
location information and discover the local spectrum access
opportunities to improve the transmission performance over each
hop. Specifically, based on location information and channel
usage statistics, a secondary user (SU) distributedly selects the
next hop relay and adapts its transmission to the dynamic spectrum access opportunities in its neighborhood. In addition, we
introduce a novel metric, namely, cognitive transport throughput
(CTT), to capture the unique properties of CRN and evaluate the
potential relay gain of each relay candidate. A heuristic algorithm
is proposed to reduce the searching complexity of the optimal
selection of channel and relay. Simulation results are given to
demonstrate that our proposed OCR well adapts to the spectrum
dynamics and outperforms existing routing protocols in CRN.
Index Terms—Cognitive radio, multi-hop transmission, opportunistic routing, dynamic spectrum access.
I. I NTRODUCTION
OGNITIVE radio network (CRN) has been emerging
as a prominent solution to improve the efficiency of
spectrum usage to meet the increasing user demand on
broadband wireless communications. In CRN, secondary users
(SUs) can utilize spectrum access opportunities for unlicensed
transmissions when primary users (PUs) do not occupy the
licensed spectrum. Therefore, the most critical issue in CRN
is the exploration and exploitation of the spectrum access
opportunities for SUs’ transmissions and in the meantime
preventing harmful interference to PUs’ transmissions [1], [2].
While most research in CRN has focused on a single-hop
wireless access network, the research community has recently
realized that cognitive paradigm can be applied in multi-hop
networks to provide great potential for unexplored services
and enable a wide range of multimedia applications with the
extended network coverage.
To fully explore the potentials of the multi-hop CRN in
support of multimedia applications, it is crucial to study routing in dynamic spectrum access system, taking into account
the unique properties of the cognitive environment. Existing
research efforts mainly focus on effective spectrum sensing
and sharing schemes in the physical and MAC layers. Some
C
Manuscript received 5 January 2012; revised 15 May 2012.
Y. Liu and X. Shen are with the Center for Wireless Communications, Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON N2L 3G1, Canada (e-mail: {y257liu, xshen}@bbcr.uwaterloo.ca).
L. X. Cai is with Department of Electrical Engineering, Princeton University, Princeton, NJ USA 08544 (e-mail: {lincai}@princeton.edu).
Digital Object Identifier 10.1109/JSAC.2012.121111.
recent studies indicate that the next major breakthrough in
CRN lies in utilizing the diversity gain of spare spectrum in the
time, frequency, and space domains to enhance transmissions
among SUs [3]. However, in multi-hop CRN, SUs distributed
at different locations may have different views of the usage
patterns of PUs over multiple frequency channels, which
makes it extremely challenging for SUs to coordinate with
each other and to exploit the multi-channel and multi-user
diversity gain. Some preliminary works on spectrum-aware
routing have been proposed for joint channel assignment
and route establishment [4], [5], [6]. However, these routing
algorithms are based on a pre-determined end-to-end routing
table, which is more suitable for static spectrum access system
where the channel conditions do not change frequently, e.g.,
in a CRN operating in TV bands [7]. In dynamic spectrum
access system, spectrum access opportunities of mobile SUs
may change over hops from time to time, which makes it
very difficult and costly to maintain a routing table. Some
recent research extends the work in a wide spectral band under
highly dynamic channel conditions other than TV bands [8],
[9]. A QoS differentiation scheme and an opportunistic relay
forwarding scheme are proposed in our previous works [9],
[10], respectively, considering heterogeneous channel usage
patterns. These works either mainly focus on the QoS provisioning in a multi-channel scenario or only exploit the
diversity of channel propagation characteristics in multi-hop
transmissions, which do not specify the impact of the channel
usage statistics on SUs’ transmissions, especially in a multihop CRN.
In this paper, we study cognitive routing in a multi-channel
multi-hop CRN, by utilizing channel usage statistics in the
discovery of spectrum access opportunities to improve transmission performance of SUs. The main contributions of this
paper are four-fold: (i) we propose an opportunistic cognitive
routing (OCR) protocol in which forwarding links are selected
based on the locally identified spectrum access opportunities.
Specifically, the intermediate SU independently selects the
next hop relay based on the local channel usage statistics so
that the relay can quickly adapt to the link variations; (ii) the
multi-user diversity is exploited in the relay process by allowing the sender to coordinate with multiple neighboring SUs
and to select the best relay node with the highest forwarding
gain; (iii) We design a novel routing metric to capture the
unique properties of CRN, referred to as cognitive transport
throughput (CTT). Based on the novel metric, we propose a
heuristic algorithm that achieves superior performance with
reduced computation complexity. Specifically, CTT represents
the potential relay gain over the next hop, which is used in
c 2012 IEEE
0733-8716/12/$31.00
LIUet al.: SPECTRUM-AWARE OPPORTUNISTIC ROUTING IN MULTI-HOP COGNITIVE RADIO NETWORKS
II. R ELATED W ORK
Routing in CRN can be formulated as a global optimization
problem with the channel-link allocation for data flows in
the network [11]. Xin et al. [4] propose a layered graph to
depict the topology of CRN in a snapshot and allocate multiple
links over orthogonal channels to enhance the traffic throughput by establishing a near-optimal topology. Pan et al. [5]
propose a joint scheduling and routing scheme according to
the long term statistics of the link transmission quality for
SUs. Gao et al. [12] develop a flow routing scheme which
mitigates the network-wide resource for multicast sessions
in multi-hop CRN. These works on cognitive routing predetermine an end-to-end relay path in CRN based on the
global network information. However, the channel conditions
of secondary links are highly dependent on PUs’ activities
in CRN. SUs usually need to track the channel status by
periodic sensing [13] or field measurements [3]. When the
channel status changes, source nodes need to re-calculate a
path. Khalif et al. [8] show that the involved computation and
communication overhead for re-building routing tables for all
flows is nontrivial, especially when the channel status changes
frequently.
Compared with centralized scheduling, distributed opportunistic routing is more suitable for a dynamic CRN since SUs
can select the next hop relay to adapt to the variations of local
channel/link conditions [14], [15]. Instead of using a fixed
relay path, a source node broadcasts its data to neighboring
nodes, and selects a relay based on the received responses
under current link conditions [14]. Liu et al. [16] propose
to apply an opportunistic routing algorithm in CRN where
the forwarding decision is made under the locally identified
spectrum access opportunities. So far, most opportunistic routing protocols have been studied in a single channel scenario.
In a multi-channel system, the channel selection and relay
link negotiation may introduce extra delay, which degrades
the performance of the network. How to extend opportunistic
routing in a multi-channel CRN is still an open research issue.
It is also recognized that with available localization services,
geographic routing can achieve low complexity and high
scalability under dynamic link conditions in various wireless
networks, such as wireless mesh networks [17], ad hoc networks [18] and vehicle communication networks [19]. With
geographic routing, a node selects a relay node that is closer
to the destination for achieving distance advances in each hop.
One hop relay
DATA
SIFS
RRSP
ACK
P
ON
C2
RREQ
SIFS
C1
SIFS
CCC
SIFS
the channel sensing and relay selection to enhance the OCR
performance; and (iv) we evaluate the performance of the
proposed OCR in a multi-hop CRN. Simulation results show
that the proposed OCR protocol adapts well to the dynamic
channel/link environment in CRN.
The remainder of the paper is organized as follows. The
related work is presented in Section II. The system model
is introduced in Section III. A multi-channel opportunistic
cognitive routing protocol is proposed in Section IV. To
maximize the relay performance of the OCR, a novel routing
metric is designed and the practical implementation issues are
discussed in Section V, followed by performance evaluation
in Section VI. Concluding remarks are given in Section VII.
1959
...
t0
ON
Fig. 1.
Active PU
t1
t2
SNSINV
in CCC
t3
Energy
detection
t4
SU TX in
data channel
The opportunistic cognitive routing timeline
Chowdhury and Felice [6] introduce geographic routing into
CRN to calculate a path with the minimal latency. However,
their work still focuses on building routing tables and thus is
not suitable for dynamic CRN. Considering the unique features
of CRN, it is essential to design a distributed opportunistic
routing algorithm by tightly coupling with physical layer
spectrum sensing and MAC layer spectrum sharing to adapt
to the network dynamics in CRN.
III. S YSTEM D ESCRIPTION
We consider a multi-hop CRN where multiple PUs and
SUs share a set of orthogonal channels, C = {c1 , c2 , ..., cm }.
SUs can exchange messages over a common control channel
(CCC)1 . Each SU is equipped with two radios: one half-duplex
cognitive radio that can switch among C for data transmissions
and the other half-duplex normal radio in CCC for signaling
exchange.
When a source SU communicates with a destination node
outside its transmission range, multi-hop relaying is required.
As shown in Fig. 1, at each hop, the sender first senses for
a spectrum access opportunity and selects a relay node in
the detected idle channel2 . We model the occupation time of
PUs in each data channel as an independent and identically
distributed alternating ON (PU is active) and OFF (PU is
inactive) process. SUs track the channel usage pattern, i.e.,
ON or OFF, and obtain the channel usage statistics through
periodic sensing operations. Generally, the statistics of channel
usage time change slowly. The parameter estimation is beyond
the scope of this paper and the details can be found in [3], [13].
With GPS or other available localization services, SUs can
acquire their own location information, and the source nodes
have the corresponding destinations’ location information,
e.g., an edge router or a gateway in the network. A summary
of main notations used in the paper is given in Table I for
easy reference.
IV. O PPORTUNISTIC C OGNITIVE ROUTING (OCR)
P ROTOCOL
In this section, an opportunistic cognitive routing (OCR)
protocol is proposed where SUs forward the packets in the
1 The CCC can be implemented by bidding on a narrow spectrum band [20]
or accessing the temporarily spare spectrum bands in a predefined frequency
hopping sequence [21].
2 In some extreme case when geographic routing fails to reach the destination, we can apply the right-hand rule for route recovery as proposed in
GPSR [17].
1960
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
TABLE I
S UMMARY OF N OTATIONS
Symbol
C = {cj }
NS
RD
c
RDj
AD (S, R)
∗
(c∗ , RcD )
c
CT T (cj , RDj )
d(S, D)
cj
cj
E[TON
](E[TOF
F ])
cj
FOF F (t)
c
c
IRj (IRj )
Tdetc
TDT X
Tinit
Trelay
TRREQ (TRRSP )
TRS
TSNS
Tswitch
t0
c
Pi j
c
j
POF
F,R (t0 , t1 )
cj
PR (t1 , t2 )
c
j
Prelay,R
i
cj
PRSf
ail
VRi
c
XRjt Rr = 1(0)
ρcj
µ
γ
Definition
Channel set, j = {1, 2, ..., m}
The SU set of the sender S’s neighbors
The set of relay candidates for the
destination D
The set of relay candidates for the
destination D in the channel cj
Relay advancement of the link SR for D
The transmission channel and the ordered
relay set selected by MAXCTT
The cognitive transport throughput (CTT)
Euclidian distance between S and D
Mean duration of a busy(idle) cj
CDF of the OFF duration of cj
SU R detects cj to be idle (busy)
Per channel energy detection delay
Per hop data packet transmission delay
Sensing initialization delay
Per hop transmission delay in OCR
RREQ (RRSP) message transmission delay
Per hop relay selection delay
Per hop sensing delay
Transceiver switching time
The latest channel status observation time
The probability that Ri is selected as
the relay in cj
The probability that cj is idle at t1 , t1 > t0 .
The probability that cj is idle during [t1 , t2 ]
at R
The probability that the relay via Ri succeeds
in cj
The probability that relay selection fails in cj
The priority of Ri in the relay selection.
SU Rt and SU Rr are (not) affected
by the same PU in cj
The chance for an idle state in cj
Backoff mini-slot
The maximum channel propagation delay
locally identified spectrum access opportunities. To adapt to
the channel dynamics, SUs opportunistically select the relay
nodes from multiple candidates according to the distance gain
and the channel usage statistics.
A. Protocol Overview
As shown in Fig. 1, the per hop relay in OCR includes
three steps, i.e., channel sensing, relay selection, and data
transmission.
In the channel sensing step, the sender searches for a temporarily unoccupied channel in collaboration with its neighbors using energy detection technique. Before sensing the data
channel, the sender broadcasts a short message, i.e., sensing
invitation (SNSINV), in the CCC to inform neighboring nodes
of the selected data channel, and the location information of
the sender and the destination. The transmission of SNSINV
message in the CCC follows the CSMA/CA mechanism as
specified in IEEE 802.11 MAC. Upon receiving the SNSINV,
the neighboring SUs set the selected data channel as nonaccessible so that no SU will transmit in the selected data
channel during the sensing period of the sender. In this
way, the co-channel interference from concurrent secondary
transmissions can be mitigated. Using the location information
in SNSINV, the neighboring SUs evaluate whether they are
eligible relay candidates, e.g., whether a relay node is closer
to the destination than the sender and thus can provide a
relay distance gain. Eligible relay candidates will collaborate
with the sender in channel sensing and relay selection. Other
SUs cannot transmit in the selected data channel during the
reserved time period specified in SNSINV. When the channel
is sensed idle, i.e., no PU activity is detected, the sender will
initiate a handshake with relay candidates in the relay selection
step. Otherwise, the sender selects another channel and repeats
the channel sensing process.
In the relay selection step, the sender selects the next hop
relay from the relay candidate SUs. Specifically, when the
channel is sensed idle, the sender first broadcasts a routing
request (RREQ) message to the relay candidates. Eligible
candidates reply routing response (RRSP) messages in a
sequence specified by the sender. A relay candidate is assigned
a higher priority to transmit RRSP after a shorter backoff
window if it has a larger link throughput [14], a greater relay
distance advancement [17], or a higher link reliability [22].
A candidate SU keeps listening to the data channel until it
overhears an RRSP or it transmits an RRSP when its backoff
timer reaches zero. The sender selects the first replying relay
candidate as the next hop relay. If the sender receives no RRSP
message, which implies no relay candidate is available in the
selected channel, it will repeat the channel sensing and relay
selection steps. After a successful RREQ-RRSP handshake,
the sender transmits data to the selected relay node in the
data transmission step.
B. Analysis of the OCR Protocol
We study the impacts of PUs’ activities on the performance
of the proposed OCR protocol. In CRN, when PUs appear in a
channel, an SU needs to stop its current transmission, update
its record of the channel status, and reselect a data channel.
Thus, PUs’ appearance will result in a larger transmission
delay, and involve extra overhead for channel sensing and
relay selection. To evaluate the impacts of PUs’ activities
on the protocol performance, we first introduce the main
performance metrics, namely, relay distance advancement and
per hop transmission delay. Based on the introduced metrics,
we then analyze the success probability in each step, i.e.,
channel sensing, relay selection, and data transmission.
1) Performance Metrics
We first introduce the relay distance advancement and
the per hop delay for performance evaluation. The relay
advancement is measured by the geographic distance gain.
For a sender S in CRN, NS is the set of SUs within its
transmission range. The neighboring relay candidate set for
the relay to the destination D is denoted by RD ⊆ NS . If an
SU R ∈ NS is selected as the relay, the relay advancement
AD (S, R) in terms of the difference in the distance between
the SU pairs, (S, D) and (R, D) can be expressed by
AD (S, R) = d(S, D) − d(R, D),
(1)
where d(S, D) and d(R, D) are the Euclidian distances between (S, D) and (R, D), respectively.
LIUet al.: SPECTRUM-AWARE OPPORTUNISTIC ROUTING IN MULTI-HOP COGNITIVE RADIO NETWORKS
The per hop transmission delay Trelay is comprised of three
parts: sensing delay (TSN S ), relay selection delay (TRS ), and
packet transmission delay (TDT X ).
The sensing delay TSN S includes the transmission time of
an SNSINV message, Tinit , and the energy detection time,
Tdetc ,
TSN S = Tinit + Tdetc .
(2)
Based on the relay capability, candidate SUs are sorted in a
given prioritized order. In the relay selection, the i-th relay
candidate Ri sends an RRSP message only when the first
i − 1 higher-priority candidates are not available. Therefore,
the relay selection delay TRS (i) is given by
TRS (i) = TRREQ + (i − 1)µ + TRRSP + 2 SIF S,
(3)
where TRREQ and TRRSP are the transmission time of an
RREQ message and an RRSP message, respectively, and µ is
the duration of one mini-slot in the backoff period. According
to [23], the length of a mini-slot can be calculated as µ =
2 · γ + tswitch , where γ is the maximum channel-propagation
delay within the transmission range, and tswitch is the time
duration that the radio switches between the receiving mode
and the transmitting mode.
Once Ri is selected, the packet transmission delay TDT X
is
TDT X = TDAT A + TACK + 2 SIF S,
(4)
which includes the packet transmission delay (TDAT A ) and
the ACK transmission time (TACK ).
The transmission delay Trelay (Ri ) via the relay at Ri is the
delay sum
Trelay (Ri ) = TSN S + TRS (i) + TDT X .
(5)
2) Channel Sensing
c
c
Denote IRj (IRj ) as the event that cj is sensed to be idle
(busy) by an SU R in the channel cj . A channel is determined
to be idle given that it is sensed idle at the starting time of
t1 and remains idle until sensing completes at t2 , as shown in
Fig.1. According to the renewal theory, the channel status can
be estimated by the distribution of the channel state duration
and the sensing history [24]. Specifically, given the channel
status (idle or busy) observed at an earlier time, e.g., t0 , we
cj
have POF
F,R (t0 , t1 ), the probability that cj is idle (OFF) at t1 ,
t1 > t0 . Assume ON and OFF durations follow exponential
cj
cj
3
] and 1/E[TOF
distributions with mean 1/E[TON
F] ,
c
=
(t0 , t1 )
P j
OF F,R
ρcj + (1 − ρcj )e−∆cj (t1 −t0 ) , if cj is OFF at t0 ,
if cj is ON at t0 ,
ρcj − ρcj e−∆cj (t1 −t0 ) ,
⎧
cj
E[T
]
OF F
⎨ ρc =
,
cj
cj
j
E[TON
]+E[TOF
F]
(6)
where
1
1
⎩∆cj =
+
.
cj
cj
E[TON ]
E[TOF F ]
Note that ρcj indicates the chance for an idle state in cj .
We then calculate the likelihood of the channel staying idle
during the sensing period. According to the renewal theory,
3 which
are commonly used in other works [3], [13]
1961
the residual time of a state in an alternating process truncated
since the time origin can be expressed by the equilibrium
distribution of the state duration [24]. Thus, the probability
that the channel at R stays in the idle state during the sensing
period [t1 , t2 ] can be calculated as
∞
cj
FOF
cj
F (u)
PR (t1 , t2 ) =
du,
(7)
cj
t2 −t1 E[TOF F ]
F
cj
(t)
where OFcFj is the probability density function (PDF) of the
E[TOF F ]
residual time of an idle channel since the time origin when
cj
it is observed as idle. FOF
F (t) is the cumulative distribution
function (CDF) of the duration of the OFF state in cj with
cj
the mean E[TOF
F ]. Then, the probability that R detects a
spectrum access opportunity in cj is given by
c
c
c
j
j
P r{IRj } = POF
F,R (t0 , t1 ) · PR (t1 , t2 ).
(8)
c
For the OCR protocol, P r{ISj } denotes the probability
of sensing success when the sender S detects cj as an idle
channel. Once the sender finds an idle channel, it will move
to the relay selection step. Otherwise, the sender will switch
to another channel and initiate the channel sensing process.
3) Relay Selection
After detecting an idle channel, the sender needs to select
a relay for data forwarding. In OCR, the prioritized RRSP
transmission enables the relay candidate of the highest relay
priority to notify the sender its availability for data forwarding.
However, active PUs may interrupt the handshaking process
and cause the failures in the relay selection when an SU
candidate cannot reply due to the detection of active PUs. Such
case is very rare, and it happens only when a nearby PU turns
on during the selection period. Since the relay selection is
very short in time, usually less than 1 millisecond, we mainly
consider the case when a candidate SU detects the selected
channel which is occupied by an active PU in the sensing. In
this case, the candidate will not respond to the RREQ. If no
relay candidate responds to the RREQ message at the moment,
the relay selection fails. Therefore, we have
c
c
c
cj
(9)
IRji ISj ,
= P r{ISj } · P r
PRSf
ail
c
Ri ∈RDj
c
where P r{ISj } indicates the probability that the sender initiates the relay selection when it detects an idle channel
as defined in Eq. (8). In cj , one feasible relay selection
c
RDj = {R1 , R2 , . . . , Rn } contains a set of SUs in RD with
c
the size of n = |RDj |. Denote VRi as the priority of Ri
c
in the RRSP transmission. RDj is sorted in the descending
order of VRi , i.e., VR1 > VR2 > . . . > VRn . The event
that no relay candidate replies in the relay selection step,
c
is equivalent to the event that all SUs in RDj sense the
channel
busy in the previous sensing with the probability
cj
c
c I
ISj .
Pr
Ri ∈RDj Ri
In the CRN, we assume that an SU is affected by at
most one active PU in one frequency band. Such assumption holds in the frequency bands such as the downstream
bands in cellular network where the adjacent cells/sectors
are usually assigned with different working frequencies to
1962
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
avoid the co-channel interference [25]. Thus, the channel
usage pattern is mainly determined by the PU activity at
c
the spot of the individual SU. Let XRjt Rr = 1 if a pair of
SUs, Rt and Rr , are affected by the same PU in cj , and
c
c
XRjt Rr = 0 otherwise. XRjt Rr can be acquired and maintained
by the periodic exchange of the channel status in the SU’s
neighborhood. A cognitive transmission is successful only
if both ends of the link are not influenced by active PUs.
For example, if the channel utilities of cj at Rt and Rr are
c
c
ρRjt and ρRjr , respectively, the link quality of the link ltr
c (1−X
c
c
cj
)
Rt Rr
canbe expressed by Pltrj = ρRjt · ρRjr
. Therefore,
cj
cj
c
Pr
IS in Eq. (9) is given by
I
Ri ∈R j Ri
D
Pr
=
=
c
c
c
P r IRj1 ISj
n
i=2
n
·
Pr
i−1
c
IRji
i=2
c
cj
XSR1 )P r IRj1
(1 −
·
c
IRji ISj
c
Ri ∈RDj
c
IRjk
∩
c
ISj
k=1
i−1
cj
k=1 (1−XRk Ri )
cj
c
(1 − XSR
)P
r
IRji
i
.
(10)
⎧
c
c
c
⎪
P r{ISj } · P r{IRj1 ISj },
for i = 1,
⎪
⎪
⎪
⎪
⎨
c
(11)
Pi j =
cj
cj
cj
i−1 cj
⎪
∩
{I
,
}
I
I
}
·
P
r
P
r{I
⎪
S
Ri
S
k=1 Rk
⎪
⎪
⎪
⎩
for 2 ≤ i ≤ n,
cj
cj
i−1 cj
where P r
can be expressed as
k=1 IRk ∩ {IRi } IS
i−1
c
c
c
Pr
c
IRju
∩ {IRji } ISj
IRjk
k=1
= Pr
c
IRj1
c
i−1
·
u=2
·
i−1
·
i−1
c
IRjk
k=1
cj
c
XSR1 )P r IRj1
c
∩ ISj
cj
c
(1 − XSR
)P
r
IRju
l
i−1
u−1
c
IRjk
∩
c
ISj
r=1
u=2
·P r IRji
= (1 −
c
ISj
u−1
r=1
c
(1−XRjr Ru )
cj
c (1−XSR
c
)
i .
(1 − XRjk Ri ) P r{IRji }
c
j
Prelay,R
i
c
c
=
j
(t3 , t4 )
Pi j · PlSR
=
c
Pi j
i
·
c
PSj (t3 , t4 )
c
· PRji (t3 , t4 )
c
j
(1−XSR
)
i
.
(13)
V. J OINT C HANNEL AND R ELAY S ELECTION
We then jointly consider the selection of the sensing channel
and relay node to improve the performance of the proposed
OCR. As many factors, including channel usage statistics, the
relay distance advances, and transmission priority of relay
candidates, may affect the relay performance, we introduce a
new metric to capture these factors and apply it in a heuristic
algorithm to select the best relay in one data channel at a
reduced computation complexity.
A. Novel OCR Metric
We design a new metric, the cognitive transport throughput
c
(CTT), CT T (cj , RDj ), to characterize the one hop relay
performance of OCR in the selected channel cj with the
c
selected relay candidate set RDj , in unit of bit·meter/second.
Suppose that the i-th relay candidate Ri in the selected relay
c
selection order RDj is available, Ri will be selected as the next
c
hop relay with the probability Pi j , given that previous i − 1
candidates are not available,
Pr
period [t3 , t4 ] in cj . Thus, the successful relay probability at
current hop via Ri can be expressed by
(12)
k=1
4) Data Transmission
Once Ri is selected, the data transmission in the link lSRi
succeeds when no active PU appears during the transmission
c
CT T (cj , RDj ) =
=
c
Aj
E L · cjD
Trelay
L · AD (S, Ri )
cj
Prelay,R
(14)
i
Trelay (Ri )
cj
Ri ∈RD
The physical meaning of the CTT defined in Eq. (14) is
the expected bit advancement per second for one hop relay of
a packet with the payload L in the channel cj . To improve
the OCR performance, we should maximize the one hop
relay performance along the path as one hop performance
improvement contributes to the end-to-end performance. In
addition, as the multi-user diversity is implicitly incorporated
in the relay selection process, we can also achieve a high
multi-user diversity gain by maximizing CTT. From Eq. (14),
we can jointly decide channel cj and the corresponding relay
c
selection order RDj to maximize CTT.
B. Heuristic Algorithm
∗
To obtain c∗ and RcD for the largest CTT, we can exhaustively search for all possible combinations of the sensing
channel and the subset of the relay candidate set. Given m
channels and up to n relay candidates, an exhaustive search
needs to find the locally optimal one in each channel by
comparing the value of CTT under all possible permutations of
the set of relay candidates. Since the CTT value is sensitive to
the set size as well as the permutation, given that k candidate
nodes are incorporated in the relay selection, 1 ≤ k ≤ n,
there are P (n, k) types of opportunistic forwarding patterns.
Therefore,
over m channels, the exhaustive search should
n
take m · k=1 P (n, k) times of the CTT calculation to
return the global
get
optimum. If n goes to infinity,
we can
n!
=
limn→∞ m · nk=1 P (n, k) = limn→∞ m · nk=0 (n−k)!
n 1
limn→∞ m · n! ·
−
1
.
Thus
the
exhaustive
search
k=0 k!
running time is O(m · n! · e), where e is the base for natural
LIUet al.: SPECTRUM-AWARE OPPORTUNISTIC ROUTING IN MULTI-HOP COGNITIVE RADIO NETWORKS
logarithms. We can see that once n becomes very large, the
exhaustive search becomes infeasible in real implementations.
To reduce the complexity, we propose an efficient heuristic
algorithm to reduce the searching space yet achieve similar
performance of the optimal solution. The performance comparison will be given in the following section.
Given independent channel usage statistics in different
channels, we can decompose the optimization problem into
two phases. First, we compare all possible relay selection
orders in each channel and find the optimal one which
maximizes the CTT. Then, we choose the relay selection
order with the largest CTT value over all channels and select
the corresponding channel as the sensing channel. Since the
number of channels is usually limited, it is more important to
reduce the searching complexity for the best relay selection
order in a single channel.
To find the optimal relay selection order, the sender should
decide both the number of the relay candidates and the relay
priority of each candidate. According to Eq. (14), a neighboring SU, Ri , is an eligible relay candidate if it contributes to a
positive relay distance advancement, AD (S, Ri ). One feasible
c
relay selection order RDj in cj is an ordered subset of RD
in the descending order of relay priority VRi . A larger size
c
of RDj includes more relay candidates and achieves a higher
diversity gain, which improves the per hop throughput at the
cost of the increased searching complexity.
To reduce the searching space and improve the algorithm
efficiency, we have the following Lemma.
c
Lemma 5.1: Given a feasible relay selection set RDj ,
cj
cj
∃Ri1 , Ri2 ∈ RD , if VRi1 > VRi2 , XRi Ri = 1, then
2
1
c
c
CT T (cj , RDj \ {Ri2 }) ≥ CT T (cj , RDj ).
c
Proof: Suppose RDj = {R1 , . . . , Ri1 , . . . , Ri2 , . . .}. Acc
cording to Eq. (11), if VRi1 > VRi2 , XRji Ri = 1 and
2
1
cj
cj
cj
XRi Ri = 1, Pi2 = 0. Thus, Prelay,Ri = 0. From Eq. (14),
1
2
2
c
CT T (cj , RDj ) =
i
2 −1
c
j
Prelay,R
r
r=1
L · AD (S, Rr )
Trelay (Rr )
c
j
Prelay,R
r
r=i2 +1
≤
cj
Prelay,R
r
r=1
c
j
Prelay,R
r
r=i2 +1
=
L · AD (S, Rr )
Trelay (Rr )
L · AD (S, Rr )
Trelay (Rr )
c
|RDj |
+
c
CT T (cj , RDj ) =
i
c
j
Prelay,R
r
r=1
c
|RDj |
+
0·
r=i+1
=
L · AD (S, Rr )
Trelay (Rr )
L · AD (S, Rr )
Trelay (Rr )
c
c
CT T (cj , RDj \ {Rk Rk ∈ RDj ,
VRk < VRi }),
which shows that the CTT performance does not change when
c
the relay candidates are removed from RDj with lower priority
than Ri .
Property 5.2 indicates that the size of the relay candidate set
can be further reduced by deleting SUs whose relay priorities
are lower than the SU that is affected by the same PU as the
sender. In other words, we can reduce the searching set without
degrading the performance of the current flow while the
deleted candidates can also participate in other transmissions,
which further improve the network performance.
As discussed above, the relay priority plays a critical role
in relay selection. It is well known that in geographic routing,
the node closest to the destination is the best next hop relay as
it provides the greatest distance gain. It is also proved that the
geographic routing approaches the shortest path routing with
the distance advance metric [26]. Therefore, we also apply
the distance advance and verify its efficiency in the proposed
OCR.
Thus, the CTT metric can be approximated as
c
i
2 −1
We observe the following property which can be used to
further reduce the searching space.
Property 5.2: (Tail Truncation Rule) Given a feasicj
c
c
= 1, then
ble relay selection RDj , ∃Ri ∈ RDj , XSR
i
c
c
c
CT T (cj , RDj ) = CT T (cj , RDj \ {Rk Rk ∈ RDj , VRk <
VRi }).
Proof: If S and Ri are affected by the same PU,
c
c
c
P r IRji ISj = 0. According to Eq. (11), Pk j = 0, ∀Rk ∈
c
RDj , VRk < VRi . Thus,
c
|RDj |
+
1963
L · AD (S, Rr )
Trelay (Rr ) − µ
c
CT T (cj , RDj \ {Ri2 }),
which shows that the CTT performance does not drop when
c
Ri2 is deleted from RDj .
Lemma 5.1 indicates that we can reduce the size of the relay
selection by excluding the relay candidates that are affected
by the same PU. The reduced set of relay candidates will not
degrade CTT. Specifically, for a given set of relay candidates,
the sender groups the SUs that are affected by the same PU,
selects the SU with the highest relay priority, and deletes other
SUs in a group from the set.
c
CT T (cj , RDj )
≃
=
c
L
Trelay
L
Trelay
|RDj |
·
Prelay,Ri AD (S, Ri )
i=1
c
· E[ADj ],
(15)
where E[ADj ] is the estimated relay advancement in cj ,
and Trelay is the estimated one hop transmission delay in
Eq. (5). To maximize the CTT in each channel, we need to
c
find an optimal relay selection to maximize E[ADj ]. When
c
opportunistic routing over independent links uses E[ADj ] as a
routing metric, [26] has proved that the optimal relay priority
should be set according to the distance of the relay candidate
c
to the destination. In addition, the maximum E[ADj ] increases
with the number of relay candidates. Therefore, we can assign
the relay priority in the descending order of AD (S, R).
We then propose a heuristic algorithm, MAXCTT, as shown
in Algorithm 1. The inputs are the channel set C, the set
of relay candidates RD , and the maximum number of relay
candidates in relay selection rmax . MAXCTT selects the
c
SUs from RD to form the relay selection order RDj and
1964
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
Algorithm 1 MAXCTT(C, RD , rmax )
TABLE II
S IMULATION PARAMETERS
∗
1: c∗ ← 0; RcD ← ∅; CT Tmax ← 0;
2: for each cj do
c
3:
N ← RD ; RE ← ∅; RDj ← ∅; Rp ← ∅; CT Tcj ← 0;
4:
while (N = ∅) do
5:
RE ← insert an SU Ri ∈ N that has max AD (S, Ri );
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
Remove Rj ∈ N with XRi Rj = 1 from N;
end while
c
while (RE = ∅ && |RDj | < rmax && XSRp = 1) do
for each SU Ri ∈ RE do
c
RT ← RDj + Ri ; Sort RT in the descending order of
AD (S, R);
Get CT T on RT according to Eq. (14);
if (CT T > CT Tcj ) then
CT Tcj ← CT T ; Rp ← Ri ;
end if
end for
c
RDj ← insert Rp in the descending order of AD (S, R);
RE ← RE − Rp ;
end while
if (CT Tcj > CT Tmax ) then
∗
c
c∗ ← cj ; RcD ← RDj ; CT Tmax ← CT Tcj ;
end if
end for
∗
return (c∗ , RcD );
Number of channels
{ρc1 , ρc2 , ρc3 , ρc4 , ρc5 , ρc6 }
Number of PUs per channel
PU coverage
E[TOF F ]
Number of SUs
SU transmission range
Source-destination distance
SU CCC rate
SU data channel rate
CBR delay threshold
Mini-slot time, µ
Per channel sensing time
Channel switching time
PHY header
rmax
6
{0.3, 0.3, 0.5, 0.5,
0.7, 0.7}
11
250 m
[100 ms, 600 ms]
[100, 200]
120 m
700 m
512 kbps
2 Mbps
2s
4 µs
5 ms
80 µs
192 µs
2
A. Simulation Settings
calculates the achieved CT Tcj in each cj . By comparing
CT Tcj over the channels, MAXCTT returns the channel c∗
that has CT Tmax and the corresponding relay selection order
∗
RcD as the algorithm output.
Specifically, an eligible relay candidate set RE is formed
by excluding the SUs affected by the same PU in cj according
to Lemma 5.1, which is a subset of RD (line 4–line 6).
c
A recursive searching [14] is then applied to obtain RDj .
cj
At the beginning of the searching step, RD contains no
c
SU. Each time, RDj includes one more relay candidate out
of the remaining SUs in RE which provides the best CTT
improvement (line 8–line 14). The selected relay candidates
c
are sorted in the descending order of AD (S, R) in RDj . The
cj
formed RD contains all relay candidates from RE , and it
satisfies the requirements of rmax and Property 5.2 (line 8).
c
The recursive searching obtains the optimal RDj in cj when
the size of the selection order is at most 2, and it achieves
almost the same performance as the optimal solution when
the final order contains more than 2 candidates according to
Lemma 5.1 in [14]. Suppose that
the largest size of RE over
n
the channels is n, at most m · k=1 k times of the CTT
calculations are required to find CT Tmax . Thus, the time
complexity of MAXCTT is O(m · n2 ), which is much lower
than exhaustive search.
VI. P ERFORMANCE E VALUATION
In this section, we evaluate the performance of the OCR
protocol by simulation under different network settings, e.g.,
channel conditions, number of SUs, and traffic loads, using
an event-driven simulator coded in C/C++ [10], [27]. The
network parameter settings are shown in Table II if no other
specification is made in the individual study.
The PU activity in each channel is modeled as an exponential ON-OFF process with parameters 1/E[TON ] and
1/E[TOF F ], and the idle rate ρ = E[TOF F ]/(E[TON ] +
E[TOF F ]) is selected accordingly. The channel status is
updated by periodic sensing and on-demand sensing before
data transmissions. We set up a CRN with multiple PUs and
SUs randomly distributed in an 800 × 800 m2 area. We set
a pair of SUs as source and destination with a distance of
700 m, and a constant bit rate (CBR) flow is associated with
the SU pair with packet size 512 bytes and flow rate of
10 packets per second (pps). The unit disc model is applied for
the data transmission. The channel switch time is 80 µs [28],
the minimum sensing duration with energy detection is 5 ms,
and a mini-slot is 4 µs [23]. We evaluate the performance of
the proposed OCR protocol in terms of the end-to-end delay,
the packet delivery ratio (PDR) and the hop count, i.e., the
total number of transmission hops between the source and
destination SUs. We run each experiment for 40 s and repeat
it 500 times to calculate the average value.
We then compare the performance of the OCR protocol
with that of SEARCH [6], based on different metrics for the
channel and relay selection, which are listed as follows.
1) SEARCH: SEARCH [6] is a representative geographic
routing protocol in CRN. It sets up a route with the
minimal latency before data transmissions. If an active
PU is detected which blocks the route, SEARCH pauses
the transmissions and recalculates the route. We modify
SEARCH by updating route periodically to adapt to the
dynamic changing spectrum access opportunities along
the route.
2) OCR (CTT): For OCR (CTT), the channel and the
relay candidate set are jointly selected by using the
proposed CTT metric and heuristic algorithm proposed
in Section V.
3) OCR (OPT): For OCR (OPT), the channel and the relay
candidate set are determined by exhaustively searching
LIUet al.: SPECTRUM-AWARE OPPORTUNISTIC ROUTING IN MULTI-HOP COGNITIVE RADIO NETWORKS
1965
1
End−to−end delay (ms)
Packet delivery ratio (PDR)
85
0.8
0.6
0.4
OCR(CTT), 10 pps
OCR(CTT), 40 pps
SEARCH, 10 pps
SEARCH, 40 pps
0.2
0
100
200
300
400
500
600
OCR(OPT)
OCR(CTT)
GOR
GR
80
75
70
100
200
E[TOFF] (ms)
500
600
TABLE III
AVERAGE NEIGHBOR DENSITY UNDER DIFFERENT SU DENSITIES
1800
End−to−end delay (ms)
400
Fig. 3. Performance comparison under different traffic loads and PU activities
(Number of SUs: 200, flow rate: 10 pps)
(a) Packet delivery ratio (PDR)
1500
Number of SUs
Average number of neighbors
Number of SUs
Average number of neighbors
1200
900
OCR(CTT), 10 pps
OCR(CTT), 40 pps
SEARCH, 10 pps
SEARCH, 40 pps
600
300
0
100
300
E[TOFF] (ms)
200
300
400
500
600
E[TOFF] (ms)
(b) End-to-end delay
Fig. 2. Performance comparison between OCR and SEARCH under different
channel conditions (Number of SUs: 200, flow rate: 10/40 pps)
for the biggest CTT over all possible channel-relay sets.
4) GOR: For geographic opportunistic routing (GOR) algorithm, the SU first selects the channel with the greatest success probability of packet transmissions; if the
channel is sensed idle, the SU then select a relay SU
over the channel. The relay selection order is based
on the location information and the relay capability of
SUs [16].
5) GR: For geographic routing (GR), an SU first selects the
channel for sensing as in GOR. If the selected channel
is sensed idle, the SU then selects the SU closest to the
destination as the next hop relay.
B. PU Activities
We first evaluate the performance of OCR under different
PU activity patterns. The average PU OFF duration E[TOF F ]
varies from 100 ms (high channel dynamics) to 600 ms
(low channel dynamics). The PDR performance of OCR
and SEARCH are compared under different traffic loads in
Fig. 2(a). A smaller E[TOF F ], e.g., 100 ms, indicates the
available time window is shorter and thus SUs’ transmissions
are more likely to be interrupted by PUs. We can see a marked
PDR improvement under dynamic channel conditions for the
100
7.0686
160
11.3097
120
8.4823
180
12.7235
140
9.8960
200
14.1372
per hop relay schemes, e.g., OCR (CTT), compared with
SEARCH which is based on the global route establishment.
In OCR (CTT), SUs are allowed to locally search and exploit
spare spectrum and select the available links for data forwarding. Thus, OCR (CTT) can adapts well in the dynamic data
channels. On the contrary, SEARCH uses a pre-determined
routing table. Once an active PU is detected along the relay
path, intermediate SUs should defer the packet relay until they
update their routing tables according to the current channel
availabilities in CRN. Since more SUs are involved in the route
establishment, the handshakes between SUs in the network to
establish the relay path introduce a large overhead and results
in a longer delay.
Fig. 2(b) and Fig. 3 compare the end-to-end delay performance. All routing protocols achieve a better delay performance when the idle channel state becomes longer, e.g., from
100 ms to 600 ms, as more packets can be transmitted during
the idle state. When the channel state change frequently,
SEARCH needs to update routing tables accordingly which
involves a long delay for route recovery. Our proposed OCR
protocols are opportunistic routing algorithms that quickly
adapt to the dynamic channel environment and achieve better
delay performance compared with SEARCH. OCR (CTT)
also outperforms GR and GOR since the latter two protocols
perform the channel and relay selection separately while
OCR (CTT) jointly considers the channel selection and relay
selection.
C. Multi-user Diversity
We investigate the impacts of node density on the relay
performance. The number of SUs in CRN varies from 100
to 200. When the number of SUs is large, the sender has
more neighbors as shown in Table III. With more SUs in
the neighborhood, the relay is more likely to find a feasible
1966
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
1
OCR(OPT)
OCR(CTT)
GOR
GR
110
Packet delivery ratio (PDR)
End−to−end delay (ms)
120
100
90
80
70
80
100
120
140
160
180
0.9
OCR(OPT)
OCR(CTT)
GOR
GR
0.8
10
200
20
Number of SUs
30
40
50
60
70
60
70
Flow rate (pps)
Fig. 4. Performance comparison of end-to-end delay under different SU
densities (flow rate: 10 pps, E[TOF F ] = 200 ms)
(a) Packet delivery ratio (PDR)
D. Effectiveness of Routing Metric
We further compare the performance of OCR (CTT) with
that of GR and GOR to evaluate the effectiveness of routing
metrics used in the channel and relay selection. We first
compare the performance under different traffic loads. We
change the traffic load by varying the flow rate from 10 pps
(light load) to 70 pps (heavy load). As shown in Fig. 5(a)
and Fig. 5(b), when the traffic load increases, the PDR and
End−to−end delay (ms)
1200
relay link with better relay distance advance, which reduces
the hop count number. The relay performance increases with
the number of SUs due to the larger diversity gain. As a
result, for all protocols, the hop count of the end-to-end
relay decreases and the PDR increases with SU density by
exploiting the multi-user diversity in CRN. The end-to-end
delay performance under different SU densities is compared
in Fig. 4. For GR and GOR, a channel is selected first, and then
SUs coordinates to serve as relay. The coordination overheads
increase with the number of SUs, which also degrades the PDR
performance. The proposed OCR (CTT) jointly considers the
channel and relay selection, and SU coordination overhead
is minimized as sender determines the relay selection order
based on the relay priority.
We also compare the performance of the heuristic algorithm
for the channel-relay selection in OCR (CTT) with the optimal
one in OCR (OPT) where the selection is based on exhaustive
search. Fig. 4 shows that OCR (CTT) achieves almost the
same performance as OCR (OPT), even when the returned
number of the selected relay candidates is only 2, according
to the value of rmax in Table II. Table III indicates that as the
SU density increases in the network, the number of neighbors
along the forwarding direction of the sender will increase
accordingly. For example, given 160 SUs over 6 channels,
the average number of neighbors of an SU is around 11. OCR
(OPT) takes over 6.5 × 108 times of the CTT calculation to
find the globally optimal solution which is infeasible for real
time implementation. In the simulated scenario, although at
most 4 neighbors are under independent PU coverage which
significantly reduces the searching space, OCR (OPT) still
takes 384 runs while OCR (CTT) only needs 60 runs, which
achieves the marked reduction at the computational expense.
OCR(OPT)
OCR(CTT)
GOR
GR
1000
800
600
400
200
0
10
20
30
40
50
Flow rate (pps)
(b) End-to-end delay
Fig. 5. Performance comparison under different traffic loads (Number of
SUs: 200, E[TOF F ] = 200 ms)
delay performance degrade. However, the decreasing rate of
OCR (CTT) is much lower than that of GR and GOR. This
is because OCR (CTT) jointly considers the optimal channel
and link selection, while the other two OCR protocols select
the channel and relay separately.
We define Pef to be the ratio of the number of successful
relay transmissions to the number of the sensing operations
performed in the data channels. Pef indicates the effectiveness of the routing metrics since the transmission relies on
detection of an idle channel and an available relay node. If
Pef approaches to 1, the selected channel for each hop relay
almost surely is available for data transmission. Fig. 6 shows
the performance of Pef under different node densities. In all
network scenarios, OCR (CTT) outperforms GR and GOR,
because CTT metric jointly considers the channel access and
relay selection.
VII. C ONCLUSIONS
In this paper, we have proposed an opportunistic cognitive
routing (OCR) protocol to improve the multi-hop transmission
performance in CRN. We have studied the impact of PU
activities on the operation of OCR in channel sensing, relay
selection and data transmission. Furthermore, we have pro-
LIUet al.: SPECTRUM-AWARE OPPORTUNISTIC ROUTING IN MULTI-HOP COGNITIVE RADIO NETWORKS
1.2
P
ef
OCR(OPT)
OCR(CTT)
GOR
GR
1.1
1
100
120
140
160
180
200
Number of SUs
Fig. 6. Performance comparison under different SU densities (flow rate:
10 pps, E[TOF F ] = 200 ms)
posed a novel metric, CTT, for the channel and relay selection.
Based on the metric, we have proposed a heuristic channel and
relay selection algorithm which approaches optimal solution.
We have compared the performance of OCR (CTT) with
that of the existing routing approaches, e.g., SEARCH, GR
and GOR and shown that the proposed OCR achieves the
highest PDR and the lowest delay. In our future work, we will
study protocol design with uncertain channel usage statistics
and the impacts of the measurement errors on the protocol
performance.
R EFERENCES
[1] H.T. Cheng and W. Zhuang, “Simple Channel Sensing Order in Cognitive Radio Networks,” in IEEE J. Sel. Areas Commun., vol. 29, no. 4,
pp.676–688, 2011.
[2] A. Alshamrani, X. Shen, and L. Xie, “QoS Provisioning for Heterogeneous Services in Cooperative Cognitive Radio Networks,” in IEEE J.
Sel. Areas Commun., Vol. 29, No. 4, pp. 819–830, 2011.
[3] M. Wellens, J. Riihijarvi and P. Mahonen, “Evaluation of Adaptive
MAC-Layer Sensing in Realistic Spectrum Occupancy Scenearios,” in
Proc. IEEE DySPAN’10, 2010.
[4] C. Xin, B. Xie, and C.-C. Shen, “A Novel Layered Graph Model
for Topology Formation and Routing in Dynamic Spectrum Access
Networks,” in Proc. IEEE DySPAN’05, 2005.
[5] M. Pan, C. Zhang, P. Li and Y. Fang, “Joint Routing and Link Scheduling
for Cognitive Radio Networks under Uncertain Spectrum Supply,” in
Proc. IEEE INFOCOM’11, Shanghai, China, April, 2011.
[6] K. R. Chowdhury, and M. D. Felice, “SEARCH: A Routing Protocol for
Mobile Cognitive Radio Ad-hoc Networks,” Computer Communications
(Elsevier), vol. 32, no. 18, pp. 1983–1997, 2009.
[7] R. Murty, R. Chandra, T. Moscibroda, and P. Bahl, “SenseLess: A
Database-Driven White Spaces Network,” IEEE Trans. Mobile Computing, Vol. 11, No. 2, pp.189–203, 2012.
[8] H. Khalif, N. Malouch, and S. Fdida, “Multihop Cognitive Radio
Networks: To Route or Not to Route,” IEEE Network, vol. 23, no. 4,
pp. 20–25, 2009.
[9] L. X. Cai, Y. Liu, X. Shen, J. W. Mark, and D. Zhao, “QoS-aware
Cognitive MAC for Multimedia over Cognitive Radio Networks,” in
Proc. IEEE Globecom’10, 2010.
[10] Y. Liu, L. X. Cai, X. Shen, and J. W. Mark, “Exploiting Heterogeneity
Wireless Channels for Opportunistic Routing in Dynamic Spectrum
Access Networks,” in Proc. IEEE ICC’11, 2011.
[11] M. Cesana, F. Cuomo, E. Ekici, “Routing in cognitive radio networks:
Challenges and solutions,” Ad Hoc Networks (Elsevier), vol. 9, no. 3,
pp. 228–248, May 2011.
1967
[12] C. Gao, Y. Shi, Y. T. Hou, H. D. Sherali, and H. Zhou, “Multicast
communications in multi-hop cognitive radio networks,” IEEE J. Sel.
Areas Commun., Vol. 29, No. 4, pp. 784–793, 2011.
[13] H. Kim and K. G. Shin, “Efficient Discovery of Spectrum Opportunities
with MAC-Layer Sensing in Cognitive Radio Networks,” IEEE Trans.
Mobile Compututing, vol. 7 no.5, pp. 533–545, May 2008.
[14] K. Zeng, Z. Yang, and W. Lou, “Location-aided Opportunistic Forwarding in Multirate and Multihop Wireless Networks,” IEEE Trans. Veh.
Technol., vol. 58, no. 6, pp. 3032–3040, July 2009.
[15] S. Biswas, R. Morris, “ExOR: Opportunistic Multi-hop Routing for
Wireless Networks,” ACMSIGCOMM Computer Communication Review, vol. 34, no. 1, pp. 133–144, 2005.
[16] Y. Liu, L. X. Cai, and X. Shen, “Joint Channel Selection and Opportunistic Forwarding in Multi-hop Cognitive Radio Networks,” in Proc.
IEEE Globecom’11, Houston, USA, Dec. 2011.
[17] B. Karp and H. T. Kung, “GPSR: Greedy Perimeter Stateless Routing
for Wireless Networks,” in Proc. ACM Mobicom, 2000.
[18] A. Abdrabou and W. Zhuang, “A Position-based QoS Routing Scheme
for UWB Ad Hoc Networks,” IEEE J. Sel. Areas Commun., vol. 24, no.
4, April 2006.
[19] A. Abdrabou and W. Zhuang, “Statistical QoS Routing for IEEE 802.11
Multihop Ad Hoc Networks,” IEEE Trans. Wireless Commun., vol. 8,
no. 3, pp. 1542–1552, March 2009.
[20] I. Pefkianakis, S. H. Y. Wong, and Songwu Lu, “SAMER: Spectrum
Aware Mesh Routing in Cognitive Radio Networks,” in Proc. IEEE
DySPAN’08, pp. 1–5, Chicago, IL, Oct. 2008.
[21] M. Timmers, S. Pollin, A. Dejonghe, L. Van der Perre and F. Catthoor,
“Distributed Multichannel MAC Protocol for Multihop Cognitive Radio
Networks,” IEEE Trans. Veh. Technol., vol. 59, no. 1, pp. 446–459, Jan.
2010.
[22] H. Khalife, S. Ahuja, N. Malouch, and M. Krunz, “Probabilistic Path
Selection in Opportunistic Cognitive Radio Networks,” in Proc. IEEE
Globecom’08, pp. 4861–4865, 2008.
[23] Y. Bi, L. X. Cai, X. Shen, and H. Zhao, “Efficient and Reliable Broadcast
in Inter-Vehicle Communications Networks: A Cross Layer Approach,”
IEEE Trans. Veh. Technol., vol. 59, no. 5, pp. 2404–2417, 2010.
[24] D. R. Cox, “Renewal Theory”, Butler and Tanner, 1967.
[25] J. P. Castro, “The UMTS Network and Radio Access Technology,” John
Wiley & Sons, 2001.
[26] K. Zeng, W. Lou, J. Yang, and D. R. Brown, “On Geographic Collaborative Forwarding in Wireless Ad Hoc and Sensor Networks,” in Proc.
WASA07, Chicago, IL, August 2007.
[27] Michigan Tech. Univ., “Cognitive Radio Cognitive Network Simulator,”
Available at
http://stuweb.ee.mtu.edu/ ljialian/index.htm.
[28] A. Sampath, L. Yang, L. Cao, H. Zheng, B.Y. Zhao, “High throughput
spectrum-aware routing for cognitive radio based ad hoc networks,” in
Proceedings of CROWNCOM’08, May 2008.
Yongkang Liu is currently working toward a Ph.D.
degree with the Department of Electrical and Computer Engineering, University of Waterloo, Canada.
He is currently a research assistant with the Broadband Communications Research (BBCR) Group,
University of Waterloo. He received the Best Paper
Award from IEEE Global Communications Conference (Globecom) 2011, Houston, USA. His general
research interests include protocol analysis and resource management in wireless communications and
networking, with special interest in spectrum and
energy efficient wireless communication networks.
1968
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 30, NO. 10, NOVEMBER 2012
Lin X. Cai received her M.A.Sc. and PhD degrees
in electrical and computer engineering from the
University of Waterloo, Ontario, Canada, in 2005
and 2010, respectively. She has been working as a
postdoctoral research fellow in electrical engineering
department at Princeton University since 2011. Her
research interests include green communication and
networking, resource management for broadband
multimedia networks, and cross-layer optimization
and QoS provisioning.
Xuemin (Sherman) Shen (IEEE M’97-SM’02F09) received the B.Sc.(1982) degree from Dalian
Maritime University (China) and the M.Sc. (1987)
and Ph.D. degrees (1990) from Rutgers University,
New Jersey (USA), all in electrical engineering.
He is a Professor and University Research Chair,
Department of Electrical and Computer Engineering,
University of Waterloo, Canada. He was the Associate Chair for Graduate Studies from 2004 to 2008.
Dr. Shen’s research focuses on resource management
in interconnected wireless/wired networks, wireless
network security, wireless body area networks, vehicular ad hoc and sensor
networks. He is a co-author/editor of six books, and has published more
than 600 papers and book chapters in wireless communications and networks,
control and filtering. Dr. Shen served as the Technical Program Committee
Chair for IEEE VTC’10 Fall, the Symposia Chair for IEEE ICC’10, the
Tutorial Chair for IEEE VTC’11 Spring and IEEE ICC’08, the Technical
Program Committee Chair for IEEE Globecom’07, the General Co-Chair for
Chinacom’07 and QShine’06, the Chair for IEEE Communications Society
Technical Committee on Wireless Communications, and P2P Communications
and Networking. He also serves/served as the Editor-in-Chief for IEEE Network, Peer-to-Peer Networking and Application, and IET Communications; a
Founding Area Editor for IEEE Transactions on Wireless Communications; an
Associate Editor for IEEE Transactions on Vehicular Technology, Computer
Networks, and ACM/Wireless Networks, etc.; and the Guest Editor for IEEE
JSAC, IEEE Wireless Communications, IEEE Communications Magazine,
and ACM Mobile Networks and Applications, etc. Dr. Shen received the
Excellent Graduate Supervision Award in 2006, and the Outstanding Performance Award in 2004, 2007 and 2010 from the University of Waterloo, the
Premier’s Research Excellence Award (PREA) in 2003 from the Province
of Ontario, Canada, and the Distinguished Performance Award in 2002 and
2007 from the Faculty of Engineering, University of Waterloo. Dr. Shen is
a registered Professional Engineer of Ontario, Canada, an IEEE Fellow, an
Engineering Institute of Canada Fellow, a Canadian Academy of Engineering
Fellow, and a Distinguished Lecturer of IEEE Vehicular Technology Society
and Communications Society.