Documente Academic
Documente Profesional
Documente Cultură
5, MAY 2013
Abstract—In this letter, a novel metric is proposed to evaluate sensing energy consumption by dividing the CTs into several
the average sensing energy-efficiency of cognitive radio (CR) subsets and activating one subset at certain period. A partial
networks. A theorem as well as the related closed-form expression spectrum sensing algorithm with decision result prediction and
for determining the optimum number of cooperative sensing
cognitive terminals is provided to optimize the proposed metric. modification techniques [7] was proposed to reduce the co-
According to the theorem, we propose an efficient terminal- operative spectrum sensing energy. However, previous works
assignment strategy for coordinated spectrum sensing. Simula- [5]– [7] merely focused on the minimization of sensing energy
tion results show that the proposed strategy can significantly consumption while failing to identify the metric that measures
improve the overall average sensing energy efficiency of the CR both sensing performance and energy consumption. The opti-
system.
mization of overall sensing energy efficiency performance in
Index Terms—Cognitive radio, energy-efficiency, coopera- multi-channel and multi-CT cases, i.e., coordinated spectrum
tive/coordinated spectrum sensing. sensing, is also scarcely discussed.
In this letter, a metric is proposed to evaluate the average
I. I NTRODUCTION sensing energy efficiency of cooperative spectrum sensing
scheme. To optimize the proposed metric, we provide a
C OGNITIVE radio (CR) [1] is a potential technique for
improving spectrum efficiency by detecting unoccupied
spectrum holes and assigning them to cognitive terminals
theorem as well as the related closed-form expression for
determining the optimum number of cooperative CTs. Fur-
(CTs). Therefore, spectrum sensing is an essential function of thermore, we extend this metric to measure the average
CR. On one hand, CT has to execute immediate and reliable sensing energy efficiency of the coordinated spectrum sensing
spectrum sensing to protect primary uses (PUs). Cooperative scheme. According to the theorem, a CT-assignment strategy
spectrum sensing has been proposed as an efficient approach for coordinated spectrum sensing is proposed to optimize the
to improve sensing reliability by fusing the sensing results overall average sensing energy-efficiency.
from different CTs and making a decision on the presence or
II. S YSTEM M ODEL
absence of a PU [2]. On the other hand, to obtain more spec-
trum opportunities, CR networks have to perform wideband In this letter, we consider a centralized CR system with
spectrum sensing, which is difficult for a single CT because of one cognitive base station (CBS) and L cooperative CTs. The
energy and hardware constraints. To address this problem, [3] CR system can opportunistically access a wideband spectrum
proposed a coordinated spectrum sensing scheme, in which the that can be divided into K non-overlapping narrow channels,
wideband spectrum is divided into several narrow channels, where L K. The CBS assigns CTs to detect different
after which CTs are assigned to detect different channels channels simultaneously, and one CT can detect one channel
simultaneously. in one sensing slot. Each CTs performs local spectrum sensing
In cooperative/coordinated sensing, higher sensing accuracy using energy detection. By combining the sensing results from
can be achieved by assigning more CTs to sense the channel. CTs, the CBS makes the final decision on the presence or
However, the participation of more CTs will result in the absence of a PU in each channel.
increased consumption of spectrum sensing and reporting Assuming that the PU signals are subject to Rayleigh
energy. Considering that the CT is typically battery-powered, fading channels, the probability of false alarm Pf a and the
energy efficiency is important for CR networks. average missed detection probability Pmd can respectively be
An optimal terminal-assignment strategy for a coordinated expressed as in [8].
sensing scheme was proposed in [4] to improve sensing Considering that [9] has proven that the OR rule always
performance. However, [4] does not consider the energy- outperforms the AND and majority rules in energy detector-
efficiency issue. The authors in [5] identified the optimal based cooperative spectrum sensing, we assume that the CBS
sensing interval and the number of sensors to minimize will apply the OR rule for fusion. Thus, the probabilities of
sensing energy consumption assuming that all CTs exhibit missed detection and false alarm of channel k are respectively
l,k
the same sensing performance. The authors in [6] reduced Qkmd = Pmd (1)
l∈Dk
Manuscript received January 17, 2013. The associate editor coordinating
the review of this letter and approving it for publication was O. Dobre. and
This work was supported in part by the Important National Science and Qkf a = 1 − (1 − Pfl,k
a ), (2)
Technology Specific Projects with No. (2010ZX03005-001-03), and the New
Century Excellent Talents in University (NCET) with NCET-11-0593. l∈Dk
The authors are with the Key Lab. of Universal Wireless Communications where Dk is the CT set assigned to detect channel k; and
for the Ministry of Education, Beijing University of Posts and Telecommu- l,k
nications, Beijing 100876, China (e-mail: huangdongyan-gua@163.com). Pmd and Pfl,k
a are the probability of missed detection and false
Digital Object Identifier 10.1109/LCOMM.2013.031913.130131 alarm of CT l for channel k, respectively.
1089-7798/13$31.00
c 2013 IEEE
HUANG et al.: ENERGY-EFFICIENT SPECTRUM SENSING STRATEGY IN COGNITIVE RADIO NETWORKS 929
In this letter, all local false alarm probabilities are set to 250 Pmd=0.1
ηEE(dB/Joule/CT)
l,k
Pmd = [Pmd ]L×K , where l = 1, 2, · · · , L, k = 1, 2, · · · , K. 100
50
III. S ENSING E NERGY-E FFICIENCY M ETRIC AND
M ATHEMATICAL A NALYSIS 0
TABLE I
s.t. Qkf a = 1 − (1 − Pf a ) ≤ Qf a , ∀ k; (12) C OMPUTATIONAL C OMPLEXITY C OMPARISON BETWEEN THE PROPOSED
xlk =1 ALGORITHM AND I TERATIVE H UNGARIAN ALGORITHM
K Algorithm Computation Complexity Instantiation
xlk ≤ 1, ∀ l, (13) (L, K, T ) =
k=1 (60, 10, 5)
Iterative Greedy O(T KL) 3000
where ηEE
k
is the average sensing energy-efficiency of channel Iterative Hungarian O(T L4 ) 64800000
k, whereas Qf a is the false alarm probability threshold.
Constraint (13) ensures that each CT can be only assigned TABLE II
SIMULATION PARAMETERS
to sense one channel in one sensing slot.
Parameter Value
B. Proposed Strategy Threshold of missed detection probability, Qf a 0.1
To solve problem (11), CBS has to determine: 1) how many Threshold of false alarm probability, Qmd 0.1
Number of channels, K 10
CTs should be assigned to sense each channel and 2) which Number of cognitive terminals, L Rang from 30 to 120
CTs should be assigned to sense a certain channel. α 0.01 J
When all CTs exhibit the same sensing performance for the Path loss exponent 3
Fading Rayleigh fading
same channel, the optimum number of assigned CTs can easily
be derived based on Theorem 1 given in Section III. However,
when different CTs have different Pmd for different channels,
the optimum solution can be achieved through an exhaustive The weaker CTs are then discarded to equate the number
search, which is impractical because of its high complexity. of assigned CTs to the theoretical optimum CT number. A
In this letter, we propose a low-complexity coordinated detailed description of Stage 3 is given as follows:
spectrum sensing strategy called Energy-Efficient (E-E) Iter- Step 1: For channel k, update Nopt k
by (9) using the
ative Greedy strategy. The proposed strategy comprises three k 1
L l,k lk
assigned CTs’ P md = N k l=1 Pmd x , where Nasgn is
k
stages: asgn
the number of CTs assigned to channel k.
Stage 1: Initialization:
Step 2: If Noptk
< Nasgnk
, then discard Nasgn
k
− Nopt
k
Input Pmd , K, L, Pf a , Qmd , Qf a .
CTs with higher Pmd from the assignment subset, update the
Each channel’s initial theoretical optimum number of CTs
corresponding element of assignment matrix X as 0, and then
N0k is calculated by (9) using average missed detection prob-
k k L go back to step 1. Otherwise, go to step 3.
ability P md , where P md = L1 l=1 Pmd l,k
. Step 3: Given that Noptk
= Nasgn
k
, k = 1, 2, · · · , K, the
max max
Initialize Nopt = min{N0 , Ncoop }, where Ncoop
k k
= algorithm terminates and outputs X as final assignment result,
log1−Pf a (1 − Qf a ) is the maximum number of CTs co- or update k = k + 1, and go to step 1.
operatively sensing one channel to maintain Qf a . In Step 2 of Stage 2, pi,· = 1 ensures that the assigned CT
Initialize the assignment matrix X = [xlk ]L×K as X = 0; is unavailable for next steps.
the channel index k = 1; the operation matrix P = [pl,k ]L×K
as P = Pmd ;
Calculate the initial assignment iterative time t = C. Complexity Analysis
max{Nopt k
}. From the description above, the computational complexity
∀k
Stage 2: Assignment, where CTs are assigned to sense of the proposed algorithm is based on the number of com-
a channel in an iterative manner. In each iterative step, the parisons. The complexities of Stages 2 and 3 are respectively
greedy algorithm is run, and one CT is allocated to each O(T KL) and O(KT log T ), where T = max{Nopt k
}. Consid-
∀k
channel in turn. The assigned CTs are unavailable in the next ering that L Nopt
k
, the complexity of the proposed strategy
steps. Provided that Nopt k
CTs have been assigned to detect is approximately O(T KL).
channel k, no CT is assigned to channel k in the next iterative Table I shows that the computation complexity of the
steps. A detailed description of Stage 2 is given as follows: proposed algorithm is significantly lower than that of the
Step 1: If Nopt
k
CTs have been assigned to channel k, then Iterative Hungarian algorithm.
go to step 3; else, run the greedy algorithm for P to assign
the CT with the lowest Pmd to sense channel k, and then go
to step 2. V. S IMULATION R ESULTS
Step 2: For every CT (denoted as i) assigned to sense Considering demand of wideband sensing and CT hardware,
channel k in previous step, set pi,· = 1 and xik = 1. the entire spectrum of interest is divided into 10 disjoint
Step 3: If k = K, go to step 4; else, update k = k + 1, and narrow channels in the simulation scenarios. We assume that
then go to step 1. only one channel is licensed to one PU.
Step 4: If t > 1, update t = t − 1 and set k = 1 then go to Based on IEEE 802.22 [10], the CTs are randomly dis-
step 1 to continue the iteration; else, set k = 1, and then go tributed within a 30 km radius of the CBS, and the CBS is
to Stage 3. approximately 150 km away from each PU. Pf a is set as 0.01
Stage 3: Modification, where the theoretical optimum num- for all the local sensing [8]. Simulation parameters are listed
ber of CTs is updated by (9) using the assigned CTs’ P md . in Table II.
HUANG et al.: ENERGY-EFFICIENT SPECTRUM SENSING STRATEGY IN COGNITIVE RADIO NETWORKS 931
70
Fig. 2 shows that the performance of the proposed strategy
Case I Case II
65 can approach that of the E-E iterative Hungarian algorithm-
Average ηEE (dB/J/CT) over the channels