IEEE COMMUNICATIONS LETTERS, VOL. 14, NO.

2, FEBRUARY 2010 133

Investigation on Multiuser Diversity in

Spectrum Sharing Based Cognitive Radio Networks
Rui Zhang, Member, IEEE, and Ying-Chang Liang, Senior Member, IEEE

Abstract—A new form of multiuser diversity, named multiuser as a function of the number of CRs, �. It is shown that due to
interference diversity, is investigated for opportunistic communi- the newly discovered MID, the MDG (as defined in this letter)
cations in cognitive radio (CR) networks by exploiting the mutual for each CR network under consideration is no smaller than
interference between the CR and the existing primary radio
(PR) links. The multiuser diversity gain and ergodic throughput that in the conventional network without the PR link, which
are analyzed for different types of CR networks and compared holds for arbitrary fading channel distributions and values of
against those in the conventional networks without the PR link. �. On the other hand, it is also shown that the MID results in
Index Terms—Cognitive radio, interference temperature, mul- the same asymptotic growth order over � for the CR network
tiuser diversity, spectrum sharing. ergodic throughput as that due to the MD for the conventional
network, as � goes to infinity.
It is noted that MD for the fading CR networks has
I. I NTRODUCTION been recently studied in, e.g., [4] and [5]. In [4], MD was
Ognitive radio (CR) is a promising technology for ef- analyzed for the fading CR MAC with a given � at the
C ficient spectrum utilization in the future wireless com-
munication systems. The main design objective for a CR
asymptotic regime where the ratio between the CR transmit
power constraint and interference power constraint goes to
network is to maximize its throughput while providing suf- infinity. In [5], MD was investigated for the fading CR PAC
ficient protection to the existing primary radio (PR) network. with the constant CR transmit power at the asymptotic region
There are two basic operation models for CRs: opportunistic with � → ∞. In this letter, we study MD for different types
spectrum access (OSA), where the CR is allowed to transmit of CR networks with arbitrary CR transmit and interference
over a frequency band only when all the PR transmissions are power constraints, and arbitrary number of CR users.
off, and spectrum sharing (SS), where the CR can transmit
concurrently with PRs, provided that it knows how to control II. S YSTEM M ODEL
the resultant interference powers at PRs below a tolerable
Consider a SS-based CR network coexisting with a PR net-
threshold [1]. On the other hand, “multiuser diversity (MD)”
work. For the propose of exposition, only one active PR link
[2], [3], as a fundamental property of wireless networks,
consisting of a PR transmitter (PR-Tx) and a PR receiver (PR-
has been widely applied for opportunistic communications in
Rx) is considered. All terminals in the network are assumed
wireless systems. The conventional form of MD is usually
to be each equipped with a single antenna. We consider the
exploited in a wireless system with multiple independent-
block-fading (BF) model for all the channels involved and
fading communication links by selecting one link with the best
coherent communications; thus, only the fading channel power
instantaneous channel condition to transmit at one time, also
gains (amplitude squares) are of interest. It is assumed that
known as dynamic time-division multiple-access (D-TDMA).
the additive noises at all PR and CR receive terminals are
In this letter, we consider SS-based fading CR networks
independent circular symmetric complex Gaussian (CSCG)
and investigate for them a new form of MD, named multiuser
random variables each having zero mean and unit variance.
interference diversity (MID), which is due to the CR and PR
We consider three types of CR networks described as follows.
mutual interference and thus different from the conventional
∙ Cognitive MAC (C-MAC), where � CRs, denoted by
MD. More specifically, “transmit MID (T-MID)” is due to
the fact that the maximum transmit powers of different CR CR1 , CR2 , . . ., CR� , transmit independent messages to
transmitters vary with their independent-fading channels to a CR base station (CR-BS). Denote �� as the power gain
a PR receiver under a given interference power constraint, of the fading channel from CR� to CR-BS, � = 1, . . . , �.
while “receive MID (R-MID)” is due to the fact that the Similarly, � � is defined for the fading channel from CR�
PR interference powers at different CR receivers vary with to PR-Rx, � is for that from PR-Tx to PR-Rx, and � is
their independent-fading channels from a PR transmitter. This for that from PR-Tx to CR-BS.
∙ Cognitive BC (C-BC), where CR-BS sends independent
letter studies the MID for various types of CR networks
with D-TDMA, including the multiple-access channel (MAC), messages to � CRs. Channel reciprocity is assumed
broadcast channel (BC), and parallel-access channel (PAC), between the C-MAC and C-BC; thus, the fading channel
and analyzes the achievable multiuser diversity gain (MDG) power gain from CR-BS to CR� in the C-BC is the same
as �� in the C-MAC. In addition, � is defined as the
Manuscript received August 7, 2009. The associate editor coordinating the fading channel power gain from CR-BS to PR-Rx, and
review of this letter and approving it for publication was B. Chen. �� is defined for the channel from PR-Tx to CR� .
The authors are with the Institute for Infocomm Research, A*STAR,
Singapore (e-mail: {rzhang, ycliang}@i2r.a-star.edu.sg). ∙ Cognitive PAC (C-PAC), where � distributed CR trans-
Digital Object Identifier 10.1109/LCOMM.2010.02.091649 mitters, denoted by CR-Tx1 , . . ., CR-Tx� , transmit inde-
c 2010 IEEE
1089-7798/10$25.00 ⃝

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
134 IEEE COMMUNICATIONS LETTERS, VOL. 14, NO. 2, FEBRUARY 2010

pendent messages to the corresponding receivers, denoted where

by CR-Rx1 , . . ., CR-Rx� , respectively. Similarly like the
�MAC = 1/(�[ℎ� ]�[min(�, Γ/�� )]) (9)
C-MAC and C-BC, we denote �� as the power gain of the
fading channel from CR-Tx� to CR-Rx� , � = 1, . . . , �; �BC = 1/(�[ℎ� ]�[1/(1 + ��� )]) (10)
� � and �� are defined for the fading channels from CR- �PAC = 1/(�[ℎ� ]�[min(�, Γ/�� )]�[1/(1 + ��� )]) (11)
Tx� to PR-Rx and from PR-Tx to CR-Rx� , respectively.
are constants, which can be shown independent of � due to the
Let � be the peak (with respect to fading states) transmit
symmetric assumptions. In order to fairly compare the MDG
power constraint at CR-BS in the C-BC, and �� be that at CR�
in each CR network with that in the conventional network,
in the C-MAC or at CR-Tx� in the C-PAC. It is assumed
we introduce a reference network by removing the PR link in
that PR-Tx transmits with a constant power �; and each
each CR network. For such reference networks, since there is
CR transmit terminal protects the PR by applying the peak
no interference power constraint nor interference from PR-Tx
interference power constraint at PR-Rx, denoted by Γ. Let ��
to CR terminals, it is easy to show that CR � with the largest
be the transmit power for CR � in the C-MAC or C-PAC, and
ℎ� among all the CRs should be scheduled for transmission at
� be that for CR-BS in the C-BC. Let �� be one realization of
each fading state due to D-TDMA. Thus, we define the MDG
� � for a particular fading state (similar notations apply for the
of the reference network as
other channels). Combining both the transmit and interference
power constraints, we obtain �¯0 (�) = �0 �[max ℎ� ] (12)
�
�� ≤ min(�� , Γ/�� ), ∀�, � ≤ min(�, Γ/�). (1) where �0 = 1/�[ℎ� ] is a constant, which is independent of �.
The maximum achievable receiver signal-to-noise ratio (SNR) From (6)-(12), it follows that the MDGs for different
of CR � can then be expressed as CR networks all differ from the conventional MDG for the
ℎ� � � ℎ� � ℎ� �� reference network. We highlight their differences as follows.
(MAC) (BC) (PAC)
�� = ,� = ,� = ∙ For the C-MAC, it is observed from (6) that the MDG is
1 + �� � 1 + ��� � 1 + ���
(2) obtained by taking the maximum product between ℎ� and
min(�, Γ/�� ) over all �’s, where the former also exists
for the C-MAC, C-BC, and C-PAC, respectively. Note that the
in the conventional MDG given in (12), while the latter
noise at each CR receive terminal includes both the additive
is a new term due to independent �� ’s over which CR
Gaussian noise and the interference from PR-Tx. We are then
transmitters interfere with PR-Rx, thus named T-MID.
interested in maximizing the long-term system throughput in
∙ For the C-BC, it is observed from (7) that the MDG is
each CR network by adopting the following D-TDMA rule: At
obtained by taking the maximum product between ℎ� and
one particular fading state, CR � is selected for transmission if
1/(1+��� ) over CRs, where the former term contributes
it has the largest achievable receiver SNR among all the CRs.
to the conventional MDG, while the latter term is a new
Let � ∗ denote the selected user at this fading state. From (1)
source of diversity due to independent �� ’s over which
and (2), it then follows that
PR-Tx interferes with CR receivers, thus named R-MID.
∗ ℎ� min(�� , Γ/�� ) ∙ For the C-PAC, it follows from (8) that in addition to
�MAC = arg max (3)
�∈{1,...,�} 1 + �� the conventional MD, there are combined T-MID and R-
∗ ℎ� min(�, Γ/�) MID.
�BC = arg max (4)
�∈{1,...,�} 1 + ��� Next, we show the following theorem, which says that the
∗ ℎ� min(�� , Γ/�� ) MDG of each CR network for a given � is lower-bounded by
�PAC = arg max (5) that of the reference network, �¯0 (�); and is upper-bounded
�∈{1,...,�} 1 + ���
by a constant (greater than one) multiplication of �¯0 (�).
for the C-MAC, C-BC, and C-PAC, respectively. By substitut-
Theorem 3.1: Under the symmetric assumptions, we have
ing � ∗ for each CR network into (2), the corresponding max-
�¯0 (�) ≤ �¯MAC (�) ≤ �MAC �¯0 (�), �¯0 (�) ≤ �¯BC (�) ≤
imum receiver SNR over CR users is obtained as �MAC (�),
�BC �¯0 (�), and �¯0 (�) ≤ �¯PAC (�) ≤ �PAC �¯0 (�), where
�BC (�), or �PAC (�), each as a function of �.
�MAC = �/�[min(�, Γ/�� )], �BC = 1/�[1/(1 + ��� )], and
III. M ULTIUSER I NTERFERENCE D IVERSITY �PAC = �MAC ⋅ �BC are all constants.
Proof: We only show the proof for the C-PAC case, while
In this section, we study the MD for different CR net-
similar proofs can be obtained for the C-MAC and C-BC and
works under a set of “symmetric” assumptions, where all
are thus omitted here. First, we consider the lower bound on
�� ’s are assumed to have the same distribution, so are � � ’s
�¯PAC (�). By denoting � ′ as the user with the largest ℎ�
and �� ’s; and �� = �, ∀�. We then define �¯MAC (�) ≜
among all the CRs, it follows that
�[�MAC (�)]/�[�MAC (1)] as the MDG for the C-MAC;
�¯BC (�) and �¯PAC (�) are similarly defined for the C-BC and (�)
�¯PAC (�) ≥ �PAC �[ℎ�′ min(�, Γ/��′ )/(1 + ���′ )]
C-PAC, respectively. From (3), (4), and (5), it follows that
(�)
�¯MAC (�) = �MAC �[max ℎ� min(�, Γ/�� )] (6) = �PAC �[ℎ�′ ]�[min(�, Γ/��′ )]�[1/(1 + ���′ )]
� (�)
�¯BC (�) = �BC �[max ℎ� /(1 + ��� )] (7) = �¯0 (�)
�
�¯PAC (�) = �PAC �[max ℎ� min(�, Γ/�� )/(1 + ��� )] (8) where (�) is due to the fact that � ′ is in general not the optimal
� � ∗ corresponding to the largest ℎ� min(�, Γ/�� )/(1 + ��� )

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.
ZHANG and LIANG: INVESTIGATION ON MULTIUSER DIVERSITY IN SPECTRUM SHARING BASED COGNITIVE RADIO NETWORKS 135

in (8); (�) is due to independence of channels ℎ�′ , ��′ , and 4.5

��′ ; and (�) is due to (11) and (12). Next, we consider the C MAC
C BC
upper bound on �¯PAC (�). The followings can be shown. 4 C PAC
Reference Network

(�)
�¯PAC (�) ≤ �PAC �[max ℎ� ⋅ max min(�, Γ/�� )/(1 + ��� )] 3.5

Normalized Ergodic Throughput

� �
R MID
(�)
= �PAC �[max ℎ� ]�[max min(�, Γ/�� )/(1 + ��� )] 3 T MID
� �
(�) Multiuser Interference Diversity (T MID & R MID)
≤ �PAC �[max ℎ� ]� 2.5
�
(�)
= �¯0 (�)�/(�[min(�, Γ/�� )]�[1/(1 + ��� )]) 2

Conventional Multiuser Diversity

where (�) is due to the fact that the user with the largest ℎ� 1.5

is not necessarily the one with the largest min(�, Γ/�� )/(1 +
��� ); (�) is due to independence of ℎ� and (�� , �� ); (�) is due 1
0 10 20 30 40 50 60 70 80 90 100
to the fact that min(�, Γ/�� )/(1 + ��� ) ≤ �, ∀�; and (�) is Number of CRs (K)

due to (11) and (12). Using the definitions of �MAC and �BC Fig. 1. Normalized ergodic throughput for CR networks with � ≤ 100.
given in Theorem 3.1, it follows that �¯PAC (�) ≤ �PAC �¯0 (�),
where �PAC = �MAC ⋅ �BC . 4

At last, we study the ergodic throughput of each D-TDMA C MAC

3.8
based CR network and its asymptotic growth order over � C BC
C PAC
as � → ∞. For the C-MAC, the ergodic throughput for a 3.6 Reference Network

given � is defined as �MAC (�) = �[log2 (1 + �MAC (�))]; Ergodic Throughput (bps/Hz)

similarly, �BC (�), �PAC (�), and �0 (�) are defined for the 3.4

C-BC, C-PAC, and reference network, respectively. According 3.2

to the extreme value theory [6, Appendix A], it is known that
�0 (�) behaves as log�0 �(�)(�) → 1 as � → ∞, where � (�)
3
2
is given by the distribution of max� �� as � → ∞ (e.g., for
2.8
“type i” distribution of �� with unit-variance, � (�) = log �
[6]). Then, from Theorem 3.1, the following corollary can be 2.6
obtained (proof is omitted here due to the space limitation).
Corollary 3.1: Under the symmetric assumptions, as � → 2.4

MAC (�) �BC (�) �PAC (�)

∞, we have � log2 � (�) = 1, log2 � (�) = 1, and log2 � (�) = 1. 2.2
2.8 2.9 3 3.1 3.2 3.3 3.4 3.5 3.6 3.7
Corollary 3.1 says that the MID results in the same ergodic- log2(log(K))

throughput asymptotic growth order over � for the CR Fig. 2. Asymptotic ergodic throughput for CR networks as � → ∞.
networks as that of the conventional MD for the reference
network, as � → ∞, regardless of the fading distribution.
V. C ONCLUSION
IV. N UMERICAL R ESULTS This letter quantified a new form of MID for SS-based CR
networks by exploiting the CR and PR mutual interference.
We assume that all the channels involved follow the stan- The MDG and ergodic throughput for opportunistic commu-
dard (unit-power) Rayleigh fading distribution. In addition, for nications in different types of CR networks were analyzed.
a fair comparison of different CR networks and the reference
network, we assume that � = � = � = Γ = 1. Fig. R EFERENCES
1 shows the ergodic throughput for different networks with
[1] R. Zhang, S. Cui, and Y. C. Liang, “On ergodic sum capacity of
� ≤ 100, after normalizing it to the ergodic throughput with fading cognitive multiple-access and broadcast channels,” IEEE Trans.
� = 1 in order to better examine the MDG. It is observed that Inf. Theory, vol. 55, no. 11, pp. 5161-5178, Nov. 2009.
the normalized ergodic throughput for different CR networks [2] R. Knopp and P. A. Humblet, “Information capacity and power control in
single-cell multi-user communications,” in Proc. IEEE Int. Conf. Comm.
is larger than that for the reference network, thanks to the (ICC), pp. 331-335, 1995.
newly discover MID. It is also observed that the combined [3] P. Viswanath, D. Tse, and R. Laroia, “Opportunistic beamforming using
T-MID and R-MID in the C-PAC result in more substantial dumb antennas,” IEEE Trans. Inf. Theory, vol. 48, no. 6, pp. 1277-1294,
June 2002.
throughput gains than T-MID in the C-MAC or R-MID in [4] T. W. Ban, W. Choi, B. C. Jung, and D. K. Sung, ”Multi-user diversity
the C-BC. Fig. 2 shows the ergodic throughput (without in a spectrum sharing system,” IEEE Trans. Wireless Commun., vol. 8,
normalization) of different networks versus log2 (log �) (Note no. 1, pp. 102-106, Jan. 2009.
[5] A. Tajer and X. Wang, “Multiuser diversity gain in cognitive networks
that � (�) = log � in this case.) for very large values of � with distributed spectrum access,” in Proc. CISS, Mar. 2009.
ranging from 103 to 106 . It is observed that all these networks [6] M. Sharif and B. Hassibi, “On the capacity of MIMO broadcast channel
have the same asymptotic growth order over � for the ergodic with partial side information,” IEEE Trans. Inf. Theory, vol. 51, no. 2,
pp. 506-522, Feb. 2005.
throughput, which is in accordance with Corollary 3.1.

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:53 UTC from IE Xplore. Restricon aply.