IEEE COMMUNICATIONS LETTERS, VOL. 17, NO. 4, APRIL 2013 705
A Jain’s Index Perspective on α-Fairness Resource Allocation over
Slow Fading Channels
Chongtao Guo, Min Sheng, Member, IEEE, Yan Zhang, Member, IEEE, and Xijun Wang, Member, IEEE
Abstract—In this letter, by focusing on α-fairness resource
allocation in single channel TDMA system with slow Rayleigh
fading, we study the impacts of α on the system efficiency and
Jain’s index which is widely used as a fairness measure. Also,
we prove that increasing α results in higher efficiency but lower
Jain’s index and reveal the role of α in controlling the tradeoff
between efficiency and fairness. Furthermore, an algorithm
is proposed to find the optimal α under a system efficiency
constraint or a Jain’s index constraint with fast convergence
and low complexity.
Index Terms—Resource allocation, α-fairness, system effi-
ciency, Jain’s index, tradeoff.
I. INTRODUCTION
R
ESOURCE Allocation in wireless communication sys-
tems aims to increase system efficiency and simultane-
ously improve user fairness. However, these two goals are
inversely related as one waxes the other wanes [1]. To deal
with this dilemma, many studies focused on maximizing α-
utility function, which further results in α-fairness resource
allocation [2]–[5].
It is intuitively accepted that larger α yields resource allo-
cations that are more fair but less efficient [2], [6]. However,
this is not always true [7], [8]. For instance, [7] gave the
counter-intuitive examples that increasing α brings about more
efficient allocations in networks under end-to-end flow control.
On the other hand, increasing α may decrease fairness if the
admissible resource allocation set is discrete [8]. Although
different outcomes arise in different scenarios as above, they
can be unified in [5]. By investigating the α-utility function,
[5] showed that the fairness-efficiency reward ratio, i.e., the
ratio between the utility function gradient in the direction of
more fairness and that in the direction of more efficiency,
is nondecreasing with α. In particular, the results in [5]
indicate three possible cases: 1) fairness increases with α while
efficiency decreases with α (e.g., [2] and [6]), 2) both fairness
and efficiency increase with α (e.g., [7]) and 3) both fairness
and efficiency decrease with α (e.g., [8]). Nevertheless, there
is a little ambiguity to apply the conclusions in [5] to specific
resource allocation problems in practical systems.
Manuscript received January 5, 2013. The associate editor coordinating the
review of this letter and approving it for publication was D. W. K. Ng.
This paper is supported by the National Natural Science Foundation of
China (61231008,61172079,61201141), the 973 Program (2009CB320404),
the 111 Project (B08038), the National S&T Major Project
(2012ZX03002009-003, 2012ZX03004002-003), and the Shaanxi Province
Science and Technology Research and Development Program (2011KJXX-
40).
The authors are with the Institute of Information Science, State Key
Laboratory of Integrated Service Networks, Xidian University, Xi’an 710071,
China (e-mail: {ctguo, msheng}@mail.xidian.edu.cn, {yanzhang, xijun-
wang}@xidian.edu.cn).
Digital Object Identifier 10.1109/LCOMM.2013.021913.130025
In the literature, proportional fairness has been exten-
sively considered in single channel Time-Division-Multiple-
Access (TDMA) system [9], [10] and Orthogonal-Frequency-
Division-Multiple-Access (OFDMA) system [11], [12]. How-
ever, in these systems, it is still unknown how α influences
system efficiency and Jain’s index under α-fairness resource
allocation, where Jain’s index is widely utilized to quantify
fairness among various fairness measures due to its highly
intuitive interpretation
1
[4]–[6], [8]. In this letter, we focus
on the tradeoff between system efficiency and Jain’s index
by maximizing α-utility function in single channel TDMA
system, where the conclusions can also be applied to multi-
channel system, such as OFDMA, with frequency-flat fading.
For frequency selective multi-channel system, one can refer
to [14] and we will concentrate upon it in the future work.
The main contributions of this letter are listed as follows.
First, the explicit expressions for system efficiency and Jain’s
index are derived under α-utility maximization. Second, we
prove that increasing α results in a strict increase in the
Jain’s index and a strict decrease in the efficiency. Finally,
a low-complexity and fast converging algorithm is proposed
to optimize α subject to a given system efficiency constraint
or a given Jain’s index constraint.
II. S
YSTEM MODEL
Consider a downlink single channel wireless communica-
tion system with one transmitter (the base station) and K
receivers (users), where users are scheduled in a TDMA
fashion with time slot lengths of T
s
. Denote user scheduling
indicator at the ith slot by I
k
(i),i.e.,I
k
(i)=1when user k
is scheduled at the ith slot and 0 otherwise.
Frequency-flat slow fading Rayleigh channels are con-
sidered. The instantaneous channel fading coefficients be-
tween the BS and users, denoted by h
k
for user k (k =
1, 2, ··· ,K), are independent and identically distributed
(i.i.d.) with CN(0, 1), and remain constant during the observa-
tion time for measuring long-term performance. For simplicity,
we let the observation time equal N time slots. It is assumed
that Channel State Information (CSI) is available at the BS.
The achievable data rate of user k is r
k
= W log(1 + γ|h
k
|
2
),
where W and γ are the system bandwidth and the transmit
signal-noise-ratio (SNR), respectively. Without loss of gen-
erality, we normalize W =1and γ =1.Letr
max
=
max {r
k
|k =1, 2, ··· ,K} and denote by K
m
the number of
users having the maximum achievable data rate.
The long-term throughput of user k during the observation
time is R
k
=
1
NT
s
N
i=1
I
k
(i)T
s
r
k
=
N
i=1
I
k
(i)
N
r
k
,where
1
The value of Jain’s index, bounded between 0 and 1, can be considered
as the fraction of favored users [13].
1089-7798/13$31.00
c
2013 IEEE