Anda di halaman 1dari 12

1460

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

SINR and Throughput Analysis for Random


Beamforming Systems with Adaptive Modulation
Chanhong Kim, Member, IEEE, Soohong Lee, and Jungwoo Lee, Senior Member, IEEE

AbstractIn this paper, we derive the exact probability


distribution of post-scheduling signal-to-interference-plus-noise
ratio (SINR) considering both user feedback and scheduling.
We also develop an optimized adaptive modulation scheme
in orthogonal random beamforming systems with M transmit
antennas and K single-antenna users. The exact probability
distributions of each users feedback SINR and the exact postscheduling SINR are derived rigorously by direct integration
and multinomial distribution. It is also shown that the derived
cumulative distribution function (CDF) of the post-scheduling
SINR happens to be identical to the the existing approximate
CDF for SINR higher than 0 dB. The closed form expressions of
system performance, such as average spectral efficiency (ASE)
and average bit error ratio (A-BER), are derived using the CDF
of the post-scheduling SINR. The optimal SINR thresholds that
maximize the ASE with a target A-BER constraint are solved
using the derived closed form CDF and a Lagrange multiplier.
Key contributions of this paper include the derivation of the
exact CDF of post-scheduling SINR by direct integration, and
its application to an optimized adaptive modulation based on
a Lagrange multiplier. Simulations show the correspondence
between theoretical and empirical CDFs, and the performance
improvement of the proposed adaptive modulation method in
terms of ASE.
Index TermsAdaptive modulation, average spectral efficiency, cumulative distribution function, orthogonal random
beamforming, post-scheduling SINR.

I. I NTRODUCTION

ODAYS wireless communication systems demand high


data rate, which poses stringent requirements on spectral
efficiency. One way to cope with the demand for high spectral
efficiency is a multiple-input multiple-output (MIMO) technique, where data rate improvement can be achieved through
spatial multiplexing scheme [1], [2], and it has been further
extended to multiuser (MU) schemes, in which simultaneous
transmission to multiple users is used. Especially, in multiple antenna broadcast channels, MU-MIMO systems such
as space division multiple access offer higher throughput
than single user (SU) MIMO time division multiple access

Manuscript received September 7, 2011; revised December 27, 2011 and


August 30, 2012; accepted January 20, 2013. The associate editor coordinating
the review of this paper and approving it for publication was Y. Jing.
C. Kim and J. Lee are with the Department of Electrical Engineering and
Computer Sciences, Seoul National University, Seoul, 151744, Korea (email: chkim@wspl.snu.ac.kr, junglee@snu.ac.kr).
S. Lee is with the Department of Mathematical Sciences, Seoul National
University, Seoul, 151744, Korea (e-mail: sh2662@gmail.com).
This research was supported in part by the Basic Science Research Program (20100013397) and Mid-career Researcher Program (20100027155)
through the National Research Foundation of Korea (NRF) funded by the
Ministry of Education, Science and Technology, Seoul R&BD Program
(JP091007, 042320090051), the Institute of New Media & Communications
(INMAC), and the BK21 program.
Digital Object Identifier 10.1109/TWC.2013.022113.111663

systems [3]. Another way is a link adaptation technique, where


transmission parameters such as modulation and coding are
dynamically adapted to varying channel conditions [4]. A
typical link adaptation technique is adaptive modulation in
which an adequate modulation level is selected according to
the current signal-to-noise ratio (SNR).
The capacity region of a Gaussian MIMO broadcast channel
can be achieved by an optimal transmission strategy called
dirty paper coding (DPC) [5], and it has also been shown that
the sum capacity of that channel with DPC scales at a rate of
M log log K as K tends to infinity, where M is the number of
transmit antennas and K is the number of users [6]. However,
the DPC scheme requires prohibitively high implementation
complexity, so several low complexity MU-MIMO schemes
such as zero-forcing beamforming (ZFBF) and orthogonal
random beamforming (ORBF) have been proposed [7], [8].
These two schemes have the same scaling law as the sumrate capacity of DPC even with limited feedback [8], [9].
Although ZFBF can achieve a large fraction of DPC capacity
with reduced complexity, it also requires accurate channel
state information at the transmitter (CSIT) as in DPC. Instead,
ORBF only requires a partial CSIT. In an ORBF system, M
orthogonal random beams are constructed and information is
transmitted to the users with the highest signal-to-interferenceplus-noise ratios (SINRs). Since each receiver is only required
to feed back its maximum SINR along with the corresponding
beam index, the amount of feedback is small. Due to this
advantage, a modified ORBF system, which is called per user
unitary and rate control (PU2RC) in the industry, was proposed
to the next-generation wireless standards [10].
Since adaptive modulation can be applied to most of wireless communication systems, it has been proposed for singleinput single-output systems [11][13], and further extended to
SU-MIMO systems [14][21] to adjust the modulation order
based on the channel condition. In [14], a unified expression
for the approximate BER was proposed, which has been used
in order to derive closed form expressions of system performance, such as average spectral efficiency (ASE) and average
bit error ratio (A-BER) [15], [18][21]. Especially, in [18] and
[21], the authors tried to obtain the optimal SNR thresholds for
selecting modulation order using a Lagrange multiplier based
on the distribution of the post-processing SNR. In MU-MIMO
systems, the probability distribution of the post-scheduling
SINR, which corresponds to the post-processing SNR of SUMIMO systems, has not been known yet except for ORBF.
Moreover, even in the ORBF literature [8], [22][25], the
approximate cumulative distribution function (CDF) of the
post-scheduling SINR has been used.

c 2013 IEEE
1536-1276/13$31.00 

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

In this paper, we try to apply adaptive modulation techniques in [18] and [21] to an ORBF system with M transmit
antennas and K single-antenna users. At first, we derive the
exact CDF of the post-scheduling SINR by direct integration,
and apply it to derive the ASE and the A-BER. We then
obtain the optimal SINR thresholds for adaptive modulation.
Although the aggregated throughput derived in [24] is similar
to the ASE in this paper, in order to obtain the optimal SINR
thresholds based on the A-BER approach, new expressions of
the ASE and the A-BER are derived. Unique contributions of
this paper include the following.
1) The CDF of each users feedback SINR is derived by
direct integration.
2) The CDF of the post-scheduling SINR is derived from
a multinomial approach. It turns out that the existing
approximate CDF [8] happens to be exact for SINR
higher than 0 dB.
3) Optimal adaptive modulation is developed to maximize
the ASE performance of an ORBF system with a finite
set of modulation order.
The remainder of this paper is organized as follows. Section II describes the signal model and the user scheduling
method. In Section III, the exact CDF of each users feedback
SINR and the exact CDF of the post-scheduling SINR are
derived. In Section IV, adaptive modulation schemes with
instantaneous BER (I-BER) and A-BER constraints are described. The closed form expressions for the ASE and the ABER are derived as well. Simulations show the correspondence
between theoretical and empirical CDFs, and performance improvement in terms of ASE in Section V. Finally, Section VI
concludes the paper.
II. S YSTEM OVERVIEW
A. Signal Model
We consider a multiple-antenna Gaussian broadcast channel
consisting of a transmitter equipped with M antennas, and K
receivers with one antenna. M orthonormal complex M 1
vectors wm (m = 1, . . . , M ) for random beamforming are
generated according to an isotropic distribution. Although
wm s are chosen randomly per given channel use, they are
fixed during that period. All of wm s are known at the
transmitter and the receivers. Let sm be the mth transmitted
symbol, which is the scheduled users QAM symbol at the
mth beam. Let Es be the symbol energy, i.e. E[|sm |2 ] = Es .
At every symbol duration, the mth vector is multiplied by the
mth transmitted symbol, so that the transmitted signal vector
is formed by the sum of all the symbols. Let the transmitted
signal vector be s which can be expressed by
s=


1
M

M


wm s m ,

(1)

m=1

1
where
M is a normalization factor to fix the average
transmitted power to Es , i.e. E[sH s] = Es . Equal power
allocation is also assumed. Note that the M elements of the
M 1 vector s will be transmitted from the M transmit
antennas simultaneously.

1461

Let the received signal of the kth user be yk , which is given


by
yk = hTk s + nk ,
k = 1, . . . , K,
(2)
where hk is an M 1 flat fading channel vector which has i.i.d.
complex Gaussian random entries with distribution CN (0, 1),
and nk is an i.i.d. complex additive white Gaussian noise
(AWGN) with distribution CN (0, N0 ). Note that ()T means
a matrix transpose.
Our system model is equivalent to that of [8] except for
power normalization. It is also assumed that all the users have
Es
, and that the kth receiver has perfect
the same average SNR N
0
estimation of hk .
B. User Scheduling and Post-scheduling SINR
Assuming that sm is the kth users desired signal and the
other sl s (l = m) are interference, yk can be written as

 
1 T
1
yk = M
hk w m s m + M
hTk wl sl + nk .
(3)
l=m

Now we can calculate M SINRs of the kth user as [8, Eq.


(5)]
 
2 
 1 T

E  M hk w m s m 
 
SINRk,m =
2 

 1 T

N0 + l=m E  M
hk w l s l 
(4)
 T
2
h wm 
k
=

 , m = 1, . . . , M,

M
 hT w l  2
+

l=m

Es
where is the average SNR defined by = N
.
0
Each receiver feeds back its maximum SINR denoted by k
along with the corresponding index ik . Thus, k and ik are
written as
(5)
k = max SINRk,m ,
1mM

ik = arg max SINRk,m .


1mM

(6)

The transmitter assigns sm to the user with the highest SINR


among candidates whose feedback index is m. Then, the postscheduling SINR, denoted by m , can be written as
m = max k ,
kKm

m = 1, . . . , M,

where Km is the user index set defined by


 

Km = k  ik = m, k = 1, . . . , K .

(7)

(8)

Finally, the transmitter sends the data to the M active users


simultaneously.
C. User Scheduling Example
A user scheduling example may be helpful to understand
the overall system. An ORBF system with M = 4 and K =
10 is considered. It is assumed that all the users SINRs are
calculated as Table I from (4). In the table, U and B denote
the user index and the beam index, respectively. Let us denote
each users feedback information as Uk (k , ik ). From (5) and
(6), Uk (k , ik )s are obtained as follows:

1462

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

- U1 (10.08, 4), U2 (2.78, 3), U3 (5.43, 2), U4 (26.73, 2),


U5 (14.94, 2)
- U6 (3.15, 2), U7 (2.18, 4), U8 (18.29, 3), U9 (7.33, 1),
U10 (15.38, 2)
Now at the transmitter, the user index set at each beam is
obtained from (8) as K1 = {9}, K2 = {3, 4, 5, 6, 10}, K3 =
{2, 8}, and K4 = {1, 7}. By the max-SINR rule of (7), a user
to be assigned at each beam is determined by B1 U9 , B2 U4 ,
B3 U8 , and B4 U1 . The post-scheduling SINR corresponding
to each beam is obtained as 1 = 7.33, 2 = 26.73, 3 =
18.29, and 4 = 10.08.
III. P ROBABILITY D ISTRIBUTION OF THE
P OST- SCHEDULING SINR
In order to analyze the performance of the system in terms
of sum-rate, throughput, and average BER, it is necessary to
know the probability distribution about the scheduled users
SINRs, i.e., m s. Unlike existing literature, the CDFs of k
and m are derived exactly by direct integration, and the final
result is compared to the existing results.
A. Cumulative Distribution Function of Each Users Feedback
SINR

2
In (4), let hTk wm  = zm , M
= c, and SINRk,m = xm
for convenience. Since SINRk,m is i.i.d. over k, the subscript
k is omitted. Now (4) can be rewritten as
zm
xm =
,
(m = 1, 2, . . . , M ). (9)
M
c + l=1,l=m zl
Since wm is an orthonormal complex vector, W =
[w1 wM ] is a unitary matrix. As mentioned in Section IIA, hk is assumed to have i.i.d. complex Gaussian random
entries with distribution CN (0, 1). Thus, hTk W is a vector
with i.i.d. CN (0, 1) entries. This implies that |hTk wm |2 are
i.i.d. over m and k with 2 (2) and variance of 12 per
each degree of freedom. Hence, the joint probability density
function (JPDF) of zm s is denoted by
f (z1 , z2 , . . . , zM )

M
e m=1 zm , zm 0, zm (m = 1, 2, . . . , M ) (10)
=
.
0,
otherwise

from
order statistics.
integrate the
 Instead, we try to
 region

M
(z1 , z2 , . . . , zM )  0 zm x c + l=1,l=m zl , m =
1, 2, . . . , M directly using the JPDF given in (10). Let the
integration region be R(c, x). Now Fk (x) can be denoted by


M
Fk (x) = e m=1 zm dz1 dz2 dzM .
(12)
R(c,x)

For M = 2, Fk (x) is obtained easily from geometry as

cx
1 2e1+x ,
x1
Fk (x) =
, (13)
2cx
cx
1x
1 2e1+x + 1x
e
,
0x<1
1+x
whose proof is shown in Appendix A. For general M , we
derived a closed form expression of Fk (x) with direction
integration by substitution as
1

(1 + x)M1
1
(14)
min(M1, x
 1) M
(r+1)cx
r
M1 1rx
(1)
e
,
(1 rx)
r+1
r=0

Fk (x) =1

where  x1  is the smallest integer not less than x1 . The detailed


derivation is rather complicated, and it is given in Appendix
B. Substituting M = 2 into (14), we can also obtain (13) as
expected.

B. Cumulative Distribution Function of the Post-Scheduling


SINR
Let pm be the probability that each users feedback beamforming vector index is equal to m and dm be the cardinality
of K
m . In other words, pm = Pr{ik = m}, dm = |Km |,
M
and
m=1 dm = K. Since k is independent over k, the
probability that |Km | = dm for m = 1, . . . , M has a
multinomial distribution as


Pr |K1 | = d1 , |K2 | = d2 , . . . , |KM | = dM
K!
= M
m=1

Since the kth users feedback SINR is chosen to be the


maximum value among the M SINRs as in (5), the CDF
of k can be denoted by


Fk (x) = Pr max(x1 , x2 , . . . , xM ) x


= Pr xm x, for m = 1, 2, . . . , M


zm
= Pr
x, for m = 1, 2, . . . , M
M
c + l=1,l=m zl




M

= Pr zm x c +
zl , for m = 1, 2, . . . , M ,
l=1
l=m

(11)
where the 
last equality comes from the fact that the denomiM
nator (c + l=1,l=m zl ) is always positive. Since xm s are not
independent from each other, Fk (x) cannot be easily obtained

M


dm ! m=1

pm d m .

(15)

Under the assumption that a certain user index set Km is


given, the conditional CDF (denoted by Fm |Km ) of the postscheduling SINR can be easily obtained from order statistics
[26, Eq. (2.1.1), p. 9] as

dm
.
Fm |Km (x) = Fk (x)

(16)

Since m s are chosen from mutually exclusive users, m is


independent over m. Thus, the conditional joint CDF (JCDF)
of m s given |Km | = dm for m = 1, . . . , M can be given by



F1 ,2 ,...,M |K1 ,K2 ,...,KM x1 , x2 , . . . , xM d1 , d2 , . . . , dM
=

M


m=1

dm
Fk (xm )
.
(17)

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

1463

TABLE I
A N EXAMPLE OF SINRk,m IN THE ORTHOGONAL RANDOM BEAMFORMING SYSTEM WITH M = 4 AND K = 10.

B1
B2
B3
B4

U1

U2

U3

U4

U5

U6

U7

U8

U9

U10

8.05
0.34
0.15
10.08

2.58
1.94
2.78
0.23

0.18
5.43
5.27
0.04

2.08
26.73
1.77
0.14

0.02
14.94
0.77
2.88

2.82
3.15
0.04
1.18

0.36
1.44
0.96
2.18

11.72
0.15
18.29
0.81

7.33
4.59
3.69
0.06

8.25
15.38
0.15
5.83

Since the JCDF of m s is given by




F1 ,2 ,...,M x1 , x2 , . . . , xM




F1 ,2 ,...,M |K1 ,K2 ,...,KM x1 , . . . , xM d1 , . . . , dM
=
m, dm Z+ {0}
M
m=1 dm =K



Pr |K1 | = d1 , |K2 | = d2 , . . . , |KM | = dM .
(18)

Substituting (15) and (17) into (18), the JCDF can be obtained
as

K
M


pm Fk (xm )
.
F1 ,2 ,...,M x1 , x2 , . . . , xM =
m=1

(19)
Since the channel is assumed to be i.i.d., all the pm s are
1
identical, i.e., m, pm = M
. By setting xm to x, and the
others to infinity, the CDF of m , which is the marginal CDF,
can be obtained as


Fm (x) = F1 ,2 ,...,M , , . . . , xm = x, . . . ,

K
(20)
Fk (x)
1
.
=
+1
M
M
Substituting (14) into (20), Fm (x) is finally given by

1

Fm (x) = 1
M (1 + x)M1
1
min(M1, x
K
 1) M
(r+1)M x
(1)r
.
(1 rx)M1 e (1rx)
r+1
r=0
(21)
When M = 2, from (13), we have

K
2x

1 e
,
x1
1+x
Fm (x) = 
.
K
2x
4x

1x
(1x)
1 e
, 0x<1
1+x + 2(1+x) e
(22)
C. Comparison with the existing results
In [8], in order to evaluate the lower and upper bounds of
the throughput, the CDF of SINRk,m denoted as Fs (x) is
shown to be [8, Eq. (15)]
M

Fs (x) = 1

e x
(1 + x)M1

(x 0),

and the CDF of max1kK SINRk,m is obtained as



K
M
e x
Fs (x) = 1
(x 0).
(1 + x)M1

(23)

(24)

This result also can be thought as the SINR of the scheduled


user is selected from E[|Km |] M 1 K candidates, where

E[|Km |] is the average cardinality of the user index set, and it


K
can be approximated as Kpm = M
[22]. In the literature [22],
[23], [27], Fs (x) is exploited to obtain the asymptotic result
and the closed form bounds of sum-rate capacity. However,
Fs (x) is not exact. Fs (x) is derived based on the assumption
that all the SINRs of users are fed back to the transmitter. In
that feedback structure, an event that a certain user is selected
for more than one signal may occur. But in the original ORBF
systems, that event never occurs because each user feeds back
only one SINR, which is the largest among the M SINRs.
Instead, there exists an event with small probability that some
beams are not used in the feedback, i.e., m, Km = [28].
It is observed that Fm (x) = Fs (x) when x 1. For 0
x < 1, however, Fs (x) is not valid any more. In (21), it can
be seen that Fm (x) has different number of terms depending
on  x1 .
IV. A DAPTIVE M ODULATION
Adaptive modulation is commonly used in wireless communication systems in order to increase the throughput of the
system while satisfying a given target BER. In a rate-adaptive
ORBF system, an appropriate modulation can also be selected
per transmitted symbol sm . Since adaptive modulation for
single user MIMO systems has been studied in the literature
[14], [18], [19], [21], we take a similar approach for multiuser MIMO systems. For analysis, a discrete rate adaptive
system is considered where modulation levels are restricted
to a finite set, M = {M0 , M1 , , ML }, with Gray coded
quadrature amplitude modulation (QAM). Each Ml denotes
the constellation size and l, Ml1 < Ml . For example, if M1
corresponds to the BPSK modulation, M1 = 2. Especially, M0
denotes no transmission, and is set to be 1 for mathematical
manipulation. The SINR range is subdivided into (L + 1) bins
bounded by the switching threshold l (l = 0, 1, . . . , L + 1)
where 0 = 0 and L+1 = . The modulation corresponding
to Ml is selected whenever l m < l+1 . If m < 1 ,
data transmission is suspended for the corresponding channel
because the target BER constraint cannot be satisfied.
A. SNR Thresholds for Instantaneous BER Constraint
An easy way to set the switching thresholds l s is to use
the instantaneous BER (I-BER). In this approach, the BER
of every reception has to be less than or equal to the target
BER 0 . In order to meet the constraint, the BER for a QAM
in AWGN channels can be used. Although the exact BER
expressions for M-QAM are shown in [29], they are not easily
inverted with respect to the SINR, so that a numerical method
is necessary. Instead, in the adaptive modulation literature
[18], [19], [21], an exponential function form is used, which

1464

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

TABLE II
C ONSTELLATION SPECIFIC CONSTANTS FOR BER APPROXIMATION IN
AWGN CHANNELS [21].
Modulation
al
cl

QPSK
0.1853
0.5397

BPSK
0.1978
1.0923

16-QAM
0.1613
0.1110

64-QAM
0.1351
0.0270

is given by
Pe (, Ml ) al exp(cl ),

(25)

where is SNR, al = 0.2 and cl is a constellation specific


constant defined as [14]

6
for rectangular QAM (odd l, l L)
l
cl = 52 34
. (26)
for square QAM (even l, l L)
2(2l 1)

it is unlikely to satisfy a given target BER when the postscheduling SINR is below 0 dB. Without loss of generality,
we assume that the transmitter is turned off in the range 0
x < 1. Thus, it is enough to deal with fm (x) only for the
range of x 1 in the analysis. By differentiating Fs (x) of
(24), i.e. {Fs (x)}K with respect to x, we have

fm (x) = K 1

K1
M
e x
(1 + x)M1


M
e x
M
(1 + x)

(M 1) +
(1 + x)M

(30)
Now by using (30), the ASE can be obtained as

If we want a more accurate form than the above approximation, we can find the modulation specific constants al and cl
numerically using a curve-fitting method [21]. Table II shows
those values of M-QAMs which are often used. Inverting (25)
with respect to , the switching threshold is determined by

al
1
l = ln
.
(27)
cl
0
Although it is simple, I-BER approach keeps the instantaneous BER at all time instants below the target BER 0 .
This is so conservative that the average BER (A-BER) tends
to be far below 0 . In order to make the A-BER be equal to
0 , SNR thresholds should be lowered. Therefore, the ASE
can be improved by adjusting the switching threshold of each
modulation.
B. SNR Thresholds for Average BER Constraint
1) Closed form expressions for average spectral efficiency
and average BER: As shown in (21), the probability distribution of the post-scheduling SINR for each beam is identical
with respect to m. Thus, the A-BER of the system is also the
same as the A-BER for one particular beam. Since the ASE
of the system is M times that of one beam, we focus on the
analysis for one beam without loss of generality. The ASE for
one beam, denoted as , is given by
=

L


b l pl ,

where fm (x) is the probability density function (PDF) of


m , which is obtained by differentiating (21) with respect to
x. Although Fm (x) is derived as a closed form expression as
(21), (21) is still too complicated (even for M = 2) to derive a
closed form expression of the A-BER because Fm (x) has M
terms in the summation for the range of 0 x < 1. Instead, if
it is assumed that 1 > 1 (0 dB), we can use only the first term
in the summation of (21), which corresponds to the CDF of
max1kK SINRk,m . Even though BPSK is used for M1 ,

L




bl

l=1


1

e l+1
(1 + l+1 )M1
 
M

e l
(1 + l )M1

K
(31)

The A-BER Pe is given by


e
N
Pe = ,
Nb

(32)

e is the average number of error bits and N


b is the
where N
b =
average number of transmitted bits. It is observed that N
e is given by
by the definition in (28) and N
e =
N

L


bl Pe (l),

(33)

l=1

where Pe (l) is the A-BER when the SINR falls into the lth
bin, given by


l+1

Pe (l) =

l=0

L



bl Fm (l+1 ) Fm (l )
l=1

(28)

where bl = log2 Ml is the number of bits corresponding to the


lth modulation. pl is the probability that the post-scheduling
SINR m falls into the lth bin, given by
 l+1
fm (x)dx,
(29)
pl =

(x 1).

Pe (x, Ml )fm (x)dx.

(34)

Using (25) and (30), Pe (l) can be obtained as (see Appendix C)


Pe (l) =



K 1
(1)s el,s (l,s )s (M 1)
s
  s=0



s , (1 + l )l,s s , (1 + l+1 )l,s + Ml,s

 



1 s , (1 + l )l,s 1 s , (1 + l+1 )l,s
,

al K

K1


(35)
where s = (M 1)(1 + s) and l,s = cl + M
(1 + s). Using

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

(32), (33), and (35), Pe is finally obtained as


Pe =

L
K1
 K 1
K
al b l
(1)s el,s (l,s )s (M 1)
s

s=0
 l=1



s , (1 + l )l,s s , (1 + l+1 )l,s + Ml,s

 



1 s , (1 + l )l,s 1 s , (1 + l+1 )l,s
.
(36)
2) Optimal SINR thresholds: In the A-BER approach, the
goal is to maximize the ASE under the constraint that the
of
A-BER is lower than or equal to 0 . Defining the set
adjustable switching thresholds as = l | l = 1, 2, . . . , L ,
the optimization problem can be formulated as
o = arg max , subject to Pe 0 .

(37)

This problem can be solved with a Lagrange multiplier. Since

Pe is defined by Ne , the constraint Pe 0 can be changed

into Ne 0 for convenience, and the Lagrangian of (37)


is defined by
e 0 ).
L(, ) = + (N

(38)

Differentiating (38) with respect to l and equating to zero,


the following relationship for l = 1, 2, . . . , L is obtained as
bl bl1
.
bl1 Pe (l , Ml1 ) bl Pe (l , Ml ) 0 (bl1 bl )
(39)
According to the relationship, once 1 is chosen, all the other
l s are uniquely determined. Thus, the optimal SNR thresholds can be found numerically by adjusting 1 only. Although
the A-BER approach has higher computational complexity
than the I-BER approach, the thresholds can be calculated
off-line so that the A-BER approach is still practical. If the
SINR range is quantized with a few bits, we can calculate
the thresholds for the representative SINR of each bin. The
quantized feedback issue is analyzed with more details in [24].
=

V. S IMULATION R ESULTS
As for wm s, the column vectors of the discrete Fourier
transform (DFT) matrix are used, which are given by
WM = [w1 w2 wM ] = (wij ),

(40)

2
ej M ij

where wij = M (i, j = 0, 1, . . . , M 1). In order to


obtain the empirical CDF of the post-scheduling SINR and
the average BER, we use 105 different realizations of hk (k =
1, 2, . . . , K).
A. Cumulative Distribution Function
In order to verify (21) numerically, two theoretical CDFs
and two empirical CDFs are plotted in Figure 1. M and K are
set to be 2 and 30, respectively. Since the two theoretical CDFs
are different each other only at 0 x < 1, the average SNR
is set to be 10 dB. In Figure 1, Theoretical (21) denotes the
curve obtained from (21) and Theoretical (24) denotes the
curve obtained from (24). Simul. (s1 ) denotes the empirical

1465

CDF of the post-scheduling SINR at the first beam, i.e., 1


and Simul. (s2 ) denotes that of 2 . Even though SINR range
is below 0 dB, as shown in Figure 1(a), the two theoretical
CDFs are not distinguishable each other so that three enlarged
figures are added. As shown in Figure 1(b) and Figure 1(c),
both of the two empirical CDFs match with the derived result
(22) and not with the approximation {Fs (x)}. However, it
is observed in Figure 1(d) that the two theoretical CDFs are
almost indistinguishable in the range higher than 6 dB. This
is because as x gets close to 1, the third term of the second
line in (22) tends to zero as
lim

x10

4x
1 x (1x)
e
= 0,
2(1 + x)

(41)

so that Fm (x) for 0 x < 1 converges to Fs (x). It is also


observed in the figures that the empirical CDFs of 1 and 2
are almost identical, which is consistent with the assumption
1
of m, pm = M
for m = 1, . . . , M .

B. Average Spectral Efficiency


In the ASE simulations, modulations are restricted to MQAM with M = {1, 2, 4, 16, 64}, where 1 means no transmission. The target BER 0 is set to be 102 . The ASE is
averaged over 105 channel realizations under a block fading
channel model, where the entries of channel matrix do not
change during the transmission of one signal vector. In the
I-BER approach, the SINR thresholds are from (27) with the
constellation specific constants defined in Table II. In the ABER approach, the constellation specific constants in Table II
are also used, and the SINR thresholds are obtained from (39)
with numerical search.
Fig. 2 shows the performance comparison between the IBER and the A-BER constraints in terms of ASE. M is fixed
to 2, and K is set to 20 or100. It is observed in the figure that
the A-BER approach has SNR gain of 15 dB compared to the
I-BER method at the same ASE. It is also observed that as the
number of users increases, the overall performance improves
due to increased multiuser diversity. Fig. 3 shows the average
BER of I-BER method and A-BER method. M and K are
set to 2 and 20, respectively. As shown in the figure, both of
the two methods satisfy the BER constraint 0 0.01 but IBER method is more conservative than A-BER method. Fig. 4
shows the ASE performance with respect to the number of
transmit antennas (beams). K is fixed to 100, and M is set to
2, 3, and 4. It is observed that the ASE performance degrades
as M increases, which mismatches with the asymptotic result,
M log log K which is valid when K . This is because,
when K is fixed to a finite number (e.g., 100 or 200), the
asymptotic scaling law does not hold any longer, and random
beamforming cannot benefit from multiplexing gain. It should
also be noted that we do not use beam selection in this paper,
which means that all the M beams are active. Thus, the
system becomes more interference-limited as M increases.
The asymptotic sum-rate with converges to a function
M
(a decreasing function of M ), which
with the factor of M1
is already known in the existing literature [24], [27].

1466

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

1
Theoretical (21)
Theoretical (24)
Simul. (s1)

0.9
0.8

Simul. (s )

Simul. (s )

0.7

0.08
Pr(SINR <= x)

Pr(SINR <= x)

Theoretical (21)
Theoretical (24)
Simul. (s1)

0.1

0.6
0.5
0.4

0.06

0.04

0.3
0.02

0.2
0.1

0
0
12

11

10

8
7
SINR (dB)

11.4

11.2

(a) The whole region


0.9

0.06

Simul. (s2)

0.84

0.82

0.8

0.02

10.2

10

9.8
9.6
SINR (dB)

9.4

9.2

0.78
6.6

6.4

(c) Enlarged region 2

6.2

6
SINR (dB)

5.8

5.6

(d) Enlarged region 3

Comparison between theoretical and empirical CDFs of the post-scheduling SINR (M = 2, K = 30, = 10 dB).

12

10

M=2,K=20,IBER
M=2,K=20,ABER
10

10

8
Average BER

Average Spectral Efficiency (bits/channel use)

10.4

0.86

0.08

0.04

Fig. 1.

10.6

Theoretical (21)
Theoretical (24)
Simul. (s )

0.88

Simul. (s2)

Pr(SINR <= x)

Pr(SINR <= x)

0.1

10.8
SINR (dB)

(b) Enlarged region 1

Theoretical (21)
Theoretical (24)
Simul. (s )

0.12

11

10

10

10

M=2,K=20,IBER
M=2,K=20,ABER
M=2,K=100,IBER
M=2,K=100,ABER

10

15

20
SNR (dB)

25

30

35

40

10

15

20
SNR (dB)

25

30

35

40

Fig. 2. Performance comparison between I-BER method and A-BER method


(M = 2 fixed).

Fig. 3. Average BER of I-BER method and A-BER method (M = 2,


K = 20)

VI. C ONCLUSION

efficiency is maximized. It is shown that the derived CDF


of the post-scheduling SINR happens to be identical to the
existing approximation for SINR higher than 0 dB, but they
are different for SINR lower than 0 dB. The closed form
expressions are used to determine the optimal SINR thresholds
that maximize the ASE under the A-BER constraint. Simula-

In this paper, we derived the exact post-scheduling SINR


distribution in ORBF systems, and developed optimized adaptive modulation schemes with a finite set of M-QAM modulation orders in ORBF systems, where the the average spectral

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

and for 0 x < 1,


 
Fk (x) = 2 Pr 0 z1

Average Spectral Efficiency (bits/channel use)

12


cx
, 0 z2 z1
1x


cx
z1
, 0 z2
c
Pr cx z1
1x
x
  cx  z1
1x
e(z1 +z2 ) dz2 dz1
=2

10

cx
1x


0

cx
2

1467

M=2,K=100,ABER
M=3,K=100,ABER
M=4,K=100,ABER
0

10

15

20
SNR (dB)

25

30

35

A PPENDIX A
P ROOF OF (13)
From (9), the two SINRs for the kth user are given by
x2 =

z2
,
c + z1

(42)

where c = 2 . From (10), the JPDF of z1 and z2 is given by

f (z1 , z2 ) =

e(z1 +z2 )
0

if z1 0 and z2 0
.
otherwise

(43)

From (11), the CDF of the kth users maximum SINR is given
by



Fk (x) = Pr max(x1 , x2 ) x


= Pr 0 z1 x(c + z2 ) and 0 z2 x(c + z1 ) ,
(44)
where the shape of the corresponding integral region changes
with respect to x, which is plotted as Figure 5.
For x 1, Fk (x) can be calculated as


z1
c
Fk (x) = 1 Pr z1 cx, 0 z2
x


z2
Pr z2 cx, 0 z1
c
x


z1
c
= 1 2 Pr z1 cx, 0 z2
x
  zx1 c
e(z1 +z2 ) dz2 dz1 ,
=12
cx

(z1 +z2 )


dz2 dz1 .
(46)

40

tion results show that the A-BER approach with the optimal
thresholds has SNR gain of 15 dB compared to the I-BER
method at the same ASE. The proposed adaptive modulation
techniques appear to be promising for next generation wireless
systems with multi-user MIMO capability.

z1
,
c + z2

Solving the above double integrals, Fk (x) can be finally given


by (13).

Fig. 4. ASE Performance with respect to the number of transmit antennas


(K = 100 fixed).

x1 =

z1
x

(45)

A PPENDIX B
P ROOF OF (14)
Let us denote the set of real numbers and the set of integers
by R and Z, respectively.
Lemma 1: Let a, b R and n Z. For a, b > 0 and n 0,

b
et (at b)n dt = n!an e a .
(47)
b
a

Proof of Lemma 1: For a > 0 and n 0, from the


definition of the Gamma function, we have

n!
eat tn dt = n+1 .
(48)
a
0
Hence, for a, b > 0 and n 0,


1 u+b n
t
n
e (at b) dt =
e a u du
b
a
0
a
1 b
= e a an+1 n!
a
b
= n!an e a .

(49)

Definition 1: Function  : R R+ {0} is defined as


follows:

x, x 0
x :=
.
(50)
0, x < 0
Lemma 2: The indefinite integral of xn is as follows:

1
xn+1 .
(51)
xn dx =
n+1
Proof of Lemma 2:

 y
1
y n+1 , y 0
n
x dx = n+1
.
0,
y<0
0

(52)

Lemma 3: Let a, b R and n Z. For a, b > 0, and


n 0,


b
n!an e a , a > 0
t
n
e at b dt =
.
(53)
0,
a0
0
Proof of Lemma 3:

1468

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

]

where

] [ ] F

Sn (a, b) := {(x1 , x2 , , xn ) | 0 xk a k,
b a x1 + x2 + + xn b} Rn .
Lemma 4: For n = 1,

F[

I1 (a, b) = b 2b a + b 2a.

] ][ F

F

F[

]

(a) x 1

]

F[

Lemma 5: For n > 1,

In (a, b) =

F

F[

F[ [

(56)

Proof of Lemma 4: Relationship between a and b can


be divided into four cases as follows:
b < 0: Since x1 0 and x1 b < 0 in (55) are contradictory each other, S1 (a, b) = . Thus, I1 (a, b) = 0.
From (50), b = b a = b 2a = 0. Therefore, the
right-hand side of (56) is also equal to zero.
0 b < a: From (55), S1 (a, b) = [0, b]. Thus, I1 (a, b) =
b. From (50), b = b, b a = b 2a = 0. Therefore,
the right-hand side of (56) is also equal to b.
a b < 2a: Since S1 (a, b) = [ba, a], I1 (a, b) = 2ab.
From (50), b = b, b a = b a, and b 2a = 0.
Therefore, the right-hand side of (56) is also equal to
2a b.
b 2a: From (55), S1 (a, b) = [0, a] [b a, b]. Since
this set is an empty set or a set having only one point
(when b = 2a, S1 (a, b) = {a}), I1 (a, b) = 0. From (50),
b = b, ba = ba, and b2a = b2a. Therefore,
the right-hand side of (56) is also equal to zero.

F

] [ ] F

(55)

]

] ][ F

In1 (a, b x)dx.

(57)

Proof of Lemma 5: Let us consider T = Sn (a, b)


{(x1 , x2 , , xn ) | xn = x} for x [0, a]. Then, we can see
that
(x1 , , xn1 , x) T
x1 , , xn1 [0, a] & x1 + + xn1
[b a x, b x]

F

(x1 , , xn1 ) Sn1 (a, b x).


(b) 0 x < 1
Fig. 5.

Therefore,

The integration region for calculating Fk (x).

In (a, b) =

From (47) in Lemma 1, it is given by n!a e .


a 0: Since at b < 0, at b = 0. Therefore, the
integral is equal to zero.

Definition 2: For non-negative real numbers a and b, the ndimensional integral In (a, b) on the region Sn (a, b) is defined
as follows:


(54)
In (a, b) := 1dx1 dx2 dxn ,
Sn (a,b)

=
0

b
n a

a > 0: From (50) in Definition 1, at


 b = 0 when
t < ab . Thus, the integral is equal to b et (at b)n dt.
a

1dx1 dxn1 dxn

(58)

In1 (a, b x)dx.

Lemma 6: For n 1,


n+1
1 
r n+1
(1)
In (a, b) =
b ran .
r
n! r=0

(59)

Proof of Lemma 6: Mathematical induction can be used:

For n = 1: (59) is equal to (56). Hence, this has already


been proved in Lemma 4.

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

Inductive step: If In1 (a, b) holds,

ship


M

zl ,
0 zm x c +

In (a, b)
 a
=
In1 (a, b x)dx

n

n
1
=
(1)r
b ra xn1 dx
r
0 (n 1)! r=0

x=a
n

n
1
1
=
(1)r
b ra xn
r
(n 1)! r=0
n
x=0


n

1
n
=
(1)r
b (r + 1)an +
r
n! r=0

n
n
1 
(1)r
b ran
r
n! r=0


n+1
1 
n
n
r
=
(1)
+
b ran
r1
r
n! r=0


n+1
1 
r n+1
=
(1)
b ran ,
r
n! r=0
(60)
a


where the third equality comes from b ra
xn1 dx = n1 b ra xn , which can be easily
obtained from (51).
Now we derive Fk (x) by using previous results. In order
to calculate (12), we change the variable zm s as follows:

z1
z2
zM1
z1 + + zM

m = 1, . . . , M

l=1
l=m

1469

y1
y2
yM1
y

Since the corresponding Jacobian matrix and its determinant


are

1 0 0 0
0 1 0 0

(61)
det ... ... . . . ... ... = 1,

0 0 1 0
1 1 1 1

M



0 (1 + x)zm x c +
zl ,

m = 1, . . . , M

l=1

0 (1 + x)ym x(c + y), m = 1, . . . , M 1 &


M1



ym x(c + y)
0 (1 + x) y
m=1

0 ym

x(c + y)
, m = 1, . . . , M 1 &

1+x

y cx
y1 + + yM1 y.
1+x

Hence, S has the same form as (55) in Definition 2, i.e.


 x(c + y) 
S = SM1
,y .
(63)
1+x
Now we can apply (59) of Lemma 6 to (62) as follows:
Fk (x)

 x(c + y) 
, y dy
ey IM1
=
1+x
0
&

M
rx(c + y) 'M1
ey 
r M
=
y
(1)
dy
r
(M 1)! r=0
1+x
0

M

M
(1)r
=

(M

1)!
r
r=0

&
rx 
rcx 'M1
y
ey 1
dy.
1+x
1+x
0
(64)
From (53) of Lemma 3, the last integral can be calculated as

rcx
1+x



rx
rx M1 1 1+x
rx
e
, 1 1+x
> 0 . (65)
(M 1)! 1 1+x
0,
rx
1 1+x
0
Since the integral has a non-zero value only when r < 1 + x1 ,
Fk (x) is finally obtained as
Fk (x)



(1)r
M
rx M1

(M 1)! 1
(M 1)! r
1+x

1
min(M,1+ x
1)

r=0

rcx

dz1 dzM = dy1 dyM1 dy. Thus, we can rewrite (12) as


follows:


Fk (x) =

R(c,x)


=

ey dy1 dy2 dyM1 dy




(62)
1dy1 dy2 dyM1 dy,

where S = {(y1 , y2 , , yM1 ) | (y1 , y2 , , yM1 , y)


R(c, x)}. From (11), we can evaluate the following relation-

11+x
rx

1+x

(1 + x)M1
1
min(M,1+ x
 1) M
rcx
(1)r
{1 (r 1)x}M1 e 1(r1)x
r
r=1

=1 +

(1 + x)M1
1
min(M1, x
 1) M
(r+1)cx
(1)r
(1 rx)M1 e 1rx ,
r+1
r=0

=1

(66)

1470

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 12, NO. 4, APRIL 2013

where x is the smallest integer not less than x.


A PPENDIX C
D ERIVATION OF Pe (l)
Substituting (25) and (30) into (34), and using the binomial
expansion, Pe (l) is calculated as
 l+1
K1
 K 1
(1)s
al ecl x K
Pe (l) =
s
l
s=0


M
(1+s)x

M
e
(M 1) +
(1 + x) dx

(1 + x)s(M1)+M
 l+1 {cl + M (1+s)}x
K1
 K 1
e

(1)s
=al K
s
(1
+ x)s(M1)+M
l
s=0


M
(1 + x) dx.
(M 1) +

(67)
Since the upper incomplete Gamma function is defined as [30,
Eq. 8.350.2, p. 899]

et t1 dt,
(68)
(, x) 


x
ex
(1+x) dx

is given by
the integral form



ex
dx = e 1 1 , (1 + ) .

(1 + x)

(69)

Let us denote s = (M 1)(1 + s) and l,s = cl + M


(1 + s)
in (67). Now using (69), Pe (l) can be finally obtained as
Pe (l) =

K1




K 1
s l,s
s
al K
(1) e (l,s ) (M 1)
s
s=0
 



s , (1 + l )l,s s , (1 + l+1 )l,s + Ml,s

 



1 s , (1 + l )l,s 1 s , (1 + l+1 )l,s
.
(70)
R EFERENCES
[1] I. E. Telatar, Capacity of multi-antenna Gaussian channels, AT & T
Bell Lab., Tech. Rep., 1995, #BL0112170-950615-07TM.
[2] G. J. Foschini, Layered space-time architecture for wireless communication in a fading environment when using multiple aantennas, Bell
Labs. Tech. J., vol. 1, no. 2, pp. 4159, 1996.
[3] P. Viswanath and D. N. C. Tse, Sum capacity of the vector Gaussian broadcast channel and uplink-downlink duality, IEEE Trans. Inf.
Theory, vol. 49, no. 8, pp. 19121921, Aug. 2003.
[4] S. Catreux, V. Erceg, D. Gesbert, and R. W. Heath Jr., Adaptive
modulation and MIMO coding for broadband wireless data networks,
IEEE Commun. Mag., vol. 40, no. 6, pp. 108115, June 2002.
[5] M. Costa, Writing on dirty paper (corresp.), IEEE Trans. Inf. Theory,
vol. 29, no. 3, pp. 439441, May 1983.
[6] G. Caire and S. Shamai, On the achievable throughput of a multiantenna Gaussian broadcast channel, IEEE Trans. Inf. Theory, vol. 49,
no. 7, pp. 16911706, July 2003.
[7] Q. Spencer, A. Swindlehurst, and M. Haardt, Zero-forcing methods
for downlink spatial multiplexing in multiuser MIMO channels, IEEE
Trans. Signal Process., vol. 52, no. 2, pp. 461471, Feb. 2004.
[8] M. Sharif and B. Hassibi, On the capacity of MIMO broadcast channels
with partial side information, IEEE Trans. Inf. Theory, vol. 51, no. 2,
pp. 506522, Feb. 2005.

[9] T. Yoo, N. Jindal, and A. Goldsmith, Finite-rate feedback MIMO


broadcast channels with a large number of users, in Proc. 2006 IEEE
Intern. Symp. Inf. Theory, pp. 12141218.
[10] Downlink MIMO for EUTRA, Samsung Electronics, Tech. Rep., Feb.
2006, 3GPP TSG RAN WG1 44/R1-060335.
[11] A. J. Goldsmith and S.-G. Chua, Variable-rate variable-power MQAM
for fading channels, IEEE Trans. Commun., vol. 45, no. 10, pp. 1218
1230, Oct. 1997.
[12] M.-S. Alouini and A. J. Goldsmith, Adaptive modulation over Nakagami fading channels, Wireless Pers. Commun., vol. 13, pp. 119143,
May 2000.
[13] S. T. Chung and A. J. Goldsmith, Degrees of freedom in adaptive
modulation: a unified view, IEEE Trans. Commun., vol. 49, no. 9, pp.
15611571, Sept. 2001.
[14] S. Zhou and G. B. Giannakis, Adaptive modulation for multi-antenna
transmissions with channel mean feedback, in Proc. 2003 IEEE Intern.
Conf. Commun., pp. 22812285.
[15] A. Maaref and S. Assa, Adaptive modulation using orthogonal STBC
in MIMO Nakagami fading channels, in Proc. 2004 IEEE Intern. Symp.
Spread Spectrum Tech. App., pp. 145149.
[16] , Rate-adaptive M-QAM in MIMO diversity systems using spacetime block codes, in Proc. 2004 IEEE Intern. Symp. Personal, Indoor
Mobile Radio Commun., pp. 22942298.
[17] Z. Zhou, B. Vucetic, M. Dohler, and Y. Li, MIMO systems with
adaptive modulation, IEEE Trans. Veh. Technol., vol. 54, no. 5, pp.
18281842, Sept. 2005.
[18] Y. Ko and C. Tepedelenlioglu, Orthogonal space-time block coded
rate-adaptive modulation with outdated feedback, IEEE Trans. Wireless
Commun., vol. 5, no. 2, pp. 290295, Feb. 2006.
[19] A. Muller and J. Speidel, Adaptive modulation for MIMO spatial multiplexing systems with zero-forcing receivers in semi-correlated Rayleigh
fading channels, in Proc. 2006 Intern. Conf. Wireless Commun. Mobile
Comput., pp. 665670.
[20] J. Huang and S. Signell, On spectral efficiency of low-complexity
adaptive MIMO systems in Rayleigh fading channel, IEEE Trans.
Wireless Commun., vol. 8, no. 9, pp. 43694374, Sept. 2009.
[21] , On performance of adaptive modulation in MIMO systems using
orthogonal space-time block codes, IEEE Trans. Veh. Technol., vol. 58,
no. 8, pp. 42384247, Oct. 2009.
[22] W. Zhang and K. Letaief, MIMO broadcast scheduling with limited
feedback, IEEE J. Sel. Areas Commun., vol. 25, no. 7, pp. 14571467,
Sept. 2007.
[23] Y. Kim, J. Yang, and D. K. Kim, A closed form approximation of the
sum rate upperbound of random beamforming, IEEE Commun. Lett.,
vol. 12, no. 5, pp. 365367, May 2008.
[24] J. Vicario, R. Bosisio, C. Anton-Haro, and U. Spagnolini, Beam selection strategies for orthogonal random beamforming in sparse networks,
IEEE Trans. Wireless Commun., vol. 7, no. 9, pp. 33853396, Sept.
2008.
[25] C. Kim and J. Lee, Adaptive modulation for multiuser orthogonal
random beamforming systems, in Proc. 2011 IEEE SPAWC.
[26] H. A. David and H. N. Nagaraja, Order Statistics, 3rd edition. John
Wiley & Sons, Inc., 2003.
[27] K.-H. Park, Y.-C. Ko, and M.-S. Alouini, Accurate approximations and
asymptotic results for the sum-rate of random beamforming for multiantenna Gaussian broadcast channels, in Proc. 2009 IEEE Veh. Technol.
Conf. Spring.
[28] C. Kim, S. Jung, and J. Lee, Post-scheduling SINR mismatch analysis
for multiuser orthogonal random beamforming systems, IEEE Trans.
Commun., vol. 59, no. 6, pp. 15091513, June 2011.
[29] M. P. Fitz and J. P. Seymour, On the bit error probability of QAM
modulation, Intern. J. Wireless Inf. Networks, vol. 1, no. 2, pp. 131
139, Apr. 1994.
[30] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
Products, 7th edition. Elsevier Academic Press, 2007.

Chanhong Kim received the B.S. degree and the


Ph. D. degree in electrical engineering from Seoul
National University, Seoul, Korea in 2004 and 2011.
He currently works as a technical staff at Samsung
Electronics, Suwon, Korea. His research interests
include link adaptation, modulation, and coding for
wireless communications, with current emphasis on
analysis of multiuser MIMO techniques.

KIM et al.: SINR AND THROUGHPUT ANALYSIS FOR RANDOM BEAMFORMING SYSTEMS WITH ADAPTIVE MODULATION

Soohong Lee will receive the B.S. degree in the


Department of Mathematics from Seoul National
University, Seoul, Korea in 2013. Currently his
research interests are algebraic geometry, noncommutative geometry, and application of mathematics
in engineering. He received silver and gold medals at
International Mathematical Olympiad in 2007, 2008
respectively.

1471

Jungwoo Lee was born in Seoul, Korea. He received a B.S. degree in Electronics Engineering from
Seoul National University, Seoul, Korea in 1988
and M.S.E. degree and Ph.D. degree in Electrical
Engineering from Princeton University in 1990 and
1994. He was a member of technical staff working
on multimedia signal processing at Sarnoff Corporation from 1994 to 1999. He has been with Wireless
Advanced Technology Lab of Lucent Technologies
since 1999, and worked on W-CDMA base station algorithm development. His research interests
include wireless communications, signal processing, communications ASIC
architecture/design, multiple antenna systems, and wireless video. He holds
12 U.S. patents. He was an associate editor for IEEE T RANSACTIONS ON
V EHICULAR T ECHNOLOGY (2008 to 2011), and he is an associate editor for
the Journal of Communications and Networks. He is also a director of KICS
and a steering committee member of JCCI.

Anda mungkin juga menyukai