I

E
E
E
P
r
o
o
f
IEEE WIRELESS COMMUNICATIONS LETTERS 1
Asymptotic Error Performance Analysis of Spatial
Modulation Under Generalized Fading
1
2
Kostas P. Peppas, Senior Member, IEEE, Martin Zamkotsian, Fotis Lazarakis, and Panayotis G. Cottis 3
AbstractThis letter presents a comprehensive framework an- 4
alyzing the asymptotic error performance of a multiple-input 5
multiple-output (MIMO) wireless system employing spatial mod- 6
ulation (SM) with maximum likelihood detection and perfect 7
channel state information. Generic analytical expressions for the 8
diversity and coding gains are deduced, which reveal fundamental 9
properties of MIMO SM systems. The presented analysis can be 10
used to obtain closed-form upper bounds for the average bit error 11
probability (ABEP) of MIMO SM systems under generalized fad- 12
ing, which become asymptotically tight in the high signal-to-noise 13
ratio (SNR) region. 14
Index TermsAsymptotic analysis, average bit error probabil- 15
ity, coding gain, diversity gain, generalized fading, multiple-input 16
multiple-output (MIMO) systems, spatial modulation (SM), space 17
shift keying (SSK) modulation. 18
I. INTRODUCTION 19
S
PATIAL MODULATION (SM) is an efcient, low- 20
complexity transmission technique for multiple-input- 21
multiple-output (MIMO) wireless systems which achieves a 22
spatial multiplexing gain, at the same time avoiding inter- 23
channel interference without requiring synchronization be- 24
tween the transmit antennas [1], [2]. A fundamental concept 25
in SM is the three-dimensional constellation diagram [2] where 26
each spatial constellation point, corresponding to the transmit 27
antenna index, denes an independent complex plane of signal 28
constellation points. When the information carrying entity is 29
solely the transmit-antenna index, SM is reduced to the space 30
shift keying (SSK) modulation, where a single transmit-antenna 31
is activated each time to transmit a symbol. 32
Several analytical frameworks assessing the error perfor- 33
mance of SM systems over fading channels are available in 34
the technical literature. For example, [3] and [4] employ a 35
moment generating function (MGF) based approach to evaluate 36
the average bit error probability (ABEP) of SSK in the presence 37
of Nakagami-m and Rice fading. The MGF-based approach 38
presented in these works is extended to the most general case of 39
SM in [5], where tight error performance bounds are deduced. 40
In a recent work [6], a generic approach for the performance of 41
Manuscript received March 3, 2014; revised May 9, 2014; accepted May 9,
2014. The associate editor coordinating the review of this paper and approving
it for publication was Y. Chen.
K. P. Peppas and F. Lazarakis are with the Laboratory of Wireless Commu-
nications, Institute of Informatics and Telecommunications, National Centre
for Scientic ResearchDemokritos, Athens 15310, Greece (e-mail:
kpeppas@iit.demokritos.gr; az@iit.demokritos.gr).
M. Zamkotsian is with the Laboratory of Wireless Communications, Insti-
tute of Informatics and Telecommunications, National Centre for Scientic
ResearchDemokritos, Athens 15310, Greece, and also with the National
Technical University of Athens, School of Electrical and Computer Engineer-
ing, Athens 15773, Greece (e-mail: mzamko@iit.demokritos.gr).
P. G. Cottis is with the National Technical University of Athens, School of
Electrical and Computer Engineering, Athens 15773, Greece (e-mail: pcottis@
central.ntua.gr).
Digital Object Identier 10.1109/LWC.2014.2326152
SSK was proposed, assuming generalized fading envelopes and 42
uniformly distributed channel phases. 43
The above cited frameworks provide an exact performance 44
analysis of SM systems over the entire signal-to-noise ratio 45
(SNR) region; however, single integrals with nite or innite 46
limits have to be readily evaluated via numerical integration to 47
this end. Moreover, these frameworks do not provide enough 48
insight into the parameters affecting system performance in 49
terms of diversity and coding gains. In an attempt to bridge this 50
gap, closed-form expressions for the asymptotic performance 51
of SM systems are provided in [5] and [7]. 52
Motivated by the above cited works, the objective of the cur- 53
rent letter is twofold: a) To deduce closed-form upper bounds 54
for the ABEP of SM MIMO systems operating over generalized 55
fading environments which become asymptotically tight in the 56
high SNR region, and b) to provide important considerations 57
about the diversity and coding gains of SM in the presence 58
of generalized fading. The proposed analysis is tested and 59
veried by numerically evaluated results accompanied with 60
Monte Carlo simulations as well as by reducing them to several 61
special cases available in the literature. 62
II. MATHEMATICAL TOOLS 63
In this section, new mathematical tools are presented that 64
simplify the performance evaluation of SM systems. According 65
to [5], the evaluation of the ABEP of SM requires the solution 66
of integrals of the form 67
J(A, L) =
1

/2
_
0
L

=1
_
/
Z

_
A
2 sin
2

__
d, A > 0 (1)
where /
Z

() denotes the MGF of the random variable Z

68
dened as Z

= [z
2,
z
1,
[
2
with z
i,
=
i,
exp(
2,
) be- 69
ing random vectors having arbitrarily distributed magnitudes 70

i,
and phases
i,
, i 1, 2. In general, closed form 71
expressions for J(A, L) are very difcult to be obtained and 72
numerical integration is used instead (see for example [3], [4] 73
and [6]). In [7], by exploiting asymptotic analysis, closed- 74
form approximations for J(A, L) are provided for high values 75
of A, assuming that
i,
are Nakagami-m distributed random 76
variables and
i,
uniformly distributed in [0, 2]. 77
In the following analysis, a generic solution of (1) for high 78
values of A will be deduced, assuming that
i,
are arbitrarily 79
distributedrandomvariables and
i,
areuniformlydistributedin 80
[0, 2]. In order to obtain such an expression, [8, Proposition 3] 81
is employed to approximate /
Z

(s) for s as
1
82
[/
Z

(s)[ = c

[s[
d

+o
_
[s[
d

_
, s (2)
1
The notation f(x) = o[g(x)] as x x
0
stands for lim
xx
0
(f(x)/
g(x)) = 0.
2162-2337 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
I
E
E
E
P
r
o
o
f
2 IEEE WIRELESS COMMUNICATIONS LETTERS
The MGF of Z

is given by [6, Eq. (9)] as 83

/
Z

(s) =
1
2s

_
0
Re

R
2
4s
_
2

i=1
1
0,R
_
f

i,
(r)
r
_
_
dR (3)
where f
a
i,
(r) is the probability density function of
i,
84
and 1
0,R
denotes the zeroth order Hankel transform [9, 85
Eq. (9.11)]. Since e
(R
2
/4s)
= o(1/s) as s , the approxi- 86
mation e
(R
2
/4s)
/2s
s
1/2s can be employed in (3) to yield 87
/
Z

(s)
s

1
2s

_
0
R
_
2

i=1
1
0,R
_
f

i,
(r)
r
_
_
dR (4)
Changing the information variable R
2
to y and comparing (4) 88
with (2), it is readily deduced that d

= 1 and 89
c

=
1
4

_
0
2

i=1
1
0,

y
_
f

i,
(r)
r
_
dy (5)
Finally, by substituting (2) and (5) into (1), it is deduced that 90
for high values of A, J(A, L) can be approximated by 91
J(A, L)
A1

2
L1

_
L +
1
2
_

(L + 1)
_
L

=1
c

_
A
L
(6)
where () is the gamma function [10, Eq. (8.310/1)]. 92
It is noted that c

and henceforth J(A, L) can be eas- 93

ily obtained in closed-form by employing Mellin transform 94
techniques, provided that a closed-form expression for 1
0,

y
95
f

i,
(r)/r is readily available. In what follows, a closed 96
form expression for c

will be deduced assuming that

i,
97
follow the Extended Generalized-/ (EGK) distribution. The 98
motivation behind the choice of this specic model is that the 99
EGK distribution exhibits good tail properties and encompasses 100
most of the well-known fading distributions either as special or 101
as limiting cases [11, Table I]. Simplied expressions for the 102
special cases of Generalized-/ and the Nakagami-m distribu- 103
tions are also deduced. 104
A. The Extended Generalized-/ Case 105
Under EGK fading, the zeroth order Hankel transform of 106
f

i,
(r)/r is determined in closed-form as [6, Eq. (11)] 107
1
0,R
_
f

i,
(r)
r
_
=
H
2,1
2,2
_
4b
s,i,
b
i,
R
2

i,

(1,1), (1,1)

_
(m
i,
)(m
s,i,
)
(7)
where H
m,n
p,q
[] is the Foxs H-function [12, Eq. (8.3.1)],
2
108

=(m
i,
, (2/
i,
)), (m
s,i,
, (2/
s,i,
)). In (7), m
i,
(0.5 < 109
m
i,
< ) and
i,
(0 <
i,
< ) represent the fading se- 110
verity and the fading shaping factor, respectively, m
s,i,
(0.5 < 111
m
s,i,
<) and
s,i,
(0<
s,i,
<) represent the shadowing 112
severity and the shadowing shaping factor, respectively, and 113

i,
= Ea
2
i,
with E denoting expectation. Moreover, 114
b
i,
= (m
i,
+ (2/
i,
))/(m
i,
) and b
s,i,
= (m
s,i,
+ 115
(2/
s,i,
))/(m
s,i,
). By substituting (7) into (5) and 116
employing [12, Eq. (2.25.1.1)] along with [12, Eq. (8.3.2.7)], 117
c

can be evaluated from 118

c

= A

H
3,3
4,4
_
x

(
1
,
1
), (
2
,
2
), (0, 1), (0, 1)
(0, 1), (M
1
,
1
), (M
2
,
2
), (0, 1)
_
(8)
where A

= (b
s,1,
b
1,
)/((m
1,
)(m
2,
)(m
s,1,
)(m
s,2,
) 119

1,
), x

=(
2,
b
1,
b
s,1,
/
1,
b
2,
b
s,2,
),
1
=1 m
2,
,
1
= 120
2/
2,
,
2
= 1 m
s,2,
,
2
= 2/
s,2,
, M
1
= m
1,
2/
1,
, 121

1
= 2/
1,
, M
2
= m
s,1,
2/
s,1,
and
2
= 2/
s,1,
. The 122
result in (8) can be reduced further by employing [12, Eq. 123
(8.3.2.6)] yielding (9), shown at the bottom of the page. 124
B. The Generalized-/ Case 125
Under Generalized-/ fading conditions, an expression for c

126
is readily obtained from (9) setting the fading shaping factor 127

i,
2 and the shadowing shaping factor
s,i,
2. (See 128
equation at bottom of page) Employing [12, Eq. (8.3.2.21)], (9) 129
yields 130
c

= B

G
2,2
2,2
_

2,
m
1,
m
s,1,

1,
m
2,
m
s,2,

1 m
2,
, 1 m
s,2,
m
1,
1, m
s,1,
1
_
(10)
where G
m,n
p,q
[] is the Meijers G-function [10, Eq. (9.301)] and 131
B

= (m
s, 1,
m
1,
/(m
1,
)(m
2,
)(m
s,1,
)(m
s,2,
)
1,
). 132
Finally, employing the identity [15, Eq. (07.34.03.0871.01)], 133
(10) can be further expressed in terms of the Gauss hyper- 134
geometric function
p
F
q
() [10, Eq. (9.14.1)] as (11), shown at 135
the bottom of the page. 136
C. The Nakagami-m Case 137
For the special case of Nakagami-mfading, an expression for 138
c

is readily obtained from (10) setting the shadowing severity 139

factor m
s,i,
. Specically, it can be shown that c

is 140
2
Note that efcient algorithms for the numerical evaluation of the H-function
are available in [13, Table 2] and [14, Appendix A].
c

=
b
s,1,
b
1,
(m
1,
)(m
2,
)(m
s,1,
)(m
s,2,
)
1,
H
2,2
2,2
_
_

2,
b
1,
b
s,1,

1,
b
2,
b
s,2,

_
1 m
2,
,
2

2,
_
,
_
1 m
s,2,
,
2

s,2,
_
_
m
1,

1,
,
2

1,
_
,
_
m
s,1,

s,1,
,
2

s,1,
_
_
_
(9)
c

=
(1 +m
2,
+m
s,1,
)(1 +m
s,2,
+m
s,1,
)(1 +m
2,
+m
1,
)(1 +m
s,2,
+m
1,
)

_
2 +

2
i=1
[m
i,
+m
s,i,
]
__

2
i=1
(m
i,
)(m
s,i,
)
_ _
m
s,1,
m
1,

1,
_
m
2,
1

_
m
2,
m
s,2,

2,
_
m
2,
2
F
1
_
1 +
2

i=1
m
i,
, 1 +m
2,
+m
s,1,
; 2 +
2

i=1
[m
i,
+m
s,i,
], 1

1,
m
2,
m
s,2,

2,
m
1,
m
s,1,
_
(11)
I
E
E
E
P
r
o
o
f
PEPPAS et al.: ASYMPTOTIC ERROR PERFORMANCE ANALYSIS OF SM UNDER GENERALIZED FADING 3
reduced to a known result. Letting m
s,i,
in (10) and 141
employing the denition of the Meijers G-function [12, Eq. 142
(8.2.1.1)], c

can be written as 143

c

=
m
1,
(2
1,
)
1
(m
1,
)(m
2,
)
_
(
_

2,
m
1,

1,
m
2,
_
u
(m
2,
u)
(m
1,
1+u)
_
lim
m
s,1,

m
u+1
s,1,
(m
s,1,
1+u)
(m
s,1,
)
_

_
lim
m
s,2,

m
u
s,2,
(m
s,2,
u)
(m
s,2,
)
_
du (12)
where ( is the Mellin-Barnes contour. Employing the identity 144
lim
x
(x
u
(x +u)/(x)) = 1 [10, Eq. (8.328)] along with 145
[12, Eq. (8.2.1.1)], (12) is written as 146
c

=
m
1,
(m
1,
)(m
2,
)
1,
G
1,1
1,1
_

2,
m
1,

1,
m
2,

1 m
2,
m
1,
1
_
.
(13)
Finally, using the identity G
1,1
1,1
[x[
a
b
] = (1 a +b)x
b
(x + 147
1)
ab1
[15, 07.34.03.0271.01], c

is given from 148

c

=
_
2

i=1
1
(m
i,
)
_
m
i,

i,
_
m
i,
_

_
1 +
2

i=1
m
i,
_

_
2

i=1
m
i,

i,
_
1

2
i=1
m
i,
(14)
which is identical to [7, Eq. (4)]. 149
III. APPLICATION TO THE PERFORMANCE ANALYSIS 150
OF SPATIAL MODULATION 151
In this section, the results reported in Section II are applied 152
to assess the asymptotic performance of SM systems. 153
A. System Model 154
A N
t
N
r
MIMO system employing SM is considered, 155
equipped with N
t
transmit and N
r
receive antennas, which 156
can send digital information via M complex symbols,
j
= 157
[
j
[e

j
, j = 1, . . . , M. In the following and without loss of 158
generality, two test cases are considered: i) A pure SSK system 159
operating under independent and identically distributed (i.i.d) 160
fading (Case I); and ii) A SM system operating under i.i.d 161
fading with constant-modulus modulation i.e. [
j
[ =
0
, j = 162
1, . . . , M (Case II). 163
1) Case I: Under the assumption of i.i.d fading, a tight 164
upper bound for the ABEP of SSK can be obtained from [3, 165
Eq. (35)], [7], as 166
P
N
t
2
PEP
SSK
(t
1
t
2
) (15)
where PEP
SSK
(t
1
t
2
) denotes the pairwise error probability 167
related to the pair of transmit antennas t
1
and t
2
, t
1
, t
2
= 168
1, 2, . . . , N
t
, and it is the same for any pair (t
1
, t
2
). The 169
PEP
SSK
(t
1
t
2
) can be evaluated as [7, Eq. (1)] 170
PEP
SSK
(t
1
t
2
) =
1

/2
_
0
N
r

=1
_
/
:

_

2 sin
2

__
d (16)
where :

= [a
t
2
,
exp(
t
2
,
) a
t
1
,
exp(
t
1
,
)[
2
, with a
t
i
,
171
and
t
i
,
being the envelopes and phases of the link dened 172
by the t
i
-th transmit antenna and the -th receive antenna. 173
Moreover, = E
s
/4N
0
is the SNR where E
s
is the symbol 174
energy and N
0
is the single-sided power spectral density of 175
the additive white Gaussian noise. For high values of , it 176
can be observed that PEP
SSK
(t
1
t
2
) can be readily evalu- 177
ated employing (6) as PEP
SSK
(t
1
t
2
)
1
J(, N
r
). Finally, 178
from (6), it is evident that the diversity gain depends only 179
on the number of the receive antennas and is independent of 180
the fading severity. This nding is in agreement with relevant 181
ndings reported in [4] and [7]. The resulting coding gain can 182
be obtained in closed-form from [8, Eq. (1)]. 183
2) Case II: The ABEP of SM can be tightly upper bounded 184
as [5, Eq. (6)] 185
P ABEP
signal
+ ABEP
spatial
+ ABEP
joint
(17)
where ABEP
signal
, ABEP
spatial
and ABEP
joint
show how the 186
error performance of SM is affected by the signal constellation 187
diagram, the spatial constellation diagram and the interaction 188
of both signal and space constellation diagrams, respectively. 189
Under generalized fading, the term ABEP
signal
when either 190
M-ary phase shift keying (M-PSK) or M-ary quadrature am- 191
plitude modulation (M-QAM) are employed, can be readily 192
evaluated using [5, Eqs. (7), (8)] and [5, Table I]. High-SNR 193
asymptotically tight expressions for ABEP
signal
can also be 194
obtained using [8]. Assuming constant modulus modulation 195
ABEP
spatial
and ABEP
joint
can be obtained from [5, Eq. (10)] 196
and [5, Eq. (11)], respectively, as 197
ABEP
spatial
=
N
t
log
2
(N
t
)
2 log
2
(N
t
M)
PEP
SM
(t
1
t
2
) (18a)
ABEP
joint
=
_
M(N
t
1) log
2
(M)+N
t
(M1) log
2
(N
t
)
2 log
2
(N
t
M)
_
PEP
SM
(t
1
t
2
) (18b)
where PEP
SM
(t
1
t
2
) can be readily obtained from (16) by 198
replacing with
0
. For high values of , the framework 199
presented in Section II can be readily employed to yield
3
200
PEP
SM
(t
1
t
2
)
1
J(
0
, N
r
). The resulting diversity gain 201
is minN
r
, Div
signal
where Div
signal
is the diversity gain of 202
ABEP
signal
[5]. 203
B. Numerical Results 204
Numerical results accompanied by computer simulations are 205
presented to study the tightness of (6) under various fading 206
conditions. In the following analysis, an 8 N
r
MIMO system 207
is considered. Fig. 1 depicts the ABEP of 8 2 and 8 3 208
MIMO SSK systems operating over EGK fading channels as 209
a function of E
s
/N
0
, assuming m
s,i,
= 2,
s,i,
= 1, m
i,
= 210
1.5,
i,
= 4 and
i,
1, 5. Fig. 1 includes upper bounds 211
for the ABEP obtained by the numerical integration of (16), 212
exact ABEP results obtained from Monte-Carlo simulation as 213
3
When non-constant modulus modulation is assumed, the framework pre-
sented in Section II can be readily applied by setting in (1)
i,
= a
t
i
,
|

|
and
i,
=
t
i
,
+

.
I
E
E
E
P
r
o
o
f
4 IEEE WIRELESS COMMUNICATIONS LETTERS
Fig. 1. ABEP of SSK for 8 2 and 8 3 MIMO EGK channels as a
function of E
s
/N
0
. Simulation Parameters: m
s,i,
= 2,
s,i,
= 1, m
i,
=
1.5,
i,
= 4 and
i,
{1, 5}.
Fig. 2. ABEP of SM-QPSK for 8 2 and 8 3 MIMO Generalized-K
channels as a function of E
s
/N
0
for various values of k. Simulation Parame-
ters: m
i,
= 1.5,
i,
= 1.
well as asymptotic ABEP results obtained employing the high 214
SNR assumption in (6) and (9). As it is evident, the proposed 215
analytical framework well predicts the diversity and coding 216
gains of the considered system and yields tight results for high 217
values of E
s
/N
0
. Moreover, it can be observed that
i,
affects 218
coding gain only and, as expected, coding gain improves as
i,
219
increases. 220
For the same antenna congurations, Fig. 2 depicts the ABEP 221
of MIMO SM-QPSK (M = 4), operating over generalized-/ 222
fading channels
4
as a function of E
s
/N
0
, assuming m
i,
= 1.5, 223

i,
= 1 and m
s,i,
= k. Different values of k are consid- 224
ered to account for two shadowing scenarios, that is frequent 225
heavy shadowing (k = 1.0931) and average shadowing (k = 226
38.0809) [16]. As for the tightness of (6), similar conclusions to 227
those reported in Fig. 1 are deduced. However, in the presence 228
4
Using [8], the diversity gain Div
signal
is deduced as min{m
s,i,
, m
i,
}.
of heavy shadowing (k = 1.0931) and for N
t
= 3 transmit an- 229
tennas, the asymptotic behavior of the ABEP-SNR curve shows 230
up at high SNR values, i.e. for E
s
/N
0
> 30 dB. Furthermore, 231
as it is expected, coding gain improves as k increases, i.e. when 232
the impact of shadowing becomes less severe. 233
IV. CONCLUSION 234
In this letter, an analytical framework for the computation of 235
the diversity and coding gains of SM systems over generalized 236
fading channels was presented. To the best of the authors 237
knowledge, the derived (5), (9), and (11) are novel and can be 238
simplied to some particular cases already reported. The newly 239
derived simplied ABEP expressions require much less time for 240
numerical evaluation compared to the exact ones, which require 241
numerical integration. It was shown that, under generalized 242
fading, the diversity gains of spatial and joint components of 243
SM do not depend on the fading severity. 244
REFERENCES 245
[1] M. D. Renzo, H. Haas, A. Ghrayeb, S. Sugiura, and L. Hanzo, Spatial 246
modulation for generalized MIMO: Challenges, opportunities, implemen- 247
tation, Proc. IEEE, vol. 102, no. 1, pp. 56103, Jan. 2014. 248
[2] M. D. Renzo, H. Haas, and P. Grant, Spatial modulation for multiple- 249
antenna wireless systems: A survey, IEEE Commun. Mag., vol. 49, 250
no. 12, pp. 182191, Dec. 2011. 251
[3] M. D. Renzo and H. Haas, Ageneral framework for performance analysis 252
of SpaceShift Keying(SSK) modulationfor MISOcorrelatedNakagami-m 253
fading channels, IEEE Trans. Commun., vol. 58, no. 9, pp. 25902603, 254
Sep. 2010. 255
[4] M. D. Renzo and H. Haas, Space Shift Keying (SSK-) MIMO over corre- 256
lated Rician fading channels: Performance analysis and a new method for 257
transmit-diversity, IEEE Trans. Commun., vol. 59, no. 1, pp. 116129, 258
Jan. 2011. 259
[5] M. D. Renzo, H. Haas, and P. Grant, Bit error probability of SM-MIMO 260
over generalized fading channels, IEEE Trans. Veh. Technol., vol. 61, 261
no. 3, pp. 11241144, Mar. 2012. 262
[6] K. Peppas, M. Zamkotsian, F. Lazarakis, and P. Cottis, Unied er- 263
ror performance analysis of space shift keying modulation for MISO 264
and MIMO systems under generalized fading, IEEE Wireless Commun. 265
Letters, vol. 2, no. 6, pp. 663666, Dec. 2013. 266
[7] M. D. Renzo and H. Haas, Bit error probability of space modulation over 267
Nakagami-mfading: Asymptotic analysis, IEEE Commun. Lett., vol. 15, 268
no. 10, pp. 10261028, Oct. 2011. 269
[8] Z. Wang and G. Giannakis, A simple and general parametrization quan- 270
tifying performance in fading channels, IEEE Trans. Commun., vol. 51, 271
no. 8, pp. 13891398, Aug. 2003. 272
[9] A. D. Poularikas, The Transforms and Applications Handbook, 2nd ed. 273
Boca Raton, FL, USA: CRC and IEEE Press, 2000. 274
[10] I. Gradshteyn and I. M. Ryzhik, Tables of Integrals, Series, Products, 275
6th ed. New York, NY, USA: Academic, 2000. 276
[11] F. Yilmaz and M.-S. Alouini, A new simple model for composite fading 277
channels: Second order statistics and channel capacity, in Proc. ISWCS, 278
York, U.K., Sep. 2010, pp. 676680. 279
[12] A. P. Prudnikov, Y. A. Brychkov, and O. I. Marichev, Integrals and Series 280
Volume 3: More Special Functions, 1st ed. New York, NY, USA: Taylor 281
& Francis, 2003. 282
[13] K. Peppas, F. Lazarakis, A. Alexandridis, and K. Dangakis, Sim- 283
ple, accurate formula for the average bit error probability of multiple- 284
input multiple-output free-space optical links over negative exponential 285
turbulence channels, Opt. Lett., vol. 37, no. 15, pp. 32433245, 286
Aug. 2012. 287
[14] F. Yilmaz and M.-S. Alouini, Product of the powers of generalized 288
Nakagami-m variates and performance of cascaded fading channels, in 289
Proc. IEEE Global Telecomm. Conf., 2009, pp. 18. 290
[15] T. Wolfram, Function site. [Online]. Availbale: http://functions.wolfram. 291
com 292
[16] K. Peppas, Accurate closed-form approximations to generalized-K sum 293
distributions and applications in the performance analysis of equal 294
gain combining receivers, IET Commun., vol. 5, no. 7, pp. 982989, 295
May 2011. 296
I
E
E
E
P
r
o
o
f
AUTHOR QUERY
NO QUERY.