Ieee Tcom Hos

IEEE TRANSACTIONS ON COMMUNICATIONS 1
On High-Order Capacity Statistics of Spectrum

Aggregation Systems Over - and -
Shadowed Fading Channels
Jiayi Zhang, Member, IEEE, Xiaoyu Chen, Kostas P. Peppas, Senior Member, IEEE, Xu Li, and Ying Liu
1 Abstract The frequency scarcity imposed by fast growing band or in different frequency bands [2]. In order to achieve a 37
2 demand for mobile data service requires promising spectrum successful spectrum aggregation, the maximum dispersion in 38
3 aggregation systems. The so-called higher order statistics (HOS) the channel capacity should be calibrated to leverage a reliable 39
4 of the channel capacity is a suitable metric on the system perfor-
5 mance. While prior relevant works have improved our knowledge transmission [3]. 40
of
6 on the HOS characterization of spectrum aggregation systems, In this context, the typical metric for performance evaluation 41
7 an analytical framework encompassing generalized fading models has been the higher-order statistics (HOS) of the channel 42
8 of interest is not yet available. In this paper, we pursue a detailed capacity, which can fully explore the reliability of the signal 43
9 HOS analysis of - and - shadowed fading channels by transmission in spectrum aggregation systems. As an useful 44
10 deriving novel and exact expressions. Furthermore, the simplified
11 HOS expressions for the asymptotically low and high signal-to- tool, the HOS can effectively describe the channel capac- 45
12 noise regimes are derived. Several important statistical measures, ity dispersion induced by the heterogeneity that inherently 46
13 such as amount of fading, amount of dispersion, reliability, exists in spectrum aggregation systems [4]. Moreover, fruitful 47
ro
14 skewness, and kurtosis, are obtained by using the HOS results. insights into the implications of the spectrum aggregation on 48
15 More importantly, the useful implications of system and fading the transmission reliability can be extracted by deriving HOS 49
16 parameters on spectrum aggregation systems are investigated for
17 channel selection. Finally, all derived expressions are validated of the channel capacity. Despite its importance, however, the 50
18 via Monte Carlo simulations. HOS of the channel capacity received relatively little attention 51
in the literature, due in part to the intractability of its analysis. 52

19 Index Terms Spectrum aggregation, higher-order statistics,
20 -, - shadowed. A number of prior works have investigated the HOS of the 53
channel capacity of different wireless systems over several flat 54
fading channels [3], [5][7]. For example, a generic framework

EP
55
21 I. I NTRODUCTION
for the asymptotic HOS of the channel capacity over inde- 56
22
23
24
W ITH the constantly growing mobile data demand
for future wireless communication systems, i.e., 5th
Generation (5G), it becomes more and more difficult to
pendent and identically distributed (i.i.d.) Nakagami-m fading
channels was provided in [5]. Yilmaz et al. [3] investigated the
HOS of the channel capacity for amplify-and-forward (AF)
57
58
59
25 allocate a wide and contiguous frequency band to each user multihop systems over gamma and generalized gamma fading 60
26 equipment (UE) and base station (BS). This has brought about channels. In addition, an MGF-based approach for the HOS 61
27 increasing scarcity in available radio spectrum. To address of the channel capacity for L-branch MRC receivers has been 62
28 these issues, the promising spectrum aggregation technique proposed in [6] with an example application of correlated 63
29 has been received much attention recently [1], [2]. Spec- Nakagami-m fading channels. Finally, [7] presented the HOS 64
30 trum aggregation refers to obtaining larger amounts of radio of the channel capacity for several diversity receivers taking 65
IEE
31 resource by aggregating possible spectrum resources that lie into account the effects of independent and non-identically 66
32 in non-adjacent frequency bands. As a successful application distributed (i.n.i.d.) Nakagami-m fading channels. 67
33 of the spectrum aggregation, the carrier aggregation (CA) The common characteristic of the above mentioned 68
34 technology has been proposed in Long-Term-Evolution works [3], [5][7], however, is that they adopt the assumption 69
35 Advanced (LTE-A) standard, increasing the usable spectrum of homogeneous fading channels. It has been proved that the 70
36 by aggregating resource blocks (RBs) either within a given homogeneous fading is often unrealistic since the surfaces are 71
spatially correlated in practical propagation environments [8]. 72

Manuscript received July 19, 2016; revised October 24, 2016; accepted
December 4, 2016. This work was supported in part by the National Yet, very few results on the HOS of the channel capacity 73
Natural Science Foundation of China (Grant Nos. 61601020 and in non-homogeneous and composite fading conditions are 74
61371068), the Fundamental Research Funds for the Central Universities available. Only recently, the HOS of the channel capacity for 75
(Grant Nos. 2016RC013, 2016JBZ003, and 2014JBZ002). The associate
editor coordinating the review of this paper and approving it for publication dispersed spectrum cognitive radio (CR) systems over i.n.i.d. 76
was M. Uysal. - fading channels was obtained in [9]. While these prior 77
J. Zhang, X. Chen, X. Li, and Y. Liu are with the School of Electronics and works have significantly improved our knowledge on the HOS 78
Information Engineering, Beijing Jiaotong University, Beijing 100044, China
(e-mail: jiayizhang@bjtu.edu.cn). of the channel capacity, a general analytic framework of spec- 79
K. P. Peppas is with the Department of Informatics and Telecommunications, trum aggregation systems which will account for more realistic 80
University of Peloponnese, 22100 Tripoli, Greece (e-mail: peppas@uop.gr). fading models seems to be missing from the open literature. 81
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. To address such non-homogeneous and composite fading 82
Digital Object Identifier 10.1109/TCOMM.2016.2637926 environments, the generalized - and - shadowed fading 83
0090-6778 2016 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.
2 IEEE TRANSACTIONS ON COMMUNICATIONS
84 channels are recently introduced in [8], [10], and [11], respec-

85 tively. Compared with classic homogeneous fading models, the
86 - and - shadowed fading models exhibit excellent agree-
87 ment with measured land-mobile satellite, underwater acoustic,
88 and body communications fading channels [11]. Moreover,
89 the - fading channel includes the Rayleigh, Rician, and
90 Nakagami-m fading channels as special cases by setting the
91 parameters and to specific real positive values [12], while
92 the - shadowed fading channel includes One-side Gaussian,
Fig. 1. Spectrum aggregation systems over generalized fading channels.
93 Rayleigh, Rician, Nakagami-m, Hoyt, -, -, and Rician
94 shadowed fading channels as special cases [13].
noncontinuous band. By assuming M available frequency 141
On the other hand, recent wireless applications become
diversity bands, the end-to-end SNR,
95
, at the output of 142
increasingly complex and require more realistic channel mod- M
each UEs receiver is given by = i=1 i . Moreover,
96
of
143
97 els for performance evaluation purposes [14][16]. Because
each frequency diversity channel is assumed to be slow and 144
of the fact that the adopted fading models can describe a
frequency non-selective. Note that the end-to-end SNR has
98
145
99 plethora of realistic fading propagation scenarios, they can
a similar form of the SNR at the output of an MRC combiner. 146
100 serve as useful tools to this end. Motivated by these important
101 observations, we herein analytically investigate the HOS of the
102 channel capacity for spectrum aggregation systems over - A. - Fading Channels 147
103 and - shadowed fading channels. In particular, the main The - distribution can be regarded as a generalization 148
104
105
106
107
108
109
110
contributions of this paper can be summarized as:
ro
We first derive exact analytical expressions for the HOS
of the channel capacity for spectrum aggregation systems
over i.i.d., i.n.i.d - and i.i.d., correlated - shad-
owed fading channels, respectively. In contrast to exsiting
works on second order statistics, the analysis of the HOS
is still limited. It is worth noting that although the sta-
tistical characteristics of general fading models are very
of the classic Rician fading model for line-of-sight (LoS)
scenarios, and has been extensively used in spatially non-
homogeneous propagation environments. The - fading sig-
nal is a composition of clusters of multipath waves with
scattered waves of identical power with a dominant component
of arbitrary power found within each cluster. Furthermore, the
parameter represents the ratio between the total power of
the dominant components and the total power of the scattered
149
150
151
152
153
154
155
EP
111 156
112 complicated, our derived results can be readily evaluated waves, while is the number of clusters. The probability 157
113 and efficiently programmed in most standard software density function (PDF) of the sum of M i.i.d. squared - 158
114 packages (e.g., MATLAB and MATHEMATICA). random variables (RVs) is given by [8, eq. (10)] 159
115 Furthermore, the asymptotically high- and low-SNR M+1

M1
expressions for the HOS of the channel capacity are also M(1 + ) 2 2 (1 + )
116
f i.i.d ( ) = M1 M+1
exp 160
117 presented to get additional insights into the impact of eM 2 (M) 2
118 system parameters, such as fading parameters and number
(1 + ) M
119 of aggregating frequency bands. More importantly, some IM1 2 (1) 161
120 of asymptotic expressions are given in terms of simple

elementary functions. M+1 M1
121
M(1 + ) 2 2 (1 + )
IEE
122 With the help of the HOS of the channel capacity, we also = M1 M+1
exp 162
eM 2 (M) 2
provided useful performance metrics in terms of ergodic
M+2i1
123

124 capacity, amount of fading (AOF), amount of disper- 1 (1+ ) M
sion (AOD), skewness, and kurtosis. Moreover, numerical , 163
i ! (M + i)
125
126 results are provided to verify our analysis. Note that the i=0
127 presented analysis is very meaningful for communication (2) 164
128 systems to aggregate best available frequency bands in

where denotes the average SNR of each - RV, Iv () is 165
129 future spectrum-limited wireless networks.
the modified Bessel function of first kind [17, eq. (8.406.1)], 166
130 This paper is organized as follows. The spectrum aggrega- and () denotes the Gamma function [17, eq. (8.31)]. From 167
131 tion system and generalized - and - shadowed fading (1) to (2), we have used the identity of [17, eq. (8.445)] and 168
132 models are introduced in Section II. In Section III, we present
carried out some algebraic manipulations. 169
133 the derivation of the HOS of the channel capacity and other The PDF of the sum of M i.n.i.d. squared - RVs is given 170
134 important metrics. In Section IV, numerical results are shown by [18, eq. (4)] 171
135 to verify our present results. Finally, Section V concludes the

2
136 paper and summarizes the key findings. e U 1 k!ck (U 1) U
fi.n.i.d ( ) = L 172
137 II. S YSTEM AND C HANNEL M ODEL (2)U (U ) k=0 (U )k k 2
As illustrated in Fig. 1, the spectrum aggregation system

e 2 (k)q q+U 1 U q
138 k
139 exploits the benefits of frequency diversity by combining the = ck , (3) 173
140 instantaneous signal-to-noise ratios (SNRs), i , from each (2)U k=0 q=0 q! (U + q) 2
ZHANG et al.: HIGH-ORDER CAPACITY STATISTICS OF SPECTRUM AGGREGATION SYSTEMS 3
174 where (a) b denotes the Pochhammer symbol [17], where only elementary functions appear. Thus, (10) can facil- 208
M
175 U = i=1 i , and the series representation of itate the calculation involved the PDF expression of - 209
176 generalized Laguerre polynomial L vk () has been used as shadowed fading channels. 210
177 [19, eq. (05.08.02.0001.01)] Furthermore, the PDF of the sum of M correlated squared 211

k - shadowed RVs is given by [21. eq. (16)] 212
(v + k + 1) (k)q y q U
L vk (y) = . (4)
178
k! q! (v + q + 1) f cor ( ) = A U 1 e
q=0 213

Moreover, the coefficients ck in (3) can be obtained as

179

Dk 1 F1 m M + k, U ; , (11)
1
k1 214
180 ck = c j dk j , k 1 (5) k=0

1 + 1
M
m
k
j =0
j +1 where denotes the average SNR, A i=1 i , 215
jU M
M

dj i ai ( ai ) j 1 i=1 i (1 + i ), and
of
181 216
2 + ai (U )
(m M+k)
i=1 k
M j Dk = m M+k 1 + 1 , (12) 217
1 ai / (U )
182 + i , j 1 (6) k+1 M
1 + (ai /) (U/ 1) m q
i=1
U k+1 = 1 k+1i , (13) 218
U 1 i ai (U )
M k +1 i
q=1 i=1
183 c0 exp
2 + am (U ) 0 = 1. (14) 219
i=1
184
185
186
187

M

i=1
ai
1 + (U/ 1)

i
,
where i 2i i and ai i /2i (1 + i ). In order to

guarantee the uniform convergence of (3), the parameters
and should be chosen appropriately [20].
ro (7)
Moreover, min (1 , 2 , , M ) is the
C .

..
1
21

12
1
.
..

. ..

minimum
1M

value of the matrix DC with D = diag mi i represents a
diagonal matrix and C denotes the M M positive definite
matrix given by

2M

.. ,
.
eigen-
(15)
220
221
222
223
224
EP
188 B. - Shadowed Fading Channels
M1 1
189 Similar to the the same multipath/shadowing scheme used
190 in the Rician shadowed model, a natural generalization of the where 0 pq 1, 1 p, q M denotes the correlation 225
191 - distribution can be obtained by an LoS shadow fading coefficient of the dominating components of - shadowed 226
192 model. Unlike the - distribution, the - shadowed model RVs. With the help of (9), we can rewrite (11) as 227
assumes that all the dominant components are subject to U

193

194 the same common fluctuation due to shadowing. With the f cor ( ) = A U 1 e Dk 228
assumption of shadowing components are correlated, while

195 k=0

q
196 multipath components are uncorrelated, the PDF of the sum (m M + k)q
197 of M i.i.d. squared - shadowed RVs is given by [21] , (16) 229
(U )q q! 1 + 1
IEE
m M M1
M (1 + ) M
q=0
m
198 fi.i.d. ( ) =
m + (M)
2 (1 + ) III. H IGHER -O RDER C APACITY S TATISTICS 230
M(1+) M

199 e
1 F1 m M, M; In this section, we present the statistical analysis for the
( + m) 231
200 (8) derivation of the HOS of the channel capacity for spectrum 232
aggregation systems. Without loss of generality, the HOS of 233
201 where m denotes the shaping parameter of shadow- the channel capacity can be defined as [6], [9] 234
202 ing, and 1 F1 () is the confluent hypergeometric function

203 [17, eq. (9.210.1)]. By utilizing the following identity n = E logn2 (1 + ) , (17) 235

(a)q x q where n N is the order of the capacity statistics, and 236
204 1 F1 (a, b; x) = , (9) E () denotes the expectation operator. Note that the first-order
(b)q q!
237
q=0 statistics of channel capacity is well-known as the ergodic 238
205 we can rewrite (8) in an alternative form as capacity. Without loss of generality, the HOS of the channel 239
m M M(1+) capacity for the spectrum aggregation systems with M non- 240
M (1 + ) M m e
adjacent frequency bands is given by
fi.i.d. ( ) = 241

206
m + (M)

2 q M
M (1+) M+q1 n = logn 1 + i
(m M)
q (+m) 242
207 , (10) 0 0 0 i=1

(M)q q! f (1 , 2 , M ) d1 d2 d M , (18)
q=0 243
244 where f (1 , 2 , M ) represents the joint pdf of the instan- This integral can be evaluated numerically in an efficient 285
245 taneous SNRs of each band. Unfortunately, it is very tedious manner by employing a N-point Gauss-Chebyshev quadrature 286
246 and computationally cumbersome to obtain the joint pdf even rule as 287
247 for the simple i.i.d. case. One possible way to solve this
2
15
problem is to use the moment generating function (MGF) tk2 2+1
248
K (, , a) = +1 wk ln 1 + t (23) 288
249 based method proposed in [6]. However, the HOS of the a a k
k=1
250 channel capacity is given in terms of a single-integral expres-
251 sion, which makes it difficult to be mathematically employed. where wk and tk are the weights and abscissae given in [22]. 289
252 Therefore, in the following, we derive the HOS of the channel Note that we only need 15 terms in (23) to converge ade- 290
253 capacity for spectrum aggregation systems over generalized quately. 291
254 fading channels by utilizing the pdf of the total SNR . It By taking n = 1 in (19), we can obtain the first-order 292
255 is worthy to mention that all our derived results are given in statistics of the channel capacity, which is the well-known 293
256 analytical form, which is easy to show the key impacts of ergodic capacity as 294
of
257 system performance. (1+)
e M ( M)i
1 = 295
lnn 2 i ! (M + i)
i=0
258 A. - Fading Channels

259 We first consider the higher-order capacity statistics of
M+i1
(1 + ) M+ik1 M + i 1

spectrum aggregation systems over i.i.d. - fading channels
296
260 k
k=0
261 as follows.

262
263
264
265
be expressed as
n = M n
e
1

ln 2
J M + i,

i=0
( M)i
i ! (M + i )
(1 + )

,n ,

(1 + ) M+i
ro
Theorem 1: The higher-order capacity statistics of spec-
trum aggregation systems over i.i.d. - fading channels can
(19)
where

a
b
= b!(ab)!
a!
G-function [17, eq. (9.301)].

3,0 (1 + )
G 2,3

1, 1
0, 0, 1 + k
, and G() denotes the Meijers
Lemma 1: For the high- and low-SNR regimes, the higher-

order capacity statistics of spectrum aggregation systems
, (24) 297
298
299
300
EP
266 301

over i.i.d. - fading channels can be respectively expressed 302
267 where the auxiliary function J () is given by (47) in the as 303
Appendix A.
(1 + ) M+i
268
1 ( M)i
269 Proof: Substituting (2) into (17), we can derive n = M n 304
e ln 2 i ! (M + i )
M+1 i=0
M(1 + ) 2
270 n = (1 + )
M1
eM 2 (M) 2 lnn 2
M+1
Q M + i 1, ,n , (25) 305
M+2i1

1 (1 + ) M
n
0 n! Sk+n (k + n + M)
n =
271
i ! (M + i) M
306
n
(k + n)! (M)
IEE
i=0
e ln 2
k=0
(1+)
lnn (1 + ) M+i1 e d . (20) n+k

272
0 1 F1 (k + n + M; M; M) , 307
(1 + )
273 To the best of the authors knowledge, the integral in (20) is
274 not included in tables of classical reference books such as [17]. (26) 308
275 Nervertheless, as shown in Appendix A, it can be computed

276 in closed form thus completing the proof. where Smn is the Stirling number of the first kind 309
277 Note that the auxiliary function J () requires is restricted [17, eq. (9.740)], and the auxiliary function Q() is given 310
278 to integer values, which assumes finite numbers of multipath by (49) in the Appendix. B. 311
279 clusters. In the most general case of real , integrals of the Proof: By taking large values of in (17) and using (2), 312
280 form the higher-order capacity is given by 313
M+1
M(1 + ) 1
ln (1 + x) exp(ax)x dx
2
281 K (, , a) = (21)
n = M1 M+1
314
0 eM 2 (M) 2 lnn 2 i ! (M + i)
i=0
282 should be evaluated. By performing the change of variables M+2i1
ax = y 2 , (21) can be expressed as (1 + ) M
283
315

2 y2
K (, , a) = +1 ln 1 + exp(y 2 )y 2+1 dy. (1+)
a 0 a lnn ( ) M+i1 e d . (27) 316
284 (22) 0
317 With the aid of [23, eq. (2.5.1.7)], the integral in (27) can be i.n.i.d. - fading channels can be expressed as 353
318 calculated as
k
1 (k)q U q
M(1 + )
M+1
n = c k 354
(2)U lnn 2 k=0 q=0 q! (U + q) 2
2
319
n = M1 M+1
eM 2(M) lnn 2
2
M+2i1 1
Q q + U 1, ,n , (33) 355
1 (1 + ) M 2
320 q
i ! (M + i)
n!
k
(k)q U
i=0 0
n = n ck 356
dn (a + 1) ln 2 q! (U + q)
321 n . (28) k=0 q=0
da ( (1 + ) /)a+1 a=M+i1
S p+n (U
n
+ q + p + n)
. (34) 357
322 Then, the high-SNR HOS (25) can be derived by using (49)
p=0
( p + n)!(2)( p+n)
323 and after some algebraic manipulation.
of
324 Moreover, the low-SNR HOS can be obtained by taking Proof: With the help of [17, eq. (3.351.3)], the proof can 358
325 0, and using the well-known expansion of the logarithm be completed by following similar steps in Lemma 1. 359
326 function as [17, eq. (9.741.2)] Note that the low-SNR HOS of the channel capacity (34) 360
is given in terms of simple elementary functions. Therefore, 361

z k+n
327
n
ln (1 + z) = n! n
Sk+n , z0 (29) we can obtain the implication that the HOS is an increasing 362
(k + n)! function in U . 363

k=0
Substituting (29) into (17), and with the aid of (2) and (17),
ro
328
329 we derive the low-SNR HOS as B. - Shadowed Fading Channels 364
M+1 Now, we move on to consider the higher-order capacity 365

M(1 + ) n! n
Sk+n
330
0
n =
2
statistics for - shadowed fading channels. As a first step, 366
M1 M+1
eM 2 (M) 2 lnn 2 k=0 (k + n)! the case of i.i.d. - shadowed fading channel is investigated. 367
M1 (1 + ) 368
331 k+n+ 2 exp trum aggregation systems over i.i.d. - shadowed fading

369
0
channels can be expressed as 370
EP
(1 + ) M m M
332 IM1 2

d . (30) M (1 + ) M m 1
n = 371
m + (M)lnn 2
To evaluate the integral in (30), we can utilize the following
2 q
333
(m M) M (1+)
identity [24, eq. (3.15.2.5)] q (+m)
334
372
(M)q q!
q=0
x q exp ( px) Iv a x d x
M (1 + )
335
0
J q + M, ,n . (35) 373
(q +v/2+1) (a/2)v v a2
336 = 1 F1 q + +1; v +1; . (31)
(v + 1) pq+v/2+1 2 4p Proof: This result is a direct consequence of taking (10) 374
into (17), and using (47). 375

IEE
337 Finally, we arrive at the desired result in (26) after some basic
338 algebra. Lemma 3: For the high- and low-SNR regimes, the higher- 376
339 Note that the auxiliary function Q() can apply for arbitrary order capacity statistics of spectrum aggregation systems 377
340 positive real values of , so the asymptotical results are over i.i.d. - shadowed fading channels can be expressed 378
generalized. It is easy to see from (25) and (26) that the HOS as 379
m M
341
342 of the channel capacity is an increasing function in the average M (1 + ) M m 1

SNR and M. n = 380
343
m + (M)lnn 2
2 q
344
(m M) M (1+)
345 trum aggregation systems over i.n.i.d. - fading channels q (+m)
381
346 can be expressed as (M)q q!
q=0

1
k
(k)q U q M (1 + )
n = ck Q q + M 1, ,n , (36) 382
347
(2)U lnn 2 k=0 q=0 q! (U + q) 2
m M
0 m n!
1 n =
J q + U, ,n .
383
348 (32) m + (M)
2
n
Sk+n (k + n + M)
349 Proof: The proof is readily completed by taking (3) 384
(k + n)! (M (1 + ) / )k+n
350 into (17), and using (47). k=0

351 Lemma 2: For the high- and low-SNR regime, the higher-
2 F1 m M, k + n + M; M; . (37) 385
352 order capacity statistics of spectrum aggregation systems over + m
386 Proof: We can obtain (36) by following similar steps in TABLE I

387 Theorem 3. While for the case of the low-SNR regime, the N UMBER OF R EQUIRED T ERMS FOR C ONVERGENCE OF THE HOS OF THE
C HANNEL C APACITY FOR S PECTRUM A GGREGATION S YSTEMS OVER
388 integral identity [24, eq. (3.35.1.2)] i.i.d. - FADING C HANNELS W ITH Pe 105 , AND M = 3

q px (q +1)
389 x e 1 F1 (a, b; x) d x = 2 F1 a, q +1; b;
0 pq+1 p
390 (38)
391 should be invoked. Note that condition on the arguments
392 of (38), p > 0, p > , and q > 1, is satisfied.
393 Considering the case where the frequency bands are corre-
394 lated, we present the following theorem.
395 Theorem 4: The higher-order capacity statistics of spec-
396 trum aggregation systems over correlated - shadowed fad- to describe how far the channel capacity lies from the ergodic 430
of
397 ing channels can be expressed as capacity. Its normalization with respect to the ergodic capacity, 431
U
q 2
A (m M + k)q AoD = 2 1 1 , is called the amount of dispersion (AoD). 432
398 n = n Dk Furthermore, we can define the reliability percentage of the
ln 2 (U )q q!
1 + 1 433

k=0

q=0 signal throughput as R = 100(1 AoD). For good channel 434
quality, AoD approaches to zero and R 100. In addition, 435

399 J q + U, , n . (39)
the skewness is a metric of the degree of asymmetry for the 436
3
distribution of the channel capacity as S = 3 1 3 . For
ro
400 Proof: The proof can be completed by following similar 437
2 21
401 steps in Theorem 1. symmetric distributions, S = 0, while S < 0 denotes the 438
402 We note from (39) that the HOS of the channel capacity distribution is skewed to the left. In addition, the kurtosis cor- 439
403 is an increasing function in the number of frequency bands responds to the degree of peakedness of the channel capacity 440
404 M and shadowing parameter m and as such it obtains its 4
maximum value for m . around the ergodic capacity as K = 4 21 2 . Within this 441
405 2 1
406 Lemma 4: For the high- and low-SNR regimes, the higher- context, it is worth mentioning that these important statistical 442
407 order capacity statistics of spectrum aggregation systems over metrics of the channel capacity can be also efficiently and 443
EP
408 correlated - shadowed fading channels can be expressed accurately computed by using the HOS expressions. 444
409 as
U
q IV. N UMERICAL R ESULTS 445
A (m M +k)q

410 n = n Dk In this section, we present various performance evaluation 446
ln 2 (U )q q! 1+1 results using the HOS of the channel capacity expressions 447
k=0 q=0
presented in Sections III for spectrum aggregation systems
448
411 Q q + U 1, , n , (40) operating over - and - shadowed fading channels,

449
respectively. To validate the accuracy of the aforementioned 450
0
S np+n Dk ( p + n + M)
412 n = An! expressions, comparisons with complementary Monte-Carlo 451
( p + n)! (/ ) p+n simulated results with 106 realizations of random variables are 452
k=0 p=0

IEE
also included in these figures. We use the approaches presented 453

1
413 2 F1 m M + k, U + p + n; , U ; . (41) in [18] and [25] to generate random variables from the squared 454
1 + 1 - and - shadowed distributions, respectively. The impact 455
414 Proof: The proof concludes by following a similar line of of system and fading parameters on the HOS performance of 456
415 reasoning as in Lemma 3. spectrum aggregation systems are discussed in detailed. 457
416 According to Lemma 4, a higher increases the HOS of

417 the channel capacity. This is anticipated, since larger values A. Convergence of Derived Results 458
418 of reduce the correlation between frequency bands, making Since the derived results are given in sum of infinite series, 459
419 receiver signal more stronger. we prove the convergence of the derived results by truncating 460
the appropriate series expressions to achieve an accuracy 461

C. Practical Implementation of HOS
up to the fifth-significant digit (e.g., Pe 105 ). Table I
420
462
421 To evaluate the performance of spectrum aggregation sys- investigates the impact of the number of moments and fading 463
422 tems, several important measures will in discussed by using the parameters and on the convergence of the HOS of the 464
423 HOS of the channel capacity presented above. These measures channel capacity for M = 3. It can be seen from Table I that all 465
424 can also serve as useful tools for the design of practical infinite series rapidly converged with the speed of convergence 466
425 dispersed spectrum cognitive radio (DS-CR) systems. First of for the scenarios of interest. Moreover, the number of terms 467
426 all, the amount of fading (AoF) of the channel capacity or increases with increasing , and average SNR , while 468
427 the so called fading figure is defined as the ratio of variance has a noticeable impact on the convergence. For high-SNR
to the square ergodic capacity as AoF =
469
428
2
2
1 [7]. The regimes, the required number of terms for ergodic capacity 470
1
429 variance of the channel capacity is denoted by var = 2 21 (n = 1) is less than the case of the 4-th statistical moment. 471
TABLE II
N UMBER OF R EQUIRED T ERMS FOR C ONVERGENCE OF THE HOS OF THE
C ORRELATED - S HADOWED FADING C HANNELS W ITH Pe
103 , M = 2, 1 = 1, 2 = 5, 1 = 1, 2 = 2, AND m = 1
of
Fig. 3. Amount of dispersion of the channel capacity against the average SNR
for spectrum aggregation systems over i.n.i.d. - fading channels (2 =
3 = 1 dB, 1 = 2.5, 2 = 3.5, 3 = 4.75, and 1 = 1, 2 = 1, 3 = 2).
the high-SNR approximation is more accurate for low order 500
Fig. 2. ro
Simulated, analytical, and asymptotic HOS of the channel
capacity against the average SNR for spectrum aggregation systems over
i.i.d. - fading channels ( = 1, = 1, and M = 3).
statistics (e.g., ergodic capacity). It is also clear from Fig. 2
that the HOS curves get closer to each other almost around
6 dB, which determines the boundary of the high- and low-
SNR regimes. Base on the interesting finding of n = 1 at
around 6 dB, we can simply model the HOS of the channel
capacity as n n1 . Due to the nth power of the ergodic
capacity, the behavior of the HOS of capacity is different in
the high- and low-SNR regimes, respectively. Furthermore, the
501
502
503
504
505
506
507
EP
508
crossing point will be shifted toward left, and therefore, the 509
472 However, only a relatively small number of terms is required ergodic capacity increasing if increasing the values of fading 510
473 for the desired accuracy. For the worst case of n = 4, parameters (e.g., and ) of each frequency band. 511
474 = 10 dB, = 1, and = 2, the maximum number of The effect of the number of frequency bands M on the AoD 512
475 terms is 23. performance of spectrum aggregation systems over i.n.i.d. 513
476 For correlated - shadowed fading channels, the effect of - fading channels is shown in Fig. 3. One can notice that 514
477 correlation coefficient on the convergence has been studied the AoD appears to increase for low and moderate SNRs, 515
478 in Table II. It is clear that the derived results are rapidly con- while it begins to decrease for the high-SNR regime for all 516
479 verged with less than 40 terms of infinite series. Note that as cases. Furthermore, it can be seen from the Fig. 3 that the 517
480 reduces, the required number of terms decreases. Furthermore, AoD plot becomes peaky at around 9 dB for M = 3, while 518
the computation time of derived analytical results is much less
IEE
481
the AoD reaches its highest value around 6 dB for M = 1. 519
482 than the one of simulations. For example, we spend less than With respect to the reliability percentage of the spectrum 520
483 20 seconds to calculate (19) by using MATHEMATICA, while aggregation system, the transmit SNR should be chosen greater 521
484 the simulation costs more than 230 seconds to derive the same than the SNR for which the AoD peaks. For example, for 522
485 result. Note that other cases have similar fact of converging the case of M = 3, the maximum AoD is 0.1919 and the 523
486 steadily and rapidly, and requiring little computational effort, reliability percentage is 90.81%, which means that the average 524
487 which are validated by our conducted numerical experiments. SNR must be chosen equal to or greater than 9 dB in order to 525
reach at least 90.81% reliable transmission. Moreover, the gap 526
488 B. Performance Analysis between different number of bands M decreases at high SNRs 527
489 For spectrum aggregation systems over i.i.d. - fading which implies that the fading effect becomes less pronounced 528
490 channels, the simulated, analytical (19), and asymptotic (25), as anticipated. 529
491 (26) HOS of the channel capacity are plotted against the Figure 4 depicts the AoF of the channel capacity for 530
492 average SNR in Fig. 2. As seen, the analytical results spectrum aggregation systems over i.i.d. - shadowed fading 531
493 perfectly match the Monte-Carlo simulations. Clearly, the channels as a function of average SNR per band for 532
494 high-SNR approximations are sufficiently tight and become different sets of frequency bands M. It is clear that the AoF 533
495 exact even at moderate SNR values, while a precise agreement decreases drastically as the value of increases. At low SNRs, 534
496 between the exact and low-SNR results can be observed. This the AoF performance of the spectrum aggregation system is 535
497 implies that they can efficiently predict the HOS of the channel significantly improved with increasing the value of frequency 536
498 capacity over a wide SNR range. Moreover, the gap between bands and M. For example, the AoF is 0.665 for the case of 537
499 the corresponding curves increases as n increases which means M = 1 at 10 dB, while it reduces to 0.148 for M = 3. 538
of
Fig. 4. Amount of fading of the channel capacity against the average SNR Fig. 6. Skewness of the channel capacity against the average SNR for
for spectrum aggregation systems over i.i.d. - shadowed fading channels spectrum aggregation systems over correlated - shadowed fading channels
( = = 2 and m = 1). (M = 3, i = 1, i = 2, and m = 1).
ro
EP
Fig. 5. Amount of fading of the channel capacity against the average SNR Fig. 7. Kurtosis of the channel capacity against the average SNR for spectrum
for spectrum aggregation systems over i.i.d. - shadowed fading channels aggregation systems over correlated - shadowed fading channels (M = 3,
(M = 2). i = 1, i = 2, and m = 1).
539 Therefore, we can use more frequency bands to combat the correlation models pq = | pq| , where = 0.1, 0.5, 0.9. 560
IEE
540 low SNRs of fading channels. It is clear to see from Figs. 6 and 7 that the skewness and 561
541 The effect of fading parameters , and m on the AoF of the kurtosis increases as average SNR of each frequency 562
542 the channel capacity for spectrum aggregation systems over band increases and/or decreases, showing that the pdf 563
543 i.i.d. - shadowed fading channels are further investigated of the channel capacity becomes more spiky with heavy 564
544 in Fig. 5. As indicated by analysis in Section III, the AoF tails and asymmetric. More importantly, the gap between the 565
545 decreases with a smaller value of (more power of LoS corresponding curves decreases as decreases which implies 566
546 components) and a higher value of (more power of clusters), that its effect becomes less pronounced. 567
547 where the fading channel becomes less deterministic. This

548 finding reveals that more scattered waves are beneficial for V. C ONCLUSION 568
549 improved AoF. One can also notice the increase of the AoF In this paper, we investigate the performance of spectrum 569
550 can be obtained for decreasing the shadowing parameter m, aggregation systems over generalized fading channels. In par- 570
551 especially for low SNRs. For example, the value of AoF ticular, we consider two recently proposed generalized fading 571
552 is 0.275 for the case of m = 1, = 1, = 2 and = 10 dB, models, namely - and - shadowed, which can model 572
553 while it reduces to 0.228 for the case of m = 2, = 1, propagation phenomena involving LoS and composite fading 573
554 = 2 and = 10 dB. The impact of fading appears to environments, respectively. Novel and exact expressions for 574
555 be particularly critical for low SNRs, while in the high-SNR the HOS of the channel capacity of spectrum aggregation 575
556 regime its impact is relatively reduced. systems are derived. Our derived expressions can extend 576
557 In Figs. 6 and 7, Skewness and Kurtosis statistics are plotted and complement existing results on classical fading models. 577
558 against the average SNR , respectively. We consider corre- Furthermore, we deduce simple HOS expressions for the 578
559 lated - shadowed fading channels with three exponential asymptotically low- and high-SNR regimes. Note that all 579
580 infinite series can be computationally efficient, accurate, and By substituting = 1 into (45), we can derive the auxiliary 616
581 requires only a relative small number of terms for yield- function J (a, b, n) as 617
582 ing accurate results. Important performance metrics, such as

J (a, b, n) 618
583 ergodic capacity, variance, AoF, AoD, skewness, and kur-
584 tosis, are also derived to show the effects of system and

a1
ak1 a 1
fading parameters on spectrum aggregation systems. Finally, = n!e b b1k
(1)
585
619
586 extensive Monte-Carlo simulations verify the accuracy of the k
k=0
587 analytical expressions and the tightness of the high- and
n+1 1 s

588 low-SNR bounds. The proposed analysis is useful for the
! "# $
589 spectrum aggregation system design engineer for performance 1, 1, , 1
n+2,0
G n+1,n+2 b . (47)
590 evaluation purposes. 0, 0, , 0, 1 + k 620
# $! "

n+1 0 s
591 A PPENDIX
of
592 A. A Useful Integral Identity B. High-Order Differentiation 621
593 Let us consider an integral in the form of With the help of Leibnizs rule [17], the nth differentiation 622
of the product of the gamma function and the power functions 623
594 (a, b) = (1 + x) x a1 ebx d x, (42) can be expressed as 624

0
d n (a + 1)
where a N, b > 0, and C. With the help of [26, eq. (39)], Q (a, b, n) = 625
da n ba+1
ro
595
596 (42) can be also expressed as

n (1)nk lnnk (b) d k (a + 1)
n
= . (48) 626
597 (a, b) = (a) U (a, a + + 1; b) k ba+1 da k
k=0
a1

a 1 (1)ak1 Utilizing the high-order differentiation of the gamma function
=e b
( + k + 1, b)
627
598
k b+k+1 [5, eq. (10)], we can derive 628
k=0
599 (43) Q (a, b, n) 629
n
EP
n (1)nk lnnk (b) k!
600 where U () denotes the Tricomi hypergeometric function = 630
k ba+1
601 [27, eq. (13.1.3)], and (, ) is the upper complementary k=0

602 incomplete gamma function [17, eq. (8.350.2)]. By using k+1 1 s
! "# $
603 Leibnizs rule [17], the nth order derivative of (43) can be k+2,0 1, 1, , 1
604 evaluated as
G k+1,k+2 1 1 + a, 0, 0, , 0 631
# $! "
d n (a, b)
k+1 0 s
605

d n =+k+1 k+1 1 s
! "# $
a1
1, 1, , 1
a 1 (1)ak1 + (1)k G 1,k+1
= eb k+1,k+2 1 1 + a, 0, 0, , 0 . (49) 632

606
k b+k+1 # $! "
IEE
k=0
n
k+1 0 s
n
d (, b)
n n p 1
ln (44)
d n =+k+1
607
p b R EFERENCES 633
p=0
a1
[1] H. Bogucka, P. Kryszkiewicz, and A. Kliks, Dynamic spectrum aggre-
a 1 (1)ak1
634
608 = n!eb gation for future 5G communications, IEEE Commun. Mag., vol. 53, 635
k b+k+1 no. 5, pp. 3543, Jun. 2015. 636
k=0
[2] R. Zhang, M. Wang, L. X. Cai, Z. Zheng, and X. Shen, LTE-unlicensed: 637
n+1 1 s The future of spectrum aggregation for cellular networks, IEEE Wireless
! "# $
638

1, 1, , 1 Commun., vol. 22, no. 3, pp. 150159, Jun. 2015. 639
n+2,0
G n+1,n+2 b , (45) [3] F. Yilmaz, H. Tabassum, and M.-S. Alouini, On the computation of 640
609
0, 0, , 0, + k
# $! " the higher order statistics of the channel capacity for amplify-and- 641

forward multihop transmission, IEEE Trans. Veh. Technol., vol. 63, 642
n+1 0 s no. 1, pp. 489494, Jan. 2014. 643
[4] M. K. Simon and M.-S. Alouini, Digital Communication Over Fading 644
610 From (44) to (45), we have used the differentiation identity Channels. Hoboken, NJ, USA: Wiley, 2005. 645
611 [19, eq. (06.06.20.0013.01)] and performed some algebraic [5] F. Yilmaz and M.-S. Alouini, Novel asymptotic results on the high- 646
order statistics of the channel capacity over generalized fading channels, 647
612 manipulations. in Proc. IEEE SPAWC, Jun. 2012, pp. 389393. 648
613 Based on (42), the nth order derivative of (a, b) can [6] F. Yilmaz and M.-S. Alouini, On the computation of the higher-order 649
614 alternatively be given by statistics of the channel capacity over generalized fading channels, IEEE 650
Wireless Commun. Lett., vol. 1, no. 6, pp. 573576, Jun. 2012.

651
[7] N. C. Sagias, F. I. Lazarakis, A. A. Alexandridis, K. P. Dangakis, and 652
615 J (a, b, n) = (1 + x) lnn (1 + x) x a1 ebx d x. (46) G. S. Tombras, Higher order capacity statistics of diversity receivers, 653
0 Wireless Pers. Commun., vol. 56, no. 4, pp. 649668, Feb. 2011. 654
655 [8] M. D. Yacoub, The - distribution and the - distribution, IEEE Xiaoyu Chen received the B.Sc. degree in commu- 730
656 Antennas Propag. Mag., vol. 49, no. 1, pp. 6881, Feb. 2007. nication engineering from Beijing Jiaotong Univer- 731
657 [9] T. Tsiftsis, F. Foukalas, G. Karagiannidis, and T. Khattab, On the sity, China, in 2015, where she is currently pursuing 732
658 higher-order statistics of the channel capacity in dispersed spectrum the M.S. degree from Wireless Self-Organized Com- 733
659 cognitive radio systems over generalized fading channels, IEEE Trans. munications Group, School of Electronic and Infor- 734
660 Veh. Technol., vol. 65, no. 5, pp. 38183823, May 2016. mation Engineering. Her research interests include 735
661 [10] J. F. Paris, Statistical characterization of - shadowed fading, IEEE the performance analysis of multi-hop and OFDM 736
662 Trans. Veh. Technol., vol. 63, no. 2, pp. 518526, Feb. 2014. communications. 737
663 [11] S. L. Cotton, Human body shadowing in cellular device-to-device
664 communications: Channel modeling using the shadowed fading model,
665 IEEE J. Sel. Areas Commun., vol. 33, no. 1, pp. 111119, Jan. 2015.
666 [12] J. Zhang, L. Dai, W. H. Gerstacker, and Z. Wang, Effective capacity of
667 communication systems over shadowed fading channels, Electron.
668 Lett., vol. 51, no. 19, pp. 15401542, Sep. 2015.
669 [13] J. Zhang, Z. Tan, H. Wang, Q. Huang, and L. Hanzo, The effective
670 throughput of MISO systems over - fading channels, IEEE Trans.
671 Veh. Technol., vol. 63, no. 2, pp. 943947, Feb. 2014.
of
672 [14] M. Dohler, R. W. Heath, A. Lozano, C. B. Papadias, and
673 R. A. Valenzuela, Is the PHY layer dead? IEEE Commun. Mag.,
674 vol. 49, no. 4, pp. 159165, Apr. 2011. Kostas P. Peppas (M09SM14) was born in 738
675 [15] K. Guan, Z. Zhong, J. I. Alonso, and C. Briso-Rodriguez, Measurement Athens, Greece, in 1975. He received the Diploma 739
676 of distributed antenna systems at 2.4 GHz in a realistic subway tunnel degree in electrical and computer engineering and 740
677 environment, IEEE Trans. Veh. Technol., vol. 61, no. 2, pp. 834837, the Ph.D. degree in wireless communications from 741
678 Feb. 2012. the National Technical University of Athens, in 1997 742
679 [16] J. Zhang, M. Matthaiou, Z. Tan, and H. Wang, Performance analysis and 2004, respectively. From 2004 to 2007, he was 743
680 of digital communication systems over composite -/gamma fading with the University of Peloponnese, Department of 744
681 channels, IEEE Trans. Veh. Technol., vol. 61, no. 7, pp. 31143124, Computer Science, Tripoli, Greece, and from 2008 745
682
683
684
685
686
687
688
689
690
Sep. 2012.
Products, 7th ed. San Diego, CA, USA: Academic, 2007.
Technol., vol. 61, no. 1, pp. 413419, Jan. 2012.
ro
[17] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
[18] K. P. Peppas, Sum of nonidentical squared - variates and applications

in the performance analysis of diversity receivers, IEEE Trans. Veh.
[19] Wolfram. The Wolfram Functions Site, accessed on Sep. 12, 2016.
[Online]. Available: http://functions.wolfram.com
[20] A. Castao-Martnez and F. Lpez-Blzquez, Distribution of a sum
of weighted noncentral chi-square variables, TEST, vol. 14, no. 2,
to 2014 he was also with the National Centre for
Scientific Research-Demokritos, Institute of Infor-
matics and Telecommunications as a Researcher.
Since 2014, he has been with the Department of Telecommunication Science
and Technology, University of Peloponnese, where he is currently a Lecturer.
He has authored over 70 journal and conference papers. His current research
interests include digital communications, MIMO systems, cooperative com-
munications, wireless and personal communication networks, cognitive radio,
digital signal processing, system level analysis, and design.
746
747
748
749
750
751
752
753
754
EP
691
692 pp. 397415, Dec. 2005.
693 [21] M. R. Bhatnagar, On the sum of correlated squared - shadowed
694 random variables and its application to performance analysis of MRC,
695 IEEE Trans. Veh. Technol., vol. 64, no. 6, pp. 26782684, Jun. 2015.
696 [22] N. M. Steen, G. D. Byrne, and E. M. Gelbard, Gaussian quadratures
( (
for the integrals 0 ex f (x)dx and 0b ex f (x)dx, Math. Comput.,
2 2
697
698 vol. 23, no. 107, pp. 661671, 1969.
699 [23] A. P. Prudnikov, I. U. A. Brychkov, and O. I. Marichev, Integrals
700 and Series: More Special Functions, vol. 3. New York, NY, USA: Xu Li received the B.Sc. degree from Jilin Uni- 755
701 Gordon and Breach, 1990. versity, China, in 1991, and the Ph.D. degree from 756
702 [24] A. P. Prudnikov, I. U. A. Brychkov, and O. I. Marichev, Integrals Northeastern University, China, in 1997. Since 2003, 757
703 and Series: Direct Laplace Transforms, vol. 4. New York, NY, USA: she has been a Professor and Head of Wireless Self- 758
704 Gordon and Breach, 1992. Organized Communications Group, School of Elec- 759
IEE
705 [25] L. Moreno-Pozas, F. J. Lopez-Martinez, J. Paris, and E. Martos-Naya, tronic and Information Engineering, Beijing Jiaotong 760
706 The - shadowed fading model: Unifying the - and - distri- University, China. She holds over 40 patents. Her 761
707 butions, IEEE Trans. Veh. Technol., vol. 65, no. 12, pp. 96309641, research interest includes wireless self-organized 762
708 Dec. 2016. communications, sensor networks, and unmanned 763
709 [26] M. Kang and M. S. Alouini, Capacity of MIMO Rician channels, aerial vehicle systems. 764
710 IEEE Trans. Wireless Commun., vol. 5, no. 1, pp. 112122, Jan. 2006.
711 [27] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
712 With Formulas, Graphs, and Mathematical Tables, 9th ed. New York,
713 NY, USA: Dover, 1964.
714 Jiayi Zhang (S08M14) received the B.Sc. and

715 Ph.D. degrees in communication engineering from
716 Beijing Jiaotong University, China, in 2007 and
717 2014, respectively. From 2012 to 2013, he was
718 a visiting Ph.D. student with the Wireless Group,
719 University of Southampton, U.K. From 2014 to Ying Liu received the B.Sc. and Ph.D. degrees from 765
720 2015, he was a Humboldt Research Fellow with Jilin University, China, in 1985 and 2001, respec- 766
721 Institute for Digital Communications, University of tively. Since 2005, she has been a Professor with the 767
722 ErlangenNuermberg, Germany. From 2014 to 2016, School of Electronic and Information Engineering, 768
723 he was a Post-Doctoral Research Associate with Beijing Jiaotong University, China, where she is 769
724 the Department of Electronic Engineering, Tsinghua the Vice Dean of School of Electronic and Infor- 770
725 University, China. Since 2016, he has been a Professor with the School mation Engineering. Her research interest includes 771
726 of Electronic and Information Engineering, Beijing Jiaotong University. His wireless self-organized communications, sensor net- 772
727 current research interests include massive MIMO, unmanned aerial vehicle works, and wireless localization systems. 773
728 systems, and performance analysis of generalized fading channels. He was an
729 Exemplary Reviewer of the IEEE C OMMUNICATIONS L ETTERS in 2015.
IEEE TRANSACTIONS ON COMMUNICATIONS 1
On High-Order Capacity Statistics of Spectrum

Aggregation Systems Over - and -
Shadowed Fading Channels
Jiayi Zhang, Member, IEEE, Xiaoyu Chen, Kostas P. Peppas, Senior Member, IEEE, Xu Li, and Ying Liu
1 Abstract The frequency scarcity imposed by fast growing band or in different frequency bands [2]. In order to achieve a 37
2 demand for mobile data service requires promising spectrum successful spectrum aggregation, the maximum dispersion in 38
3 aggregation systems. The so-called higher order statistics (HOS) the channel capacity should be calibrated to leverage a reliable 39
4 of the channel capacity is a suitable metric on the system perfor-
5 mance. While prior relevant works have improved our knowledge transmission [3]. 40
of
6 on the HOS characterization of spectrum aggregation systems, In this context, the typical metric for performance evaluation 41
7 an analytical framework encompassing generalized fading models has been the higher-order statistics (HOS) of the channel 42
8 of interest is not yet available. In this paper, we pursue a detailed capacity, which can fully explore the reliability of the signal 43
9 HOS analysis of - and - shadowed fading channels by transmission in spectrum aggregation systems. As an useful 44
10 deriving novel and exact expressions. Furthermore, the simplified
11 HOS expressions for the asymptotically low and high signal-to- tool, the HOS can effectively describe the channel capac- 45
12 noise regimes are derived. Several important statistical measures, ity dispersion induced by the heterogeneity that inherently 46
13 such as amount of fading, amount of dispersion, reliability, exists in spectrum aggregation systems [4]. Moreover, fruitful 47
ro
14 skewness, and kurtosis, are obtained by using the HOS results. insights into the implications of the spectrum aggregation on 48
15 More importantly, the useful implications of system and fading the transmission reliability can be extracted by deriving HOS 49
16 parameters on spectrum aggregation systems are investigated for
17 channel selection. Finally, all derived expressions are validated of the channel capacity. Despite its importance, however, the 50
18 via Monte Carlo simulations. HOS of the channel capacity received relatively little attention 51
in the literature, due in part to the intractability of its analysis. 52

19 Index Terms Spectrum aggregation, higher-order statistics,
20 -, - shadowed. A number of prior works have investigated the HOS of the 53
channel capacity of different wireless systems over several flat 54
fading channels [3], [5][7]. For example, a generic framework

EP
55
21 I. I NTRODUCTION
for the asymptotic HOS of the channel capacity over inde- 56
22
23
24
W ITH the constantly growing mobile data demand
for future wireless communication systems, i.e., 5th
Generation (5G), it becomes more and more difficult to
pendent and identically distributed (i.i.d.) Nakagami-m fading
channels was provided in [5]. Yilmaz et al. [3] investigated the
HOS of the channel capacity for amplify-and-forward (AF)
57
58
59
25 allocate a wide and contiguous frequency band to each user multihop systems over gamma and generalized gamma fading 60
26 equipment (UE) and base station (BS). This has brought about channels. In addition, an MGF-based approach for the HOS 61
27 increasing scarcity in available radio spectrum. To address of the channel capacity for L-branch MRC receivers has been 62
28 these issues, the promising spectrum aggregation technique proposed in [6] with an example application of correlated 63
29 has been received much attention recently [1], [2]. Spec- Nakagami-m fading channels. Finally, [7] presented the HOS 64
30 trum aggregation refers to obtaining larger amounts of radio of the channel capacity for several diversity receivers taking 65
IEE
31 resource by aggregating possible spectrum resources that lie into account the effects of independent and non-identically 66
32 in non-adjacent frequency bands. As a successful application distributed (i.n.i.d.) Nakagami-m fading channels. 67
33 of the spectrum aggregation, the carrier aggregation (CA) The common characteristic of the above mentioned 68
34 technology has been proposed in Long-Term-Evolution works [3], [5][7], however, is that they adopt the assumption 69
35 Advanced (LTE-A) standard, increasing the usable spectrum of homogeneous fading channels. It has been proved that the 70
36 by aggregating resource blocks (RBs) either within a given homogeneous fading is often unrealistic since the surfaces are 71
spatially correlated in practical propagation environments [8]. 72

Manuscript received July 19, 2016; revised October 24, 2016; accepted
December 4, 2016. This work was supported in part by the National Yet, very few results on the HOS of the channel capacity 73
Natural Science Foundation of China (Grant Nos. 61601020 and in non-homogeneous and composite fading conditions are 74
61371068), the Fundamental Research Funds for the Central Universities available. Only recently, the HOS of the channel capacity for 75
(Grant Nos. 2016RC013, 2016JBZ003, and 2014JBZ002). The associate
editor coordinating the review of this paper and approving it for publication dispersed spectrum cognitive radio (CR) systems over i.n.i.d. 76
was M. Uysal. - fading channels was obtained in [9]. While these prior 77
J. Zhang, X. Chen, X. Li, and Y. Liu are with the School of Electronics and works have significantly improved our knowledge on the HOS 78
Information Engineering, Beijing Jiaotong University, Beijing 100044, China
(e-mail: jiayizhang@bjtu.edu.cn). of the channel capacity, a general analytic framework of spec- 79
K. P. Peppas is with the Department of Informatics and Telecommunications, trum aggregation systems which will account for more realistic 80
University of Peloponnese, 22100 Tripoli, Greece (e-mail: peppas@uop.gr). fading models seems to be missing from the open literature. 81
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. To address such non-homogeneous and composite fading 82
Digital Object Identifier 10.1109/TCOMM.2016.2637926 environments, the generalized - and - shadowed fading 83
0090-6778 2016 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.
84 channels are recently introduced in [8], [10], and [11], respec-

85 tively. Compared with classic homogeneous fading models, the
86 - and - shadowed fading models exhibit excellent agree-
87 ment with measured land-mobile satellite, underwater acoustic,
88 and body communications fading channels [11]. Moreover,
89 the - fading channel includes the Rayleigh, Rician, and
90 Nakagami-m fading channels as special cases by setting the
91 parameters and to specific real positive values [12], while
92 the - shadowed fading channel includes One-side Gaussian,
Fig. 1. Spectrum aggregation systems over generalized fading channels.
93 Rayleigh, Rician, Nakagami-m, Hoyt, -, -, and Rician
94 shadowed fading channels as special cases [13].
noncontinuous band. By assuming M available frequency 141
On the other hand, recent wireless applications become
diversity bands, the end-to-end SNR,
95
, at the output of 142
increasingly complex and require more realistic channel mod- M
each UEs receiver is given by = i=1 i . Moreover,
96
of
143
97 els for performance evaluation purposes [14][16]. Because
each frequency diversity channel is assumed to be slow and 144
of the fact that the adopted fading models can describe a
frequency non-selective. Note that the end-to-end SNR has
98
145
99 plethora of realistic fading propagation scenarios, they can
a similar form of the SNR at the output of an MRC combiner. 146
100 serve as useful tools to this end. Motivated by these important
101 observations, we herein analytically investigate the HOS of the
102 channel capacity for spectrum aggregation systems over - A. - Fading Channels 147
103 and - shadowed fading channels. In particular, the main The - distribution can be regarded as a generalization 148
104
105
106
107
108
109
110
contributions of this paper can be summarized as:
ro
We first derive exact analytical expressions for the HOS
of the channel capacity for spectrum aggregation systems
over i.i.d., i.n.i.d - and i.i.d., correlated - shad-
owed fading channels, respectively. In contrast to exsiting
works on second order statistics, the analysis of the HOS
is still limited. It is worth noting that although the sta-
tistical characteristics of general fading models are very
of the classic Rician fading model for line-of-sight (LoS)
scenarios, and has been extensively used in spatially non-
homogeneous propagation environments. The - fading sig-
nal is a composition of clusters of multipath waves with
scattered waves of identical power with a dominant component
of arbitrary power found within each cluster. Furthermore, the
parameter represents the ratio between the total power of
the dominant components and the total power of the scattered
149
150
151
152
153
154
155
EP
111 156
112 complicated, our derived results can be readily evaluated waves, while is the number of clusters. The probability 157
113 and efficiently programmed in most standard software density function (PDF) of the sum of M i.i.d. squared - 158
114 packages (e.g., MATLAB and MATHEMATICA). random variables (RVs) is given by [8, eq. (10)] 159
115 Furthermore, the asymptotically high- and low-SNR M+1 M1

expressions for the HOS of the channel capacity are also M(1 + ) 2 (1 + )
2
116
f i.i.d ( ) = M1 M+1
exp 160
117 presented to get additional insights into the impact of eM 2 (M) 2
118 system parameters, such as fading parameters and number
(1 + ) M
119 of aggregating frequency bands. More importantly, some IM1 2 (1) 161
120 of asymptotic expressions are given in terms of simple

elementary functions. M+1 M1
121
M(1 + ) 2 2 (1 + )
IEE
122 With the help of the HOS of the channel capacity, we also = M1 M+1
exp 162
eM 2 (M) 2
provided useful performance metrics in terms of ergodic
M+2i1
123

124 capacity, amount of fading (AOF), amount of disper- 1 (1+ ) M
sion (AOD), skewness, and kurtosis. Moreover, numerical , 163
i ! (M + i)
125
126 results are provided to verify our analysis. Note that the i=0
127 presented analysis is very meaningful for communication (2) 164
128 systems to aggregate best available frequency bands in

where denotes the average SNR of each - RV, Iv () is 165
129 future spectrum-limited wireless networks.
the modified Bessel function of first kind [17, eq. (8.406.1)], 166
130 This paper is organized as follows. The spectrum aggrega- and () denotes the Gamma function [17, eq. (8.31)]. From 167
131 tion system and generalized - and - shadowed fading (1) to (2), we have used the identity of [17, eq. (8.445)] and 168
132 models are introduced in Section II. In Section III, we present
carried out some algebraic manipulations. 169
133 the derivation of the HOS of the channel capacity and other The PDF of the sum of M i.n.i.d. squared - RVs is given 170
134 important metrics. In Section IV, numerical results are shown by [18, eq. (4)] 171
135 to verify our present results. Finally, Section V concludes the

2
136 paper and summarizes the key findings. e U 1 k!ck (U 1) U
fi.n.i.d ( ) = L 172
137 II. S YSTEM AND C HANNEL M ODEL (2)U (U ) (U )k k 2
k=0
As illustrated in Fig. 1, the spectrum aggregation system

e 2 (k)q q+U 1 U q
138 k
139 exploits the benefits of frequency diversity by combining the = ck , (3) 173
instantaneous signal-to-noise ratios (SNRs), i , from each (2)U q! (U + q) 2

140 k=0 q=0
174 where (a) b denotes the Pochhammer symbol [17], where only elementary functions appear. Thus, (10) can facil- 208
M
175 U = i=1 i , and the series representation of itate the calculation involved the PDF expression of - 209
176 generalized Laguerre polynomial L vk () has been used as shadowed fading channels. 210
177 [19, eq. (05.08.02.0001.01)] Furthermore, the PDF of the sum of M correlated squared 211
- shadowed RVs is given by [21. eq. (16)]

(v + k + 1)
k 212
(k)q y q U
L vk (y) = . (4)
178
k! q! (v + q + 1) f cor ( ) = A U 1 e
q=0 213

Moreover, the coefficients ck in (3) can be obtained as

179

Dk 1 F1 m M + k, U ; , (11)
1
k1 214
180 ck = c j dk j , k 1 (5) k=0

1 + 1
M
m
k
j =0
j +1 where denotes the average SNR, A i=1 i , 215
jU M
M

dj i ai ( ai ) j 1 i=1 i (1 + i ), and
of
181 216
2 + ai (U )
(m M+k)
i=1 k
M j Dk = m M+k 1 + 1 , (12) 217
1 ai / (U )
182 + i , j 1 (6) k+1 M
1 + (ai /) (U/ 1) m q
i=1
U k+1 = 1 k+1i , (13) 218
U 1 i ai (U )
M k +1 i
q=1 i=1
183 c0 exp
2 + am (U ) 0 = 1. (14) 219
i=1
184
185
186
187

M
ai
1 + (U/ 1)
i=1

i
,
where i 2i i and ai i /2i (1 + i ). In order to

guarantee the uniform convergence of (3), the parameters
and should be chosen appropriately [20].
ro
(7)
Moreover, min (1 , 2 , , M ) is the
C .

..
1
21

12
1
.
..

. ..

minimum
1M

value of the matrix DC with D = diag mi i represents a
diagonal matrix and C denotes the M M positive definite
matrix given by

2M

.. ,
eigen-
(15)
220
221
222
223
224
EP
B. - Shadowed Fading Channels .
188

M1 1
189 Similar to the the same multipath/shadowing scheme used
190 in the Rician shadowed model, a natural generalization of the where 0 pq 1, 1 p, q M denotes the correlation 225
191 - distribution can be obtained by an LoS shadow fading coefficient of the dominating components of - shadowed 226
192 model. Unlike the - distribution, the - shadowed model RVs. With the help of (9), we can rewrite (11) as 227
assumes that all the dominant components are subject to U

193

194 the same common fluctuation due to shadowing. With the f cor ( ) = A U 1 e Dk 228
assumption of shadowing components are correlated, while

195 k=0

q
196 multipath components are uncorrelated, the PDF of the sum (m M + k)q
197 of M i.i.d. squared - shadowed RVs is given by [21] , (16) 229
(U )q q! 1 + 1
IEE
m M M1
M (1 + ) M
q=0
m
198 fi.i.d. ( ) =
m + (M)
2 (1 + ) III. H IGHER -O RDER C APACITY S TATISTICS 230
M(1+) M

199 e
1 F1 m M, M; In this section, we present the statistical analysis for the
( + m) 231
200 (8) derivation of the HOS of the channel capacity for spectrum 232
aggregation systems. Without loss of generality, the HOS of 233
201 where m denotes the shaping parameter of shadow- the channel capacity can be defined as [6], [9] 234
202 ing, and 1 F1 () is the confluent hypergeometric function

203 [17, eq. (9.210.1)]. By utilizing the following identity n = E logn2 (1 + ) , (17) 235

(a)q x q where n N is the order of the capacity statistics, and 236
204 1 F1 (a, b; x) = , (9) E () denotes the expectation operator. Note that the first-order
(b)q q!
237
q=0 statistics of channel capacity is well-known as the ergodic 238
205 we can rewrite (8) in an alternative form as capacity. Without loss of generality, the HOS of the channel 239
m M M(1+) capacity for the spectrum aggregation systems with M non- 240
M (1 + ) M m e
adjacent frequency bands is given by
fi.i.d. ( ) = 241

206
m + (M)

2 q M
M (1+) M+q1 n = logn 1 + i
(m M)
q (+m) 242
207 , (10) 0 0 0 i=1

(M)q q! f (1 , 2 , M ) d1 d2 d M , (18)
q=0 243
244 where f (1 , 2 , M ) represents the joint pdf of the instan- This integral can be evaluated numerically in an efficient 285
245 taneous SNRs of each band. Unfortunately, it is very tedious manner by employing a N-point Gauss-Chebyshev quadrature 286
246 and computationally cumbersome to obtain the joint pdf even rule as 287
247 for the simple i.i.d. case. One possible way to solve this
2
15
problem is to use the moment generating function (MGF) tk2 2+1
248
K (, , a) = +1 wk ln 1 + t (23) 288
249 based method proposed in [6]. However, the HOS of the a a k
k=1
250 channel capacity is given in terms of a single-integral expres-
251 sion, which makes it difficult to be mathematically employed. where wk and tk are the weights and abscissae given in [22]. 289
252 Therefore, in the following, we derive the HOS of the channel Note that we only need 15 terms in (23) to converge ade- 290
253 capacity for spectrum aggregation systems over generalized quately. 291
254 fading channels by utilizing the pdf of the total SNR . It By taking n = 1 in (19), we can obtain the first-order 292
255 is worthy to mention that all our derived results are given in statistics of the channel capacity, which is the well-known 293
256 analytical form, which is easy to show the key impacts of ergodic capacity as 294
of
257 system performance. (1+)
e M ( M)i
1 = 295
lnn 2 i ! (M + i)
i=0
258 A. - Fading Channels

259 We first consider the higher-order capacity statistics of
M+i1
(1 + ) M+ik1 M + i 1

spectrum aggregation systems over i.i.d. - fading channels
296
260 k
k=0
261 as follows.
(1 + )
262
263
264
265
be expressed as
n = M n
e
1

ln 2
J M + i,

i=0
( M)i
i ! (M + i )
(1 + )

,n ,

(1 + ) M+i
ro
trum aggregation systems over i.i.d. - fading channels can
(19)
where

a
b
= b!(ab)!
a!
G 3,0
G-function [17, eq. (9.301)].

2,3

1, 1
0, 0, 1 + k
, and G() denotes the Meijers
Lemma 1: For the high- and low-SNR regimes, the higher-

order capacity statistics of spectrum aggregation systems
, (24) 297
298
299
300
EP
266 301

over i.i.d. - fading channels can be respectively expressed 302
267 where the auxiliary function J () is given by (47) in the as 303
Appendix A.
(1 + ) M+i
268
1 ( M)i
269 Proof: Substituting (2) into (17), we can derive n = M n 304
e ln 2 i ! (M + i )
M+1 i=0
M(1 + ) 2
270 n = (1 + )
M1
eM 2 (M) 2 lnn 2
M+1
Q M + i 1, ,n , (25) 305
M+2i1

1 (1 + ) M
n
0 n! Sk+n (k + n + M)
n =
271
i ! (M + i) M
306
n
(k + n)! (M)
IEE
i=0
e ln 2
k=0
(1+)
lnn (1 + ) M+i1 e d . (20) n+k

272
0 1 F1 (k + n + M; M; M) , 307
(1 + )
273 To the best of the authors knowledge, the integral in (20) is
274 not included in tables of classical reference books such as [17]. (26) 308
275 Nervertheless, as shown in Appendix A, it can be computed

276 in closed form thus completing the proof. where Smn is the Stirling number of the first kind 309
277 Note that the auxiliary function J () requires is restricted [17, eq. (9.740)], and the auxiliary function Q() is given 310
278 to integer values, which assumes finite numbers of multipath by (49) in the Appendix. B. 311
279 clusters. In the most general case of real , integrals of the Proof: By taking large values of in (17) and using (2), 312
280 form the higher-order capacity is given by 313
M+1
M(1 + ) 1
ln (1 + x) exp(ax)x dx
2
281 K (, , a) = (21)
n = M1 M+1
314
0 eM 2 (M) 2 lnn 2 i ! (M + i)
i=0
282 should be evaluated. By performing the change of variables M+2i1
ax = y 2 , (21) can be expressed as (1 + ) M
283
315

2 y2
K (, , a) = +1 ln 1 + exp(y 2 )y 2+1 dy. (1+)
a 0 a lnn ( ) M+i1 e d . (27) 316
284 (22) 0
317 With the aid of [23, eq. (2.5.1.7)], the integral in (27) can be i.n.i.d. - fading channels can be expressed as 353
318 calculated as
k
1 (k)q U q
M(1 + )
M+1
n = c k 354
(2)U lnn 2 k=0 q=0 q! (U + q) 2
2
319
n = M1 M+1
eM 2 (M) 2 lnn 2
M+2i1 1
Q q + U 1, ,n , (33) 355
1 (1 + ) M 2
320 q
i ! (M + i)
n!
k
(k)q U
i=0 0
n = n ck 356
dn (a + 1) ln 2 q! (U + q)
321 n . (28) k=0 q=0
da ( (1 + ) /)a+1 a=M+i1
S p+n (U
n
+ q + p + n)
. (34) 357
322 Then, the high-SNR HOS (25) can be derived by using (49)
p=0
( p + n)!(2)( p+n)
323 and after some algebraic manipulation.
of
324 Moreover, the low-SNR HOS can be obtained by taking Proof: With the help of [17, eq. (3.351.3)], the proof can 358
325 0, and using the well-known expansion of the logarithm be completed by following similar steps in Lemma 1. 359
326 function as [17, eq. (9.741.2)] Note that the low-SNR HOS of the channel capacity (34) 360
is given in terms of simple elementary functions. Therefore, 361

z k+n
327
n
ln (1 + z) = n! n
Sk+n , z0 (29) we can obtain the implication that the HOS is an increasing 362
(k + n)! function in U . 363

k=0
Substituting (29) into (17), and with the aid of (2) and (17),
ro
328
329 we derive the low-SNR HOS as B. - Shadowed Fading Channels 364
M+1 Now, we move on to consider the higher-order capacity 365

M(1 + ) n! n
Sk+n
330
0
n =
2
statistics for - shadowed fading channels. As a first step, 366
M1 M+1
eM 2 (M) 2 lnn 2 k=0 (k + n)! the case of i.i.d. - shadowed fading channel is investigated. 367
M1 (1 + ) 368
331 k+n+ 2 exp trum aggregation systems over i.i.d. - shadowed fading

369
0
channels can be expressed as 370
EP
(1 + ) M m M
332 IM1 2

d . (30) M (1 + ) M m 1
n = 371
m + (M)lnn 2
To evaluate the integral in (30), we can utilize the following
2 q
333
(m M) M (1+)
identity [24, eq. (3.15.2.5)] q (+m)
334
372
(M)q q!
q=0
x q exp ( px) Iv a x d x
M (1 + )
335
0
J q + M, ,n . (35) 373
(q +v/2+1) (a/2)v v a2
336 = 1 F1 q + +1; v +1; . (31)
(v + 1) pq+v/2+1 2 4p Proof: This result is a direct consequence of taking (10) 374
into (17), and using (47). 375

IEE
337 Finally, we arrive at the desired result in (26) after some basic
338 algebra. Lemma 3: For the high- and low-SNR regimes, the higher- 376
339 Note that the auxiliary function Q() can apply for arbitrary order capacity statistics of spectrum aggregation systems 377
340 positive real values of , so the asymptotical results are over i.i.d. - shadowed fading channels can be expressed 378
generalized. It is easy to see from (25) and (26) that the HOS as 379
m M
341
342 of the channel capacity is an increasing function in the average M (1 + ) M m 1

SNR and M. n = 380
343
m + (M)lnn 2
2 q
344
(m M) M (1+)
345 trum aggregation systems over i.n.i.d. - fading channels q (+m)
381
346 can be expressed as (M)q q!
q=0

1
k
(k)q U q M (1 + )
n = ck Q q + M 1, ,n , (36) 382
347
(2)U lnn 2 k=0 q=0 q! (U + q) 2
m M
0 m n!
1 n =
J q + U, ,n .
383
348 (32) m + (M)
2
n
Sk+n (k + n + M)
349 Proof: The proof is readily completed by taking (3) 384
(k + n)! (M (1 + ) / )k+n
350 into (17), and using (47). k=0

351 Lemma 2: For the high- and low-SNR regime, the higher-
2 F1 m M, k + n + M; M; . (37) 385
352 order capacity statistics of spectrum aggregation systems over + m
386 Proof: We can obtain (36) by following similar steps in TABLE I

387 Theorem 3. While for the case of the low-SNR regime, the N UMBER OF R EQUIRED T ERMS FOR C ONVERGENCE OF THE HOS OF THE
388 integral identity [24, eq. (3.35.1.2)] i.i.d. - FADING C HANNELS W ITH Pe 105 , AND M = 3

q px (q +1)
389 x e 1 F1 (a, b; x) d x = 2 F1 a, q +1; b;
0 pq+1 p
390 (38)
391 should be invoked. Note that condition on the arguments
392 of (38), p > 0, p > , and q > 1, is satisfied.
393 Considering the case where the frequency bands are corre-
394 lated, we present the following theorem.
395 Theorem 4: The higher-order capacity statistics of spec-
396 trum aggregation systems over correlated - shadowed fad- to describe how far the channel capacity lies from the ergodic 430
of
397 ing channels can be expressed as capacity. Its normalization with respect to the ergodic capacity, 431
U
q 2
A (m M + k)q AoD = 2 1 1 , is called the amount of dispersion (AoD). 432
398 n = n Dk Furthermore, we can define the reliability percentage of the
ln 2 (U )q q! 1 + 1
433

k=0

q=0 signal throughput as R = 100(1 AoD). For good channel 434
quality, AoD approaches to zero and R 100. In addition, 435

399 J q + U, , n . (39)
the skewness is a metric of the degree of asymmetry for the 436
3
distribution of the channel capacity as S = 3 1 3 . For
ro
400 Proof: The proof can be completed by following similar 437
2 21
401 steps in Theorem 1. symmetric distributions, S = 0, while S < 0 denotes the 438
402 We note from (39) that the HOS of the channel capacity distribution is skewed to the left. In addition, the kurtosis cor- 439
403 is an increasing function in the number of frequency bands responds to the degree of peakedness of the channel capacity 440
404 M and shadowing parameter m and as such it obtains its 4
maximum value for m . around the ergodic capacity as K = 4 21 2 . Within this 441
405 2 1
406 Lemma 4: For the high- and low-SNR regimes, the higher- context, it is worth mentioning that these important statistical 442
407 order capacity statistics of spectrum aggregation systems over metrics of the channel capacity can be also efficiently and 443
EP
408 correlated - shadowed fading channels can be expressed accurately computed by using the HOS expressions. 444
409 as
U
q IV. N UMERICAL R ESULTS 445
A (m M +k)q

410 n = n Dk In this section, we present various performance evaluation 446
ln 2 (U )q q! 1+1 results using the HOS of the channel capacity expressions 447
k=0 q=0
presented in Sections III for spectrum aggregation systems
448
411 Q q + U 1, , n , (40) operating over - and - shadowed fading channels,

449
respectively. To validate the accuracy of the aforementioned 450
0
S np+n Dk ( p + n + M)
412 n = An! expressions, comparisons with complementary Monte-Carlo 451
( p + n)! (/ ) p+n simulated results with 106 realizations of random variables are 452
k=0 p=0

IEE
also included in these figures. We use the approaches presented 453

1
413 2 F1 m M + k, U + p + n; , U ; . (41) in [18] and [25] to generate random variables from the squared 454
1 + 1 - and - shadowed distributions, respectively. The impact 455
414 Proof: The proof concludes by following a similar line of of system and fading parameters on the HOS performance of 456
415 reasoning as in Lemma 3. spectrum aggregation systems are discussed in detailed. 457
416 According to Lemma 4, a higher increases the HOS of

417 the channel capacity. This is anticipated, since larger values A. Convergence of Derived Results 458
418 of reduce the correlation between frequency bands, making Since the derived results are given in sum of infinite series, 459
419 receiver signal more stronger. we prove the convergence of the derived results by truncating 460
the appropriate series expressions to achieve an accuracy 461

C. Practical Implementation of HOS
up to the fifth-significant digit (e.g., Pe 105 ). Table I
420
462
421 To evaluate the performance of spectrum aggregation sys- investigates the impact of the number of moments and fading 463
422 tems, several important measures will in discussed by using the parameters and on the convergence of the HOS of the 464
423 HOS of the channel capacity presented above. These measures channel capacity for M = 3. It can be seen from Table I that all 465
424 can also serve as useful tools for the design of practical infinite series rapidly converged with the speed of convergence 466
425 dispersed spectrum cognitive radio (DS-CR) systems. First of for the scenarios of interest. Moreover, the number of terms 467
426 all, the amount of fading (AoF) of the channel capacity or increases with increasing , and average SNR , while 468
427 the so called fading figure is defined as the ratio of variance has a noticeable impact on the convergence. For high-SNR
to the square ergodic capacity as AoF =
469
428
2
2
1 [7]. The regimes, the required number of terms for ergodic capacity 470
1
429 variance of the channel capacity is denoted by var = 2 21 (n = 1) is less than the case of the 4-th statistical moment. 471
TABLE II
N UMBER OF R EQUIRED T ERMS FOR C ONVERGENCE OF THE HOS OF THE
C ORRELATED - S HADOWED FADING C HANNELS W ITH Pe
103 , M = 2, 1 = 1, 2 = 5, 1 = 1, 2 = 2, AND m = 1
of
Fig. 3. Amount of dispersion of the channel capacity against the average SNR
for spectrum aggregation systems over i.n.i.d. - fading channels (2 =
3 = 1 dB, 1 = 2.5, 2 = 3.5, 3 = 4.75, and 1 = 1, 2 = 1, 3 = 2).
the high-SNR approximation is more accurate for low order 500
Fig. 2. ro
Simulated, analytical, and asymptotic HOS of the channel
capacity against the average SNR for spectrum aggregation systems over
i.i.d. - fading channels ( = 1, = 1, and M = 3).
statistics (e.g., ergodic capacity). It is also clear from Fig. 2
that the HOS curves get closer to each other almost around
6 dB, which determines the boundary of the high- and low-
SNR regimes. Base on the interesting finding of n = 1 at
around 6 dB, we can simply model the HOS of the channel
capacity as n n1 . Due to the nth power of the ergodic
capacity, the behavior of the HOS of capacity is different in
the high- and low-SNR regimes, respectively. Furthermore, the
501
502
503
504
505
506
507
EP
508
crossing point will be shifted toward left, and therefore, the 509
472 However, only a relatively small number of terms is required ergodic capacity increasing if increasing the values of fading 510
473 for the desired accuracy. For the worst case of n = 4, parameters (e.g., and ) of each frequency band. 511
474 = 10 dB, = 1, and = 2, the maximum number of The effect of the number of frequency bands M on the AoD 512
475 terms is 23. performance of spectrum aggregation systems over i.n.i.d. 513
476 For correlated - shadowed fading channels, the effect of - fading channels is shown in Fig. 3. One can notice that 514
477 correlation coefficient on the convergence has been studied the AoD appears to increase for low and moderate SNRs, 515
478 in Table II. It is clear that the derived results are rapidly con- while it begins to decrease for the high-SNR regime for all 516
479 verged with less than 40 terms of infinite series. Note that as cases. Furthermore, it can be seen from the Fig. 3 that the 517
480 reduces, the required number of terms decreases. Furthermore, AoD plot becomes peaky at around 9 dB for M = 3, while 518
the computation time of derived analytical results is much less
IEE
481
the AoD reaches its highest value around 6 dB for M = 1. 519
482 than the one of simulations. For example, we spend less than With respect to the reliability percentage of the spectrum 520
483 20 seconds to calculate (19) by using MATHEMATICA, while aggregation system, the transmit SNR should be chosen greater 521
484 the simulation costs more than 230 seconds to derive the same than the SNR for which the AoD peaks. For example, for 522
485 result. Note that other cases have similar fact of converging the case of M = 3, the maximum AoD is 0.1919 and the 523
486 steadily and rapidly, and requiring little computational effort, reliability percentage is 90.81%, which means that the average 524
487 which are validated by our conducted numerical experiments. SNR must be chosen equal to or greater than 9 dB in order to 525
reach at least 90.81% reliable transmission. Moreover, the gap 526
488 B. Performance Analysis between different number of bands M decreases at high SNRs 527
489 For spectrum aggregation systems over i.i.d. - fading which implies that the fading effect becomes less pronounced 528
490 channels, the simulated, analytical (19), and asymptotic (25), as anticipated. 529
491 (26) HOS of the channel capacity are plotted against the Figure 4 depicts the AoF of the channel capacity for 530
492 average SNR in Fig. 2. As seen, the analytical results spectrum aggregation systems over i.i.d. - shadowed fading 531
493 perfectly match the Monte-Carlo simulations. Clearly, the channels as a function of average SNR per band for 532
494 high-SNR approximations are sufficiently tight and become different sets of frequency bands M. It is clear that the AoF 533
495 exact even at moderate SNR values, while a precise agreement decreases drastically as the value of increases. At low SNRs, 534
496 between the exact and low-SNR results can be observed. This the AoF performance of the spectrum aggregation system is 535
497 implies that they can efficiently predict the HOS of the channel significantly improved with increasing the value of frequency 536
498 capacity over a wide SNR range. Moreover, the gap between bands and M. For example, the AoF is 0.665 for the case of 537
499 the corresponding curves increases as n increases which means M = 1 at 10 dB, while it reduces to 0.148 for M = 3. 538
of
Fig. 4. Amount of fading of the channel capacity against the average SNR Fig. 6. Skewness of the channel capacity against the average SNR for
for spectrum aggregation systems over i.i.d. - shadowed fading channels spectrum aggregation systems over correlated - shadowed fading channels
( = = 2 and m = 1). (M = 3, i = 1, i = 2, and m = 1).
ro
EP
Fig. 5. Amount of fading of the channel capacity against the average SNR Fig. 7. Kurtosis of the channel capacity against the average SNR for spectrum
for spectrum aggregation systems over i.i.d. - shadowed fading channels aggregation systems over correlated - shadowed fading channels (M = 3,
(M = 2). i = 1, i = 2, and m = 1).
539 Therefore, we can use more frequency bands to combat the correlation models pq = | pq| , where = 0.1, 0.5, 0.9. 560
IEE
540 low SNRs of fading channels. It is clear to see from Figs. 6 and 7 that the skewness and 561
541 The effect of fading parameters , and m on the AoF of the kurtosis increases as average SNR of each frequency 562
542 the channel capacity for spectrum aggregation systems over band increases and/or decreases, showing that the pdf 563
543 i.i.d. - shadowed fading channels are further investigated of the channel capacity becomes more spiky with heavy 564
544 in Fig. 5. As indicated by analysis in Section III, the AoF tails and asymmetric. More importantly, the gap between the 565
545 decreases with a smaller value of (more power of LoS corresponding curves decreases as decreases which implies 566
546 components) and a higher value of (more power of clusters), that its effect becomes less pronounced. 567
547 where the fading channel becomes less deterministic. This

548 finding reveals that more scattered waves are beneficial for V. C ONCLUSION 568
549 improved AoF. One can also notice the increase of the AoF In this paper, we investigate the performance of spectrum 569
550 can be obtained for decreasing the shadowing parameter m, aggregation systems over generalized fading channels. In par- 570
551 especially for low SNRs. For example, the value of AoF ticular, we consider two recently proposed generalized fading 571
552 is 0.275 for the case of m = 1, = 1, = 2 and = 10 dB, models, namely - and - shadowed, which can model 572
553 while it reduces to 0.228 for the case of m = 2, = 1, propagation phenomena involving LoS and composite fading 573
554 = 2 and = 10 dB. The impact of fading appears to environments, respectively. Novel and exact expressions for 574
555 be particularly critical for low SNRs, while in the high-SNR the HOS of the channel capacity of spectrum aggregation 575
556 regime its impact is relatively reduced. systems are derived. Our derived expressions can extend 576
557 In Figs. 6 and 7, Skewness and Kurtosis statistics are plotted and complement existing results on classical fading models. 577
558 against the average SNR , respectively. We consider corre- Furthermore, we deduce simple HOS expressions for the 578
559 lated - shadowed fading channels with three exponential asymptotically low- and high-SNR regimes. Note that all 579
580 infinite series can be computationally efficient, accurate, and By substituting = 1 into (45), we can derive the auxiliary 616
581 requires only a relative small number of terms for yield- function J (a, b, n) as 617
582 ing accurate results. Important performance metrics, such as

J (a, b, n) 618
583 ergodic capacity, variance, AoF, AoD, skewness, and kur-
584 tosis, are also derived to show the effects of system and

a1
ak1 a 1
fading parameters on spectrum aggregation systems. Finally, = n!e b b1k
(1)
585
619
586 extensive Monte-Carlo simulations verify the accuracy of the k
k=0
587 analytical expressions and the tightness of the high- and
n+1 1 s

588 low-SNR bounds. The proposed analysis is useful for the
! "# $
589 spectrum aggregation system design engineer for performance 1, 1, , 1
G n+2,0 .
590 evaluation purposes. n+1,n+2 b 0, 0, , 0, 1 + k (47) 620
# $! "

n+1 0 s
591 A PPENDIX
of
592 A. A Useful Integral Identity B. High-Order Differentiation 621
593 Let us consider an integral in the form of With the help of Leibnizs rule [17], the nth differentiation 622
of the product of the gamma function and the power functions 623
594 (a, b) = (1 + x) x a1 ebx d x, (42) can be expressed as 624

0
d n (a + 1)
where a N, b > 0, and C. With the help of [26, eq. (39)], Q (a, b, n) = 625
da n ba+1
ro
595
596 (42) can be also expressed as

n (1)nk lnnk (b) d k (a + 1)
n
= . (48) 626
597 (a, b) = (a) U (a, a + + 1; b) k ba+1 da k
k=0
a1

a 1 (1)ak1 Utilizing the high-order differentiation of the gamma function
=e b
( + k + 1, b)
627
598
k b+k+1 [5, eq. (10)], we can derive 628
k=0
599 (43) Q (a, b, n) 629
n
EP
n (1)nk lnnk (b) k!
600 where U () denotes the Tricomi hypergeometric function = 630
k ba+1
601 [27, eq. (13.1.3)], and (, ) is the upper complementary k=0

602 incomplete gamma function [17, eq. (8.350.2)]. By using k+1 1 s
! "# $
603 Leibnizs rule [17], the nth order derivative of (43) can be k+2,0 1, 1, , 1
604 evaluated as
G k+1,k+2 1 1 + a, 0, 0, , 0 631
# $! "
d n (a, b)
k+1 0 s
605

d n =+k+1 k+1 1 s
! "# $
a1
1, 1, , 1
a 1 (1)ak1 + (1)k G 1,k+1
= eb k+1,k+2 1 1 + a, 0, 0, , 0 . (49) 632

606
k b+k+1 # $! "
IEE
k=0
n
k+1 0 s
n
d (, b)
n n p 1
ln (44)
d n =+k+1
607
p b R EFERENCES 633
p=0
a1
[1] H. Bogucka, P. Kryszkiewicz, and A. Kliks, Dynamic spectrum aggre-
a 1 (1) ak1 634
608 = n!eb gation for future 5G communications, IEEE Commun. Mag., vol. 53, 635
k b+k+1 no. 5, pp. 3543, Jun. 2015. 636
k=0
[2] R. Zhang, M. Wang, L. X. Cai, Z. Zheng, and X. Shen, LTE-unlicensed: 637
n+1 1 s The future of spectrum aggregation for cellular networks, IEEE Wireless
! "# $
638

1, 1, , 1 Commun., vol. 22, no. 3, pp. 150159, Jun. 2015. 639
G n+2,0 , [3] F. Yilmaz, H. Tabassum, and M.-S. Alouini, On the computation of

n+1,n+2 b 0, 0, , 0, + k (45) 640
609

# $! " the higher order statistics of the channel capacity for amplify-and- 641

forward multihop transmission, IEEE Trans. Veh. Technol., vol. 63, 642
n+1 0 s no. 1, pp. 489494, Jan. 2014. 643
[4] M. K. Simon and M.-S. Alouini, Digital Communication Over Fading 644
610 From (44) to (45), we have used the differentiation identity Channels. Hoboken, NJ, USA: Wiley, 2005. 645
611 [19, eq. (06.06.20.0013.01)] and performed some algebraic [5] F. Yilmaz and M.-S. Alouini, Novel asymptotic results on the high- 646
order statistics of the channel capacity over generalized fading channels, 647
612 manipulations. in Proc. IEEE SPAWC, Jun. 2012, pp. 389393. 648
613 Based on (42), the nth order derivative of (a, b) can [6] F. Yilmaz and M.-S. Alouini, On the computation of the higher-order 649
614 alternatively be given by statistics of the channel capacity over generalized fading channels, IEEE 650
Wireless Commun. Lett., vol. 1, no. 6, pp. 573576, Jun. 2012.

651
[7] N. C. Sagias, F. I. Lazarakis, A. A. Alexandridis, K. P. Dangakis, and 652
615 J (a, b, n) = (1 + x) lnn (1 + x) x a1 ebx d x. (46) G. S. Tombras, Higher order capacity statistics of diversity receivers, 653
0 Wireless Pers. Commun., vol. 56, no. 4, pp. 649668, Feb. 2011. 654
655 [8] M. D. Yacoub, The - distribution and the - distribution, IEEE Xiaoyu Chen received the B.Sc. degree in commu- 730
656 Antennas Propag. Mag., vol. 49, no. 1, pp. 6881, Feb. 2007. nication engineering from Beijing Jiaotong Univer- 731
657 [9] T. Tsiftsis, F. Foukalas, G. Karagiannidis, and T. Khattab, On the sity, China, in 2015, where she is currently pursuing 732
658 higher-order statistics of the channel capacity in dispersed spectrum the M.S. degree from Wireless Self-Organized Com- 733
659 cognitive radio systems over generalized fading channels, IEEE Trans. munications Group, School of Electronic and Infor- 734
660 Veh. Technol., vol. 65, no. 5, pp. 38183823, May 2016. mation Engineering. Her research interests include 735
661 [10] J. F. Paris, Statistical characterization of - shadowed fading, IEEE the performance analysis of multi-hop and OFDM 736
662 Trans. Veh. Technol., vol. 63, no. 2, pp. 518526, Feb. 2014. communications. 737
663 [11] S. L. Cotton, Human body shadowing in cellular device-to-device
664 communications: Channel modeling using the shadowed fading model,
665 IEEE J. Sel. Areas Commun., vol. 33, no. 1, pp. 111119, Jan. 2015.
666 [12] J. Zhang, L. Dai, W. H. Gerstacker, and Z. Wang, Effective capacity of
667 communication systems over shadowed fading channels, Electron.
668 Lett., vol. 51, no. 19, pp. 15401542, Sep. 2015.
669 [13] J. Zhang, Z. Tan, H. Wang, Q. Huang, and L. Hanzo, The effective
670 throughput of MISO systems over - fading channels, IEEE Trans.
671 Veh. Technol., vol. 63, no. 2, pp. 943947, Feb. 2014.
of
672 [14] M. Dohler, R. W. Heath, A. Lozano, C. B. Papadias, and
673 R. A. Valenzuela, Is the PHY layer dead? IEEE Commun. Mag.,
674 vol. 49, no. 4, pp. 159165, Apr. 2011. Kostas P. Peppas (M09SM14) was born in 738
675 [15] K. Guan, Z. Zhong, J. I. Alonso, and C. Briso-Rodriguez, Measurement Athens, Greece, in 1975. He received the Diploma 739
676 of distributed antenna systems at 2.4 GHz in a realistic subway tunnel degree in electrical and computer engineering and 740
677 environment, IEEE Trans. Veh. Technol., vol. 61, no. 2, pp. 834837, the Ph.D. degree in wireless communications from 741
678 Feb. 2012. the National Technical University of Athens, in 1997 742
679 [16] J. Zhang, M. Matthaiou, Z. Tan, and H. Wang, Performance analysis and 2004, respectively. From 2004 to 2007, he was 743
680 of digital communication systems over composite -/gamma fading with the University of Peloponnese, Department of 744
681 channels, IEEE Trans. Veh. Technol., vol. 61, no. 7, pp. 31143124, Computer Science, Tripoli, Greece, and from 2008 745
682
683
684
685
686
687
688
689
690
Sep. 2012.
Products, 7th ed. San Diego, CA, USA: Academic, 2007.
Technol., vol. 61, no. 1, pp. 413419, Jan. 2012.
ro
[17] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and
[18] K. P. Peppas, Sum of nonidentical squared - variates and applications

in the performance analysis of diversity receivers, IEEE Trans. Veh.
[19] Wolfram. The Wolfram Functions Site, accessed on Sep. 12, 2016.
[Online]. Available: http://functions.wolfram.com
[20] A. Castao-Martnez and F. Lpez-Blzquez, Distribution of a sum
of weighted noncentral chi-square variables, TEST, vol. 14, no. 2,
to 2014 he was also with the National Centre for
Scientific Research-Demokritos, Institute of Infor-
matics and Telecommunications as a Researcher.
Since 2014, he has been with the Department of Telecommunication Science
and Technology, University of Peloponnese, where he is currently a Lecturer.
He has authored over 70 journal and conference papers. His current research
interests include digital communications, MIMO systems, cooperative com-
munications, wireless and personal communication networks, cognitive radio,
digital signal processing, system level analysis, and design.
746
747
748
749
750
751
752
753
754
EP
691
692 pp. 397415, Dec. 2005.
693 [21] M. R. Bhatnagar, On the sum of correlated squared - shadowed
694 random variables and its application to performance analysis of MRC,
695 IEEE Trans. Veh. Technol., vol. 64, no. 6, pp. 26782684, Jun. 2015.
696 [22] N. M. Steen, G. D. Byrne, and E. M. Gelbard, Gaussian quadratures
( (
for the integrals 0 ex f (x)dx and 0b ex f (x)dx, Math. Comput.,
2 2
697
698 vol. 23, no. 107, pp. 661671, 1969.
699 [23] A. P. Prudnikov, I. U. A. Brychkov, and O. I. Marichev, Integrals
700 and Series: More Special Functions, vol. 3. New York, NY, USA: Xu Li received the B.Sc. degree from Jilin Uni- 755
701 Gordon and Breach, 1990. versity, China, in 1991, and the Ph.D. degree from 756
702 [24] A. P. Prudnikov, I. U. A. Brychkov, and O. I. Marichev, Integrals Northeastern University, China, in 1997. Since 2003, 757
703 and Series: Direct Laplace Transforms, vol. 4. New York, NY, USA: she has been a Professor and Head of Wireless Self- 758
704 Gordon and Breach, 1992. Organized Communications Group, School of Elec- 759
IEE
705 [25] L. Moreno-Pozas, F. J. Lopez-Martinez, J. Paris, and E. Martos-Naya, tronic and Information Engineering, Beijing Jiaotong 760
706 The - shadowed fading model: Unifying the - and - distri- University, China. She holds over 40 patents. Her 761
707 butions, IEEE Trans. Veh. Technol., vol. 65, no. 12, pp. 96309641, research interest includes wireless self-organized 762
708 Dec. 2016. communications, sensor networks, and unmanned 763
709 [26] M. Kang and M. S. Alouini, Capacity of MIMO Rician channels, aerial vehicle systems. 764
710 IEEE Trans. Wireless Commun., vol. 5, no. 1, pp. 112122, Jan. 2006.
711 [27] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
712 With Formulas, Graphs, and Mathematical Tables, 9th ed. New York,
713 NY, USA: Dover, 1964.
714 Jiayi Zhang (S08M14) received the B.Sc. and

715 Ph.D. degrees in communication engineering from
716 Beijing Jiaotong University, China, in 2007 and
717 2014, respectively. From 2012 to 2013, he was
718 a visiting Ph.D. student with the Wireless Group,
719 University of Southampton, U.K. From 2014 to Ying Liu received the B.Sc. and Ph.D. degrees from 765
720 2015, he was a Humboldt Research Fellow with Jilin University, China, in 1985 and 2001, respec- 766
721 Institute for Digital Communications, University of tively. Since 2005, she has been a Professor with the 767
722 ErlangenNuermberg, Germany. From 2014 to 2016, School of Electronic and Information Engineering, 768
723 he was a Post-Doctoral Research Associate with Beijing Jiaotong University, China, where she is 769
724 the Department of Electronic Engineering, Tsinghua the Vice Dean of School of Electronic and Infor- 770
725 University, China. Since 2016, he has been a Professor with the School mation Engineering. Her research interest includes 771
726 of Electronic and Information Engineering, Beijing Jiaotong University. His wireless self-organized communications, sensor net- 772
727 current research interests include massive MIMO, unmanned aerial vehicle works, and wireless localization systems. 773
728 systems, and performance analysis of generalized fading channels. He was an
729 Exemplary Reviewer of the IEEE C OMMUNICATIONS L ETTERS in 2015.

Ieee Tcom Hos

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ieee Tcom Hos

Uploaded by

Copyright:

Available Formats

IEEE TRANSACTIONS ON COMMUNICATIONS 1

On High-Order Capacity Statistics of Spectrum

in the literature, due in part to the intractability of its analysis. 52

channel capacity of different wireless systems over several flat 54

fading channels [3], [5][7]. For example, a generic framework

spatially correlated in practical propagation environments [8]. 72

84 channels are recently introduced in [8], [10], and [11], respec-

115 Furthermore, the asymptotically high- and low-SNR M+1

120 of asymptotic expressions are given in terms of simple 

128 systems to aggregate best available frequency bands in

180 ck = c j dk j , k 1 (5) k=0

where i 2i i and ai i /2i (1 + i ). In order to

assumes that all the dominant components are subject to U 

assumption of shadowing components are correlated, while

aggregation systems. Without loss of generality, the HOS of 233

202 ing, and 1 F1 () is the confluent hypergeometric function

q=0 statistics of channel capacity is well-known as the ergodic 238

207 , (10) 0 0 0 i=1

G-function [17, eq. (9.301)].

, and G() denotes the Meijers

Lemma 1: For the high- and low-SNR regimes, the higher-

267 where the auxiliary function J () is given by (47) in the as 303

275 Nervertheless, as shown in Appendix A, it can be computed

280 form the higher-order capacity is given by 313

is given in terms of simple elementary functions. Therefore, 361

(k + n)! function in U . 363

329 we derive the low-SNR HOS as B. - Shadowed Fading Channels 364

M+1 Now, we move on to consider the higher-order capacity 365

into (17), and using (47). 375

342 of the channel capacity is an increasing function in the average M (1 + ) M m 1

386 Proof: We can obtain (36) by following similar steps in TABLE I

quality, AoD approaches to zero and R 100. In addition, 435

411 Q q + U 1, , n , (40) operating over - and - shadowed fading channels,

 respectively. To validate the accuracy of the aforementioned 450

also included in these figures. We use the approaches presented 453

416 According to Lemma 4, a higher increases the HOS of

the appropriate series expressions to achieve an accuracy 461

the high-SNR approximation is more accurate for low order 500

reach at least 90.81% reliable transmission. Moreover, the gap 526

547 where the fading channel becomes less deterministic. This

582 ing accurate results. Important performance metrics, such as

594  (a, b) = (1 + x) x a1 ebx d x, (42) can be expressed as 624

596 (42) can be also expressed as

Products, 7th ed. San Diego, CA, USA: Academic, 2007.

Technol., vol. 61, no. 1, pp. 413419, Jan. 2012.

[18] K. P. Peppas, Sum of nonidentical squared - variates and applications

714 Jiayi Zhang (S08M14) received the B.Sc. and

On High-Order Capacity Statistics of Spectrum

in the literature, due in part to the intractability of its analysis. 52

channel capacity of different wireless systems over several flat 54

fading channels [3], [5][7]. For example, a generic framework

spatially correlated in practical propagation environments [8]. 72

84 channels are recently introduced in [8], [10], and [11], respec-

115 Furthermore, the asymptotically high- and low-SNR M+1 M1

120 of asymptotic expressions are given in terms of simple 

128 systems to aggregate best available frequency bands in

instantaneous signal-to-noise ratios (SNRs), i , from each (2)U q! (U + q) 2

- shadowed RVs is given by [21. eq. (16)]

180 ck = c j dk j , k 1 (5) k=0

where i 2i i and ai i /2i (1 + i ). In order to

assumes that all the dominant components are subject to U 

assumption of shadowing components are correlated, while

aggregation systems. Without loss of generality, the HOS of 233

202 ing, and 1 F1 () is the confluent hypergeometric function

120 of asymptotic expressions are given in terms of simple

where i 2i i and ai i /2i (1 + i ). In order to

assumes that all the dominant components are subject to U

respectively. To validate the accuracy of the aforementioned 450

594 (a, b) = (1 + x) x a1 ebx d x, (42) can be expressed as 624

120 of asymptotic expressions are given in terms of simple

instantaneous signal-to-noise ratios (SNRs), i , from each (2)U q! (U + q) 2

where i 2i i and ai i /2i (1 + i ). In order to

assumes that all the dominant components are subject to U

respectively. To validate the accuracy of the aforementioned 450

594 (a, b) = (1 + x) x a1 ebx d x, (42) can be expressed as 624

G n+2,0 , [3] F. Yilmaz, H. Tabassum, and M.-S. Alouini, On the computation of