2464 IEEE COMMUNICATIONS LETTERS, VOL. 24, NO.

11, NOVEMBER 2020

Performance Analysis of Intelligent Reflecting Surface

Aided Communication Systems
Qin Tao, Junwei Wang, and Caijun Zhong , Senior Member, IEEE

Abstract— This letter presents a detailed performance analysis

of the intelligent reflecting surface (IRS) aided single-input single-
output communication systems, taking into account of the direct
link between the transmitter and receiver. A closed-form upper
bound is derived for the ergodic capacity, and an accurate
approximation is obtained for the outage probability. In addition,
simplified expressions are presented in the asymptotic regime.
Numerical results are provided to validate the correctness of the
theoretical analysis. It is found that increasing the number of Fig. 1. IRS-aided communication systems.
reflecting elements can significantly boost the ergodic capacity
and outage probability performance, and a strong line-of-sight efficiency of the MISO systems was studied, assuming that the
component is also beneficial. In addition, it is desirable to deploy channel between the transmitter and IRS is deterministic. In
the IRS close to the transmitter or receiver, rather than in the
[12], [13], the outage probability and achievable rate of single-
middle.
input single-output (SISO) system was considered, assuming
Index Terms— Intelligent reflecting surface, ergodic capacity, that the direct link between the transmitter and receiver does
outage probability, Rician fading.
not exist.
In practice, the IRS is usually deployed in a position with
I. I NTRODUCTION line-of-sight (LOS) to both transmitter and receiver, it is
desirable to adopt the Rician fading model. Motivated by this,
T HE intelligent reflecting surface (IRS), which can manip-
ulate the propagation channel into a favorable shape,
has been regarded as a promising technology for the next
in this letter, we present a detailed performance analysis of
IRS-aided SISO systems in Rician fading channels, taking
into account of the direct channel between the transmitter and
generation wireless communication systems. As such, it has
receiver. Closed-form expressions are derived for the ergodic
received considerable interests from both the industry and
capacity upper bound and outage probability approximation of
academia [1], [2].
the system. In addition, simplified expressions are obtained in
Thus far, most of the works on IRS focus on the design of
the asymptotic regime. The findings of the letter suggest that
the phase shift matrix and the transmit beamformer [3]–[9].
the IRS can provide an effective signal-to-noise ratio (SNR)
For multiple-input single-output (MISO) systems, the joint
gain of N 2 , where N is the number of reflecting elements,
active and passive beamfomer design problem was studied
and a diversity order of N + 1 can be achieved.
in [3], and the impact of discrete phase shift was further
considered in [4]. Meanwhile, the energy efficiency of the
systems was characterized in [5]. Later on, a statistical II. S YSTEM M ODEL
channel state information (CSI) based design framework was We consider a three-node communication system consisting
proposed in [6]. The more general multiple-input multiple- of a single antenna transmitter T, a single antenna receiver R
output (MIMO) system was considered in [7], while the and an IRS with N reflecting elements, as illustrated in Fig. 1.
multi-cast MISO scenario was addressed in [8], [9]. Since the receiver can overhear the signal from both the
While the aforementioned works have improved our knowl- T-IRS-R and T-R links, the received signal y at the R can be
edge of IRS-aided communication systems, few works have expressed as
studied the analytical performance of IRS aided systems. √  
y = P hT2 Φh1 + g x + n, (1)
In [10], [11], the outage probability and ergodic spectral
where x denotes the normalized transmit signal with unit
Manuscript received June 29, 2020; accepted July 20, 2020. Date of energy, g denotes the T-R channel, the N × 1 vectors h1
publication July 24, 2020; date of current version November 11, 2020.
This work was supported in part by the National Key R&D Program of and h2 denote the T-IRS and IRS-R channels, respectively.
China 2019YFB1803400, the National Natural Science Foundation of China Also, the N × N matrix Φ  diag{θ} denotes the phase shift
under Grant 61922071 and 61725104. The editor coordinating the review matrix, where θ = [ejθ1 , ejθ2 , . . . , ejθN ] with θn ∈ [0, 2π)
of this letter and approving it for publication was G. Alexandropoulos.
(Corresponding author: Caijun Zhong.)
representing the phase shift of the n-th IRS reflecting element.
The authors are with the College of Information Science and Electronic Finally, n represents the additive white Gaussian noise with
Engineering, Zhejiang University, Hangzhou 310027, China, and also with the zero mean and variance N0 .
Zhejiang Provincial Key Laboratory of Information Processing, Communica- Since the IRS is deployed at a position that has LOS to both
tion and Networking, Zhejiang University, Hangzhou 310027, China (e-mail:
caijunzhong@zju.edu.cn). T and R, h1 and h2 are modeled by Rician fading. In contrast,
Digital Object Identifier 10.1109/LCOMM.2020.3011843 there is no LoS path between T and R, hence, g is modeled
1558-2558 © 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on February 08,2023 at 13:56:49 UTC from IEEE Xplore. Restrictions apply.
TAO et al.: PERFORMANCE ANALYSIS OF INTELLIGENT REFLECTING SURFACE AIDED COMMUNICATION SYSTEMS 2465

by Rayleigh fading. Therefore, the channel coefficients can be the ergodic capacity upper bound Cup is a symmetric function
expressed as with respect to the K1 and K2 , indicating the identical impact
   of the two hop channels.
1 Kl 1
hl =  αl h̄l + h̃l , l = 1, 2,
dl Kl + 1 Kl + 1 B. Outage Probability
1 In this subsection, we analyze the outage probability of the
g =  α3 h̃3 ,
d3 system, which is defined as the probability of the instantaneous
SNR that falls below a pre-defined threshold γth . Mathemati-
where dl and αl , l ∈ {1, 2, 3}, denote the distance and path
cally, it is given by
loss exponent of the corresponding channel, while Ki , i ∈
{1, 2} denotes the Rician factor. Also, h̄l , l ∈ {1, 2}, denotes Pout = Prob (γmax ≤ γth ) . (5)
the normalized LoS component, and h̃l , l ∈ {1, 2, 3} denotes
the normalized non-LOS component. Since the exact distribution of γmax is unknown, we resort
Assuming perfect CSI at the IRS, applying the optimal to tight approximations of the outage probability, and we have
phase shift matrix Φ give in [14], the maximum SNR of the the following important result:
system can be obtained as Theorem 2: When N → ∞, the outage probability can be
 N 2 approximated as Eq. (7) shown on the bottom of the next page,

γmax = γ0 |h2,n ||h1,n | + |g| , (2) where
n=1 π
μ =  α1 α2 L 1 (−K1 )L 12 (−K2 ),
where γ0 = P/N0 represents the transmit SNR. 4 d1 d2 (K1 + 1)(K2 + 1) 2
and
III. P ERFORMANCE A NALYSIS ⎡  2 ⎤
2
In this section, we provide a detailed analysis of the achiev- 1 ⎢ π L (−K1 )L (−K2 ) ⎥
1 1

σ2 =
2 2
able systems performance. Specifically, two important metrics, ⎣1 − ⎦. (6)
dα1 α2
1 d2 16(K1 + 1)(K2 + 1)
i.e., ergodic capacity and outage probability, are considered.
We start with the ergodic capacity.
Proof: See Appendix B.
A. Ergodic Capacity Theorem 2 presents a closed-from approximation for the
outage probability, which consists of only elementary func-
The ergodic capacity of the system can be expressed as
tions, hence can be efficiently evaluated. Also, we observe
C = E {log2 (1 + γmax )} . (3) that both μ and σ 2 are symmetric functions with respect to
K1 and K2 , which implies K1 and K2 have identical impact
Since the exact distribution of γmax is intractable, it is chal- on the outage probability. In addition, although Theorem 2 is
lenging to characterize the exact ergodic capacity. Therefore, obtained with the assumption of large N , as will be shown
we resort to a tractable bound with the help of Jensen’s through numerical results, the approximation turns out to be
inequality, and have the following key results: sufficiently tight even for moderate N .
Theorem 1: The ergodic capacity of the system is upper To gain further insight, we now look into the high SNR
bounded by Cup given in Eq. (4), shown on the bottom of the regime, and we have the following important result:
next page, where L 12 (·) denotes the Laguerre polynomial [15]. Theorem 3: When γ0 → ∞, the outage probability can be
Proof: See Appendix A. accurately approximated by
Theorem 1 provides a closed-form expression involving √ 
only elementary functions, which is applicable for arbitrary −(N +1)
πaN dα
1
1 N α2 N α3
d2 d3 2γ0
system configurations, thereby enabling efficient evaluation Pout,2 ≈   , (8)
Γ N + 32 (N + 1)! γth
of the ergodic capacity performance. In addition, it also
facilitates the characterization of the impact of key parameters where
on the ergodic capacity. In particular, we have the following 
observations: a = (K1 + 1)(K2 + 1)e−(K1 +K2 ) Ei(K1 ) + Ei(K2 ) − 2γ
Remark 1: The ergodic capacity upper bound increases
  
monotonically with N . For sufficiently large N , the ergodic (K1 + 1)(K2 + 1)
capacity Cup is mainly dominated by the term log2 γ1 N 2 , − ln K1 − ln K2 + 2K0 2 ,
2 dα 1 α2
1 d2 γ0
L 1 (−K1 )L 1 (−K2 )π
where γ1 = γ0 2
α α
2
, indicating an effective
16d1 1 d2 2 (K1 +1)(K2 +1) γ denotes the Euler’s constant, Ei(x) is the exponential integral
2
SNR gain of N . This can be explained by the fact that IRS [16], and Kn (x) is the modified Bessel function of the second
not only achieves the beamforming gain, but also attains the kind [16].
inherent aperture gain by collecting more signal power. Proof: See Appendix C.
Remark 2: Observing that L 12 (−K) is a monotonically Theorem 3 suggests that a diversity order of N + 1 is
increasing function of K, which suggests that a strong LOS achieved. Moreover, the impact of LOS component is mainly
component would enhance the ergodic capacity. In addition, reflected on the achievable coding gain of the system.

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on February 08,2023 at 13:56:49 UTC from IEEE Xplore. Restrictions apply.
2466 IEEE COMMUNICATIONS LETTERS, VOL. 24, NO. 11, NOVEMBER 2020

Fig. 2. Impact of K and N on the ergodic capacity. Fig. 3. Impact of IRS location on the ergodic capacity.

IV. S IMULATION R ESULTS

In this section, numerical results are presented to validate
the theoretical analysis in the previous section. Unless oth-
erwise specified, the following set of parameters are used in
the simulations. As in [5], we focus on the sub-6G scenario,
the system has a bandwidth of 180KHz, and the noise power
spectrum density is −173 dBm/Hz. The distances are set to
be d1 = 150 m, d2 = 150 m and d3 = 200 m, with path loss
exponents given by α0 = 3.5, α1 = α2 = 2.0, respectively.
The reference path loss at the reference distance d0 = 1 m is
set to be −30 dB and the outage threshold γth = 10 dB. Also,
we set K1 = K2 = K.
Fig. 4. Outage approximation Pout,1 with K = 1.
Fig. 2 illustrates the ergodic capacity of the system with
different N . As can be readily observed, the ergodic capacity
upper bound is tight for all configurations, indicating the
accuracy of the closed-form expression in Theorem 1. In addi-
tion, we observe the intuitive result that the ergodic capacity
increases monotonically with the number of IRS reflecting
elements N and Rician factor K.
Fig. 3 shows the impact of IRS location on the ergodic
capacity assuming the IRS is deployed between the line
segment of T and R, i.e., d1 + d2 = 300 m. As can be readily
observed, the ergodic capacity is a symmetric function with
respect to d1 and d2 . The minimum is achieved when the IRS
is deployed in the middle of the T and R. The above result
indicates that it better to deploy the IRS in the vicinity of
Fig. 5. High SNR outage approximation Pout,2 with K = 1.
either T or R to obtain higher capacity.
Fig. 4 plots the outage probability of the system with
K = 1. It can be observed that the approximation works well, Fig. 5 validates the accuracy of Pout,2 in the high SNR
even for moderate value of N . In addition, we see that the regime when K = 1. As readily can be seen, the high SNR
outage probability decreases sharply when the SNR increases, approximation is quite accurate, thus validates the correctness
and the decreasing rate is faster with a larger N . of Pout,2 . Also, a diversity order of N + 1 is observed,

 
1 N
Cup = log2 1 + γ0 + α1 α2
dα
3
3
d1 d2
 
N (N − 1)π 2  2 N 2 π3
+ L 12 (−K1 )L 12 (−K2 ) + L 12 (−K1 )L 12 (−K2 ) (4)
16d1 d2 (K1 + 1)(K2 + 1)
α1 α2
8dα
1 d2 d3 (K1 + 1)(K2 + 1)
1 α2 α3

⎛γ ⎞ ⎛  ⎞⎡ ⎛  ⎞⎤
1 1 ⎝ γ0
th
1 − Nμ −( γγth0 − N μ)2 γ0 − N μ
γth

Pout,1 ≈ + erf √ ⎠−  exp ⎝ ⎠ ⎣1 + erf ⎝  ⎠⎦ (7)

2 2 2(1/dα 2
2 1 + N σ 2 dα 3 + Nσ ) 2N σ 2 (1 + N σ 2 dα )
2 3
2N σ 3
3
3
3

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on February 08,2023 at 13:56:49 UTC from IEEE Xplore. Restrictions apply.
TAO et al.: PERFORMANCE ANALYSIS OF INTELLIGENT REFLECTING SURFACE AIDED COMMUNICATION SYSTEMS 2467

indicating the great benefit of increasing N in terms of outage and observing the fact that h1 and h2 are independent,
performance. we obtain
⎧ ⎫
⎪
⎪ ⎪
⎪
V. C ONCLUSION ⎨ 
N N ⎬
E |h2,n ||h1,n ||h2,j ||h1,j |
This letter has studied the ergodic capacity and outage prob- ⎪
⎪ ⎪
⎪
⎩n=1 j=1 ⎭
ability of IRS-aided SISO systems under mixed Rayleigh and j=n
Rician fading channels. Closed-from expressions are derived N (N − 1)π 2  2  2
for the ergodic capacity upper bound and outage approxima- = L 12 (−K1 ) L 12 (−K2 ) .
16dα
1 d2 (K1
1 α2
+ 1)(K2 + 1)
tion, which provide efficient means to evaluate the system (13)
performance. Moreover, concise expressions are obtained in
the asymptotic regime, which sheds lights on the impact of 3) Computing x3 : For the Rayleigh variable |g|, we have
key parameters on the system performance. It is revealed that 
π
the use of IRS contributes to an effective SNR gain of N 2 , and E{|g|} = . (14)
the diversity order can be increased to N +1. Also, the position 2dα3
3

of IRS also has significant impact on the system performance, As such, x3 can be calculated as
it is desirable to deploy the IRS close to the transmitter or 
receiver, and with strong LOS path. π3 N 2
x3 = L 1 (−K1 )L 12 (−K2 ).
8d1 d2 d3 (K1 + 1)(K2 + 1) 2
α1 α2 α3

A PPENDIX A (15)
P ROOF OF T HEOREM 1
To this end, pulling x1 , x2 and x3 together yields the desired
Applying the Jensen’s inequality, we have the upper bound result.
of the capacity as
A PPENDIX B
C ≤ Cup = log2 (1 + E {γmax }) . (9) P ROOF OF T HEOREM 2
The remaining task is to compute E {γmax }. Using the rela- According to the definition, the outage probability can be
tionship in (2) and applying the binomial expansion theorem, transformed into
we have   
γth
⎧ 2 ⎫ Pout (γth ) = Prob z ≤ , (16)
 2 ⎨  N ⎬ γ0
E {γmax } = E |g| + E |h2,n ||h1,n | -N
! "# $ ⎩ ⎭ where z  u+|g| with u  n=1 |h2,n ||h1,n |. For sufficiently
n=1
x1 ! "# $ large N , u can be accurately approximated by the normal

x2
 distribution according to the central limit theorem, i.e., u ∼

N N (N μ, N σ 2 ), where
+ 2E |h2,n ||h1,n ||g| . (10)
n=1 μ = E{|h2,n ||h1,n |}
! "# $ π
x3 =  α1 α2 L 1 (−K1 )L 12 (−K2 ), (17)
4 d1 d2 (K1 +1)(K2 +1) 2
We now calculate x1 , x2 and x3 one-by-one.
1) Computing x1 : Obviously, we have x1 = dα13 . and
3
2) Computing x2 : Using the binomial expansion, x2 can be 1
σ 2 = D{|h2,n ||h1,n |} = α1 α2
further expanded as d1 d2
  2 
N  π2
 × 1− L 12 (−K1 )L 12 (−K2 ) .
x2 = E 2
|h2,n | |h1,n |2 16(K1 +1)(K2 +1)
(18)
n=1
⎧ ⎫
⎪
⎪ ⎪
⎪ Since z is the sum of independent random variable u and |g|,
⎨N 
N ⎬
+E |h2,n ||h1,n ||h2,j ||h1,j | . (11) its cumulative distribution function (CDF) can be calculated
⎪
⎪ ⎪
⎪ via
⎩n=1 j=1 ⎭
j=n 1 ∞
,- . Fz (z) ≈ fu (z − x)F|g| (x)dx, (19)
N 2 2
It is easy to see that E n=1 |h2,n | |h1,n | = α1N α2 . −∞
/ 0 d1 d2 2 3
-N -N where fu (x) = √ 1 exp −(x−N μ)2
is the probability
Regarding E n=1 j=1 |h2,n ||h1,n ||h2,j ||h1,j | , noticing 2πN σ2 2N σ2  α3 2
j=n −d3 x
that, for the Rician variable |hl,n |, l ∈ {0, 1}, density function (PDF) of u and F|g| (x) = 1 − exp 2
 denotes the CDF of Rayleigh variable |g|.
π To this end, the desired result can be obtained after some
E{|hl,n |} = L 1 (−Kl ), (12)
4dα
l
l
(K l + 1) 2 algebraic manipulations.

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on February 08,2023 at 13:56:49 UTC from IEEE Xplore. Restrictions apply.
2468 IEEE COMMUNICATIONS LETTERS, VOL. 24, NO. 11, NOVEMBER 2020

A PPENDIX C 3) For n = p, we observe that Pn=p is dominated by the

P ROOF OF T HEOREM 3 term n = p = 0, thus we have
Recall γmax in Eq. (2), it can be interpreted as the effec- Pn=p ≈ 2(K1 + 1)(K2 + 1)e−(K1 +K2 )
tive SNR of an equal gain combining SIMO system. Then,
according to [17], in order to obtain the outage approximation
in the high SNR regime, it is sufficient to characterize the
behavior of the probability density function (PDF) of the SNR
of individual branch near the origin. Specifically, let β denote
the SNR of the branch, f (β) denote the PDF of β. If
lim fβ (β) = aβ t + O(β t+ε ),
β→0
where O(x) is the big O notation, then the high SNR outage
probability can be accurate characterized by the parameters a
and t.
Therefore, the main task is to obtain the behavior of the PDF
of |g|2 and |h2,n |2 |h1,n |2 near the origin. Since |g|2 follows
the exponential distribution, its behavior near the origin can
be easily obtained. Hence, we focus on |h2,n |2 |h1,n |2 in the
following.
Capitalizing on the results of [18], the PDF of random
2 2
variable β = dα 1 d2 |h2 | |h1 | can be derived as
1 α2
∞
 n p 2
 K12 K22
fβ (β) = 2(K1 + 1)(K2 + 1)e−(K1 +K2 )
n,p=0
n!p!
 
× K0 2 (K1 + 1)(K2 + 1)β . (27)
Combining the above three parts, we have t = 0 and

a = (K1 + 1)(K2 + 1)e−(K1 +K2 ) Ei(K1 ) + Ei(K2 ) − 2γ
  
(K1 +1)(K2 +1)
(20) − ln K1 −ln K2 +2K0 2 . (28)
dα1 α2
1 d2 γ0
To this end, invoking [17, Eq. (12)], the desired results can
be obtained after some algebraic manipulations.
R EFERENCES
[1] C. Huang et al., “Holographic MIMO surfaces for 6G wireless net-
works: Opportunities, challenges, and trends,” 2019, arXiv:1911.12296.
[Online]. Available: http://arxiv.org/abs/1911.12296
[2] C. Huang, R. Mo, C. Yuen, and S. Member, “Reconfigurable intelligent
surface assisted multiuser MISO systems exploiting deep reinforce-
ment learning,” 2020, arXiv:2002.10072. [Online]. Available: http://
arxiv.org/abs/2002.10072
[3] Q. Wu and R. Zhang, “Intelligent reflecting surface enhanced wireless
network via joint active and passive beamforming,” IEEE Trans. Wireless
Commun., vol. 18, no. 11, pp. 5394–5409, Nov. 2019.
[4] Q. Wu and R. Zhang, “Beamforming optimization for wireless network
aided by intelligent reflecting surface with discrete phase shifts,” IEEE

n+p Trans. Commun., vol. 68, no. 3, pp. 1838–1851, Mar. 2020.
× (K1 + 1)(K2 + 1)β Kn−p [5] C. Huang, A. Zappone, G. C. Alexandropoulos, M. Debbah, and
 
C. Yuen, “Reconfigurable intelligent surfaces for energy efficiency in
× 2 (K1 + 1)(K2 + 1)β . wireless communication,” IEEE Trans. Wireless Commun., vol. 18, no. 8,
pp. 4157–4170, Aug. 2019.
With the following relationships of bessel-K function [16, [6] Y. Han, W. Tang, S. Jin, C.-K. Wen, and X. Ma, “Large intelli-
gent surface-assisted wireless communication exploiting statistical CSI,”
Eq. 10.30.2] IEEE Trans. Veh. Technol., vol. 68, no. 8, pp. 8238–8242, Aug. 2019.
1 1 [7] S. Zhang and R. Zhang, “Capacity characterization for intelligent reflect-
lim Kν (z) ≈ Γ(ν)( z)−ν , ν > 0, (21) ing surface aided MIMO communication,” IEEE J. Sel. Areas Commun.,
z→0 2 2 to be published. [Online]. Available: http://arxiv.org/abs/1910.01573
Kν (z) = K−ν (z), (22) [8] X. Hu, C. Zhong, Y. Zhu, X. Chen, and Z. Zhang, “Programmable
metasurface based multicast systems: Design and analysis,” IEEE J. Sel.
where ν represents the real part of ν. We find it is convenient Areas Commun., to be published.
to consider three separate cases depending on the relationship [9] G. Zhou, C. Pan, H. Ren, K. Wang, and A. Nallanathan, “Intel-
ligent reflecting surface aided multigroup multicast MISO com-
of n and p, namely, munication systems,” 2019, arXiv:1909.04606. [Online]. Available:
http://arxiv.org/abs/1909.04606
lim fβ (β) = Pn>p + Pn<p + Pn=p . (23) [10] Y. Jia, C. Ye, and Y. Cui, “Analysis and optimization of an intel-
β→0
ligent reflecting surface-assisted system with interference,” 2020,
1) For n > p, we have arXiv:2002.00168. [Online]. Available: http://arxiv.org/abs/2002.00168
∞ 
∞ [11] C. Guo, Y. Cui, F. Yang, and L. Ding, “Outage probability analysis and
 minimization in intelligent reflecting surface-assisted MISO systems,”
Pn>p ≈ 2(K1 + 1)(K2 + 1)e−(K1 +K2 ) (24) IEEE Commun. Lett., vol. 24, no. 7, pp. 1563–1567, Jul. 2020.
p=0 n=p+1 [12] D. Kudathanthirige, D. Gunasinghe, and G. Amarasuriya, “Perfor-
 2 mance analysis of intelligent reflective surfaces for wireless com-
n
K12 K22
p

2p munication,” 2020, arXiv:2002.05603. [Online]. Available: http://
× (K1 + 1)(K2 + 1)β arxiv.org/abs/2002.05603
n!p! [13] I. Trigui, W. Ajib, and W.-P. Zhu, “A comprehensive study of
1 reconfigurable intelligent surfaces in generalized fading,” 2020,
× Γ(n − p). arXiv:2004.02922. [Online]. Available: http://arxiv.org/abs/2004.02922
2 [14] J. Gao, C. Zhong, X. Chen, H. Lin, and Z. Zhang, “Unsupervised
It is easy to see that Pn>p is mainly determined by p = 0, learning for passive beamforming,” IEEE Commun. Lett., vol. 24, no. 5,
hence we have pp. 1052–1056, May 2020.
[15] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions.
∞
−(K1 +K2 ) K1n New York, NY, USA: Dover, 1974.
Pn>p ≈ 2(K1 + 1)(K2 + 1)e . (25) [16] F. W. Olver, D. W. Lozier, R. F. Boisvert, and C. W. Clark, NIST
n=1
2nn! Handbook of Mathematical Functions. New York, NY, USA: Cambridge
Univ. Press, 2010.
2) For n < p, exploiting the symmetric property in Eq. (22), [17] Z. Wang and G. B. Giannakis, “A simple and general parameterization
we can similarly derive quantifying performance in fading channels,” IEEE Trans. Commun.,
∞ vol. 51, no. 8, pp. 1389–1398, Aug. 2003.
 K2p [18] N. O’Donoughue and J. M. F. Moura, “On the product of independent
Pn<p ≈ 2(K1 + 1)(K2 + 1)e−(K1 +K2 ) . (26) complex gaussians,” IEEE Trans. Signal Process., vol. 60, no. 3,
p=1
2pp! pp. 1050–1063, Mar. 2012.

Authorized licensed use limited to: TONGJI UNIVERSITY. Downloaded on February 08,2023 at 13:56:49 UTC from IEEE Xplore. Restrictions apply.