Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

2013 International Conference on Communication Systems and Network Technologies

Gaussian Q-function and its Approximations

Sirisha Malluri, Member, IEEE, Vinay Kumar Pamula, Member, IEEE,


Department of ECE, Department of ECE,
DVR & Dr.HS MIC College of Technology, University College of Engineering, JNTUK,
Kanchikacherla, India 521180. Kakinada, India 533003.
sirishamalluri@ieee.org pamulavk@ieee.org

Abstract—This paper presents some useful The Q-function and complementary error function are related
approximations to the Gaussian Q-function with an by [1]
application to computation of bit error probability
(BEP) of M-ary phase shift keying (MPSK) modulation 1 x
scheme. Computational results are provided to compare Q( x) = erfc ( ). (2)
various approximations to Gaussian Q-function. 2 2

Keywords- Absolute error; bit error probability; Nakagami A pair of upper and lower bounds on the Gaussian Q-
fading; phase shift keying; Q-function. function are presented in [3]. The following are the tight
bounds on the Q(x), valid only for all x  0. The Upper
I. INTRODUCTION bound and lower bound are, respectively, given by [3]
The Gaussian Q-function, popular in communication
theory literature, is frequently found in the analysis of multi- − x2 1 − x 2 /2 1
Q( x) ≥ e . +e . (3)
antenna communication systems over fading channels [1]. 50 2( x + 1)
These functions are tabulated, and often available as built-in
functions in mathematical software tools. However, in many and
cases it is useful to have closed-form bounds or
approximations instead of the exact expression. In fact,
these approximations are particularly useful in evaluating the − x2 1 − x 2 /2 1
Q( x) ≤ e . +e . . (4)
bit error probability (BEP) in many communication theory 12 2π ( x + 1)
problems.
The Gaussian Q-function’s definition in the form of an
improper integral makes it hard to conduct exact analyses for Another approximation to the Q-function is given in terms of
communication systems. Thus it would be highly desirable erfc in [4, (14)] as
to obtain a closed-form using elementary functions.
However no such solution is possible. The only option has − x2 4
e 1 − x2
been to approximate. A number of approximations have erfc ( x ) ≈ + e 3
(5)
been proposed by mathematicians, but the search continuous. 6 2
The remaining part of the paper is organized as:
section II provides some approximations to Q-function. In i.e.,
Section III, an example on the application of the above
approximations is discussed and the average BEP of MPSK − x 2 /2 2
− x2
e 1
is computed. In section IV, the absolute error for the Q( x) ≈ + e 3
. (6)
approximations is discussed evaluated. And concluding 12 4
remarks are offered in section V.
The approximation (5) works well for some problems, but
II. Q-FUNCTION APPROXIMATIONS large errors on small arguments limit its application, which
The Gaussian Q-function, which relates the was denoted as the CDS approximation in [5]. The
complementary error function erfc(x), is important in approximation (7) is called GKAL approximation of
performance analysis of wireless communication systems Karagiannidis & Lioumpas [6] gives smaller approximation
over fading channels. The Q-function and erfc are related by error and is given by
[2, (4.1.1)]
§ · − x 2 /2
∞ ¨1− 1 e− ax ¸ e
1 −ξ 2 ¨ ¸
Q( x) = ³e dξ . (1) Q( x) ≈
© 2 ¹
(7)
2π x
b 2π x

978-0-7695-4958-3/13 $26.00 © 2013 IEEE 74


DOI 10.1109/CSNT.2013.25

Authorized licensed use limited to: TEL AVIV UNIVERSITY. Downloaded on February 25,2023 at 18:36:36 UTC from IEEE Xplore. Restrictions apply.
where a = 1.98 and b = 1.135. The accuracy with which (7) Pe = AQ ( Bλ ) (11)
represents the actual Q-function is quite remarkable.
However the presence of x in the denominator of (7) makes it
where A and B are constants. The value of A is typically a
difficult to evaluate the probability of error [7]. Another
approximation to Q-function [8, (8)] is function of number of symbols and B is proportional to
minimum Euclidean distance between the source symbols.
The values of A and B for MPSK are 2 and
− x 2 /2
e
Q( x) ≈ . (8) 2 §π ·
2 2 sin ¨ ¸ respectively. The average BEP with Q-
1.64 + 0.76 x + 4 ©M ¹
function approximation (3) is given by [10, (3)]
And one better known approximation is the Chernoff bound
[9] Bλ
§ − Bλ − ·
¸§§ m · λ

m m −1 − m λ ·
¨
2
e e
Q ( x ) ≈ 1 e− x
2
Pe = ³ + ¨¨ ¨ ¸ e λ ¸d λ
¸
/2 .
2
(9) ¨
0 ¨ 12 ¸
2π ( Bλ + 1) ¸ © © λ ¹ Γ ( m ) ¹
© ¹
III. APPLICATION TO FADING CHANNELS
∞ § e − Bλ · § § m · m −1 − m λ ·
m
λ
= ³¨ ¨
¸ ¨ ¸ e λ ¸d λ
0 © 50 ¹ ¨ © λ ¹ Γ ( m ) ¸
A. Importance of Nakagami fading channel
There are several probability distributions that have been © ¹
used to model the statistical characteristics of the fading Bλ

channel. When there are a large number of scatterers in the ∞ § § m ·m λ m −1 − m λλ ·
e 2
+³ ¨¨ ¸ e ¸¸ d λ
0 2( Bλ + 1) ¨ © λ ¹ Γ ( m )
channel that contribute to the signal at the receiver, as is the
case in ionospheric or tropospheric signal propagation, © ¹
application of the central limit theorem leads to a Gaussian . (12)
process model for the channel impulse response. If the
process is zero-mean, then the envelope of the channel
The first term of (12) is reduced into closed-form as [12]
impulse response at any instant has a Rayleigh probability m
distribution and an alternative statistical model to represent A§ m ·
channel response in the Nakagami-m distribution [10]. ¨ ¸ . The second term of (12) will not reduce
50 © m + B λ ¹
In contrast to the Rayleigh distribution, which has a
single parameter that can be used to match the fading- into closed-form, in terms of elementary functions.
channel statistics, the Nakagami-m is a two parameter
distribution, with the parameter distribution, with the Numerical integral technique will be used to solve it. The
parameters m and . As a consequence this distribution Average BEP with lower bound (4) is given by
provides more flexibility and accuracy in matching the

observed single statistics. The Nakagami-m distribution can § − Bλ − ·
¸§§ m · λ

m m −1 − m λ ·
¨
2
be used to model fading channel conditions that are either e e
Pe = ³ + ¨¨ ¨ ¸ e λ ¸d λ
more or less severe than the Rayleigh distribution as a 0¨ ¸
2π ( Bλ + 1) ¸ © © λ ¹ Γ ( ) ¸
¨ 12 m ¹
special case (m = 1) and this distribution provides the best © ¹
fit for data signals received in urban radio channels [10].
∞ § e − Bλ · § § m · m −1 − m λ ·
m
As an application to fading channel, Nakagami fading λ
= ³¨ ¨
¸ ¨ ¸ e λ ¸d λ
0 © 12 ¹ ¨ © λ ¹ Γ ( m ) ¸
will be considered in this paper. The probability density
function of instantaneous received SNR  on Nakagami © ¹
fading channel is given by [6, (13)] Bλ

∞ Ae § § m ·m λ m−1 − m λλ ·
2
λ +³ ¨¨ ¸ e ¸¸ d λ
2π ( Bλ + 1) ¨© © λ ¹ Γ ( m )
m
m −1 − m
§m· λ ¹
e λ
0
f (λ ) = ¨ ¸ (10)
© ¹
λ Γ ( m ) . (13)
The closed-form expression to the first term of (12) is
where m is fading factor, λ is average output received SNR. A§ m ·
m

¨ ¸ . And there is no closed-form solution for


B. Average bit error probability 12 © m + B λ ¹
The average BEP for MPSK will be derived in this the second term in terms of elementary functions, so the
section. The conditional probability of error for MPSK is second term integration is solved by numerical integration
given by [11]

75

Authorized licensed use limited to: TEL AVIV UNIVERSITY. Downloaded on February 25,2023 at 18:36:36 UTC from IEEE Xplore. Restrictions apply.
m
technique. The average BEP using approximation (6) is A§ m ·
Pe = . (19)
given by 2 ¨© m+ B λ /2 ¸¹

§ − B2λ −
2 Bλ
· m m −1 − m λ
The average BEP of MPSK signals over the Nakagami
¨e

¸§ m · λ
3 channel using all the Q-function approximations (3), (4),
e
Pe = A ³ + ¨ ¸ e λ dλ
0 ¨ 12 4 ¸¸ © λ ¹ Γ ( m )
(6), (7), (8), (9) is plotted in Fig. 1. The average BEP of
¨ MPSK signals over the Nakagami-m channel using all the
© ¹
Q-function approximations, (3), (4), (6), (7), (8), (9) is
§ − Bλ · plotted in Fig. 1. Variation of the average BEP for all
¸§§ m · λ
∞ e 2
m m −1 − m λ ·
= A³ ¨
approximations for a range of values of average signal-to-
¨ e λ ¸d λ
¸¸ ¨ ¨© λ ¸¹ Γ ( m )
. (14)
0¨ ¸ noise ratio (SNR) is shown. The average BEP decreases
¨ 12 © ¹
© ¹ with increase in average SNR. Average BEP with
approximation (3) has small value at origin when compared
§ − 2 B3λ ·
¸§§ m · λ

m m −1 − m λ · to remaining approximations and it is large with
+A³ ¨
e
e λ ¸d λ
¸ ¨¨ ¨© λ ¸¹ Γ ( m )
approximation (8). It is also clear that average BEP
¨ ¸
0
¨ 4 ¸© ¹ computed with (3), (4), (6) and (8) is almost similar to that
© ¹ of exact average BEP computed using numerical integration
(NI) method. Average BEP using approximations (6) and
The closed-form representation of (14) is (8) is exactly same.
m m IV. CALCULATION OF ABSOLUTE ERROR
§ · § ·
A¨ m ¸ A¨ ¸ Although all the approximations to Q-function are may be
m accurate, but cannot be applied practically, because in some
Pe = ¨ ¸ + ¨ ¸ . (15)
12 ¨ Bλ ¸ 4¨ 2Bλ ¸ applications accuracy is not a desired parameter of interest.
¨m+ ¸ ¨m+ ¸ And in some applications accuracy becomes primary
© 2 ¹ © 3 ¹ parameter of interest but at a cost of complexity in the
approximate function. One of the measures of accuracy of
The average probability of error using Q-function these approximations is absolute error. The absolute error
approximation (7) is given by can be calculated using [13]

A §m·
m

m −1
(1 − e − a Bλ
2 )e (
− B 2+ m λ λ )
'
Qabs = Q ( x ) − Q ( x ) (20)
Pe = ¨ ¸ ³λ dλ .
Γ(m) © λ ¹ 0 b 2π Bλ

(16)
The integration in (16) will not be reduced into closed-form
for all values of m. This can also be solved by numerical
integration technique. The average probability of error using
(8) is given by

§ B m·
m −¨ + ¸λ
©2 λ¹
A §m· ∞
m −1 e
Pe = ¨ ¸ ³λ d λ . (17)
Γ(m) © λ ¹ 0 1.64 + 0.76 Bλ + 4

The integration in (17) will not be reduced into closed-form


for all values of m and hence numerical integration has to be
used. The average BEP, using Q-function approximation
(9), is given as

· § § m · λ m−1 − m λ ·
λ
§ Ae − Bλ
m

Pe = ³ ¨ ¸¨¨ ¸ e
¸¸
dλ . (18)
0© 2 ¹ ©¨ © λ ¹ Γ ( m ) ¹
Figure 1. Average BEP Vs Average SNR (dB) with m = 1, 10.
The closed-form expression to (18) is given by

76

Authorized licensed use limited to: TEL AVIV UNIVERSITY. Downloaded on February 25,2023 at 18:36:36 UTC from IEEE Xplore. Restrictions apply.
REFERENCES
[1] J.G. Proakis, Digital communications, McGrawHill, Newyork, 2001.
[2] M. K.Simon and M.S. Alouini, Digital communications over fading
channels: A unified approach to performance analysis, John Wiley &
Sons, 2000.
[3] G.T. Freitas, “Supertight algebraic bounds on the Gaussian Q-
function,” IEEE Trans. Vehic. Tech., 2009.
[4] R. Nicole, “New exponentional bounds and approximations for the
computation of error probability in fading channels,” IEEE Trans.
Wireless Commun., vol.2, pp.840-845, Jul. 2003.
[5] D.N. Phong and N. Hoai, “Evolving the best known approximations
to Q function,” GECCO’12, Jul. 7-11, 2012.
[6] G.K. Karagiannidis and A.S.Lioumpas, “An improved approximation
for the Gaussian Q-function,” IEEE Commun. Lett., vol. 11, no. 8, pp.
644–646, Aug. 2007.
[7] Y. R. B. Isukapalli, "An Analytically Tractable Approximation for the
Gaussian Q-Function," IEEE Communications Letters, vol.12, no.9,
pp.669-671, Sep. 2008.
[8] N. Kingsbury, “Approximation formulae for the Gaussian, error
integral,” work produced by the connexions project. Available:
http://cnx.org/content/m11067/latest/
[9] M. Ambramowitz and I.A. Stagun, “Handbook of Mathematical
Figure 2. Absolute Error Vs x.. functions with formulas, graphs and mathematical tables,” p.257, 940,
941, National bureau of standards, 1964.
V. CONCLUSION [10] E. Biglieri, J.G. Proakis, and S. Shamai “Fading channels:
Information-Theoretic and Communication Aspects,” IEEE Trans.
Some useful approximations to Gaussian Q-function are Inform. Theory, vol. 44, no. 6, Oct. 1998.
presented in this paper. As an application of Q-function [11] P. Sharma, “Selection of diversity and modulation parameters for
approximations, the average BEP of MPSK modulation Nakagami fading channels to jointly satisfy outage and bit error
scheme over Nakagami-m fading channels is computed. requirements,” vol. V, no.6, IEEE Trans. Wireless Commun., Jun.
Absolute error is also computed to study the accuracy of 2006.
these approximations. [12] I.S. Gradshteyn, I.M.Ryzhik, and Alan Jeffery “Table of integrals,
series and products,” Acemademic Press, San Diego, CA, 1994.
[13] S.A.Dyer and J.S.Dyer “Approximations to error functions,” IEEE
Instr. Mag., Dec. 2007.

77

Authorized licensed use limited to: TEL AVIV UNIVERSITY. Downloaded on February 25,2023 at 18:36:36 UTC from IEEE Xplore. Restrictions apply.

You might also like