IEEE COMMUNICATIONS LETTERS, VOL. 18, NO.

9, SEPTEMBER 2014 1495

A Clarification of the Proper-Integral Form for the Gaussian Q-Function and

Some New Results Involving the F -Function
F. Javier Lopez-Martinez, R. F. Pawula, Eduardo Martos-Naya, and Jose F. Paris

Abstract—We discuss the origin of the well-known integral form However, many authors [4]–[6] (including Simon himself) have
for the Gaussian Q-function. Although this form is originally later reported that the form (1) was implied in earlier works
attributed to Craig in his 1991 paper, it actually was inferred by Weinstein [7], and explicitly stated by Pawula, Rice and
in previous communications theory references in the literature Roberts [8]. Notwithstanding these prior information and com-
by Weinstein, and explicitly stated by Pawula, Rice and Roberts.
Specifically, this form can be seen as a particular case of the munication theory papers, the term “Craig’s form” had already
general F -function, that is related to the distribution of the phase become the de facto standard nomenclature for integrals with
angle between two vectors perturbed by Gaussian noise. We show the form in (1). For this reason, the first objective of this paper
that the F -function includes as particular cases the one- and is to shed light on the origin of the well-known proper-integral
two-dimensional Gaussian Q-functions, and is connected with a form of Q(x) in (1). This has been a long-standing issue in
class of confluent hypergeometric functions through the multi- Communication Theory.
(n)
variate Φ1 function. This connection generalizes the existing As a direct consequence of the present study, we demonstrate
relation between the Gaussian Q-functions and this family of that the Gaussian Q-function is a particular case of the F -
hypergeometric functions.
function introduced in [8] and further simplified in [9] when
Index Terms—Communication theory, F -function, gaussian characterizing the distribution of the phase angle between two
Q-function, hypergeometric functions, special functions. vectors perturbed by Gaussian noise. The F -function for this
I. I NTRODUCTION case is, for 0 ≤ |ψ| < π,
√   
sgnψ π−|ψ| ρ1 ρ2 sin2 ψ
T HE Gaussian Q-function Q(x) = (1/2) erfc(x/ 2), de-
fined as the complementary cdf of the normal distribution,
is perhaps the most studied special function in communication
F (ψ) = −
2π 0
exp −
ρ1 sin2 θ+ρ2 sin2 (θ+|ψ|)
dθ,
(2)
theory as it is used to characterize the performance of com- where ρ1 and ρ2 are the signal-to-noise ratios (SNR) of the two
munication systems affected by additive white Gaussian noise vectors. This simplified form comes from the application of the
(AWGN). Because it is an improper integral, erfc(x) is not Euler-Legendre change of variables over the original expression
easy to evaluate. Serendipitously, a better, proper integral for in [8].
it was discovered over 40 years ago in a table of sonar integrals In fact, the two-dimensional Gaussian Q-function [10] de-
[1]. However, beginning around 1991 the proper integral form fined as the joint bivariate complementary cdf of two Gaussian
has been incorrectly credited to Craig in his paper presented at random variables is also shown to be connected with the F -
MILCOM’91 [2] when analyzing the probability of error for function. Finally, we also prove that the existing connections
two-dimensional signal constellations. This expression is between the one and two-dimensional Gaussian Q-functions
and the family of confluent hypergeometric functions [11] can
 π
1 2 x2 be generalized at a higher level, showing that the F -function is a
Q(x) = e− 2 sin2 θ dθ, (1) particular case of one of the confluent forms of the multivariate
π 0
hypergeometric function Φ1 [12].
for x ≥ 0, and satisfies Q(x) = 1 − Q(|x|) for x < 0. More-
over, ever since the later book by Simon, Hinedi and Lindsey II. A H ISTORICAL P ERSPECTIVE ON THE
in 1995 [3], the integral in (1) has been referred to as “Craig’s P ROPER I NTEGRAL (1)
form” and the paper [2] has been cited more than 500 times.
Equation (1) follows directly from [2, eq. 9], being a special-
ization of the more general expression [2, eq. 5]

Manuscript received February 20, 2014; accepted June 11, 2014. Date of 1 π−Φ −ρ sin22 Φ
publication July 11, 2014; date of current version September 8, 2014. This PM (ρ) = e sin θ dθ (3)
work was supported by Junta de Andalucia (P11-TIC-7109), the Spanish π 0
Government-FEDER (TEC2010-18451, TEC2011-25473), and by the Univer-
sity of Malaga and EU under Marie-Curie COFUND U-mobility program which gives the error probability of M -ary phase-shift keying
(ref. 246550). The associate editor coordinating the review of this paper and (PSK) modulation in AWGN channel, with ρ representing the
approving it for publication was T. Tsiftsis. symbol energy divided by the noise density. This problem had
F. J. Lopez-Martinez is with the Electrical Engineering, Stanford University,
Stanford, CA 94305 USA (e-mail: [email protected]).
been thoroughly studied by many authors (e.g. [13]), although
R. F. Pawula resides in San Diego, CA 92117 USA. Craig’s derivation is simpler than previous results.
E. Martos-Naya and J. F. Paris are with Departimento Ingenieria de Commu- Nine years before the publication of Craig’s paper, Pawula,
nicaciones, Universidad de Malaga, Malaga, Spain. Rice and Roberts published the paper “Distribution of the
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. phase angle between two vectors perturbed by Gaussian noise”
Digital Object Identifier 10.1109/LCOMM.2014.2331056 [8]. In this paper, the authors introduced the F -function as

1089-7798 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1496 IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 9, SEPTEMBER 2014

a means to characterize the distribution of such phase in the Compared to (1), equation (2) has been used in many other
most general situation. One of the scenarios considered in [8] communication-theoretic problems, including the performance
is the symbol error rate analysis of MPSK, resulting in the of MPSK and MDPSK [8], binary continuous-phase frequency-
following expression for the probability of error for equiprob- shift-keying (CPFSK) [9], digital FM and multi-frequency FM
able, equal energy MPSK signals using the optimum receiver systems using different bit detection strategies such as limiter
(cfr. [8, eq. 71]) discriminator (L/D) with integrate-and-dump [16], L/D with
 π  1  π/2−π/M sin2 π/M
sample-and-hold or differential detection [17]. This approach
PMPSK = F (π)−F = e−ρ cos2 θ dθ, (4) has also facilitated the analysis of double error rates ([18]),
M π −π/2 and the effects of power imbalances, phase mismatch or noise
correlation on receiver performance. While keeping the same
where F (·) denotes the F -function given by (2), with ρ1 → advantages of (1), it is clear that (2) finds more applications
∞ and ρ2 = ρ. It is noted that (1) and (4) are the same except owing to its more general form.
for a phase shift of π/2, that does not have any impact on its It should also be mentioned that even though the proper-
evaluation. integral form of the Gaussian Q-function is extensively used in
It is worth mentioning that the authors in [8] realized that communications, no reference to this form appears in relevant
a similar expression was presented by Weinstein in [7] in 1974 references such as [19], [20] or in Wolfram/Mathematica.
when characterizing the cdf of the phase of a sine wave affected
by AWGN. In fact, a slight manipulation of [7, eq. 1] yields IV. C ONNECTIONS WITH G AUSSIAN Q-F UNCTIONS
directly (3). Besides, the connection between (4) and the error
function was also established in [7, eq. 2]. In the literature, the Gaussian Q-functions are connected
Improperly giving credit to Craig’s 1991 paper instead of with other families of more general special functions. Specif-
Weinstein’s 1974 paper [7] and Pawula, Rice, and Robert’s ically, the connection between Q(x) and the confluent hy-
1982 paper [8] can be traced back to Section 3.2.8 of Simon, pergeometric function 1 F1 was reported in [19, 7.1.21], and
Lindsey and Hinedi’s widely-read 1995 book [3] where a new connection between the two-dimensional Gaussian
Craig’s clever derivation is praised, but Craig’s failure to cite Q-function Q(x, y; ρ) and the confluent hypergeometric func-
(2)
significantly-relevant, years-earlier, peer-reviewed works is not tion of two variables Φ1 was recently established in [11].
mentioned. Even though two different expressions for the SER However, the relation between the more general F -function and
of MPSK can be found in [3] (cfr. pp. 137–140 and p. 209), the other families of special functions has not been addressed in the
connections with [8] were not included. literature to the best of our knowledge.
The connections between the F -function (2) and the Rice
III. T HE F -F UNCTION : A M ORE G ENERAL Ie-function and Marcum Q-function were addressed in [21],
P ROPER I NTEGRAL [9]. Thus, a direct application of the recent results in [22]
shows that (2) is connected with the confluent hypergeometric
We will now pay attention to the more general proper in- function of two variables Φ3 .
tegral defined in (2), referred to as case-II F -function in [8]. Here, we will show that the case-IV F -function can be
This expression is counterintuitive as ρ1 and ρ2 do not seem expressed in terms of the confluent hypergeometric function
interchangeable. This fact has been unnoticed for many years; (2)
of two variables Φ1 , while the case-I F -function in [15] is
however, using a change of variables θ = (π − ψ) − θ suffices connected with the confluent hypergeometric function of three
to complete the proof, showing that the SNRs can indeed be (3)
exchanged. variables Φ1 . Both confluent hypergeometric functions admit
This work [8], included in The Best of the Best: Fifty Years a simple evaluation in terms of a single integral of Euler-
of Communications and Networking Research [14], has been type, and appear in classical books of integrals [20], transforms
(n)
referenced by almost 400 research papers and books, allowing [23] and special functions [12]. Since the Φ1 function also
the performance analyses of a number of digital communication appears in numerous scenarios in the literature of communi-
systems. Just like expression (1), one of the most appealing cation theory [11], [24]–[26], this connection may be useful
properties of the F -function relies on its very simple compu- for future updates of mathematical software packages. Using
tation: the different cases for the F -function included in [8], these connections, we present alternative expressions for the
[15] are each given in terms of a single integral with finite inte- F -function in terms of the Φ1 function.
gration limits and smooth Gaussian-like integrand. This enables
further averaging in scenarios affected by fading, or facilitates A. Definitions
the extraction of analytical insights. The usefulness of the Definition 1: Complete integral F(ψ, u).
F -function as an auxiliary function from which probabilities  π
2 −u(1−ξ cos θ)
Δ
like error rates are calculated is detailed in [8]. F(ξ, u) = e (1−ξ cos θ) dθ, (5)
Specifically, the form in (2) was presented in a later work 0
by Pawula [9], although it is equivalent to the original for- where u ≥ 0 and ξ ∈ [0, 1).
mula derived in [8]. Expression (2) is remarkably simple but Definition 2: Incomplete integral Δ(x, v)
general: setting ρ1 = ρ2 = ρ yields the error probability of  π
M -ary differential PSK (MDPSK) with coherent detection, 2 x2
e− 2 sin2 θ dθ,
Δ
whereas ρ1 → ∞ and ρ2 = ρ leads to the error probability of
Δ(x, v) =  1+v
(6)
arctan 1−v
MPSK, i.e., reduces to (1). Both cases were studied in [8], and
correspond to the case-I and case-IV F -function, respectively. where −1 ≤ v ≤ 1 and x ≥ 0.
LOPEZ-MARTINEZ et al.: A CLARIFICATION OF THE PROPER-INTEGRAL FORM FOR THE GAUSSIAN Q-FUNCTION 1497

Definition 3: Multivariate confluent hypergeometric func-

(n)
tion Φ1 (a; b1 . . . bn−1 ; c; x1 . . . xn ), n ≥ 2 [12, eq. 14].
(n)
Φ1 (a; b1 . . . bn−1 ; c; x1 . . . xn ) =
 ∞
(a)m1 +...+mn (b1 )m1 . . . (bn−1 )mn−1 x1 xn
... ,
m ...m
(c) m 1 +...+m n
m 1 ! m n!
1 n=0 Fig. 1. New connections between F -function and other special functions.
(7)
where (·)m denotes the Pochhammer symbol
Proposition 2: The integral Δ(x, v) can be expressed in terms
Definition 4: Euler-type integral form of the multivariate
(n) of the confluent hypergeometric function of two variables as
confluent hypergeometric function Φ1 .
 
Γ(c) 1−v − x2 (2) 1 3 v−1 x2 v−1
(n)
Φ1 (a; b1 . . . bn−1 ; c; x1 . . . xn ) = Δ(x, v) = e 2 Φ1 ; 1; ; , , (14)
Γ(a)Γ(c − a) 1+v 2 2 v+1 2 v+1
 1
n−1
Proof: Specializing the results in [11, eq. 7], we directly
× ta−1 (1 − t)c−a−1 etxn (1 − xi t)−bi dt (8) obtain (14). 
0 i=1

We also include the definitions of the F -function listed in C. Alternative Expressions for the F -Function
[15]. For convenience of notation, we introduce a superindex in
the F -function to identify the special cases of this function. We Theorem 1: The function F (I) (ψ, r) can be expressed in
also consider the phase arguments of the F -function belong to terms of the complete integral in (5) as
the interval (0, π]; results for the case [−π, 0) are direct since sin ψ
the F -function is defined as an odd function F (I) (ψ, r) = − F(cos ψ, r). (15)
2π
Definition 5: Case-I F -Function [15, eq. 12]
 π Proof: Identifying the terms in the integral expression (5),
sin ψ 2 e−r(1−cos ψ cos θ) 
F (I) (ψ, r) = − dθ, (9) we directly obtain (15).
2π 0 (1 − cos ψ cos θ) Theorem 2: The function F (IV ) (ϑ, r) can be expressed in
where 0 ≤ ψ ≤ π and r ≥ 0. terms of the complete integral in (2) as
Definition 6: Case-IV F -Function [15, eq. 24]. 1 κϑ 
 F (IV ) (ϑ, r) = − Q(x) − Δ(x, v) (16)
sgnϑ π−ϑ −r sin22 ϑ 2 π
F (IV )
(ϑ, r) = − e sin θ dθ, (10) √
2π 0 where x = 2r sin ϑ, v = − cos 2ϑ, κϑ = sgn{ϑ − π/2} and
where 0 ≤ ϑ ≤ π and r ≥ 0. Q(·) is the Gaussian Q-function.
Proof: See Appendix (A).
In this theorem, the case-IV F -function is expressed in terms
B. New Results
of the Gaussian Q-function and the integral function Δ(x, v),
Proposition 1: The integral F(ξ, u) can be expressed in (2)
given in (14) in terms of Φ1 function. Since Δ(x, v) is
terms of the confluent hypergeometric function of three vari- connected with the two-dimensional Gaussian Q-function [11]
ables as through Δ(x, v) = Q(x) − Q(x, x; v), this implies that the
 
1 3 case-IV F -function is indeed related to the one and two-
F(ξ, u) = 2e−u Φ1
(3)
1; 1, ; ; ξ, −1, uξ . (11) dimensional Gaussian Q-functions.
2 2
Corollary 1: The function F (IV ) (ϑ, r) reduces to the ex-
Proof: Using the change of variable t = cos θ and after pression for the Gaussian Q-function in [2] when ψ = π/2.
some manipulations, we obtain  
 1 π x2
Q(x) = −2F (IV )
, . (17)
F(ξ, u) = e−u euξt (1 − ξt)−1 (1 + t)−1/2 (1 − t)−1/2 dt. 2 2
0
(12) Proof: Setting ϑ = π/2 in (16) and using Δ(x, 1) = 0
Identifying this equation with the Euler-form integral of the directly yields (17). 
confluent hypergeometric function given in [24, eq. 22] Corollary 1 shows that the so-called Craig’s formula for the
Gaussian Q-function presented in 1991 is actually a particular
(3) Γ(c)
Φ1 (a; b1 , b2 ; c; x1 , x2 , x3 ) = case of the more general F -function. Although the definition
Γ(a)Γ(c − a) of the case-IV F -function was introduced in [15], it was in-
 1
2
cluded in the original paper by Pawula, Rice and Roberts [17,
× ta−1 (1 − t)c−a−1 etx3 (1 − xi t)−bi dt, (13) eq. 69–71] as a degenerate form of the case-II F -function. We
0 i=1
summarize the connections here unveiled in Fig. 1.
where Γ(·) is the Gamma function, {a} > 0, {c − a} > 0, In Fig. 2, we represent the F -function defined in (2) for
and setting a = 1, c = 3/2, b1 = 1, b2 = 1/2, x1 = ξ, x2 = different values of ρ1 and ρ2 , stating its relationship with the
−1 and x3 = uξ, we obtain (11).  Gaussian Q-function through (17).
1498 IEEE COMMUNICATIONS LETTERS, VOL. 18, NO. 9, SEPTEMBER 2014

where κϑ = sgn{ϑ − π/2} affects the sign of the integral

depending on whether ϑ is over or below
√ π/2. Identifying this
integral with (6) as πI(ϑ, r) = κϑ Δ( 2r sin ϑ, − cos 2ϑ), we
obtain (16).

R EFERENCES
[1] A. H. Nuttall, “Some Integrals Involving the Q-Function,” Naval Under-
water Systems Center Newport, Newport, RI, USA, Tech. Rep. TR4297,
pp. 1–41, 1972.
[2] J. W. Craig, “A new, simple and exact result for calculating the proba-
bility of error for two-dimensional signal constellations,” in Proc. IEEE
MILCOM Conf., Nov. 1991, vol. 2, pp. 571–575.
[3] M. K. Simon, S. M. Hinedi, and W. C. Lindsey, Digital Communication
Techniques: Signal Design and Detection. Englewood Cliffs, NJ, USA:
Prentice-Hall, 1994.
[4] M. K. Simon and M.-S. Alouini, “A unified approach to the performance
analysis of digital communication over generalized fading channels,”
Proc. IEEE, vol. 86, no. 9, pp. 1860–1877, Sep. 1998.
[5] M. K. Simon and M.-S. Alouini, Digital Communication Over Fading
Channels. Hoboken, NJ, USA: Wiley, 2004.
[6] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Cambridge
Univ. Press, 2005.
[7] F. Weinstein, “Simplified relationships for the probability distribution of
Fig. 2. F -function in (2) for different values of ρ1 and ρ2 . Red marker the phase of a sine wave in narrow-band normal noise (corresp.),” IEEE
represents the specific value related with the Gaussian Q-function in (17). Trans. Inf. Theory, vol. IT-20, no. 5, pp. 658–661, Sep. 1974.
[8] R. F. Pawula, S. O. Rice, and J. Roberts, “Distribution of the phase
angle between two vectors perturbed by Gaussian noise,” IEEE Trans.
V. C ONCLUSION Commun., vol. COM-30, no. 8, pp. 1828–1841, Aug. 1982.
[9] R. F. Pawula, “Generic error probabilities,” IEEE Trans. Commun.,
We addressed a long-standing issue in Communication The- vol. 47, no. 5, pp. 697–702, May 1999.
[10] M. Simon and D. Divsalar, “Some new twists to problems involving the
ory, related with the origin of the incorrectly denominated Gaussian probability integral,” IEEE Trans. Commun., vol. 46, no. 2,
“Craig’s form” of the Gaussian Q-function. This expression is pp. 200–210, Feb. 1998.
[11] J. F. Paris, F. J. Lopez-Martinez, and E. Martos-Naya, “On the con-
traced back to Weinstein’s work in 1974, and is in fact a very nection between Gaussian Q-Functions and a class of hypergeometric
particular case of the expression derived by Pawula, Rice and functions: Application to LCR of lognormal processes,” IEEE Commun.
Roberts in 1982, regarded as F -function. Lett., vol. 17, no. 8, pp. 1493–1496, Aug. 2013.
[12] H. Srivastava and P. W. Karlsson, Multiple Gaussian Hypergeometric
We proved that the family of F -functions, defined in a Series. Hoboken, NJ, USA: Wiley, 1985.
communication-theoretic context to characterize the phase dis- [13] H. Voelcker, “Phase-shift keying in fading channels,” Proc. Inst. Elect.
tribution of two vectors perturbed by Gaussian noise, is con- Eng.—Part B: Electronic and Communication Engineering, vol. 107,
no. 31, pp. 31–38, Jan. 1960.
nected with a more general family of hypergeometric functions, [14] W. Tranter, The Best of the Best: Fifty Years of Communications and
which have been extensively characterized in classical books of Networking Research. Hoboken, NJ, USA: Wiley, 2007.
special functions, integrals and transforms. These results may [15] R. F. Pawula, “Distribution of the phase angle between two vectors per-
turbed by Gaussian noise II,” IEEE Trans. Veh. Technol., vol. 50, no. 2,
be useful for the communication theory community, as well pp. 576–583, Mar. 2001.
as for considering future extensions of the built-in functions [16] M. K. Simon, “The impact of mismatch on the performance of coded
included in mathematical software packages. narrow-band FM with limiter/discriminator detection,” IEEE Trans. Com-
mun., vol. COM-31, no. 1, pp. 28–36, Jan. 1983.
[17] R. F. Pawula, “On the theory of error rates for narrow-band digital FM,”
IEEE Trans. Commun., vol. COM-29, no. 11, pp. 1634–1643, Nov. 1981.
[18] R. F. Pawula, “Digital FM double errors,” IEEE Trans. Veh. Technol.,
A PPENDIX vol. 50, no. 2, pp. 570–575, Mar. 2001.
P ROOF OF T HEOREM 2 [19] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions
with Formulas, Graphs, Mathematical Tables. New York, NY, USA:
The case-IV F -function can be split into two integrals as Dover, 1964.
[20] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Prod-
⎧ ⎫ ucts, 7th ed. New York, NY, USA: Academic, 2007.
⎪
⎪ ⎪
⎪ [21] R. F. Pawula, “Relations between Rice Ie-function and Marcum
⎪
⎪ π/2  π/2 ⎪
⎪ Q-function with applications to error rate calculations,” Electron. Lett.,
1 ⎨ sin 2
ϑ sin 2
ϑ ⎬ vol. 31, no. 20, pp. 1717–1719, Sep. 1995.
−r −r
F (IV )
(ϑ, r) = − e dθ − e dθ . [22] D. Morales-Jimenez, F. J. Lopez-Martinez, E. Martos-Naya, J. F. Paris,
2π⎪⎪
2
sin θ ⎪
2
⎪
sin θ
⎪
⎪0    π−ϑ  ⎪
⎪
and A. Lozano, “ Connections Between the Generalized Marcum
⎩ √
I
⎭ Q-Function and a Class of Hypergeometric Functions,” IEEE Trans. Inf.
Q( 2r sin ϑ) ϑ,r
Theory, vol. 60, no. 2, pp. 1077–1082, Feb. 2014.
(18) [23] A. Erdélyi, W. Magnus, F. Oberhettinger, and F. G. Tricomi, Tables of
Integral Transforms, vol. I. New York, NY, USA: McGraw-Hill, 1954.
The first integral can be easily identified with the Gaussian [24] N. Ermolova, “Useful integrals for performance evaluation of commu-
nication systems in generalised η − μ and κ − μ fading channels,” IET
Q-function √in (17). In the second integral, we use the transfor- Commun., vol. 3, no. 2, pp. 303–308, Feb. 2009.
mation x = 2r sin ϑ and tan2 (π−ϑ) = 1+v/1−v, yielding [25] G. Alfano, A. De Maio, and A. M. Tulino, “A theoretical framework for
LMS MIMO communication systems performance analysis,” IEEE Trans.
Inf. Theory, vol. 56, no. 11, pp. 5614–5630, Nov. 2010.
 π
2
 π
[26] D. Dixit and P. Sahu, “Performance of QAM signaling over TWDP
2 2 x2
−r sin 2 ϑ dθ
Iϑ,r = e sin θ = κϑ  e− 2 sin2 θ dθ (19) fading channels,” IEEE Trans. Wireless Commun., vol. 12, no. 4,
arctan(tan(π−ϑ)) 1+v pp. 1794–1799, Apr. 2013.
1−v