1544 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO.

3, MARCH 2010
[22] X.-D. Wang and H.V. Poor, Wireless Communication Systems: Advanced
Techniques for Signal Reception. Upper Saddle River, NJ: Prentice-Hall,
2004.
[23] C.-C. Tu and B. Champagne, Subspace blind MIMO-OFDM channel
estimation with short averaging periods: Performance analysis, in Proc.
IEEE Wireless Commun. Netw. Conf., Mar. 2008, pp. 2429.
[24] E. de Carvalho, J. Ciof, and D. Slock, Cramer-Rao bounds for
blind multichannel estimation, in Proc. IEEE Global Commun. Conf.,
Dec. 2000, vol. 2, pp. 10361040.
[25] 3rd Generation Partnership Project, Technical Specication Group
Radio Access Network; Spatial Channel Model for Multiple-Input
Multiple Output Simulations (Rel. 6), 3GPP TR 25.996 Technical Report
V6.0.0 (2003-09), 2003.
[26] D. A. Harville, Matrix Algebra From a Statisticians Perspective.
New York: Springer-Verlag, 1997.
Performance Analysis of Fixed-Gain
Amplify-and-Forward Relaying With MRC
Nam H. Vien, Student Member, IEEE, and
Ha H. Nguyen, Senior Member, IEEE
AbstractRelay transmission has recently attracted much attention
since it can offer spatial diversity with single antenna terminals. This
paper addresses the performance of a multiple-relay systemwith xed-gain
amplify-and-forward (AF) relaying in Nakagami-mfading. A tight upper
bound on the average symbol error probability (SEP) is obtained for a
system with K relays and when the maximal ratio combining (MRC)
is used at the destination. Based on the obtained bound, a maximum
diversity order of m(K+ 1), where mis the fading parameter, is shown.
Moreover, the problem of power allocation (PA) to minimize the SEP
upper bound is investigated. Numerical results illustrate signicant gains
provided by the proposed PA over equal PA (EPA) under various channel
conditions.
Index TermsAmplify-and-forward (AF) protocol, diversity order,
maximum ratio combining (MRC), Nakagami-m fading, performance
analysis, power allocation (PA), relay communications.
I. INTRODUCTION
Recently, relay communication has attracted a lot of research in-
terest due to its ability to offer spatial diversity while still satisfying
the size and power constraints of mobile devices [1][9]. The benet
comes from the cooperation of relays in a network to assist trans-
mission from the source to the destination. This is because, with the
assistance of relays, the transmission from the source to the destination
can be performed over a virtual antenna array [3], [4].
The most popular signal processing methods at relays are decode-
and-forward (DF) and amplify-and-forward (AF). For DF, cooperative
relays rst try to decode the received information and then regenerate
a new version to transmit to the destination [3], [10][12]. On the
other hand, for AF, the relays retransmit scaled versions of the received
information to the destination without decoding them. Therefore, AF
Manuscript received May 27, 2009; revised September 27, 2009 and
November 26, 2009. First published January 19, 2010; current version pub-
lished March 19, 2010. This work was supported by a Natural Sciences and
Engineering Research Council of Canada Discovery Grant. The review of this
paper was coordinated by Dr. C. Yuen.
The authors are with the Department of Electrical and Computer Engineer-
ing, University of Saskatchewan, Saskatoon, SK S7N 5A9, Canada (e-mail:
nam.vien@usask.ca; ha.nguyen@usask.ca).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TVT.2010.2040755
needs no sophisticated processing at the relays or the destination [3],
[5], [13]. To limit the transmit power at the relays, the received signal at
each relay can be amplied with a varying or xed gain. The varying-
gain relaying scheme maintains the constant transmit power at the
relays at all times, but it requires knowledge of the instantaneous
channel gains of all the sourcerelay links. To reduce the complexity
at the relays, the xed-gain relaying scheme has been proposed, which
maintains the long-term average transmit power at each relay [6].
For AF relaying, optimal maximal ratio combining (MRC) has
been considered in [4], [6], and [14][16]. In [14] and [15], the
performance of a single-relay system with a multiple-antenna desti-
nation is considered. In [14], under Rayleigh fading, the maximum
diversity orders with varying and xed gains are shown to be 2N
and N + 1, respectively, where N is the number of antennas at the
destination. For Nakagami-m fading, the exact MRC performance
for the xed-gain relay system is derived with the help of Kampe
de Feriets function in [15]. For multiple-relay systems with a single
antenna in both transmitter and receiver, the work in [4] investigates
the symbol error probability (SEP) for AF with the varying gains
at relays and with the varying MRC at the destination. In [16], the
performance of relay systems with xed-gain relays over generalized
fading channels is determined by using the geometric-mean (G-M)
bound on the instantaneous signal-to-noise ratio (SNR). This bound
is then used to evaluate the outage and error probabilities by using
the well-known moment generating function (MGF) approach. Using
the same technique as in [4], the authors in [6] consider the outage
probability behaviors of the relay systems with both varying and xed
gains in Nakagami-m fading at the high SNR region. The work shows
that when the fading severity difference between the source-to-relay
and the relay-to-destination channels exists, both strategies can achieve
the same diversity gain, whereas the xed-gain AF relaying strictly
loses some coding gains. However, the limitation of the method used
in [4] and [6] is that it can only predict the asymptotic performance of
the systems at the high SNR region, whereas in some applications, the
performance at low and medium SNRs might be more important [17].
It should be pointed out that the foregoing works on AF relay
systems assume that the total transmit power is uniformly allocated
over the source and the relays, i.e., equal power allocation (EPA).
Obviously, the performance of such AF systems with EPA is inferior
to that with optimal power allocation (OPA) among the source and all
the cooperating relays [18]. Such OPA schemes have recently been
considered in [18][22]. However, all the previously proposed OPA
schemes are restricted to Rayleigh fading channels. It is well known
that the Nakagami-m distribution provides a much better tting for
the fading channel distributions than the Rayleigh distribution in many
scenarios [23]. In fact, it includes the Rayleigh distribution (m = 1)
as a special case [23].
Motivated by the foregoing observations, this paper rst obtains a
tight upper bound on the SEP for xed-gain AF multiple-relay systems
with MRC at the destination. It is illustrated that the obtained bound is
effective over a wide range of channel settings. It is also shown that the
MRC can achieve the maximum diversity order of m(K + 1). Based
on the minimization of the obtained bound on the SEP performance,
a novel power allocation (PA) scheme is proposed. In particular,
under the assumption that all the channel statistics are available at the
destination, the proposed PA coefcients can easily be calculated at the
destination and then fed back to the source and relays during a startup
phase. The proposed PA demonstrates signicant performance gains
over EPA.
The rest of this paper is organized as follows. Section II describes
the system model under consideration. The upper bound on the overall
0018-9545/$26.00 2010 IEEE
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010 1545
SEP is derived in Section III. The proposed PA is described in
Section IV. Conclusions are given in Section V.
Notation: x

is the complex conjugate of x, and Re(x) takes

the real part of complex number x. For a random variable (RV)
X, p
X
() denotes its probability density function (pdf), and E{X}
denotes its expectation. The circularly symmetric complex Gaussian
RV with variance
2
is denoted by CN(0,
2
). The Q-function
is dened as Q(x) = (1/

2)

x
exp(t
2
/2)dt. The notation
E
1
(x) is used to denote the exponential integral, i.e., E
1
(x) =

x
(exp(t)/t)dt. (x) represents the gamma function, i.e., (x) =

0
exp(t)t
x1
dt, Re(x) > 0. The incomplete gamma function is
denoted by (, x), i.e., (, x) =

x
0
exp(t)t
1
dt, Re() > 0.
II. SYSTEM MODEL
Consider a multirelay system with a source S, K relays (R
k
)
K
k=1
,
and a destination D. All nodes are equipped with a single-antenna
transmitter and a receiver. Information is sent from S to D with the
help of (R
k
)
K
k=1
. The system is half-duplex, in which (R
k
)
K
k=1
cannot
transmit and receive at the same time. Frequency-at slow fading is
assumed throughout this paper. The channel coefcients (a
SR
k
)
K
k=1
,
(a
R
k
D
)
K
k=1
, and a
SD
represent independent Nakagami-m fading with
mean-square values (
SR
k
)
K
k=1
, (
R
k
D
)
K
k=1
, and
SD
, respectively.
Therefore, the normalized received SNR of the PQ link without power
scaling, i.e.,
PQ
= |a
PQ
|
2
, (PQ) {(SR
k
)
K
k=1
, (R
k
D)
K
k=1
, SD}
with = E
s
/N
0
, has the following pdf [24, Eq. (14.4-35)]:
p

PQ
(
PQ
) =
m
m

m1
PQ

m
PQ
(m)
exp

m
PQ

PQ

(1)
where m is the Nakagami-m fading parameter and assumed to be an
integer.
1
The average SNR of the PQ link, i.e.,
PQ
, is equal to
PQ
.
Transmission to D is done in (K + 1) time slots. In the rst time
slot, S broadcasts the signal s with the average transmitted power

s
E
s
to (R
k
)
K
k=1
and D, where 0
s
K + 1 is the scaling
power coefcient at S. The received signals at R
k
, y
SR
k
, and at D,
y
SD
, are, respectively, given as
y
SR
k
=

s
E
s
a
SR
k
s + z
SR
k
(2)
y
SD
=

s
E
s
a
SD
s + z
SD
(3)
where z
SR
k
and z
SD
represent additive white Gaussian noise
(AWGN) and are modeled as independent identically distributed
CN(0, N
0
) RVs.
For AF with xed-gain relays, R
k
applies a xed scaling to the
received signal from, S regardless of the fading amplitude of the rst
hop to maintain the long-term average transmitted power at R
k
. The
received signals y
R
k
D
at D from time slot 2 to time slot (K + 1) are
given as
y
R
k
D
= G
k
a
R
k
D

s
E
s
a
SR
k
s + z
SR
k

+ z
R
k
D
k = 1, . . . , K (4)
where z
R
k
D
is CN(0, N
0
) and represents AWGN at D. The xed gain
G
k
can be expressed as [13]
G
2
k
=
k
E
s
/

s
E
s

SR
k
+ N
0

=
k
/

s

SR
k
+ 1

(5)
1
For simplicity of analysis and presentation, it is assumed that the Nakagami
fading parameter m is the same for all links. The analysis method in this paper,
however, can be extended to the case where all the links have unequal fading
parameters.
which makes the long-term average SNR at R
k
be
k
, where
0
k
K + 1 is the power scaling coefcient at R
k
. Therefore,
the total power E
tot
used in the system is computed as
E
tot
=

s
+
K

k=1

E
s
. (6)
For convenience, the corresponding SNR is dened as

tot
= E
tot
/N
0
. Let us normalize (
s
+

K
k=1

k
) to be (K + 1) so
that the total power is equal to (K + 1)E
s
, which is the same as that in
EPA in which all the source and relays transmit at the same power E
s
.
The destination implements the optimal MRC detector as in [14],
[21], and [22]. The corresponding SNR at the output of the detector is
=
s

SD
+
K

k=1
G
2
k

SR
k

R
k
D
/

+ G
2
k

R
k
D

. (7)
III. ERROR PERFORMANCE ANALYSIS
For simplicity of presentation, the following analysis considers
M phase-shift-keying (M-PSK) constellation. However, the method
and the main results can easily be generalized to an arbitrary 2-D
constellation [such as M quadratic-amplitude modulation (M-QAM)].
The average SEP for M-PSK is given by [23, Eq. (8.22)]
P
s
=
1

(M1)/M

0
M

sin
2
(/M)/ sin
2

d (8)
in which
M

(t) = E

{exp(t)} (9)
is the MGF of the instantaneous received SNR at D given in (7). To
calculate P
s
as in (8), one needs to take the expectation of exp(t)
over all the channel SNRs, i.e.,
SD
, (
SR
k
)
K
k=1
, and (
R
k
D
)
K
k=1
,
and then compute the integral [see (8)] over nite limits. Due to the
complexity of such an exact computation, the interest here is to obtain
a tight upper bound on the SEP. It is given in the following lemma.
Lemma 1: For the K-relay system with the received SNR at D
given in (7), M-PSK modulation, and Nakagami-m fading, the
average SEP P
s
is upper bounded as
P
s

1

(M1)
M

1
(t
s

SD
/m + 1)
m
K

k=1
(A
k,1
(t) A
k,2
(t)
+ B
k,1
(t))

t=sin
2
(/M)/ sin
2

d (10)
where A
k,1
(t), A
k,2
(t), and B
k,1
(t) are given as
A
k,1
(t) =
1
(m)

2m
2

tG
2
k

s

SR
k

R
k
D

2m
tG
2
k

s

R
k
D

m1
exp

2m
2

tG
2
k

s

SR
k

R
k
D

E
1

2m
2

tG
2
k

s

SR
k

R
k
D

+
m1

l=1

2m
tG
2
k

s

R
k
D

m1l

m1
l

1546 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010

1
l!
l

p=1
(p 1)!

m

SR
k

lp

2m
2

tG
2
k

s

R
k
D

p
+
(m/
SR
k
)
l
l!
exp

2m
2

tG
2
k

s

SR
k

R
k
D

E
1

2m
2

tG
2
k

s

SR
k

R
k
D

(11)
A
k,2
(t) =
exp

m
G
2
k

R
k
D

2
(m)

m
2

SR
k

R
k
D

m1

l=0
( /G
2
k
)
l
l!

2
tG
2
k

ml

m1

p=0
pml

m1
p

2m
tG
2
k

s

R
k
D

m1p

p+lm

q=0

p+lm
q

2m
tG
2
k

s

R
k
D

p+lmq

q!

t
s
2
+
m

SR
k

q+1
m1

p=0
p=ml1

m1
p

2m
tG
2
k

s

R
k
D

m1p
exp

t
s
2
+
m

SR
k

2m
tG
2
k

s

R
k
D

E
1

t
s
2
+
m

SR
k

2m
tG
2
k

s

R
k
D

+
m1

p=0
pml2

m1
p

2m
tG
2
k

s

R
k
D

m1p

1
(mpl1)!
mpl1

q=0
(q1)!

t
s
2

m

SR
k

mplq1

2m
tG
2
k

s

R
k
D

q
+

t
s
2

m

SR
k

mpl1
(mpl1)!
exp

t
s
2
+
m

SR
k

2m
tG
2
k

s

R
k
D

E
1

t
s
2
+
m

SR
k

2m
tG
2
k

s

R
k
D

(12)
B
k,1
(t) =

SR
k

t
s
2
+
m

SR
k

m, m /

G
2
k

R
k
D

(m)
. (13)
Proof: See Appendix A.
A closed-form expression for the upper bound in (10) seems hard
to obtain. However, for the simplest case when there is only one relay
R with binary phase-shift keying (BPSK) modulation and Rayleigh
fading (m = 1), the bound can be simplied by using the Chernoff
bound as follows:
P
s

1
2

2
G
2

s

RD

SR
exp

2
G
2

s

RD

SR

E
1

2
G
2

s

RD

SR

2 exp

G
2

RD

G
2

s

SR

RD
exp

s
2
+
1

SR

2
G
2

s

RD

E
1

s
2
+
1

SR

2
G
2

s

RD

+
exp

G
2

RD

1 +

s

SR
2

1 +

s

SD
2
. (14)
The xed-gain G is computed as G =
1
/(
s

SR
+ 1), where
s
and
1
are the scaling coefcients at S and R, respectively. Note that
symbol error probability is identical with bit error probability (BEP)
for the case of BPSK.
For the case of EPA, i.e.,
s
=
1
= 1, and in the high SNR region,
(14) can further be approximated as
P
s

1
2

2

RD

SR

SR
1

+
2

RD

SR

SR

2 exp

SR

RD

SR

RD

SR
1
1 +

1
2
+
1

SR

2


SR

RD
+
exp

SR

RD

1 +

SR
2

1 +

SD
2

2

SD

1

RD
+
exp(
SR
/
RD
)
RD

SR
(
SR
+
RD
)

(15)
where the inequality 1/(1 + x) < exp(x)E
1
(x) 1/( + x) x >
0, 0 < < 1 [25] has been used. The approximated bound in (15)
is strikingly simple when compared with the original bound in (10)
while still maintaining its tightness (as seen later in Figs. 1 and 2).
It can clearly be seen from the bound in (15) that the overall perfor-
mance slope decays with the exponent of 2 in the loglog scale. The
expression also shows how the channel qualities, i.e.,
SR
,
RD
, and

SD
, contribute to the overall performance in this special case.
For the general case, the upper bound can easily be calculated
by numerical integration techniques since the right-hand side (RHS)
of (10) only involves a single integral with nite limits. Figs. 13
show the tightness of the derived bound on the performance of MRC
with BPSK modulation.
2
The total power
tot
is equally distributed
2
For K = 2, it is assumed that
SR
1
=
SR
2
=
SR
and that
R
1
D
=

R
2
D
=
RD
.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010 1547
Fig. 1. Simulation results, upper bounds, and G-M lower bounds on the SEP
performance: (
SR
,
RD
,
SD
) = (0, 0, 0) dB.
Fig. 2. Simulation results, upper bounds, and G-M lower bounds on the SEP
performance for one relay: (
SR
,
RD
,
SD
) = (20, 0, 0) dB.
among the source and relays in these special cases. For the case
that (
SR
,
RD
,
SD
) = (0, 0, 0) dB, Fig. 1 reveals that the derived
bound and its simplied version (for the case of (m = 1, K = 1))
are reasonably tight when compared with the simulation results.
For all the four cases of (m = 1, K = 1), (m = 2, K = 1), (m =
1, K = 2), and (m = 2, K = 2), all the corresponding bounds lie
within 0.7 dB of the simulation curves. Fig. 1 also plots the G-M
lower bound obtained in [16]. It can be seen that the G-M lower bound
is also quite tight over the whole range of SNR. The tightness of the
G-M lower bound in this case is a consequence of the fact that the
equality condition of the G-M lower bound holds when
SR
i
=
R
i
D
,
i = 1, . . . , K, i.e., balanced channels. However, the G-M lower bound
becomes very loose when the channels are unbalanced, i.e.,
SR
i
=

R
i
D
, i = 1, . . . , K, as can clearly be observed in Figs. 2 and 3. In
contrast, our derived bound can maintain its tightness over a wide
range of channel settings. As shown in Figs. 2 and 3, the proposed
bounds and its simplied version [for the case of (m = 1, K = 1)]
still lie within about 1.5 dB of the simulation results for all the four
cases of (m = 1, K = 1), (m = 2, K = 1), (m = 1, K = 2), and
(m = 2, K = 2) when (
SR
,
RD
,
SD
) = (20, 0, 0) dB.
3
3
The tightness of the proposed bound can also be seen with other channel
conditions, such as (
SR
,
RD
,
SD
) = (0, 20, 0) dB. However, due to
space limitations, the gures are omitted.
Fig. 3. Simulation results, upper bounds, and G-M lower bounds on the SEP
performance for two relays: (
SR
,
RD
,
SD
) = (20, 0, 0) dB.
To give an insight into the SEP performance, the following propo-
sition establishes the achievable diversity order of the relay system by
further examining its bound in Lemma 1.
Proposition 1: For the K-relay system with the received SNR at
D given in (7), M-PSK modulation, and Nakagami-m fading, the
average SEP P
s
is bounded as
(k
s
)
m(K+1)

P
s

(

k
s
)
m(K+1)
(16)
where

k
s
and k
s
are nonnegative constants that depend on
(
SR
k
,
R
k
D
)
K
k=1
and
SD
.
Proof: See Appendix B.
Since both upper and lower bounds in Proposition 1 are proportional
with (1/ )
m(K+1)
, one can conclude that MRC with xed-gain relays
can offer the maximum diversity order of m(K + 1) in the high SNR
region. Furthermore, the achievable diversity order does not depend on
the locations of (R
k
)
K
k=1
.
IV. POWER ALLOCATION AND ILLUSTRATIVE RESULTS
Exercising PA to optimize the performance (such as error or outage
probabilities) of relaying systems has extensively been studied in the
literature (see, e.g., [10], [18], [19], [21], and [22]). In most cases,
using the exact performance analysis as the optimization objective
for the PA problem is not tractable. As such, it is common to rely
on the performance bounds, e.g., upper bounds on the average SEP
[10] or on the pair-wise error probability [19], [21], as they are more
favorable to analyze. Given the tightness of the bounds, it is expected
that the PA obtained based on the bounds also works well for the actual
performance.
This section investigates the effects of PA on the SEP performance
of the relay system under consideration. Here, the goal is to minimize
the SEP upper bound, which is given in (10). The proposed PA scheme
only requires that the destination knows the channel statistics, i.e.,
(
SR
k
)
K
k=1
, (
R
k
D
)
K
k=1
, and
SD
, to compute the SEP upper bound.
Having all the needed information, the destination then computes
the power coefcients for the source and relays and noties all the
nodes of their transmission powers via low-rate feedback channels.
The optimization problem is formulated as follows.
Problem Statement 1: For the given channel statistics (
SR
k
)
K
k=1
,
(
R
k
D
)
K
k=1
, and
SD
and the total power constraint E
tot
, determine
the PA coefcients (
OPA
s
, (
OPA
k
)
K
k=1
) to minimize the upper bound
on P
s
given in (10) while satisfying the total power constraint E
tot
=
(
OPA
s
+

K
k=1

OPA
k
)E
s
= (K + 1)E
s
.
1548 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010
Fig. 4. Power gains of the proposed PA over EPA for one relay (predicted
by the SEP bound). Three cases of
SR
= {20, 0, 20} dB correspond to the
(+, o, ) markers, respectively.
Unfortunately, an analytical solution for the foregoing PA problem
appears to be intractable, even with high SNR approximations. Here,
we resort to numerical search based on the SEP upper bound obtained
in Lemma 1. It should be emphasized that the bound contains simple
nonlinear functions, i.e., the gamma function (x) and the exponential
integral function E
1
(x), which are readily available in many com-
puting softwares (e.g., Matlab). Common computing softwares (e.g.,
Matlab) provide both Gaussian quadrature and simple techniques, such
as the NewtonCotes formula. These techniques can readily be adapted
to calculate the integral expression in Lemma 1. Once the SEP upper
bound is computed, the numerical search for the OPA can easily be
carried out. For the simplest case of one-relay system, the line search
method [26] is used to nd the optimal values numerically. However,
when there are more than one relay in the system, we rely on the
brute-force search to nd out the optimal power coefcients for S
and (R
k
)
K
k=1
, i.e.,
OPA
s
and (
OPA
k
)
K
k=1
. Furthermore, in practical
implementation, the OPA coefcients can be obtained in advance for
typical channel statistics (
SR
k
)
K
k=1
, (
R
k
D
)
K
k=1
, and
SD
and then
built into a lookup table.
As a rst example, consider a single-relay system, i.e., K = 1.
To cover a wide range of possible channel settings, i.e., the triple
mean-squared values (
SR
,
RD
,
SD
), we rst normalize
SD
to
be 0 dB. For the SR link, three possible cases are the following:
1) R is farther to S than D; 2) R and D are equidistant to S; and
3) R is closer to S than D. These three settings are assumed to
correspond to
SR
= 20, 0, and 20 dB, respectively. For each value
of
SR
, a range of
RD
, i.e.,
RD
(
SR
20 dB,
SR
+ 20 dB),
is investigated.
4
Fig. 4 shows the power gains of the proposed PA over
EPA for m = 1 and 2, respectively, under the target BEP of 10
4
.
For a wide range of the relative gap between
RD
and
SR
, i.e.,

RD-SR
= (
RD

SR
)dB, the predictable gains are promising,
i.e., up to 2.3 dB for m = 1 and 2.7 dB for m = 2. When the RD link
is better than the SR link, i.e.,
RD-SR
= 20 dB, the proposed PA
can offer gains of at least 1.5 and 2.3 dB for m = 1 and 2, respectively.
As shown in Fig. 4, the proposed PA also provides power gains when

SR
is 20 dB and
RD-SR
is low, e.g.,
RD-SR
= 20 dB. This
is explained as follows. Due to the very poor quality of the indirect
channels when compared with that of the direct link (
SD
= 0 dB
4
Other settings reported in [7], [10], and [21] only consider the case that the
three nodes S, R, and D are in a 2-D space, which puts a constraint on the set
(
SR
,
RD
,
SD
), for example, [7, Eq. (22)]. On the other hand, we study in
this paper the more general case, i.e., three nodes are in 3-D space, which has
no constraint on the set (
SR
,
RD
,
SD
).
Fig. 5. Performance comparison of one relay with the proposed PA (dashed
line) and EPA (solid line).
RD-SR
= 20 dB. The cases of
SR
=
(20, 0, 20) dB correspond to the (+, o, ) markers, respectively.
here), it is helpful to assign a large amount of power to the source
in the rst time slot rather than distribute power equally between
the source and the relays. The overall performance is then mainly
determined by the direct transmission during the rst time slot. As a
result, the performance gain offered by the proposed PA over EPA can
be large. On the other hand, when the indirect links are getting better
(say
RD-SR
= 0 dB), the imbalance between the indirect and direct
channels qualities is decreasing. The proposed PAtends to converge to
EPA, and consequently, there is not much performance improvement
when using the proposed PA instead of EPA. A similar explanation
applies for the other cases that
SR
is 0 or 20 dB.
Similarly, for the two-relay system, i.e., K = 2, our results
5
also
show that the proposed PA can offer even more power gains over EPA
when compared with the single-relay systems, e.g., 3.5 and 4.5 dB
for m = 1 and m = 2, respectively. Such larger gains are reasonable
since with more relays in the system, there is more freedom to allocate
power between the relays and the source.
Next, simulation results are provided to conrm the power gains
predicted by the bound (presented in Fig. 4). For each specic value
of
SR
, various values of
RD-SR
are investigated. The particularly
important values are those corresponding to the boundary and middle
values, i.e.,
RD-SR
= 20, 0, and 20 dB. Therefore, the following
simulation results shall focus to validate the performance gains at these
values.
For the single- and two-relay systems, Figs. 5 and 6 compare the
performance of EPA and the proposed PA for the two values of m
when the RD channel is better than the SR channel (
RD-SR
=
20 dB). The OPA coefcients for the source and relays are presented
in Table I(a). For the single-relay system under Rayleigh fading, i.e.,
m = 1, and in the three cases in which
SR
= 20, 0, and 20 dB,
the proposed PA can offer gains of about 2, 1.6, and 1.2 dB, respec-
tively, over the EPA. These gains are in good agreement with the gains
predicted in Fig. 4 (based on the bound). For m = 2, the improvement
of the proposed PA over EPA is also very promising, i.e., about 2.5,
2, and 1.8 dB for
SR
= 20, 0, and 20 dB, respectively, similar to
that predicted in Fig. 4. For the two-relay system, the benets of the
proposed PA are illustrated in Fig. 6. As can be seen from Fig. 6, for
both values of m, the gains are at least 3 dB, as predicted.
Fig. 7 demonstrates the performance gains by the proposed PA
when
SR
= 20 dB and
RD-SR
= 20 dB. The optimal power
coefcients for the sources and relays are given in Table I(b). The
5
Due to space limitations, the gure is omitted.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010 1549
Fig. 6. Performance comparison of two relays with the proposed PA (dashed
line) and EPA (solid line):
RD-SR
= 20 dB. The cases of
SR
=
(20, 0, 20) dB correspond to the (+, o, ) markers, respectively.
TABLE I
PA COEFFICIENTS (
OPA
s
/
OPA
1
AND
OPA
s
/
OPA
1
/
OPA
2
FOR K = 1
AND 2, RESPECTIVELY). (a)
RD-SR
= 20 dB. (b)
RD-SR
= 20 dB
Fig. 7. Performance comparison of SEP performance between the (dashed
line) proposed PA and (solid line) EPA.
SR
= 20 dB, and
RD-SR
=
20 dB.
channel settings investigated in Fig. 7 represent the case that the
RD and SR fading channels are very poor when compared with
the direct link SD. However, one can still obtain signicant gains of
about 2.5, 2.8, 3, and 4.5 dB for (K = 1, m = 1), (K = 1, m = 2),
(K = 2, m = 1), and (K = 2, m = 2), respectively. These gains
also agree with that predicted based on the SEP bound (see Fig. 4).
It is worthwhile to point out that the cases of (m = 2, K = 1) and
(m = 2, K = 2) have almost the same performance with the proposed
PA in the SNR range shown in Fig. 7. It can be veried that at the
very high SNR region the two curves depart and have their decaying
exponents to be 4 and 8, respectively, as proven in Proposition 1.
Fig. 8. Performance comparison of SEP performance between varying- and
xed-gain relaying systems.
Finally, Fig. 8 compares the performance of the optimal PA
proposed by Zhao et al. [22] with our proposed PA. It should be
noted that the authors in [22] consider the varying-gain relay system
where the transmit power at the relays is maintained constant at all
time. The varying gain requires the relays to know the instantaneous
channel gains of all the sourcerelay links, whereas the xed gain only
requires the sourcerelay statistics available at the relays. Since the
OPA algorithm in [22] predetermines the power for S, i.e.,
s
E
s
, in the
rst time slot, the comparison only makes sense if the system has more
than one relay, and the relays are at different locations, i.e., K > 1,

SR
i
=
SR
j
, or
R
i
D
=
R
j
D
, where i = j, and i, j = 1, . . . , K.
For the two-relay system, consider the case where R
1
is closer to S
than D, whereas R
2
is closer to D than S. The case is assumed to
correspond to
SR
1
=
R
2
D
= 0 dB and
SR
2
=
R
1
D
= 20 dB.
The OPA coefcients, i.e.,
OPA
s
/
OPA
1
/
OPA
2
, are found to be
6
1.8/0.1305/1.0695 and 1.8/0.3/0.9 for the varying- and xed-gain
relay systems, respectively. Fig. 8 shows that, in this setting, the
varying-gain system performs better than the xed-gain system, as
expected. It is interesting to see that at BEP = 10
3
, the gains offered
by our algorithm and by Zhao et al. over the EPA are the same, i.e.,
about 0.5 dB. It should be noted that the performance gains depend
on the channel setting. When the channels from S to R
1
and R
2
to D
are worse, the performance of the xed-gain relaying approaches that
of the varying-gain relaying. Both algorithms tell us that less power
should be allocated to R
1
and more power to R
2
. For example, for the
case
7
of
SR
1
=
R
2
D
= 20 dB and
SR
2
=
R
1
D
= 20 dB, one
obtains
OPA
s
/
OPA
1
/
OPA
2
= 2.1/0.0135/0.8865 and 2.1/0.1/0.8 for
the varying- and xed-gain relaying, respectively. Both varying- and
xed-gain relaying performance curves are almost identical, and the
gain over the EPA is about 2 dB at BER = 10
3
.
V. CONCLUSION
In this paper, an upper bound on the SEP has been derived for the
multiple-relay systems with xed-gain AF relaying and MRC. It was
shown that the tightness of the bound is maintained over a wide range
6
Note that the performance of the OPA in [22] depends on the predetermined
power coefcient
s
at the source. However, the issue of optimizing
s
is not
addressed in [22]. As such, in comparison between the OPA in [22] and our
proposed PA, the same power allocation obtained from our proposed PA is also
applied for the case of OPA.
7
Again, due to space limitations, the gure for such a channel setting is
omitted.
1550 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010
of channel settings, even when the sourcerelay and relaydestination
channels are unbalanced. The bound also shows that MRC can achieve
the maximum diversity order of m(K + 1). Based on the obtained
bound, a novel PA scheme was proposed. Extensive simulation was
conducted, and the results are in good agreement with the gains
predicted based on the bound and shown to provide a signicant saving
of total transmitted power.
APPENDIX A
PROOF OF LEMMA 1
To calculate the upper bound on the average SEP P
s
, one needs
to evaluate the MGF M

(t) rst. Then, the resulting MGF can be

applied into (8). The MGF M

(t) is computed as
M

(t) =E

SD
{exp(t
s

SD
)}
K

k=1
E

SR
k
,
R
k
D

exp

tG
2
k

SR
k

R
k
D
/

+ G
2
k

R
k
D

=
K

k=1
(A
k
(t) + B
k
(t)) /[t
s

SD
/m + 1]
m
. (17)
The term A
k
(t) on the right-hand side of (17) is calculated as
A
k
(t) =

0
/G
2
k

0
exp

tG
2
k

SR
k

R
k
D

+ G
2
k

R
k
D

R
k
D
(
R
k
D
)p

SR
k
(
SR
k
) d
R
k
D
d
SR
k
(18)

0
/G
2
k

0
exp

tG
2
k

SR
k

R
k
D
/(2 )

R
k
D
(
R
k
D
)p

SR
k
(
SR
k
) d
R
k
D
d
SR
k
=

0
m
m
(m)
m
R
k
D

(m)

tG
2
k

SR
k
/(2 ) + m/(
R
k
D
)

m
exp

G
2
k

tG
2
k

SR
k
2
+
m

R
k
D

m1

l=0
(m) ( /G
2
k
)
l
(l + 1)

tG
2
k

SR
k
(2 )
+
m

R
k
D

ml

SR
k
(
SR
k
) d
SR
k
=A
k,1
(t) A
k,2
(t). (19)
After some manipulations on the right-hand side of (19), A
k
(t) can be
bounded as
A
k
(t) A
k,1
(t) A
k,2
(t) (20)
where A
k,1
(t) and A
k,2
(t) are specied in (11) and (12), respectively.
To arrive at (20), one can use [27, eqs. (3.351.1), (3.351.3), (3.352.4),
and (3.353.2)].
The term B
k
(t) in (17) is expressed as
B
k
(t) =

/G
2
k
exp

tG
2
k

SR
k

R
k
D

+ G
2
k

R
k
D

R
k
D
(
R
k
D
)p

SR
k
(
SR
k
)d
R
k
D
d
SR
k
(21)

/G
2
k
exp

t
s

SR
k
2

R
k
D
(
R
k
D
)p

SR
k
(
SR
k
) d
R
k
D
d
SR
k
=

SR
k
(t
s
/2 + m/
SR
k
)

m,
m
G
2
k

R
k
D

(m)
=B
k,1
(t) (22)
where (22) is computed with the aid of [27, eqs. (3.351.2) and
(3.351.3)].
Substituting (17), (20), and (22) into (8) completes the proof of
Lemma 1.
APPENDIX B
PROOF OF PROPOSITION 1
The Chernoff bound can be used to upper bound the probability P
s
by setting in (10) to /2 as follows:
P
s

M 1
M

sin
2
(/M)
s

SD
/m + 1

k=1
(A
k,1
(t) A
k,2
(t) + B
k,1
(t))

t=sin
2
(/M)

M 1
M

sin
2
(/M)
s

SD
/m + 1

k=1
(A
k,1
(t) + B
k,1
(t))

t=sin
2
(/M)
. (23)
The term A
k,1
(t) in (19) can be calculated and bounded as
A
k,1
(t) =
m
2m
(m)
m
SR
k

m
R
k
D

2
tG
2
k

m1
SR
k

SR
k
+
2m
tG
2
k

s

R
k
D

m
exp

m
SR
k

SR
k

d
SR
k
<
m
2m
(m)
m
SR
k

m
R
k
D

2
tG
2
k

0
exp

m
SR
k

SR
k

d
SR
k

SR
k
+
2m
tG
2
k

s

R
k
D

(24)
=
1
(m)

2m
2

tG
2
k

s

SR
k

R
k
D

m
exp

2m
2

tG
2
k

s

SR
k

R
k
D

E
1

2m
2

tG
2
k

s

SR
k

R
k
D

(25)
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010 1551
<
1
(m)

2m
2

tG
2
k

s

SR
k

R
k
D

tG
2
k

s

SR
k

R
k
D
2m
2
+ tG
2
k

s

SR
k

R
k
D
(26)
where the inequality (24) follows from the fact that

m1
SR
k

SR
k
+
2m
tG
2
k

s

R
k
D

m
<

SR
k
+
2m
tG
2
k

s

R
k
D

m1

SR
k
+
2m
tG
2
k

s

R
k
D

m
=1

SR
k
+
2m
tG
2
k

s

R
k
D

. (27)
The equality (25) is evaluated with the help of [27, eq. (3.532.4)],
and (26) comes from the following inequality: 1/(1 + x) <
exp(x)E
1
(x) 1/( + x), x > 0, 0 < < 1 [25].
Substituting B
k,1
(t) and the upper bound on A
k,1
(t) in (22) and
(26), respectively, to (23) results in (16) in Proposition 1.
For the lower bound, one also needs to evaluate the MGF M

(t)
and then apply the resulting MGF into (8). Using the result in (17), the
MGF M

(t) is calculated and lower bounded as

M

(t) =E

SD
{exp(t
s

SD
)}
K

k=1
E

SR
k
,
R
k
D

exp

tG
2
k

SR
k

R
k
D
/

+ G
2
k

R
k
D

=
K

k=1
[A
k
(t) + B
k
(t)]
[t
s

SD
/m + 1]
m

K

k=1
A
k
(t)
[t
s

SD
/m + 1]
m
(28)
where A
k
(t) and B
k
(t) are given in (18) and (21), respectively. The
inequality [see (28)] is due to the fact that B
k
(t) 0.
The term A
k
(t) is then lower bounded as
A
k
(t)

0
/G
2
k

0
exp

t
s

SR
k
/2

R
k
D

R
k
D

SR
k

SR
k

d
R
k
D
d
SR
k
=

0
exp

t
s

SR
k
/2

(m)

m,
m
G
2
k

R
k
D

m
m

m1
SR
k

m
SR
k
(m)
exp

m
SR
k

SR
k

d
SR
k
(29)
=
m
m

m, m /G
2
k

R
k
D

(m)
m
SR
k

t
s
2
+
m

SR
k

m
(30)
where (29) and (30) are computed with the help of [27, eqs. (3.531.1)
and (3.531.3)], respectively.
Substituting (28) and (30) into (8), one has the lower bound on the
average SEP for M-PSK as follows:
P
s

m
m(K+1)
[(m)]
K

m
SD
K

k=1

m,
m
G
2
k

R
k
D

m
SR
k

(M1)
M

t
s
+
m

SD

m
K

k=1
1

t
s
2
+
m

SR
k

t=
sin
2
(/M)
sin
2

d.
(31)
As clearly seen in (31), one can conclude that the lower bound on
P
s
is proportional to (1/ )
m(K+1)
in the high SNR region. Therefore,
the proof of the lower bound in (16) is completed.
REFERENCES
[1] A. Sendonaris, E. Erkip, and B. Aazhang, User cooperation
diversityPart 1: System description, IEEE Trans. Commun., vol. 51,
no. 11, pp. 19271938, Nov. 2003.
[2] J. N. Laneman and G. W. Wornell, Distributed spacetime-coded proto-
col for exploiting cooperative diversity, IEEE Trans. Inf. Theory, vol. 49,
no. 10, pp. 24152425, Oct. 2003.
[3] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, Cooperative diversity
in the wireless networks: Efcient protocols and outage behavior, IEEE
Trans. Inf. Theory, vol. 50, no. 12, pp. 30623080, Dec. 2004.
[4] A. Ribeiro, X. Cai, and G. Giannakis, Symbol error probabilities for
general cooperative links, IEEE Trans. Wireless Commun., vol. 4, no. 3,
pp. 12641273, May 2005.
[5] D. Chen and J. N. Laneman, Modulation and demodulation for coopera-
tive diversity in wireless systems, IEEE Trans. Wireless Commun., vol. 5,
no. 7, pp. 17851794, Jul. 2006.
[6] Y. Li and S. Kishore, Asymptotic analysis of amplify-and-forward relay-
ing in Nakagami fading environments, IEEE Trans. Wireless Commun.,
vol. 6, no. 12, pp. 42564262, Dec. 2007.
[7] W. Cho, R. Cao, and L. Yang, Optimum resource allocation for amplify-
and-forward relay networks with differential modulation, IEEE Trans.
Signal Process., vol. 56, no. 11, pp. 56805691, Nov. 2008.
[8] V. K. Bhargava and K. B. Letaief, Special issue on cooperative commu-
nications, IEEE Trans. Wireless Commun., vol. 7, no. 5, pp. 17771778,
May 2008.
[9] D. S. Michalopoulos, G. K. Karagiannidis, and G. S. Tombras, Sym-
bol error probability of decode and forward cooperative diversity in
Nakagami- m fading channels, J. Franklin Inst., vol. 345, no. 7, pp. 723
728, Oct. 2008.
[10] W. Cho and L. Yang, Optimum resource allocation for relay networks
with differential modulation, IEEE Trans. Commun., vol. 56, no. 4,
pp. 531534, Apr. 2008.
[11] T. Wang, G. Giannakis, and R. Wang, Smart regenerative relays for link-
adaptive cooperative communications, IEEE Trans. Commun., vol. 56,
no. 11, pp. 19501960, Nov. 2008.
[12] N. Vien, H. Nguyen, and T. Le-Ngoc, Diversity analysis of smart relay-
ing, IEEE Trans. Veh. Technol., vol. 58, no. 6, pp. 28492862, Jul. 2009.
[13] R. U. Nabar, H. Bolcskei, and F. W. Kneubhler, Fading relay channels:
Performance limits and spacetime signal design, IEEE J. Sel. Areas
Commun., vol. 22, no. 6, pp. 10991109, Aug. 2004.
[14] H. Mheidat and M. Uysal, Impact of receive diversity on the performance
of amplify-and-forward relaying under APS and IPS power constraints,
IEEE Commun. Lett., vol. 10, no. 6, pp. 468470, Jun. 2006.
[15] H. Shin and J. Song, MRC analysis of cooperative diversity with xed-
gain relays in Nakagami-m fading channels, IEEE Trans. Wireless
Commun., vol. 7, no. 6, pp. 20692074, Jun. 2008.
[16] G. Karagiannidis, Performance bounds of multihop wireless communi-
cations with blind relays over generalized fading channels, IEEE Trans.
Wireless Commun., vol. 5, no. 3, pp. 498503, Mar. 2006.
[17] A. Avestimehr and D. Tse, Outage capacity of the fading relay channel in
the low-SNR regime, IEEE Trans. Inf. Theory, vol. 53, no. 4, pp. 1401
1415, Apr. 2007.
[18] A. Host-Madsen and J. Zhang, Capacity bounds and power allocation
for wireless relay channels, IEEE Trans. Inf. Theory, vol. 51, no. 6,
pp. 20202040, Jun. 2005.
1552 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 59, NO. 3, MARCH 2010
[19] Y. Jing and B. Hassibi, Distributed spacetime coding in wireless relay
networks, IEEE Trans. Wireless Commun., vol. 5, no. 12, pp. 35243536,
Dec. 2006.
[20] Y. Yao, X. Cai, and G. Giannakis, On energy efciency and optimum
resource allocation of relay transmissions in the low-power regime, IEEE
Trans. Wireless Commun., vol. 4, no. 6, pp. 29172927, Nov. 2005.
[21] M. Fareed and M. Uysal, BER-optimized power allocation for fading
relay channels, IEEE Trans. Wireless Commun., vol. 7, no. 6, pp. 2350
2359, Jun. 2008.
[22] Y. Zhao, R. Adve, and T. Lim, Improving amplify-and-forward relay net-
works: Optimal power allocation versus selection, IEEE Trans. Wireless
Commun., vol. 6, no. 6, pp. 31143123, Aug. 2007.
[23] M. K. Simon and M.-S. Alouini, Digital Communication Over Fading
Channels, 1st ed. New York: Wiley-IEEE Press, 2000.
[24] J. G. Proakis, Digital Communications, 4th ed. New York: McGraw-
Hill, 2001.
[25] T. Wang, A. Cano, and G. B. Giannakis, Link-adaptive cooperative
communications without channel state information, in Proc. IEEE Mil.
Commun. Conf., Oct. 2006, pp. 17.
[26] S. Boyd and L. Vandenberghe, Convex Optimization. Cambridge, U.K.:
Cambridge Univ. Press, Mar. 2004.
[27] L. S. Gradshteyn and L. M. Ryzhik, Tables of Integrals, Series and
Products, 7th ed. New York: Academic, 2007.
Nakagami-mPhaseEnvelope Joint Distribution:
A New Model
Michel Daoud Yacoub, Member, IEEE
AbstractThe aim of this paper is to propose an improvement on the
Nakagami-mcomplex signal model presented elsewhere in the literature.
By introducing what here is named a phase parameter, a more general and
realistic framework is obtained. However, with an additional parameter,
the formulations are still found in simple and closed-form expressions.
With another physically based consideration, which, incidentally, further
simplies the formulations, the phase parameter is written as a function
of the fading parameter, and the expressions are found in an even more
compact manner. With the proposed fading model, higher order statistics
are then derived, particularly those related to the phase crossing rate.
Index TermsHigher order statistics, Nakagami-mdistribution, phase
crossing rate (PCR), phaseenvelope distribution, phase statistics.
I. INTRODUCTION
The derivation of the envelope probability density function (pdf)
of the basic fading models, namely, Rayleigh [1], Hoyt [2], and
Rice [3], produced, as an intermediate step, the corresponding joint
phaseenvelope pdf. In contrast to this, for the envelope-based models,
such as, for instance, Nakagami-m [4], - [5], and - [5], no
information on the phase distribution was provided, since the aim
was purely the signal envelope. Thus, for these cases, the phase
characterization is an open matter and, therefore, subject to debate.
In an attempt to ll such a gap, models for the complex signals in
the Nakagami-m [6], - [7], and - [8] environments have been
proposed, and the corresponding phase and envelope pdfs, both jointly
and marginally, have been obtained.
Manuscript received July 11, 2009; revised September 10, 2009,
November 19, 2009, and January 7, 2010. First published January 15, 2010;
current version published March 19, 2010. The review of this paper was
coordinated by Dr. K. T. Wong.
The author is with the Wireless Technology Laboratory, Department of Com-
munications, School of Electrical and Computer Engineering, State University
of Campinas, 13083-852 Campinas-SP, Brazil (e-mail: michel@wisstek.org).
Digital Object Identier 10.1109/TVT.2010.2040641
The envelope and phase distributions are useful in the study of a
variety of topics in communications systems, such as radar clutter
modeling, signal detection, and others [9][15]. More specically,
the phase statistics have found applications, for instance, in the es-
timation of probabilities of error for M-phase signaling over fading
channels using diversity [16] and in the detection of M-ary phase-
shift keying signal constellations [17]. In some situations, the joint
phaseenvelope distribution may be used to facilitate the determination
of high-order statistics in multibranch diversity scenarios [18]. In
coherent communication systems, error performance is drastically
inuenced by the carrier phase estimation. In such a case, rather than
laying hold of the exact distribution of the phase, the distribution
of the difference between the true phase and its estimation may be
used. This approach has been applied in [19][21] for Ricean and
Nakagami-m applications, in which cases, the phase-difference pdfs
were arbitrarily considered to be Gaussian or Von Mises (Tikhonov
[22]) distributed. A phase-difference distribution for the Nakagami-m
case was derived in [23], where the phases of the Nakagami-m signals
were also arbitrarily assumed to be uniform. Recently, a simulator
for the phase pdf, as derived in [6], has been proposed in [24] that
yielded accurate autocorrelated properties. The exact Ricean phase
pdf was used in [25], in which the optimum scheme that maximizes
the ergodic capacity of a multiple-inputsingle-output system was
investigated. The phase distribution for fading signals is indeed a very
interesting and challenging topic and has long been attracting the
attention of several researchers. (See [26], [27], and references therein
for a discussion on the appropriateness of the use of a uniform phase
distribution in the sum of random vectors.)
As is widely known, Nakagami-m is an empirical model that
is proposed as an approximation to an extremely intricate problem
concerning the statistics of a signal, which is the resultant of the super-
position of complex random vectors whose distributions are unknown.
In [4] and [28][30], such a superposition has been interpreted as a
random walk process in the complex plane. Departing from this most
general mathematical condition and then carrying out approximations
sufciently good enough for engineering problems [4, p. 15], the
Nakagami-m envelope pdf was derived. The phase pdf of the sum
of the vectors in the complex plane was then obtained in [29], [31],
and [32]. It turns out that the resulting pdf is rather intricate, being
given in terms of the statistics of the quadrature components, namely,
mean values, variances, and correlation coefcient. The simplications
in [4] were a mathematical manipulation directly carried out in the
derivation process of the envelope pdf, i.e., it had no connection with
the physical model of the complex signal. Therefore, no information
on the corresponding resultant phase can be extracted from this
process. Departing from the general phase pdf reported in [29], [31],
and [32], and trying to simplify it in an attempt to follow a path
similar to that of the envelope case, a substantial number of simplied
scenarios are possible (e.g., [29]). All these scenarios would involve
considerations on the pertinence or relevance of the parameters, in
terms of which the phase pdf is described. It is noted, however, that
if those envisaged scenarios are then mapped back onto the denition
of the complex signal, then attaining the exact Nakagami-m envelope
pdf from the modied model is certainly compromised. Of course,
one may always propose the trivial solution, namely, assuming 1) the
envelope to be Nakagami-m distributed and 2) an independent phase
distribution of any given choice, from which in-phase and quadrature
components can be found. An innite number of possible solutions
can be found in this case. In [6], a novel approach for the Nakagami-
m complex signal was proposed. It is based on the widely used
fading model that describes the Nakagami-m envelope in terms of
0018-9545/$26.00 2010 IEEE