amount of fading revisited
TRANSCRIPT
-
7/29/2019 Amount of Fading Revisited
1/23
1
Amount of Fading Revisited
Yan Li and Shalinee Kishore
Abstract
The fading statistics in various diversity systems are uniquely determined by the distribution
of the signal-to-noise ratio (SNR) of combined signals. In this study, the amount of fading
(AF), introduced in [1] as a unified measure of the severity of fading, is revisited and utilized to
parameterize the SNR distribution. Specifically, two approaches are proposed, approximating the
SNR as Gaussian and Gamma distributions. The approximated SNR distributions are then used
to evaluate various conventional performance measures: Shannon capacity, outage probability
and average error rates. The accuracy of these approximations based on the AF implies that
systems with the same AF have roughly the same performance, given the receiver is optimally
designed. Therefore a complicated system can be equivalent to a much simpler diversity system.
This equivalence can be used to simplify certain theoretical analysis. As illustrations, applications
to receive antenna diversity and multiuser diversity are investigated; the results confirm this claim
and establish the broader utility of the AF.
I. Introduction
The random, time-varying channel characteristics between a transmitter and a receiver
pose many challenges to reliable wireless communications. A key challenge lies in variable
fading of the received signal power which can lead to variable performance over the commu-
nication link. Diversity methods attempt to counter this disparity; in particular, they focus
on reducing the fluctuations of received signal power to overcome the unreliability caused by
poor fading conditions. In order to evaluate the performance of various diversity schemes,
different fading measures have been introduced and investigated.
The authors are with the Department of Electrical and Computer Engineering, Lehigh University, Bethlehem, PA18015, USA. (Email: {liyan,skishore}@lehigh.edu). This research was supported by the National Science Foundationunder CAREER Grant CCF-03-46945.
-
7/29/2019 Amount of Fading Revisited
2/23
2
Average signal-to-noise ratio (SNR) at the diversity combiner output is a useful and
perhaps the simplest performance measure. However, this performance criterion does not
adequately capture diversity benefits as it ignores the variation of the received SNR. To
account for its variance, we need to look at performance measures which take into account
higher moments of the SNR.
In 1979, Charash introduced a measure of the severity of the fading channel [1], referred
to as the amount of fading (AF), which can be computed using only the first and second
central moments of the SNR at the diversity combiner output. This measure was gener-
alized by Simon and Alouini when describing the behavior of diversity combining systems
over correlated log-normal fading channels [2]. Another measure, diversity factor (DF) wasintroduced by Awoniyi, Mehta and Greenstein [3] to study the impact of dispersive channels
on the throughput of code division multiple access (CDMA) schemes. Calculable for any
channel delay profile, it captures the effective number of equal mean gain paths offered
by a delay profile to a RAKE receiver. In [5] and [6], this definition was used to quantify
the uplink and downlink user capacities of a hierarchical cellular CDMA system in various
multipath environments. We note here that the DF is simply the reciprocal of the AF. The
normalized standard deviation, defined by Win and Winters in [4], is another quantity with
a definition similar to that of the AF.
Although the AF is widely used as an alternative performance criterion (e.g. [7] [11]),
there is, to the best of our knowledge, no answer to the following questions: Does the AF,
which depends only on the first two moments of the combined output SNR, accurately
reflect the behavior of the diversity combining techniques and capture the statistics of the
fading channels? If the answer is yes, then what is the relationship between the AF andother conventional performance measures, such as Shannon capacity, outage probability,
etc.? We try to give some answers to these questions in the paper. Specifically, we answer
these questions by proposing approximations to the distribution of the output SNR in terms
of AF which are valid in a variety of scenarios. With these approximations at hand, we then
-
7/29/2019 Amount of Fading Revisited
3/23
3
compute other conventional performance measures as functions of AF.
The rest of the paper is outlined as follows: Section II describes the system model and
provides the definition of AF. Section III contains the theoretical results of approximations
to the distribution of the output SNR. Based on these results, we study the relationships
between AF and other performance criteria in Section IV. As illustrations to our theoretical
results, some applications are investigated and the numerical results are given in Section V.
We summarize the contributions of this paper in Section VI.
II. System Model
In this study, we use the following standard notations: superscripts T for vector transpose,
for conjugate and for transpose conjugate; E[] for expectation and V ar[] for variance.The complement of a set S will be denoted by S.
We consider a simplified fading channel model with L-branch diversity
Yl = alX+ Nl, l = 1, 2,...,L, (1)
where Yl is the processed received signal of the l-th sub-channel or diversity branch (out of
L); X is the scalar transmitted signal with power constraint
E[|X|2] = P; (2)
{al}Ll=1 are the mutually independent fading coefficients; and {Nl}Ll=1 are the independent,identically distributed (i.i.d.) complex Gaussian noise (plus interference in multiuser environ-
ments under the standard Gaussian assumption) components with zero-mean and variance
N0/2 per dimension. In addition, we assume the channel state information (CSI) is known
at the receiver thus enabling coherent detection.
For simplicity, we can rewrite (1) in the vector form
Y = AX+N, (3)
-
7/29/2019 Amount of Fading Revisited
4/23
4
where Y= [Y1, Y2,...,YL]T, A = [a1, a2,...,aL]
T and N= [N1, N2,...,NL]T.
Among different types of diversity combiners, e.g., generalized selection combiner (GSC),
equal-gain combiner (EGC), etc., the optimal one is the maximal ratio combiner (MRC).
The output Z of the MRC can be written as
Z = AY
=Ll=1
al Yl. (4)
In this paper, we focus primarily on this type of an MRC system.
A. Definition of the AF
Let denote the total instantaneous SNR at the combiner output,
=Ll=1
|al|2PN0
=Ll=1
l, (5)
where l =
|al|2P
N0 is the instantaneous SNR on the l-th sub-channel. The AF is defined as
AF V ar[]
E2[]
=
Ll=1 V ar[|al|2]Ll=1 E[|al|2]2 . (6)
Note that AF 1 and that it is the reciprocal of the diversity factor defined in [3].
III. Approximations to the Distribution of the Output SNR
The distribution of the MRC output SNR depends on the fading environment. As an
example, let us consider the simplest case in which all L subchannels experience Rayleigh
fading. The distribution of is given in [14, p.847] and depends on whether the L branches
experience i.i.d. fading or have different means. Let E[] = and let k denote the mean
-
7/29/2019 Amount of Fading Revisited
5/23
5
SNR of branch k.
1) When {k}Lk=1 are all the same, we can denote c = k = /L. Then, we have theprobability density function (pdf) of as
f() =1
(L)LcL1e/c
G(; L, c), (7)
where () is a complete gamma function defined as
(a)
0
ta1etdt. (8)
The pdf in (7) describes the density of a Gamma random variable with the shape parameter
L and the scale parameter c. It is straightforward to show that E[] = and V ar[] = 2/L,
implying that AF = 1/L for this system.
2) When {k}Lk=1 are distinct, the pdf of is
f() =
Lk=1
kk
e/k , (9)
where
k =L
i=1,i=k
kk i . (10)
Under such fading conditions, it can be shown that AF > 1/L (see Appendix I for proof).
In comparing case 1 and 2 above, we find that in Rayleigh channels, uniform i.i.d. fading
over the L branches leads to lower AF than the non-uniform case. Therefore, for the same ,
uniform fading has lower variation about this mean SNR, implying a smaller probability of
deep fades and a smaller probability of high instantaneous SNRs than non-uniform fading.
In many diversity schemes (e.g., RAKE reception), smaller values of AF are desirable for
a fixed as they yield a fewer instances of deep fades. On the other hand, in multiuser
diversity systems with some M users, larger AF values are desirable for a fixed value of
-
7/29/2019 Amount of Fading Revisited
6/23
6
at the receiver. The reason is because multiuser diversity capacity grows with max, the
instantaneous value of the maximum received SNR among the M users. Specifically, when the
value of AF is large so is the probability that max is high, leading to a higher overall capacity
for the system. We see therefore that the parameter AF is in fact related to the performance
of a diversity scheme. It is the goal of this paper to give more explicit descriptions of the
relationship between AF and conventional performance metrics (like capacity).
To do so, we begin by developing approximations to the output SNR distributions as
functions of AF. It is difficult to solely use AF to characterize the exact distribution of the
output SNR since it is comprised only of the first and second moments of the SNR. However,
we argue here that the AF is accurate enough to reflect the statistics of the fading channelunder various conditions. For example, when L is large, the central limit theorem (CLT)
applies and we have:
Proposition 1 (Gaussian Approximation (GA)): The distribution of the MRC combiner
output SNR tends to a Gaussian with mean and variance AF 2, s.t.
f() 1
2 AF exp
( )2
2AF 2
N(; , AF 2) (11)
given the Lindeberg condition [15, p.256] is satisfied. The condition requires that the indi-
vidual variances {V ar[k]}Lk=1 are small as compared to their sum AF 2, i.e., that for any > 0
V ar[k]
AF 2 < , k = 1,...,L, (12)
for L sufficiently large.
For other diversity combining techniques (for instance, GSC), various CLTs involving the
asymptotic normality ofL-statistics or trimmed sum of order statistics are needed; they are
usually complicated and beyond the scope of our discussion here.
-
7/29/2019 Amount of Fading Revisited
7/23
7
As implied by the Lindeberg condition, GA works well for nearly equal-power and diversity-
rich environments. However, in many fading environments the powers on different sub-
channels may vary dramatically. Furthermore, the supported levels of diversity may be
restricted, for instance, due to the affordable receiver complexity. Under such conditions when
the CLT does not hold, the Gamma distribution is found to be a better approximation to the
output SNR for Nakagami-m fading channels than the Gaussian distribution. Nakagami-m
distribution, named and proposed by Nakagami (1943) from his large-scale experiments on
rapid fading in long-distance propagation, has a pdf
f(r) =2mmr2m1
(m)m
e(m/)r2
M(r; m, ), (13)
where r is the received signal amplitude, = E[r2], and m is called the fading figuredefined
as
m =2
E[(r2 2)2] 1
2, always. (14)
To come up with the Gamma approximation, we use the following lemma proposed by
Nakagami [13].
Lemma 1 (Sum of Squares of m-Variables): Let {ri}ni=1 be a sequence of independent Nak-agami m-variables distributed according to M(ri, mi, i) and define
R2 =Li=1
r2i . (15)
Then, we have approximately
f(R) M
R; oM,ni=1
i
, (16)
where
oM =
ni=1
i
2/
ni=1
imi
. (17)
-
7/29/2019 Amount of Fading Revisited
8/23
8
For more details on Lemma 1, readers are referred to [13, p.20].
Now consider a diversity system where all sub-channels experience Nakagami-m fading so
that k Mk, mk, k. Since =Lk=1 k, we have via Lemma 1,
f(
) M
; oM,
Lk=1
k
= M (
; oM, ) , (18)
where
oM = L
k=1k2
/
Lk=1
kmk . (19)
From (14) we have
mk =k
E[(k k)2]=
kV ar[k]
. (20)
By substituting (20) into (19) we have
oM =
Lk=1
k
2/
Lk=1
V ar[k]
= 2/V ar[]
= 1/AF. (21)
Finally, we obtain the approximate pdf of as
g() = f()2
=1
(AF1)
AF 1/AF
exp
AF
= G(; 1/AF, AF ). (22)
-
7/29/2019 Amount of Fading Revisited
9/23
9
G(; 1/AF, AF ) is a Gamma distribution with the shape parameter 1/AF and the scaleparameter AF . Recall that the output SNR follows the Gamma distribution with theshape parameter L and the scale parameter /L when all the L sub-channels experience
i.i.d. Rayleigh fading. Subsequently, we have the following proposition:
Proposition 2 (Gamma Approximation (GaA)): A MRC diversity combining system over
Nakagami-m fading channel with the amount of fading AF can be approximated by a
Rayleigh fading channel with DF i.i.d. paths, where DF = 1/AF. In other words, the
distribution of the combiner output SNR can be approximated by a Gamma distribution
with the shape parameter DF and the scale parameter /DF.
IV. Interrelations between the AF and Other Performance Criteria
In this section, we study the relationships between the AF and other performance measures
by applying our propositions.
Given the channel realization A is known at the receiver and the channel input X is
a circularly symmetric complex Gaussian with zero-mean and variance P, the maximum
mutual information between the input X and the combiner output Z can be achieved as
Imax = maxE[|X|2]P
I(X; Z)
= log(1 + ). (23)
Two important information theoretical measures are related to Imax: Shannon capacity and
outage probability.
A. Shannon Capacity
When the channel is ergodic, the Shannon capacity is simply the expectation of Imax.
From Shannons capacity function we have
C = E[log(1 + )] . (24)
-
7/29/2019 Amount of Fading Revisited
10/23
10
The first order approximation
C1 log(1 + ), (25)
is the simplest approximation. It can, for example, be used as the capacity upper bound,
which directly comes from Jensens inequality
E[log(1 + )] log (1 + ) . (26)
The equality in (26) holds if and only if = with probability of 1 and corresponds to a
non-fading AWGN environment.
Given that the tail probability of is small enough, we can obtain a reliable second order
approximation:
C2 log(1 + ) AF2
1 +
2. (27)
In Appendix II the derivation of (27) is given, which shows the second order approximation
is applicable for any combining scheme and accurate enough in most practical fading cases
in which AF 1. We verify this via numerical results in Section V.Alternatively, if we approximate as a Gaussian, we have
CGA =
R+
log(1 + )N(; , AF 2)d. (28)
If we approximate as a Gamma, we have
CGaA = R+log(1 + )G(; AF1, AF )d. (29)
All the capacity approximations (27)(29) yield functions of AF.
B. Outage Probability
When the fading is slow varying and can be considered as fixed for all uses of the channel,
the maximum mutual information is not equal to the channel capacity anymore. In this
-
7/29/2019 Amount of Fading Revisited
11/23
11
case the Shannon capacity in the strict sense is zero: no matter how small the rate is,
there is a non-zero probability that the channel is incapable of supporting it even for very
long code length (outage event). There is a tradeoff between the supportable data rate R
(bits/channel use) and the outage probability Pout(R). Based on GA, the outage probability
can be approximated as
Pout(R) = Pr(log2(1 + ) R)
= Q
+ 1 2R
AF
, (30)
where Q() is defined as
Q(x)
x
12
et2/2dt. (31)
Based on GaA, the outage probability can be approximated as
Pout(R) =
2R10
G(, AF1, AF , )d
=
1
AF,
2R 1AF
, (32)
where (a, x) is a incomplete Gamma function defined as
(a, x) 1
(a)
x0
ta1etdt. (33)
C. Average Error Probability
Average error probability is another widely used performance measure. Given the channel
realization, the conditional error probability can be typically formulated as
Ps() = MQ
M
, (34)
where M and M depend on the type of the modulation.
-
7/29/2019 Amount of Fading Revisited
12/23
12
The average error probability is simply
Ps = E[Ps()]
= E
MQ
M
. (35)
By applying an alternate representation of Q function obtained by Craig [16],
Q(x) =1
/20
exp
x22sin2
d, (36)
we have
Ps =M
/20
E
exp M2sin2 d
=M
/20
M
M2sin2
d, (37)
where M(s) E[es] is the moment generating function (MGF) [12] of .
Based on GA, M(s) can be approximated as
M(s) = exps +AF 2s2
2 , (38)and hence we have
Ps =M
/20
exp
M2sin2
+AF 2M2
8sin4
d. (39)
Based on GaA we have
M(s) = 1/ (1 AF s)1/AF , (40)
and
Ps =M
/20
1 1
1 + 2sin2
AFM
1/AFd. (41)
Thus, we can evaluate the average error probability by a single integral with finite limits.
-
7/29/2019 Amount of Fading Revisited
13/23
13
It is worth discussing here the limiting behavior of our approximations to the tail error
probabilities. First, our approximations asymptotically converge to the true values as the
average SNR increases, since both the approximations and the true values tend to be
zero. However, this asymptotic equality does not hold in the exponent. For example, as a
well-known result, the outage or error probability for Rayleigh fading is asymptotically equal
to 1/L. We can easily show that the approximation based on GaA is asymptotically equal
to 1/DF = 1/1/AF 1/L. For this reason, we cannot expect to have tight approximationfor high SNRs in the logarithm domain.
V. Applications and Numerical Results
A. Spatially Correlated Antenna Diversity
For simplicity, we consider a single-input multiple-output (SIMO) system equipped with
one transmit antenna and L receive antennas. Extensive discussions on the AF in multiple-
input multiple-output (MIMO) systems can be found in [11]. Following conventional nota-
tions, we denote the channel coefficients as H= [h1, h2, ...hL]T. When the receive antennas
are spatially correlated, H can be represented as a product form
H = R1/2Hw, (42)
where Hw is a circularly symmetric complex Gaussian vector with i.i.d. zero-mean and unit
variance entries, andR is the covariance matrix representing the receive antenna correlations.
In particular, we assume that uniform linear arrays (ULA) are employed at the receiver end.
Thus, for a rich scattering environment, the receive covariance matrix R has the following
Toeplitz structure
R =
1 R (2) (L1)
1 (L2)
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
L1 L2 L3 1
.
-
7/29/2019 Amount of Fading Revisited
14/23
14
The AF can be obtained by
AF =V ar [|H|22]E2 [|H|22]
, (43)
where | |2 denotes 2-norm. Given the AF, we can then apply our theoretical results.Fig. 1 shows the simulated Shannon capacity of the SIMO channels as (a) the function
of the number of antennas for various values of SNR per antenna and a fixed correlation
coefficient = 0.5; (b) the function of the correlation coefficient for SNR per antenna = 5
dB and various values ofL. In addition, we plot in both cases the capacity estimated by the
GA, GaA and second order approximation respectively. The curves match the simulation
results very well, except for the gaps between the simulation and the GA approximations
when is large or L is small. On the other hand, the GaA works well even if L is small, as
long as the sub-channel fading characteristics can be described by Nakagami-m distribution.
This advantage is more clearly demonstrated in Fig. 2, in which we plot the simulated and
approximated outage probability as the function of the SNR per antenna for L = 2 and 8,
respectively. When L = 2, the GA fails, but the GaA still matches the simulation results.
In Fig. 1 we also note that the second order approximation to the Shannon capacity tightly
follows the GaA results.
B. Multiuser Diversity Scheduling Gain
The basic idea of multiuser diversity (MD) scheduling is to assign radio resources (power,
bandwidth, antenna, etc.) to the sole user with the highest Imax, i.e., the user that can best
utilize these resources to achieve the highest individual throughput.
Our baseline reference is a Round-Robin (RR) scheduler. Given K users, the maximum
average data rate that can be achieved is
DRR = E
Kk=1 I
(i)max
K
=
Kk=1 C
(i)
K, (44)
-
7/29/2019 Amount of Fading Revisited
15/23
15
where the superscript (i) denotes the i-th user.
With an MD scheduler, the maximum average data rate that can be achieved is
DMD = E
max(I(i)
max : i = 1,...,K)
. (45)
Since I(i)max = log(1 + (i)) is a strictly increasing function of (i), we have
max(I(i)max) = log
1 + max((i))
. (46)
Let
max
max({i} : i = 1,...,K). (47)
Thus, we have
DMD = E[log(1 + max)] . (48)
By applying our propositions, the approximations to the pdf fi and the cumulative dis-
tribution function (cdf) Fi of (i) are obtained. Then, the pdf of max can be expressed
as
f(max) =Ki=1
wi(max)fi(max), (49)
where
wi(max) =K
j=1,j=i
Fj(max). (50)
Knowing the pdf ofmax, we can substitute it into (48) to approximate DMD. We then define
the multiuser diversity gain as
GMD =DMDDRR
. (51)
The accuracy of our proposed approximations to the multiuser diversity gain can be seen
-
7/29/2019 Amount of Fading Revisited
16/23
16
in Fig. 3. To simplify simulation, we have chosen all users have the same multipath Rayleigh
fading channels with the classic exponentially decaying multipath intensity profile (MIP) of
the form
i =eiLl=1 e
l , (52)
where is a small positive constant that determines the steepness of the power decay.
Fig. 3 shows the multiuser diversity gain as a function of the number of total users for (a)
various values of and a fixed number of resolvable paths, L = 10; (b) various values of L
and fixed = 0.2. We again see the approximations based on GaA matches the simulation
curves very well. The approximation curves based on GA diverge when the is large, e.g.,
= 1.2, or L is small, e.g., L = 2. The reason for this divergence is that the Lindeberg
condition is not satisfied for large or small L.
VI. Conclusions
We used the AF to parameterize two types of approximate distributions of the com-
bined signal SNR for the optimal diversity combining receiver. Specifically, the GA works
for nearly equal-power and diversity-rich environments. A typical working scenario is the
asymptotic analysis of the multiple antenna system. The GaA requires that all sub-channels
experience Nakagami-m fading. Since the Nakagami-m distribution is a generalization of
many frequently-used fading models (such as Rayleigh fading, Ricean fading and so on),
the GaA is valid and applicable for a wide-ranging cases. Given these approximated SNR
distributions, we then evaluated various conventional performance measures. For the cal-
culation of Shannon capacity, we proposed an additional approach which computes the
second order approximation of C as a function of AF. Our approximations based on the
AF imply that systems with the same AF have roughly the same performance, assuming
the receiver is optimally designed. Thus, certain theoretical analysis of complicated systems
can be simplified by examining equivalent, simpler diversity systems. Although most of our
results have been derived for MRC receivers, we note that the reliability of the second
-
7/29/2019 Amount of Fading Revisited
17/23
17
order approximation to capacity implies that receiver optimality assumption can be relaxed
in some circumstances. As illustrations, applications to the receive antenna diversity and
multiuser diversity were investigated. The results further confirm our claim and establish
the broader utility of the AF.
Appendix I
For the Rayleigh fading channel with L-branch diversity, the amount of fading is given as
AF 1/L with equality if and only if the fades on all diversity branches are independent,identically distributed.
Proof: From Schwartzs inequality we have
(T 1)2 ||22 |1|22, (53)
where 1 is a L 1 vector with each element equal to 1 and = [1, 2,..., k]T. The equalityholds if and only if = 1 with a constant. With this observation we can bound AF as
AF =
Ll=1 l
2
Ll=1 l2
=||22
(T 1)2
1|1|22= 1/L. (54)
Thus, AF = 1/L when the L fading paths have equal mean powers and AF > 1/L, otherwise.
-
7/29/2019 Amount of Fading Revisited
18/23
18
Appendix II
Derivation of the second order approximation to the Shannon capacity
Let us define
E1 C C1
=
R+
log
1 +
1 +
f()d (55)
and
E2
t
log
1 +
1 +
f()d, (56)
where
t =
R+| 1 + 2 . (57)Then, it is easy to show
| |1 +
1 if t< 1 otherwise.
Thus, we have
E1 E2 t
1 +
12
1 +
2f()d (58)
and
E2 >
t
1 +
12
1 +
2f()d
E3, (59)
where the approximation and the inequality follow the facts that log(1 + x) x 12
x2 for
|x| < 1 and log(1 + x) > x 12x2 for x 1.
-
7/29/2019 Amount of Fading Revisited
19/23
19
Given that the tail probability of is small such that |E2| C and |E3| C, we have
C = C1 + (E1 E2) + E2
C1 + (E1 E2) + E3
C1 +R+
1 +
12
1 +
2f()d
= log(1 + ) AF2
1 +
2 C2. (60)
Hence, we obtain the second order approximation.
Intuitively, the tail probability Pr(t) is small if the AF is small since the standard
deviation of is
AF. However, even when AF = 1, which corresponds the Rayleigh
flat fading case, we observe that |E2| C and |E3| C.For Rayleigh fading, is exponentially distributed with the pdf
f() =1
exp
. (61)
Thus, we have
C = exp
1
Ei
1
, (62)
E2 = exp
1 + 2
log2 exp
1
Ei
2(1 + )
, (63)
and
E3 =
12
1 + 2
exp
1 + 2
, (64)
where Ei() is the exponential integral function defined as
Ei(x) x
etdt
t. (65)
-
7/29/2019 Amount of Fading Revisited
20/23
20
Both |E2|/C and |E3|/C are bounded functions of : E2/C = E3/C = 0 when = 0 or = +, max(|E2|/C) 0.085 when is around 2.92 dB and max(|E3|/C) 0.063 when is around 1.34 dB.
Thus, in most practical fading cases in which AF < 1, the condition that |E2| C and|E3| C is satisfied. This condition ensures the accuracy of the second order approximation.In addition, the derivation does not make use of the MRC assumption, which implies that
the second order approximation is applicable for any diversity scheme.
References
[1] U. Charash, Reception through Nakagami fading multipath channels with random delays, IEEE Trans.
Commun., vol. 27, pp. 657670, Apr. 1979.
[2] M. Alouini and M. Simon, Dual diversity over log-normal fading channels, Proc. IEEE Int. Conf. Commun.ICC01, Helsinki, Finland, June 2001.
[3] O. Awoniyi, N. B. Mehta and L. J. Greenstein, Characterizing the orthogonality factor in WCDMA downlinks,
IEEE Trans. Wireless Commun., vol. 2, pp. 621625, July 2003.
[4] M. Win and J. Winters, Analysis of hybrid selection/maximal-ratio combining in Rayleigh fading,IEEE Trans.
Commun., vol. 47, pp. 17731776, Dec. 1999.
[5] S. Kishore, et al, Uplink user capacity in a CDMA system with hotspot microcells: effects of finite transmit
power and dispersion, To appear IEEE Trans. Wireless Commun., 2005.
[6] S. Kishore, et al, Downlink user capacity in a CDMA macrocell with a hotspot microcell, Proc. of Global
Telecommunications Conf. GLOBECOM03,, vol. 3, pp. 15, Dec. 2003.
[7] C. Chen and L. Wang,A unified capacity analysis for wireless systems with joint antenna and multiuser diversity
in Nakagami fading channels, Proc. IEEE Int. Conf. Commun. ICC04, vol. 27, no 1, pp. 35233527, June 2004.
[8] K. Yao, M. Simon and E. Biglieri, A unified theory on the modeling of wireless communication fading channel
statistics, Proc. Conf. Info. Sci. Sys. CISS04, Princeton, NJ, Mar. 2004.
[9] N.C. Sagias, et al, Performance analysis of switched diversity receivers in Weibull fading, Electronics Letters,
vol. 39, pp. 14721474, Oct. 2003.
[10] N.C. Sagias, et al, Performance analysis of dual selection diversity in correlated Weibull fading channels, IEEE
Trans. Commun., vol. 57, pp. 10631067, July 2004.
[11] B. Holter, et al, On the amount of fading in MIMO diveristy systems, IEEE Trans. Wireless Commun., to
appear.
[12] M. Simon and M. Alouini, Digital Communications over Fading Channels: A Unified Approach to Performance
Analysis, John Wiley & Sons, Inc. 2000.[13] M. Nakagami, The m-distribution: A general formula of intensity distribution of rapid fading, Statistical
Methods in Radio Wave Propagation, W. C. Hoffman, Ed. Pergamon Press Inc., 1960.
[14] J. Proakis, Digital Communications, Fourth Edition, McGraw-Hill, Inc., 2000.
[15] W. Feller, An Introduction to Probability Theory and Its Applications, Volume II, John Wiley & Sons, Inc., 1966.
[16] J. Craig, A new, simple, and exact result for calculating the probability of error for two-dimensional signal
constellations, Proc. IEEE Milit. Commun. Conf. MILCOM91, McLean, VA, Oct. 1991.
-
7/29/2019 Amount of Fading Revisited
21/23
21
1 2 3 4 5 6 7 8 9 101
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
L
Shannoncapacity,nats/channeluse
Simulation
Second Order
Gaussian Approximation
Gamma Approximation
15 dB
10 dB
5 dB
(a)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.92.2
2.4
2.6
2.8
3
3.2
3.4
3.6
3.8
4
annoncapacity,nats
c
anneluse
SimulationSecond OrderGaussian ApproximationGamma Approximation
L=16
L=8
L=4
(b)
Fig. 1. Antenna diversity: (a) Shannon capacity versus L for = 0.5 and various values of SNR; (b) Shannoncapacity versus for SNR per antenna = 5 dB and various values of L.
-
7/29/2019 Amount of Fading Revisited
22/23
22
10 9 8 7 6 5 4 3 2 1 00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SNR per antenna, dB
utageprobability
SimulationGaussian ApproximationGamma Approximation
L=8
L=2
Fig. 2. Antenna diversity: Outage probability versus SNR/antenna for = 0.5, R = 1bit/channel use.
-
7/29/2019 Amount of Fading Revisited
23/23
23
1 2 3 4 5 6 7 8 9 100.9
1
1.1
1.2
1.3
1.4
1.5
1.6
Number of Users
MultiuserDiversityGain
SimulationGaussian Approximation
Gamma Approximation
= 0.2
= 0.7
= 1.2
(a)
1 2 3 4 5 6 7 8 9 100.9
1
1.1
1.2
1.3
1.4
1.5
1.6
Number of Users
MultiuserDiversityGain
Simulation
Gaussian Approximation
Gamma Approximation
L=2
L=4
L=8
(b)
Fig. 3. Multiuser diversity: Multiuser diversity gain versus number of users for SNR = 5 dB, (a) L = 10 and variousvalues of; (b) = 0.2 and various values ofL.