Anda di halaman 1dari 11

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO.

2, FEBRUARY 2000 231

A General Analysis of Signal Strength Handover


Algorithms with Cochannel Interference
Fortunato Santucci, Senior Member, IEEE, Marco Pratesi, Marina Ruggieri, Senior Member, IEEE, and
Fabio Graziosi, Member, IEEE

Abstract—In this paper, the problem of computing cochannel strengths the MS receives from the BS’s, and the bit-error rate
interference statistics in signal strength measurements in mo- [8]–[10]. Relative signal strength measurement is a commonly
bile radio systems is presented. In a cellular environment with used criterion. As it is based on the sequence of signal strength
cross-correlated log-normal shadowings, extensions of Wilkinson’s
method and Schwartz and Yeh’s method are proposed for evalu- measurements taken on the radio links, cochannel interference
ating the statistics of differences between signal strengths that a is also important in analyzing performance of handover algo-
mobile terminal measures on the links to any pair of base stations rithms. The problem of modeling cochannel interference in han-
in the presence of cochannel interferers. The derived statistics are dover algorithms has been addressed in [11] and [12]. In [11], a
then used in the performance analysis of relative signal strength simplified scenario of two cells was considered, with an inter-
handover algorithms. Results provided by the two methods are
compared with simulation results, in order to assess their accu- ferer for each cell. In [12], the authors considered the problem
racy, and computational issues are addressed. From numerical of a generic cellular layout, with a number of BS’s and their
results, it is also seen that handover algorithm performance has a complete set of interferers of the first tier. Moreover, an angle
noticeable dependence on both cross correlation among shadow dependent cross-correlation law between shadow fadings be-
fadings and the interference level. Finally, it is seen that previous longing to any pair of links was modeled. A modified version
approaches to derive cochannel interference statistics in the
presence of log-normal shadowing can be obtained as particular of the Schwartz and Yeh’s approximation (SYA) was proposed
cases. to calculate cochannel interference statistics for the handover
Index Terms—Cochannel interference, land mobile radio
problem. From the numerical results, it was seen that handover
cellular systems, land mobile radio interference, log-normal algorithm performance may depend markedly on cross-correla-
distributions, mobile communication. tion parameters and on the cluster size.
In the present paper, the authors propose an investigation of
cochannel interference statistics that are suitable for handover
I. INTRODUCTION
algorithm analysis. In particular, the SYA method [13] and

T HE EFFECTS of cochannel interference in cellular mo-


bile radio systems have been investigated in the presence
of fading or shadowing alone, as well as in the presence of com-
Wilkinson’s approximation (WA) method [14] are considered.
A generalization of both methods is developed in order to
account for the general model which is here adopted, as
bined fading and shadowing [1]–[7], and design criteria of a fre- it will be pointed out in the following sections. Moreover,
quency reuse system have been defined. computational issues are addressed, and a new bound is found
The achievement of high capacities in cellular systems also for the truncation error of the numerical series involved in the
depends on the deployment of small cells. This corresponds to SYA method. Finally, it is seen that the typical problem of
increasing the spectral efficiency while keeping constant the fre- calculating the outage probability in the presence of log-normal
quency reuse factor (i.e., the cluster size). However, as cell di- interferers can be solved as a particular case of the problem
mensions are reduced, the handover rate increases and the effi- addressed here.
ciency of handover algorithms is expected to play a key role The effects of log-normal shadowing on cellular systems have
in overall system performance. Handover algorithms may be long been investigated. In [1], [5], [15], and [16], it was assumed
based on various criteria, such as the distances between the mo- that the shadowings are statistically independent. In some situ-
bile station (MS) and surrounding base stations (BS), the signal ations, however, they may be correlated. For example, the sig-
nals an MS receives from surrounding BS’s may be shadowed
Paper approved by D. P. Taylor, the Editor for Signal Design, Modulation, by the same obstacles in the neighborhood of the receiver. In
and Detection of the IEEE Communications Society. Manuscript received fact, cross correlation between the shadowings has been found
November 26, 1997; revised January 15, 1999 and July 19, 1999. This work by measurements [17], [18] and reasonably modeled in previous
was supported by MURST of Italy in the frame of a National Research Project.
This paper was presented in part at IEEE Pacific Rim Conference, Victoria, works [10], [12], [19]. In particular, in [17] and [18], it was
BC, August 1997, and at the IEEE International Conference on Universal reported that the cross correlation depends on the geometrical
Personal Communications, San Diego, CA, October 1997. angle between the links. In [10] and [12], an angle dependent
F. Santucci and F. Graziosi are with the Department of Electrical Engineering,
University of L’Aquila, I-67040 L’Aquila, Italy (e-mail: santucci@ing.univaq.it; cross-correlation law was actually used in performance calcula-
graziosi@ing.univaq.it). tions. Hence, in general, the local means of any pair of signals
M. Pratesi and M. Ruggieri are with the Department of Electronics En- received at the mobile station are correlated. Moreover, correla-
gineering, University of Roma “Tor Vergata,” I-00133 Roma, Italy (e-mail:
pratesi@nausicaa.eln.uniroma2.it; ruggieri@uniroma2.it). tion coefficients for any two distinct pair of links may not have
Publisher Item Identifier S 0090-6778(00)01569-5. the same value.
0090–6778/00$10.00 © 2000 IEEE
232 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO. 2, FEBRUARY 2000

In the study of signal strength-based handover algorithms, hexagonal cells is considered. Cells are assumed to have equal
the characterization of crossing events of a certain threshold by size. The distance between any two BS’s is denoted by . The
low-pass filtered versions of the power sums of Gaussian pro- cluster size, denoted by , is a parameter of the model and
cesses is needed. In analysis, it is useful to deal with Gaussian can be expressed in terms of the shift parameters and as
processes. Hence the moment matching approximation (MMA) [22]. MS moves from (current BS)
[7], the cumulant matching approximation (CMA) [7], and the towards the cell served by (target BS) along a straight line,
SYA [13] are considered in the present work. In the evaluation the latter forming an angle with the line connecting to
of crossing probabilities, it is important to have accurate approx- . An appropriate spatial reference system is introduced,
imations of the mean and the standard deviation of the process through the definition of an abscissa along the direction
(for instance, the tail of the distribution is not as important as in of motion. Hence, the distance ( ,
the evaluation of the outage probability). Hence, for the MMA ) of the MS from can be expressed in
and CMA, only their first regions matchings are considered. As terms of and . The signal level the MS receives from each
an alternative, the SYA is used along with the extension pro- of the surrounding BS’s is here considered to be the sum (in
vided in [20]. For the problem at hand, however, it is required decibels) of the mean signal strength and shadow fading. The
to extend both methods in order to calculate the statistics of dif- mean signal strength has a log-dependence of the –
ferences between the power sums of Gaussian processes. These distance , while shadow fading is a zero-mean Gaussian
differences are used in relative signal strength handover algo- process with exponentially decaying autocorrelation func-
rithms, where they are compared with a hysteresis margin after tion [23]. Hence, let (in decibels) be the signal level the
averaging. As a preliminary consideration on the accuracy of MS receives from . For a given , it is expressed as
the considered methods, it can be observed from [1], [2], and
[13] that the WA is valid only for restricted values of the stan- (1)
dard deviation and of the number of RV’s. However, in [21], it
is shown that the WA method may give much better estimates where and are the parameters of the mean signal strength
of the distribution in some practical problems, for example, in for the – link. and
the evaluation of outage probability. denote the standard deviation and the correlation decay length of
The paper is organized as follows. In Section II, the system the shadowing process, respectively. In a uniform propagation
model is described. In this frame, the cross correlation is intro- environment, the same values will be assumed for all links. In
duced to model shadow fadings. In Section III, statistics of the the following, the distance dependence of propagation laws is
power sums of cross-correlated cochannel interferers are com- expressed in terms of for a given .
puted. The analysis is carried out by resorting to the Gaussian For a given , the cross-correlation coefficient between
approximation and is presented in two steps. In Section III-A, and , denoted by , is reasonably assumed to depend on
the WA method is extended for computing the mean value and the angle between the links – and – [10],
the variance of power sums, along with the cross-correlation co- [12]. In [10], the suitability of a general cross-correlation model
efficients among them. In Section III-B, an extension of the SYA is proven. It relies on a Fourier-series expansion with nonneg-
method is used to evaluate the same statistical parameters. In ative coefficients. In the present work, only the first two terms
Section IV, the series involved in the SYA method are consid- are taken into account, thus the following cross-correlation law
ered and related computational issues are addressed. The con- is considered for numerical computation:
vergence of the series is further investigated, and an efficient
truncation criterion is achieved. Furthermore, evaluation of the (2)
series is discussed and an improved approximation is proposed.
In Section V, the handover algorithm analysis is carried out. Al- The model above can be effectively exploited if a model related
gorithm performance is derived in terms of the average number to experimental measurements is not available. In fact, due to its
of handovers and the handover delay. Section VI is split into simple form, (2) can be utilized to understand the effects of cross
two parts. In Section VI-A, analytical results provided by the correlation on cochannel interference by varying the values of
extended versions of the SYA and WA methods are presented the parameters and .
and compared with those obtained through simulations. In Sec- Due to the presence of cochannel interferers, measurements
tion VI-B, typical performance curves of handover algorithms of the signal strengths performed by the MS on the links to
are displayed. The tradeoff is discussed for optimal design of surrounding BS’s of the central cluster need to be character-
the algorithms. Finally, conclusions and future perspectives are ized through the sum of all cochannel signals. It is here as-
given in Section VII. sumed that a conventional frequency-division multiple-access
(FDMA) or time-division multiple-access (TDMA) technique
is used and all potential interfering channels are active. Let
II. SYSTEM MODEL ( ) denote the power sum (in decibels) of the
A cellular network is considered, which consists of desired signal the MS receives from and of the interfering
base stations belonging to a central cluster, signals from . It can be written
i.e., , and their first tier of inter-
fering BS’s, which are denoted by , (3)
. A typical cellular environment with
SANTUCCI et al.: SIGNAL STRENGTH HANDOVER ALGORITHMS WITH COCHANNEL INTERFERENCE 233

The deterministic mean and the zero-mean random component where


of are denoted by and , respectively.
(6)
III. ANALYSIS OF COCHANNEL INTERFERENCE
The presence of multiple correlated signals is modeled by re- Matching the second moment yields
sorting to a log-normal approximation for the sum of the log-
normal RV’s involved in (3). Therefore, for a fixed , the var- (7)
ious ’s ( ) can be considered jointly where
Gaussian RV’s, even in the presence of cross correlation among
shadowings [7], [13], [20]. Further insight on the accuracy of
the Gaussian approximation is provided in the appendix. In par-
ticular, moving from the approach used in [24] and [25], it is
shown that the accuracy of the Gaussian approximation worsens
for increasing values of and decreasing values of cross-cor-
relation coefficients.
In order to simplify the notation, the dependence on is not (8)
inserted explicitly. Moreover, the signal levels the MS receives
from surrounding BS’s are expressed in nats, through the no- By solving (5) and (7) for and , one obtains
tation ( , ),
where . The mean of is denoted by ,
and it is given by , while the standard deviation is denoted
by . The power sum (nats) of the desired signal the MS (9)
receives from and the interfering signals it receives from
where and are in nats and hence must be divided by to
is denoted by ( ). It
compute and .
can be written
Now Wilkinson’s method is extended in order to estimate
the statistics of ( ). It is necessary to evaluate
(4) the cross-correlation coefficient ( ). It is computed
by matching the second moment of with
the second moment of . The
In the following, the log-normal approximation will be used for second moments of and have been matched
, i.e., will be retained as a Gaussian RV, whose mean with and . Hence, it is only needed to match the cor-
and standard deviation will be denoted by and , respec- relation between and with the correlation be-
tively. The statistics of the differences ( ) are to be tween and , which is denoted
computed for analyzing handover algorithms based on relative by , i.e.,
signal strength measurements. In particular, the cross-correla-
tion coefficient between and (or equivalently, between
and ), which is denoted by , is to be evaluated.
It can be observed that solving the above problem for
also provides the solution to the evaluation of signal-to-inter- (10)
ference ratio statistics and of the outage probability, as well.
In fact, it is sufficient to replace with (i.e., the desired where the last equality holds true from (5), and
signal strength from ) and with the power sum of only
cochannel interferers (i.e., in (3), the sum index must range from
to ) and then to set .

A. Extension of Wilkinson’s Method


The parameters and are computed through
Wilkinson’s approach [14], along with the generalization
provided in [7]. Let and denote the first moment
and the second moment of , respectively. The
parameters and can be obtained by matching the first (11)
moment and the second moment of with and
. By matching the first moment, one obtains
By solving (10) with respect to , one obtains

(5) (12)
234 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO. 2, FEBRUARY 2000

where the expression to be used for is given in (11). The functions , , have the same expressions given in
[20], i.e.,
B. Extension of Schwartz and Yeh’s Method
The SYA method [13] was proposed for computing the power
sum of two uncorrelated Gaussian RV’s with unequal means and
(19)
variances. Following an approach similar to the one developed
in [24] and [25], exact expressions for the first two moments are
obtained. Furthermore, through a recursive procedure, the sta-
tistics of the sum of more than two RV’s can be derived. In [20],
the SYA method is extended in order to compute the first two
moments of the power sum of correlated log-normal RV’s. In
the following, the SYA method [13] is used along with the gen-
eralization in [20] in order to derive statistics of the difference
between two power sums of correlated log-normal interferers.
The iterative approach moves from the definition of partial (20)
sums, as follows:

(21)
(13) where

whose mean and standard deviation are denoted by and (22)


, respectively. Moreover, denotes the cross-corre-
lation coefficient between and for , and
(23)
denotes the cross-correlation coefficient between and
for and , or . The RV
is also introduced in order to get compact expressions. The fol-
lowing expressions hold for its mean and variance, respectively: (24)

The recursive procedure starts with , which implies


. Successive iterations are performed at . The
(14)
whole procedure is applied for . Finally, since
, the following expressions result:
Expressions of , , , and in terms of the
corresponding parameters of the lower order partial sum
are then obtained as (25)

(15) (26)
and can be obtained by expressing and in deci-
bels

(16) (27)

In the above equations, the explicit dependence of has been


omitted. However, it is to be remarked that , , and
depend on , due to the ’s and ’s dependence on the
MS position.

(17) IV. COMPUTATIONAL ISSUES


In this section, the convergence and the truncation of series
involved in , , and are considered, along
with their efficient evaluation. In [24], the convergence and trun-
cation issues were addressed, by resorting to upper and lower
bounds for (which are asymptotic approximations valid
(18) for a sufficiently large ) and proving that the terms of series
SANTUCCI et al.: SIGNAL STRENGTH HANDOVER ALGORITHMS WITH COCHANNEL INTERFERENCE 235

decrease monotonically for a sufficiently large . In the present By using (30) with for large values of
work, exact expressions of the generic terms of series are con-
sidered. First of all, it is proven that their absolute values de-
crease monotonically for every value of . Hence, series can be (33)
truncated when the last added term is less or equal to the max-
imum truncation error, even if is small. This allows a reduc- Let denote the error in the asymptotic evaluation
tion of the computing time for given values of the truncation of . By using (32), it can be stated that
errors. Furthermore, for large values of , the terms of series
are evaluated by using an asymptotic approximation which per- (34)
forms better than the one used in [24]. The related errors are
markedly reduced. They can be upper bounded through the in-
tegral criterion and shown to be negligible with respect to trun- In practice, we use the approximation (33) as long as ,
cation errors for typical values of the parameters involved. with a fixed value of (accuracy is traded off with com-
Let us observe from (24) that both and decrease putational effort). This improves the approximation used in [24],
with , while is constant and equal to 1. Hence, the where is considered (i.e., the correction terms corre-
convergence of series involved in (19)–(21) can be proven, sponding to are left aside). The related errors, which can
according to Leibniz’s criterion, by showing that be upper bounded through the integral criterion, are very small.
decreases with for any positive value of and for all values In fact, they are seen to be negligible with respect to the trunca-
of . With and as parameters, let us consider as a real tion errors.
variable rather than an integer one. Then, can be differenti-
ated with respect to

(28) V. ANALYSIS OF HANDOVER ALGORITHMS

where . If , then and the In the algorithm under study, the MS takes signal strength
second member of (28) turns out to be negative. If , measurements on both the current link and the links to adjacent
then results and the following inequality can be exploited (noninterfering) BS’s at regular time intervals . At the end of
([26, p. 932, eq. (26.2.12)] with ): each interval, the handover decision is taken according to a rel-
ative signal strength criterion: if the channel corresponding to
(29) a BS different from the current one provides an averaged level
which exceeds by at least dB (hysteresis margin) the level pro-
Then, it is proven that the second factor of (28) is negative for vided by the current channel, a handover is initiated. The han-
all values of . Hence, decreases as increases, and the con- dover is assumed to be executed instantaneously. The quantities
vergence of series can be stated. Moreover, as series have alter- of interest are defined as in [10]. In particular, since the MS
nate sign elements, it results that the truncation error is upper moves with constant speed and the effect of mobile speed is
bounded by the absolute value of the last term considered in the not under study here, the analysis is performed in the space-do-
computation of series. main, by setting [9], [10]. Hence, handover decisions
The second problem addressed here is the computation of the are considered at the end of each interval , which
expression for in (23). In fact, the evaluation of the is denoted as the th interval, with .
two factors and may be It has been remarked that , , and depend on , due to
prohibitive at large values of . Indeed, for increasing values the ’s and ’s dependence on the MS position. As a con-
of , the first factor approaches , while the second factor ap- sequence, the generic process is neither stationary in the mean
proaches 0. However, at large values of , an approximation can nor in the covariance. However, this process can be considered
be found for by resorting to the following asymptotic locally stationary, if it is reasonably assumed that the path-loss
approximation of at large values of its argument [26, p. exhibits no significant variations within distances on the order
932, eq. (26.2.12)]: of tens of meters (i.e., in the typical range of shadowing auto-
(30) correlation decay distance for a cellular environment and the
averaging window length which is considered in the following).
where Similar arguments lead one to consider that the ’s are fairly
constant within the same distance range [10]. Moreover, ac-
cording to [11], the autocorrelation function of the zero-mean
(31) component can be still assumed to have an exponentially
decaying trend with decay distance , so that
and . Let denote the error in the (35)
asymptotic evaluation of considered in (30). It can be Hence, within the above distance ranges, the ’s can be con-
proven that [26] sidered jointly stationary Gaussian processes with a good de-
gree of approximation for the problem at hand. They can be then
(32)
considered in handover algorithm analysis, in the frame of the
236 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO. 2, FEBRUARY 2000

approach developed in [10]. To this end, let us define the dif- Hence, the following expressions result for the joint prob-
ferences between the signal strengths the MS receives on the abilities :
various links, for a given

(40)
(36)

The zero-mean random component of is denoted by


for (41)
and can be expressed as . Its
variance is given by
Now, the probability of the MS being assigned to BS during the
th interval turns out to be
(37)
(42)
Since the considered handover algorithm is based on the average
of relative signal strength measurements, the following process while the probability of handover is given as
results from low-pass filtering through an expo-
nential window to obtain: (43)

The joint probabilities and the state probabilities


are computed for each by iterative solution, in the same way
(38)
as in [10], with the reasonable initial condition that
and for . It is worth noting that the transition
where is the window decay distance. The standard devi- from to (i.e., a handover) is assumed to take place
ations of and (where denotes the first-order when a down-crossing of occurs only for the with
derivative of ) are denoted by and , whose . Actually, in the case of contemporary down-crossing
dependence on results as a consequence of the nonstationarity of by both and an with , the handover algo-
of with the assumed cross-correlation model. They can rithm should select as “target” BS, among the two “candidate”
be in turn assumed to be “locally” constant and result in the ap- BS’s, the one providing the highest signal level. However, as a
proximate expressions [10] small is chosen, the probability of having a crossing is quite
low. Hence, the probability of having contemporary crossings
by different processes can be neglected. Simulation results con-
(39) firm the validity of the approximation.
From the above probabilities, final performance measures of
handover algorithms can be derived in terms of the average
Asymptotic results of the level crossing theory of Gaussian
number of handovers along the mobile trajectory and the han-
processes can be now used to derive expressions for the
dover delay. In the present case, for a generic angle , the av-
probabilities of crossing of the hysteresis band, as well as
erage number of handovers is defined as
to characterize the sequences of crossing events [9], [27].
The probability that up-crosses the threshold
number of handovers (44)
and the probability that down-crosses the
threshold , respectively, during the th decision interval
where ( stands for the lower integer part
can be calculated (see [10] for expressions of crossing proba-
of ) and corresponds to the point at minimum distance
bilities). For the sequence of up-crossing events of the upper
between the MS and BS . For , the value of is
hysteresis threshold and of down-crossing events of the lower
, as in the case of two BS’s. Moreover, in order to account for
threshold, the Poisson point process model is assumed to
the handover delay, the crossover point is defined as the point
hold true. This is a reasonable approximation as long as the
where becomes lower than the probability of the MS being
probability of having one crossing in an interval of length
connected to BS with .
is low and the probability of having more than one crossing
is negligible. Let denote the probability of the MS
being connected to BS (state probability) during the th VI. NUMERICAL RESULTS
interval, and let ( ) denote the In this section, numerical results are presented. In numerical
joint probability of the MS being connected to BS during computations, the following values were chosen for the main
the th interval and to BS during the th interval. system parameters: m, , and for
Under the Poisson assumption, an up-crossing event of , , m. The section is
by during the th interval is independent of what split in two parts. Section VI-A addresses the basic behavior of
happens in the other intervals. Moreover, up-crossings of the methods used to approximate the sum of log-normal compo-
and down-crossings of are independent of each other. nents. Results provided by the extended versions of the SYA and
SANTUCCI et al.: SIGNAL STRENGTH HANDOVER ALGORITHMS WITH COCHANNEL INTERFERENCE 237

Fig. 1.  versus the BS –MS distance for various values of  , as Fig. 2.  versus the BS –MS distance for two values of  , as obtained
obtained from SYA, WA, and simulations. from SYA, WA, and simulations.

WA methods are presented in terms of standard deviations and


cross-correlation coefficients of the processes resulting from the
power sums of log-normal components. The accuracy provided
by the two methods at various values of channel parameters and
cluster size is an important issue of the discussion. In Section
VI-B, the focus is on the computation of typical performance
measures of handover algorithms.

A. Statistical Parameters of the Power Sums


Attention here is restricted to the properties of processes
and , as they are the most important to the analysis of
handover algorithms. The MS moves along a straight line at
, so that and may be considered as belonging to
the links to the source BS and to one of the two equiprobable Fig. 3.  versus the BS –MS distance for various values of  , as
target BS’s, respectively. However, the considerations that will obtained from SYA, WA, and simulations.
be made do apply to all other processes .
In Fig. 1, the standard deviation is plotted versus the distance. In particular, the WA method tends to overestimate
– distance at various values of the shadowing standard at small distances, while it provides an underestimate at large
deviation . The cluster size is and the cross-correla- distances.
tion parameters , are chosen so that a nonnegative cross-cor- Finally, Fig. 3 shows the standard deviation , which is a
relation coefficient results for any pair of links along the mobile key parameter, plotted at various values of . It can be noticed
path. As a consequence of the power sum, is lower than , that sensibly varies along the path, mainly at the largest
and it can be noticed that decreases as the – distance values of . It is important to observe that may get its
increases. This results as a consequence of the variation of both largest value when the MS is closer to the boundary region be-
path-losses and cross correlation among the various log-normal tween the source cell and the target cell, where the probability of
components along the MS path. In particular, interference has a handover also increases as a consequence of the small difference
negligible effect when the MS is very close to , i.e., when the between the average signal strengths. The figure also shows that
mean signal strength from this BS is much larger than any inter- the SYA method definitively outperforms the WA method. As
fering signal. From the same figure, it can be seen that the SYA an example, at large values of the – distance, the value
method provides results that are in close agreement with simu- provided by the WA method for dB approaches the one
lation results, while the WA method provides noticeably poorer that is actually obtained for dB. At smaller values of
performance as either the – distance increases or in- , however, the WA method may provide satisfactory accu-
creases. For example, for dB, the WA method provides racy, with the advantage of requiring a moderate computational
a 15% overestimate of at a large – distance. effort if compared to the SYA method.
In Fig. 2, the cross-correlation coefficient is plotted versus In Figs. 4–6, the effects of cross-correlation law parameters
the – distance at dB and dB. It can be are analyzed at dB and . In particular, two rep-
seen that varies along the mobile path, as a consequence of resentative choices of and are considered. As previously
variation in the path-losses and the cross correlation between the stated, leads to nonnegative cross-correlation co-
links contributing to the power sums. It can be seen again that at efficients between any pair of links. Instead, the choice ,
dB, the accuracy gap between the SYA method and the may lead to a negative cross-correlation coefficient for a
WA method is noticeable at the largest values of the – pair of links that are more than 90 apart. From Fig. 4, it can be
238 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO. 2, FEBRUARY 2000

(a)
Fig. 4.  versus the BS –MS distance for two sets of values of
cross-correlation parameters, as obtained from SYA, WA, and simulations.

(b)

Fig. 7. Power sums statistical parameters versus the BS –MS distance for
Fig. 5.  versus the BS –MS distance for two sets of values of various cluster sizes, as obtained from SYA and simulations: (a)  and (b)
cross-correlation parameters, as obtained from SYA, WA, and simulations.  .

clearly noticed that the WA is unable to predict the larger vari-


ation that may occur along the mobile path when the cross-cor-
relation coefficient between any pair of links can span the entire
range . As a consequence, from Fig. 6 it can be seen that
the WA method provides an overestimate of the standard devi-
ation , which is significant in the handover region and is
evidenced when the cross-correlation coefficient is allowed to
vary within a larger range. From all three figures, it can be seen
that both the SYA and the WA methods perform better when the
cross-correlation law allows only positive cross-correlation co-
efficients. In Fig. 6, results obtained from the SYA method are
also plotted in the absence of cross correlation, in order to ev-
idence the effects of cross-correlation parameters. It can be af-
firmed that along most of the mobile path, a larger standard de-
Fig. 6.  versus the BS –MS distance for various sets of values of viation is observed when negative cross-correlation coefficients
cross-correlation parameters, as obtained from SYA, WA, and simulations.
are allowed.
Fig. 7 address the effect of cluster size. In this frame, only
seen that the decrease of with the – distance is more the results obtained from the SYA method and simulations are
marked at , . In particular, for nonnegative cross plotted. The cluster size has a negligible effect on at small
correlations, shows a smaller deviation from the 6-dB ref- values of the – distance, where the useful signal is much
erence value obtained in the absence of interference. However, larger than any interfering signal. By the contrary, when the mo-
only the SYA method performs with sufficient accuracy. Instead, bile terminal is closer to the boundary, the standard deviation is
the WA method is quite insensitive to the choice of cross-cor- seen to vary noticeably with the cluster size. In Fig. 7(a) and (b),
relation parameters, thus performing with poor accuracy in the the cross-correlation coefficient and the standard deviation
considered case. The same remark holds true in Fig. 5. It can be are plotted for various values of the cluster size, at
SANTUCCI et al.: SIGNAL STRENGTH HANDOVER ALGORITHMS WITH COCHANNEL INTERFERENCE 239

dB and , . It can be seen that is even more sen-


sitive than to the cluster size, the sensitiveness being higher
when the MS is closer to . It is to be remarked that even at
, varies along the MS path as a consequence of the
cross-correlation dependence of the angle . Furthermore, at
the lowest values of , may not have monotonic behavior,
and the range of its values decreases as decreases. It can be
also observed that the accuracy of the SYA method worsens re-
markably at , mainly for larger values of the –
distance. Nonetheless, this is not true for , which is of ul-
timate interest in our application. Indeed, the agreement with
simulation results in Fig. 7(b) is quite good at every cluster size.

B. Performance of Handover Algorithms (a)


In the analysis of handover algorithms, the sampling interval
is set to 1 m. Moreover, the decay distance of the averaging
filter impulse response is set to 30 m. Hence, algorithm
performance will be only shown as a function of the hysteresis
margin . Although the model can work with a generic , it
has been already pointed out [10] that can provide satis-
factory accuracy in predicting handover performance for every
along the considered MS path. By considering the first tier of
interfering BS’s, 21 BS’s are considered in numerical computa-
tions.
In Fig. 8(a) and (b), results are shown in terms of the behavior
of the handover probability along the mobile path. It can be seen
that the WA method tends to overestimate the handover prob-
ability, as a consequence of the overestimate of . By the
(b)
contrary, the SYA method is in close agreement with simula-
tions. The accuracy gap between the two methods is significant Fig. 8. Probability of handover versus the BS –MS distance, as obtained from
at larger shadowing variance. From a closer examination of the SYA, WA, and simulations for: (a)  = 4 dB and (b)  = 6 dB.
figure, it can also be observed that the maximum value of the
probability predicted by the WA method falls at the left-hand
side of the maximum value predicted by both the SYA method
and simulations. This is a consequence of a poorer estimation
of the mean value of the power sums. A relevant consequence
in the handover algorithm is that the crossover point abscissa is
underestimated, as will be shown.
In Fig. 9, typical design curves are shown with ,
dB, , and . Exhaustive results are only
shown for , as obtained from both model (SYA, WA) and
simulations. It can be seen that a good agreement is found for the
SYA method, the model providing slightly pessimistic perfor-
mance. In the same figure, the “optimal” design points are also
reported at various values of , according to the SYA method.
They have been selected from the knee of related performance Fig. 9. Crossover point versus average number of handovers at K = 7, as
curves. It can be seen that approximately the same performance obtained from SYA (“optimal” performance is also shown at other values of
can be achieved, however at different values of . In the figure, K ), WA, and simulations.
the optimal point for is obtained at dB. As
decreases, lower hysteresis values are required, and the optimal also shown in Fig. 9. In the previous section, it has been ob-
point for is obtained at dB. Moreover, in Table I, served that the WA method tends to overestimate the probability
the effect of interferers is evidenced in terms of sensitivity of of handover, and consequently, a larger number of handovers is
algorithm performance ( stands for the crossover point ab- predicted. Furthermore, is can be seen that the WA method un-
scissa) to the hysteresis margin (1-dB step increase). It can be derestimates the crossover point abscissa. The curves obtained
seen that performance variation, corresponding to 1-dB step in through the SYA and WA methods quite overlap each other.
the hysteresis margin, increases as decreases, which means However, a given point of the curve is obtained at quite different
that performance sensitivity increases as interference intensity values of by the two methods. In particular, the WA method
increases. The design curve obtained from the WA method is overestimates the nearly “optimal” of about 25% all along the
240 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 48, NO. 2, FEBRUARY 2000

TABLE I problem show that the Gaussian approximation becomes poorer


SENSITIVITY OF ALGORITHM PERFORMANCE TO CLUSTER SIZE for increasing values of and for decreasing values of the
cross-correlation coefficients.

REFERENCES
[1] R. Prasad and A. Kegel, “Improved assessment of interference limits
in cellular radio performance,” IEEE Trans. Veh. Technol., vol. 40, pp.
412–419, May 1991.
[2] J.-P. M. G. Linnartz, “Exact analysis of the outage probability in mul-
curves. Hence, only the SYA allows to accurately predict the tiple-user mobile radio,” IEEE Trans. Commun., vol. 40, pp. 20–23, Jan.
1992.
“optimal” . [3] A. A. Abu-Dayya and N. C. Beaulieu, “Outage Probabilities of diversity
cellular systems with cochannel interference in Nagakami fading,” IEEE
VII. CONCLUSIONS Trans. Veh. Technol., vol. 41, pp. 343–355, Nov. 1992.
[4] A. A. Abu-Dayya and N. C. Beaulieu, “Outage probabilities of cellular
Extensions of the Wilkinson’s method and of the Schwartz mobile radio systems with multiple Nagakami interferers,” IEEE Trans.
Veh. Technol., vol. 40, pp. 757–768, Nov. 1991.
and Yeh’s method have been presented in order to compute [5] Y. S. Yeh and S. C. Schwartz, “Outage probability in mobile telephony
cochannel interference statistics in signal strength measure- due to multiple log-normal interferers,” IEEE Trans. Commun., vol.
ments in mobile radio systems. In a rather general propagation COM-32, pp. 380–388, Apr. 1984.
[6] A. Şafak, “Optimal channel reuse in cellular radio systems with multiple
environment, with cross-correlated shadow fading components, correlated log-normal interferers,” IEEE Trans. Veh. Technol., vol. 43,
it is seen that only the extended Schwartz and Yeh’s method pp. 304–312, May 1994.
generally provides satisfactory accuracy. [7] A. A. Abu-Dayya and N. C. Beaulieu, “Outage probabilities in the pres-
ence of correlated lognormal interferers,” IEEE Trans. Veh. Technol.,
The methods are proposed to accurately predict the perfor- vol. 43, pp. 164–173, Feb. 1994.
mance of relative signal strength handover algorithms. In fact, [8] M. Gudmundson, “Analysis of handover algorithms,” in Proc. IEEE 41st
numerical results show that the cluster size and cross-correla- Vehicular Technology Conf., St. Louis, MO, May 1991, pp. 537–542.
[9] R. Vijayan and J. M. Holtzman, “A model for analyzing handoff algo-
tion parameters are to be carefully considered for an “optimal” rithms,” IEEE Trans. Veh. Technol., vol. 42, pp. 351–356, Aug. 1993.
choice of handover algorithm parameters. [10] F. Graziosi, M. Pratesi, M. Ruggieri, and F. Santucci, “A multicell model
Future work will concern the application of the extended of handover initiation in mobile cellular networks,” IEEE Trans. Veh.
Technol., vol. 48, pp. 802–814, May 1999.
methods to an improved outage analysis in cellular mobile [11] M. D. Austin and G. L. Stüber, “Cochannel interference modelling for
systems with generically correlated log-normal shadowings. signal strength based handoff analysis,” Electron. Lett., vol. 30, pp.
1914–1915, Nov. 1994.
[12] M. Pratesi, M. Ruggieri, F. Graziosi, and F. Santucci, “Performance
APPENDIX of signal strength handover algorithms with interference and correlated
shadowings,” in Proc. IEEE Vehicular Technology Conf., Phoenix, AZ,
This appendix deals with the validity of the Gaussian approx- May 1997, pp. 530–534.
imation for the power sum of correlated Gaussian components. [13] S. C. Schwartz and Y. S. Yeh, “On the distribution function and moments
For simplicity, the power sum of two correlated Gaussian RV’s of power sums with log-normal components,” Bell Syst. Tech. J., vol. 61,
pp. 1441–1462, Sept. 1982.
with equal means is considered. Furthermore, in a uniform prop- [14] N. A. Marlow, “A normal limit theorem for power sums of normal
agation environment, shadowing components can be modeled random variables,” Bell Syst. Tech. J., vol. 46, pp. 2081–2089, Nov.
through RV’s with equal variances. Following the same nota- 1967.
[15] R. Muammar and S. C. Gupta, “Cochannel interference in high capacity
tion of [24], let and have a bivariate Gaussian distribution mobile radio systems,” IEEE Trans. Commun., vol. COM-30, pp.
with zero means. Let denote their variances 1973–1982, Aug. 1982.
and their correlation coefficient. Their power sum is [16] D. C. Cox, “Co-channel Interference considerations in frequency
reuse small-coverage area radio systems,” IEEE Trans. Commun., vol.
(A.1) COM-30, pp. 135–142, Jan. 1982.
[17] V. Graziano, “Propagation correlations at 900 MHz,” IEEE Trans. Veh.
In the limit case , and are equal with unitary prob- Technol., vol. VT-27, pp. 182–189, Nov. 1978.
ability, hence, it can be affirmed that their power sum is just a [18] J. Van Rees, “Cochannel measurements for interference limited small
Gaussian RV, with mean and variance and , respectively. cell Planning,” Arch. Elek. Ubertragung., vol. 41, pp. 318–320, 1987.
[19] A. J. Viterbi, A. M. Viterbi, K. S. Gilhousen, and E. Zehavi, “Soft
For , the moment generating function of , can handoff extends CDMA cell coverage and increases reverse link
be obtained from [25, eqs. (2.1–2.4)] for capacity,” IEEE J. Select. Areas Commun., vol. 12, pp. 1281–1288,
Oct. 1994.
[20] A. Şafak, “Statistical analysis of the power sum of multiple correlated
log-normal components,” IEEE Trans. Veh. Technol., vol. 42, pp. 58–61,
Feb. 1993.
(A.2) [21] N. C. Beaulieu, A. A. Abu-Dayya, and P. J. McLane, “Estimating the
distribution of a sum of independent lognormal random variables,” IEEE
where , i.e., is half the variance of Trans. Commun., vol. 43, pp. 2869–2873, Dec. 1995.
. For (i.e., ), the series above is equal to , [22] W. C. Y. Lee, Mobile Communications Engineering. New York: Mc-
hence, the expression of becomes in fact the moment Graw-Hill, 1982.
[23] M. Gudmundson, “Correlation model for shadow fading in mobile radio
generating function of a Gaussian RV with mean and variance systems,” Electron. Lett., vol. 27, pp. 2145–2146, Nov. 1991.
and , respectively. It can be seen that we run away from [24] J. I. Naus, “The distribution of the logarithm of the sum of two log-
this particular case when increases, i.e., when the variance normal variates,” J. Amer. Statist. Assoc., vol. 64, pp. 655–659, June
1969.
of and increases or the correlation coefficient between [25] M. A. Hamdan, “The logarithm of the sum of two correlated log-normal
and decreases. In fact, the numerical results obtained in our Variates,” J. Amer. Statist. Assoc., vol. 66, pp. 105–106, Mar. 1971.
SANTUCCI et al.: SIGNAL STRENGTH HANDOVER ALGORITHMS WITH COCHANNEL INTERFERENCE 241

[26] M. Abramovitz and I. A. Stegun, Handbook of Mathematical Functions, Marina Ruggieri (S’84–M’85–SM’94) was born in
with Formulas, Graphs, and Mathematical Tables, 9th ed. New York: Naples, Italy, in 1961. She graduated cum laude in
Dover, 1972. electronics engineering in 1984 at the University of
[27] H. Cramèr and M. R. Leadbetter, Stationary and Related Stochastic Pro- Roma La Sapienza, Rome, Italy.
cesses. New York: Wiley, 1967. She was with FACE-ITT in the high-frequency
[28] R. Prasad and A. Kegel, “Correction to improved assessment of interfer- division from 1985 to 1986 and was trained on
ence limits in cellular radio performance,” IEEE Trans. Veh. Technol., GaAs monolithic design and fabrication techniques
vol. 41, p. 551, Nov. 1992. at GTC-ITT, Roanoke, VA, in 1985. She was a
Research and Teaching Assistant with the Elec-
tronics Engineering Department at the University
of Roma “Tor Vergata,” Rome, Italy, from 1986 to
1991. She was an Associate Professor of Signal Theory with the Department
Fortunato Santucci (S’93–M’95–SM’00) was born of Electrical Engineering at the University of L’Aquila, L’Aquila, Italy, from
in L’Aquila, Italy, in 1964. He received the Laurea 1991 to 1994. Since 1994, she has been an Associate Professor of Digital
degree (cum laude) and the Ph.D. degree in elec- Signal Processing at the University of Roma Tor Vergata. In 1998, she was
tronic engineering from the University of L’Aquila, appointed Expert Member in the Appointments Board for the assignment of the
L’Aquila, Italy, in 1989 and 1994, respectively. Professor Chair in Radio Communications at the Lund Institute of Technology,
In 1989, he was with Selenia Spazio S.p.a., Rome, Lund, Sweden. In this frame, she cooperated with the space industry and
Italy. From 1991 to 1992, he was at the IESS-CNR, the European Space Agency. She is the Principal Investigator of a Satellite
Rome, Italy, doing research on superconductor Communications Mission (DAVID) selected by the Italian Space Agency,
receivers for millimeter wave satellite systems. in the frame of a national contest, to be developed and launched in 2003.
Since 1994, he has been an Assistant Professor Her research interests mainly include satellite communications systems, but
in Telecommunications with the Department of they also include mobile networks, wide-band wireless local area networks,
Electrical Engineering, University of L’Aquila, In 1996, he was a Visiting multimedia communications, and satellite system reliability.
Researcher at the Department of Electrical and Computer Engineering, Dr. Ruggieri was awarded the 1990 Piero Fanti International Prize for the
University of Victoria, Victoria, BC, Canada, where he did research on CDMA contribution “A New Millimetre Wave Satellite System for Land Mobile Com-
networks. His current research activity is focused on mobility management, munications,” together with Profs. F. Valdoni, A. Paraboni, and F. Vatalaro. She
traffic modelling, and multiple-access techniques for mobile communications received a nomination for the Harry M. Mimmo Award in 1996 for her contribu-
networks. tion “A Reliability Model for Active Phased Arrays in Satellite Communication
Systems.”In 1998, she was the chairperson of the 3rd European Workshop on
Mobile/Personal Satcoms (EMPS).

Marco Pratesi was born in Teramo, Italy, in 1971. Fabio Graziosi (S’96–M’97) was born in L’Aquila,
He received the Laurea degree (cum laude) in Italy, in 1968. He received the Laurea degree (cum
electronics engineering from the University of laude) and Ph.D. degree in electronic engineering
L’Aquila, L’Aquila, Italy, in 1995, and the Ph.D. from the University of L’Aquila, L’Aquila, Italy, in
degree in telecommunications from the University 1993 and 1997, respectively.
of Roma “Tor Vergata,” Rome, Italy, in 1999. Since February 1997, he has been an Assistant
Currently, he is involved in the satellite communi- Professor in Telecommunications at the Department
cations mission DAVID (data and video distribution), of Electrical Engineering, University of L’Aquila.
which is supported by the Italian Space Agency, to His current research interests include wireless
be developed and launched in 2003. His current re- communication systems with emphasis on mobility
search interests include handover algorithms perfor- management, handover techniques, multiple-access
mance analysis and on outage probability evaluation in mobile radio networks. techniques, and performance evaluation for personal communication systems.

Anda mungkin juga menyukai