iCAR : an Integrated Cellular and Ad-hoc Relay System*

Hongyi Wu

Chunming Qiao

Abstract

In this paper, we study the important problem of how to

evolve from the existing, heavily-invested cellular infrastructure to wireless systems that scale well with the number of mobile hosts. We propose to integrate the cellular
infrastructure with modern Ad-hoc relaying technologies to
achieve dynamic load balancing among different cells in a
cost-effective way. The basic idea of the proposed iCAR
(integrated Cellular and Ad-hoc Relay) system is to place
a number of Ad-hoc Relay Stations (or ARSs) at strategic locations, which can be used to relay signals between
MHs and base stations. By using ARSs, its possible to
divert traffic in one (possibly congested) cell to another
(non-congested) cell. This helps circumvent congestion, and
make it possible to maintain (or hand-off) calls involving
MHs (especially a high-pri,ority call) that are moving into a
congestedcell, or to accept new call requests involving MHs
that are in a congested cell. Other benefits include enhanced
coverage and reliability (or fault-tolerance) of the system,
and potential improvement in MHs battery life and transmission rate.
The rest of this paper is organized as follows. Section
2 describes the basic operations and principle benefits of
the proposed next generation wireless system which include
the concept of primary, secondary and cascaded relaying via
ARSs in both Congested-induced (CI) and Non-congestedinduced (NCI) situations. Section 3 discusses two important
system design issues, namely, the number and placement of
ARSs. In particular, the proposed seed-growing approach
for placing a limited number of ARSs is discussed and an
upper bound on the number of ARSs needed is derived. Additional analysis and simulation results of the iCAR system
are given in Section 4.Related works are in 5. Finally, Section 6 concludes this paper.

Ever increasing data traffic and limited capacity are major

causes for congestion in current cellular systems. This paper
presents a new architecture for the next generation wireless
systems based on the integration of the cellular infrastructure
and modem Ad-hoc relaying technologies. The new architecture can efficiently balance traffic loads between cells by
using Ad-hoc relay stations (ARS) to relay traffic from one
cell to another cell dynamically. This can not only increase a
systems capacity cost-effectively, but also reduce transmission power for mobile hosts, and provide services for shadow
areas. In this paper, we present the architectural concept including its basic operations and principle benefits. We also
propose a seed-growing approach for ARS placement, and
discuss the upper bound on the number of seed ARSs needed
in the system. We evaluate the performance improvement of
the new architecture through analysis and simulations.

1 Introduction
Ever increasing data traffic and limited capacity are major
causes for congestion in current cellular systems. It is also
expected that congestion will occur in peak usage hours even
in the third generation systems. By congestion, we mean
that in some cells, there will be no more capacity left. More
specifically, in a congested cell, there will be no more available data channels (or DCHs) for use by additional MHs in
the cell. Note, however, that control channels (or CCHs) for
signaling (or paging) may still be accessible by all MHs in
a congested cell.
Adding to the problem of limited capacity in existing wireless systems is the presence of unbalanced trafic. Specifically, some cells may be heavily congested (called hot spots),
while the other cells still have available DCHs. In other
words, even though the traffic load doesnt reach the maximum capacity of the entire system, call blocking and dropping may occur due to localized congestion. Since the locations of hot spots vary from time to time (e.g. downtown
areas in Monday morning, or amusement parks in Sunday afternoon), its difficult, if not impossible, to provide the guarantee of a sufficient amount of resources in each cell in an
cost-effective way. Other problems in cellular systems include the presence of shadows where MHs cannot receive
strong enough signals, and the need for MHs to transmit at
high power when they are farther away from base stations
thus limiting their battery life.

2 Basic Operation and Principle Benefits of iCAR

In this section, we describe the basic operations and principle benefits of the new architecture. To simplify the following presentation, we will focus on cellular systems where
each BTS is controlled by an Mobile Switching Center (or
MSC) (which is sometimes referred to as Mobile Telephone
Switching Office as well) [ l , 21. The major differences between BTSs and the proposed ARSs are as follows. Once
a BTS is installed, its location is fixed since it often has a
wired interface to an MSC (and a backbone network). On the
other hand, an ARS is a wireless communication device deployed by a network operator. It has its own controller with

*the authors are with Dept. of Computer Science and Engineering, University at Buffalo (SUNY). Email: {qiao,hongyiwu} @cse.buffalo.edu

0-7803-6494-5/00/$10.00 02000 IEEE

154

a much lower complexity and fewer functionality than that

needed for a BTS. In addition, it may have a limited mobility (which, unlike that of an MH, is under the control of an
MSC), and can communicate directly with an MH, a BTS or
another ARS through air interfaces.
An example of relaying is illustrated in Figure 1 where
MH X in cell B (congested) communicates with the BTS
in cell A (or BTS A, which is non- congested) through two
ARSs (there will be at least one ARS along what we call a
relaying route). Note that each ARS has two air interfaces,
the C (for cellular) interface for communications with a BTS
and the R (for relaying) interface for communicating with an
MH or another ARS. In the following discussion, we will
assume that the C interface operates at around 1900 MHz
(PCS), and the R interface uses an unlicensed band at 2.4
GHz though our concept also applies when different bands
are used (for example, 850 MHz for the C interface as in 2G
systems or 2 GHz for 3G systems). The R interface (as well
as the medium access control (MAC) protocol used) is similar to that used in wireless LANs or Ad-hoc networks (see
for example [3,4,5]).Note that because multiple ARSs can
be used for relaying, the transmission range of each ARS using its R interface can be much shorter than that of a BTS,
which means that an ARS can be much more smaller and less
costly than a BTS. At the same time, it is possible for ARSs
to communicate with each other and with BTSs at a higher
data rate than MHs can due to limited mobility of A R S s and
specialized hardware (and power source).

use the MH (and its R interface) to relay signals between another MH and a BTS as in Ad-hoc networks or the so-called
Opportunity Driven Multiple Access (ODMA) proposal (see
http://www.etsi.org), issues such as security (authentication,
privacy), billing, and unpredictable movement of the MHs
make such an approach difficult to implement. In fact, the
main challenges in the design and operation of Ad-hoc networks stem from the possibility of rapid movements of the
MHs as well as the lack of a centralized control and managemenr entity [ 6 ] .In the proposed system, ARSs (approximate) locations and their (potential) movement are under the
control of MSCs; and as a result, a relaying route with satisfactory QoS parameters (if it exists) can be established more
quickly, and once established, maintained with a higher degree of stability, making our approach more suitable for real
time applications.

2.1 Congestion-Induced (CI) Relaying

A principle benefit of the proposed integration of the cellular and Ad-hoc relaying technologies is that both the blocking probability of new calls to/from a congested cells, and
the call dropping probability during hand-offs to a congested
cell, can be drastically reduced via what we call congestioninduced (or CI) relaying. This is illustrated below.

2.1.1 New Call

In an existing cellular system, if MH X is involved in a new
call (as a caller or callee) but it is in a congested cell B, the
new call will be blocked. In the proposed next generation
wireless system with integrated cellular and relaying technologies, the call does not have to be blocked. More specifically, MH X which is in the congested cell B, can switch
over to the R interface to communicate with an ARS in cell
A, possibly through other ARSs in cell B (see Figure 1 for
an example). We call this strategy that establishes a relaying route between MH X (in a congested cell) and a BTS in
a nearby non- congested cell primary relaying.
With primary relaying, MH X can communicate with BTS
A, albeit indirectly (i.e. through relaying). Hereafter, we
will refer to the process of changing from the C interface
to the R interface (or vice versa) as switching-over, which
is similar to (but different from) frequency-hopping [I]. Of
course, MH X may also be relayed to another nearby noncongested cell other than cell A. Finally, a relaying route between MH X and its corresponding (i.e., caller or callee) MH
X may also be established, (in which case, both MHs need
to switch over from their C interfaces to their R interfaces),
even though the probability that this happens could be very
low.
If primary relaying is not possible because, for example
in Figure 1, ARS 1 is not close enough to MH X to be a
proxy (and there are no other nearby ARSs), one may resort
to secondary relaying so as to free up a DCH from BTS B for
use by MH X. Two basic cases are illustrated in Figure 2 (a)
and (b), respectively, where MH Y denotes any MH in cell
B which is currently involved in a call. More specifically,
as shown in Figure 2 (a), one may establish a relaying route
between MH Y and BTS A (or any other cell). In this way,

Figure 1: A relaying example where MH X communicates

with BTS A through two Ad- hoc Relaying Stations (ARSs)
(it may also communicate with MH X through ARS 1)
Among the ARSs involved in relaying, we may call an
ARS which directly communicates with an MH (e.g. ARS
1 in Figure 1) a proxy, and an ARS which directly communicates with a BTS (e.g. ARS 2 in Figure 1) a gateway (an
ARS can serve as both a proxy and a gateway at the same
time as illustrated in Figure 3 (a)). When and only when an
ARS serves as a gateway, it uses the C and R interfaces concurrently. Other ARSs along a relaying route use the R interface only. This means that an ARS does not use any DCH
unless it is serving as a gateway between an MH and a BTS,
in which case, a DCH will be allocated to the ARS dynamically by a MSC).
Note that, to enable relaying, a MH also needs to have
the R interface to communicate with an ARS, in addition to
having the C interface used to communicate directly with a
BTS under the normal situation (i.e. without relaying). Although it is possible to treat an MH just as an ARS, that is, to

155

after MH Y switches over, the DCH used by MH Y can now

be used by MH X. Similarly, as shown in Figure 2 (b), one
may establish a relaying route between MH Y and its corresponding MH Y in cell B or in cell C, depending on whether
MH Y is involved in an intra-cell call or an inter-cell call.
Note that, given that cell B is congested which means that
there are a lot of on-going calls (or candidates for MH Y),
the chance that case (b) in Figure 2 could occur should be
better than that a relaying route between MH X and MH X
can be established using primary relaying (as in Figure 1). In
addition, although the concept of having such an MH-to-MH
call via ARSs only (i.e. no BTSs are involved) is similar to
that in Ad-hoc netwonking, a distinct feature (and advantage)
of the proposed integrated system is that an MSC can perform (or at least assist in performing) critical call management functions such as authentication, billing, locating the
two MHs and finding andor establishing a relaying route
between them, as mentioned earlier. Such a feature is also
important to ensure that switching-over of the two MHs (this
concept is not applicable to Ad-hoc networks) is completed
fast enough so as not to disconnect the on-going call involving the two MHs or not to cause severe QoS degradation
(though the two MHs may experience a glitch or jitter).

sent as soon as the power level from BTS A received by MH

X goes below a certain threshold (and that from BTS B is becoming higher). A successful hand-off will take place, usually within a few hundred milliseconds (depending on the
moving speed of the MH) before the received power from
BTS A reaches an unacceptable level [ 1,2].
If cell B is congested, the hand-off request may be queued
(that is, the call may be blocked) for a short period of time,
e.g., up to a few tens of milliseconds as long as the received
power is still above the unacceptable level. If the congestion
in cell B persists, that is, there are still no DCHs available in
cell B after this short period of blocking time, the call will be
dropped.
In the proposed integrated system, MSC may apply the
primary relaying strategy to establish a relaying route between MH X to a BTS in a nearby non-congested cell (similar to Figure 1) or the secondary relaying strategies and cascaded relay to free up DCHs in cell B for use by MH X (similar to Figure 2). In this way, the handoff call can take place
successfully.
Note that by applying the relaying strategies (primary and
secondary) to establish a relaying route between an MH in a
congested cell B and a BTS in another cell (not necessarily an
immediate neighbor), new calls involving MHs in cell B and
hand-off calls involving MHs moving into cell B can now be
supported, it is as if cell B has borrowedsome DCHs from
other cells. In other words, the capacity of cell B has been effectively increased, thus eliminating (or at least alleviating)
congestion.

2.2 Noncongestion-Induced(NCI) Relaying

1II

lbl

Clearly, relaying can also be used to pro-actively balance

load among different cells by transferring calls from a
heavily-loaded cell to other lightly-loaded, and possibly remote cells (for example, two cells such as B and D in Figure 2
(c), between which there are no relaying routes available).
This is one of the main advantages of the proposed approach
over channel borrowing via cell sectorization whereby a cell
can only borrow a pre- determined set of channels (into one
of its sectors) from its immediate neighbors [2, 7, 81.
As a result of being able to distribute traffic evenly in an
entire system using the proposed approach, the call blocking
probability will be minimized. More formally, we have the
following theorem.

Figure 2: Secondary relaying to free up a channel for MH

X. (a) MH Y to BTS A, (b) MH Y to MH Y ,or (c) cascaded
secondary relaying (i.e. MH Y to BTS C and MH 2 to either
MH Z or BTS D).
If neither primary relaying, nor basic secondary relaying
as shown in Figure 2 (a) and (b) works, the new call may still
be supported. More specifically, assume that there is a relaying route, which can be either primary or secondary relayed,
between MH X and ARS, say G (for gateway),in a nearby
cell C which unfortunately is congested. As shown in Figure 2(c), one may apply any of the two basic secondary relaying strategies described above in the congested cell C (i.e.
in a cascaded fashion) as if ARS G is being handed-over
(see discussion below). Hence, if a relaying route between
an MH (say MH Z) in cell C and either another BTS in a noncongested cell or MH Z can be established, ARS G can be
allocated the DCH previously used by MH Z in cell C, and
in turn MH X can be allocated the DCH previously used by
MH Y in cell B if the route between MH X and ARS G is set
up by secondary relay.

Theorem 1 Assume that the total trafJic in a n-cell system is

T Erlangs, then the (system wide) call blocking probability
is minimized when the trafJic in each cell is T l n Erlangs.
The proof for this theorem is in the Appendix A. Note that
when no cells are currently congested, and relaying is used
to, for example, balance load as described above, we might
call this type of relaying noncongestion-induced (or NCI).
NCI-relaying is also useful to overcome so-called shadows
where no coverage by a BTS is available because either there
are buildings surrounding an MH, which completely block
signals from a nearby BTS to the MH, or no BTSs are close
enough to the MH. This is illustrated in Figure 3 (a) and

2.1.2 Hand-Off
In an existing cellular system, if an MH X involved in a call
moves from cell A to cell B, a request for hand-off will be

156

number of boundary cells is

rRI2 - r(R - 2 R ) 2
J = 1 4 6 - 4J

Figure 3: Non-congestion-induced (NCI) relaying to overcome shadows

Since each boundary cell has at least two non-shared edges,

the number of shared edges is at most E = 6 n - 2 1 4 6 41. Hence, the maximum number of seed ARSs for a nearcircular shaped n-cell system is E / 2 or,

3 n - L 4 f i - 4J

(2)
0

(b) respectively. In Figure 3 (a), an MH behind a building

(or buildings) may still receive the signal from a BTS due
to multi-path propagation of radio signals, though the signal
could be very weak [ 11. In such a case, NCI- relaying may
improve the signal strength and other QoS performances. As
an added benefit of relaying, either NCI or CI, one may reduce the power consumption of an MH since the distance
between the MH and the proxy ARS can be much shorter
than that between the MH and the BTS. More specifically,
given that the typical transmission ranges of a cell and a ARS
are 2km (at 1900 MHz) and 500m (at 2.4 GHz) respectively,
the maximum path losses are 104.04 dB and 94.02 dB, respectively (according to the free space propagation model
where the path loss is equal to 32.44 20Zog(frequency)
20log(distance)). This means that MHs using relay consume almost 10 times less power in Watts.

(1)

r R2

Proposition 2 For a n-cell system, the maximum number of

seed ARSS needed is 3 n - L4fi - 41.
Proof: Obviously, fewer boundary cells imply fewer nonshared edges, or in other words, more shared edges and thus
more required seed ARSs. According to the basic geometry
theory, for a given coverage area equal to that covered by n
cells, a circle will have the shortest perimeter. This means a
circle will have the fewest boundary cells among all possible
shapes. Thus, the upper bound on the number of seed ARSs
needed in a circular-shaped system is also the upper bound
on the number of seed ARSs needed in any shaped system.
From Proposition 1, we know this upper bound is

S, = 3 n - L4fi - 4J
0
Proposition 2 means that any other shaped system will require fewer seed ARSs. For example, Figure 4 (c) shows a
rectangular shaped system with 20 cells, which requires only
43 seed ARSs, (which is the same as the circular- shaped
system with 19-cells shown in Figure 4 (b), and certainly less
than the number of ARSs required in a circular-shaped system with 20 cells). In fact, for any rectangular-shaped system with n = n, x n b cells, we can come up with the exact
number of seed ARSs. More specifically, the total number
of non-shared edges in such a system is 4 n , 4nb - 2. So
the number of seed ARSs needed is S = 3n,nb - 2 ( n ,
nb)
1. Since nanb = n and n, n b 2 2 6 , we have
372,726 - 2 ( n , nb) 1 5 3 n - 4 f i
1,or S 5 S,.

3 Number and Placement of ARSs

In [9], we discussed the maximum number of relay stations
needed so as to ensure that a relaying route can be established between any BTS and an MH located anywhere in any
cell. However, when only a limited number of ARSs are
deployed in a cell, a natural question is how many ARSs
are reasonable and where to place them. Here, we propose a
Seed Growing approach whereby one seed ARS is placed on
each pair of shared edges (denoted by a bold line in Fig 4
(a)) along the border between two cells. Additional ARSs
will grow from the seeds as to be discussed later (See Figure 5). Since a seed ARS is shared by two cells, and each
cell has 6 edges, it is obvious that at most 3 n seed ARSs
are needed in a n-cell system. In fact, the total number of
seed ARSs required will be less because it makes no sense
to put any ARSs on a non-shared edge (denoted by a dotted line in Fig 4 (a)) of a boundary cell. More formally, we
have the following propositions stating the upper bound on
the number of seed ARSs.

Proposition 1 For a n-cell system where the cells are arranged in a near-circular shape (as in Fig 4 (b)),the number
of seed ARSS needed is at most 3 n - L4fi - 41.

Figure 4: The number of seed ARSs with seed-growing

approach. (a) One seed ARS for two shared edges. (b)
a circular-shaped system with 19 cells. (c) a rectangularshaped system with 20 cells.

Proof : Suppose that the radius of each cell is R, then the

total coverage of the n cells is about n x r R 2 .Since the cells
are organized as a near-circle whose radius RI satisfies the
equation rRl2 M n x rR2,or RM fi x R, the estimated

We now discuss how the topology of the ARSs is managed, and how a relaying route is set up. Note that, with the

157

help of MSCs, topology maintenance (as well as mobility

management of MHs and possible ARSs too) in iCAR is
easier than in most Ad-hoc networks studied so far. More
specifically, each ARS, triggered by a timer or an event (such
as a power on), only needs to collect neighbor information
(via its R-interface), and report to a MSC via its C-interface
(i.e., through a BTS using one of its control channels). The
topology map can then be calculated by MSCs, and only
the (updated) part relevant to the ARSs in a cell is broadcast by the corresponding BTS to all ARSs in the cell. This
means that the topology updating process only involves four
one-hop handshakings (two between a ARS and its neighbors, and two between the ARS and a MSCBTS), and thus
is faster and more reliable than topology updates in purely
Ad hoc networks.
Based on the broadcasted topology map, each ARS creates a routing table containing one entry for every BTS that
is reachable from that ARS via relay (i.e., other ARSs). This
routing table specifies information about each possible relaying route to a BTS, such as the next hop (i.e., the next A R S ) ,
and the number of hops as well as the total power consumption.
Whenever a MSC determines that a MH needs a relaying
route, it chooses one or more BTSs which have free channels, and informs the MH. The MH can then broadcast a
route probe message to all neighbor ARSs asking for information on the relaying route(s) to any one of these BTSs.
The ARSs which have a relaying route to any of these BTSs
replies with the information contained in their routing tables.
The MH can choose the best ARS whose relaying route is the
shortest (and/or has the lowest power consumption) as the
proxy ARS by sending a route set-up message to the proxy
ARS. The proxy ARS then consults its routing table, and
relays the route set-up message to a gateway ARS through
other intermediate ARSs. The gateway ARS then requests
for a DCH to be allocated between itself and a BTS, and upon
a success, sends an acknowledgment back to the MH to complete the set-up of a relaying route. As soon as the MH gets
the acknowledgment, it can switch over to the R-interface
and start data transmission over the relaying route.

zysg;;sd

position

a . Coverage and position

of additional ARSs
grown from IIK
s e d ARS

Figure 5: Simulation Environment

ployed. Such performance improvement stems largely from
the ability of the proposed system to balance loads among
cell. We will assume that, by default, the number of DCHs
in a cell, or equivalently, the number of concurrent calls that
can be supported in a cell without relaying) is A4 = 100.
Also by default, the radius of a cell and a ARS are R = 2km
and r = 500m, respectively. For simplicity, we assume that
an ARS may relay as many calls as needed and a gateway
ARS relaying m calls will be assigned m,channels for use
on the C-interface.

4.1 Analysis
Two main factors affecting the performance are the probability of setting up a relaying route from a congested cell, say
A, to another cell, say Bi,where 1 <_ i <_ 6, and the availability of the DCHs in B,. In this subsection, we focus on
the former.
Suppose that the ARSs are randomly distributed in each
cell, then the probability that a MH in a cell is covered by
ARSs, p, is given by
=

the coverage of ARSs in the cell

th,e coverage of the cell.

(3)

If we consider the situation where only the seed ARSs are

placed at the border of two cells, the coverage of the ARSs
in any given cell is
= 37rr2.Thus, by treating each cell
as a circle, we have

4 Performance Evaluation
Now, lets analyze the probability of being able to set up
a relaying route via primary and secondary relay.
Primary Relay: Recall that a primary relay route can be established from the MH involved in a new call in cell A to
a neighbor cell Bi as long as the MH is covered by a seed
ARS. Accordingly, the probability of finding a primary relaying route, Pprimary,is the same as p . With the default
values of r and R being 500m and R = 2000m, respectively, Pprim,aryis about 19% (in case we only have the
seed ARSs). This probability will increase when additional
ARSs are placed in cell A.
Secondary Relay: Recall that a secondary relay route can be
established as long as one of the active MHs in cell A is covered by a ARS (and thus can be relayed to another cell and its
channel on the C-interface released). If m active MHs are in

In this section, we evaluate the proposed iCAR system via

analysis and simulation. In particular, we consider a system adopting the seed-growing approach mentioned earlier,
whereby a seed ARS is placed at every border of two adjacent cells, and thus can communicate with the two BTSs in
the cells on the C- interface (see Figure 5 for an illustration).
Each of the additional ARSs will grow from the seeds to
form a cluster involving at least one seed, or in other words,
any additional ARSs can communicate with at least one seed
via relay on the R-interface (also see Figure 5).
We evaluate the performance improvement of the proposed systems over a conventional cellular system (without
ARSs) in terms of the reduced new call blocking probability, and increased number of calls that can be simultaneously
supported in a cell as a function of the number of ARSs de-

158

cell A (where m is likely to be M = 100 since cell A is congested), the (conditional) probability of finding at least one
secondary relaying route (assuming that a primary relaying
route cannot be found) is given by
PSecondary = (l-P,,i,,,,).[l-(l-p)ml

= (l-P)[l-(l-P)ml

Obviously, when M is large enough (e.g., when M =

In addition, the probability of setting up a relaying route via either primary or secondary relaying is P P r i m a r y P S , o c o n d a r y r which will approach 1, meaning that we can relay a new call from cell A
to B, with a extremely high probability of success.
Note that, the new calls relayed to Bi can be supported if
B, has at least one available channel. If this is not the case,
then cascaded relaying may be used to find a relaying route
from an active MH in Bi to Cj . The probability of finding a
relaying route from an active MH in Bi to Cj can be determined in a similar way in which P S e c o n d a r y was determined,
except that if there are m active MHs in Bi, then this probability is a half of P S e c o n d a r y ( m ) since each B, can relay to
only three neighboring Cjs (e.g., when i = 1,j = 1 , 2 , 1 2 )
whereas cell A can relay to six neighboring Bis.
As for the call blocking probability in cell A, exact analysis will be difficult if not impossible to obtain, since it will
depend on the traffic load distribution among the neighboring cells. Accordingly, we will rely on simulation.

loo), PSecondo.ry approaches (1- p ) .

4.2

call is blocked or not). We repeat each experiment for several hundreds of times to compute the average value.
The simulation results are plotted in Figure 6 and Figure 7. As can be seen from Figure 6, the number of concurrent new (or additional) calls which can be supported at
any given time increases almost linearly with the number of
ARSs in cell A using either primary relay only or primary
and secondary relay. The latter results in between 30 and 75
more new calls when the number of ARSs increases from
6 to 25 (because more existing MHs in cell A are covered
by the ARSs, hence can assist in secondary relay). When
the number of ARSs in cell A is equal to 25, the number of
new calls that can be supported reaches 375 and 450, respectively, which means that the capacity of cell A has been increased effectively by 3.75 and 4.5 times. In addition, Figure 7 shows that with primary relay only, the blocking probability decreases from 100% to 79% when there are only six
seed ARSs. This means that the probability a call can be
relayed to a neighbor cell is 21%, which is close to the analytical result presented earlier (which is 19%). The blocking probability will further be reduced to 25% when there are
19 additional ARSs. Furthermore, Figure 7 shows that with
both primary and secondary relay, the blocking probability
is near 0, which again agrees with our analysis.

Simulation

In this subsection, we present simulation results using the default values of the parameters M , T and R. The following
scenarios are considered where, unlike in Theorem 1, we assume that the existing calls will last forever, that is, calls will
have an infinite holding time, and thus the traffic is incremental.

4.2.1 Scenario 1: Primary and Secondary Relay

In the first scenario, we consider a subsystem consisting of
seven cells A and Bi ( 1 <_ i 5 6 ) as shown in Figure 5.
Assume that cell A is currently congested (i.e., it already has
M = 100 calls, and thus there is no free DCHs available in
A), while cells Bis are not (in fact, we assume that as many
DCHs as needed are available in each Bi). Seed ARSs are
placed at the borders between A and 6,&
as described earlier. Additional ARSs grow from the seeds within cell A,
and the total number of U S S in cell A will vary between 6
(with 0 additional ARS) and 25 (with 19 additional ARSs).
There are two metrics being considered : (1) the number
of additional calls which can be supported simultaneously
in cell A, and (2) new call blocking probability in cell A.
For the first metric, 500 new calls are generated such that
the MHs involved in these calls (one MH per call) are randomly located within cell A. We determine the number of
these new calls which can be supported (i.e., relayed to one
of the neighboring cell B,). To obtain the second metric, a
new call is generated such that the MH involved is randomly
located within cell A, and we determine whether the new call
can be relayed to a neighboringcell Bi (i.e., whether the new

Figure 6: Supported new calls in cell A as a function of the

number of ARSs in cell A

4.2.2 Scenario 2: Cascaded Relay with Location Dependent Traffic Load

In this scenario, we consider a subsystem with 19 cells as
shown in Figure 5(a). 24 seed ARSs are placed at the borders between A and B1-6, and between B1--6 and Cl,la.
Additional ARSs (0 to 133) are grown from these seeds
within A and B1-6 (and accordingly, the average number of
additional ARSs in each of these cells varies from 0 to 19,
and the total number of ARSs in the 19 cells varies from 24
to 157).
In real networks, traffic load is likely to be locationdependent. More specifically, the cells closer to a hot spot
(congested cell) are more likely to experience a heavier traffic load than those further from the hot spot. A common assumption is that if the traffic load in the hot spot cell is t H ,

159

Figure 7: New call blocking probability in cell A as a function of the number of ARSs in cell A

Figure 8: Supported new calls in cell A as a function of the

number of ARSs in cell A and B1-0

the traffic load in a cell d cells away is t = t H - d x k ,

where k is a constant scaling factor (0 5 k < 1). In this scenario, we assume that the traffic load in cell A (the hot spot)
is around t A = 0.94(+0.04) (i.e., there are 90 98 existing calls), and k = 0.4. Thus, the number of existing calls
in each B,(15 i 5 6) is about 54 (since d = l),and that in
each Cj(1 5 j 5 12) is about 14 (since d = 2).
A major difference between this scenario and Scenario 1
is that here, a new call (the first in addition to the 94 existing calls) in cell A can always be supported in cell A without
relay, and hence we will focus only on the number of concurrent new calls involving MHs in cell A that can be supported. Although the respective number of ARSs and existing MHs in cell A, as well as the number of new calls
to be generated concurrently in simulation, in this scenario
are about the same as those in Scenario 1, we expect that the
number of new calls that can be supported via primary and
secondary relay will be different. This is because in this scenario, even though a MH involved in a new call has a relaying route (via primary or secondary relay) to Bi,the new call
may be rejected by Bi because Bi has reached its capacity.
In fact, the maximum number of new calls that can be supported via primary and secondary relay only cannot exceed
6 + 6. 46 (which is about 280). Of course, additional calls
can be supported via cascaded relay, and the maximum number of which depends on how many existing MHs in each Bi
(out of the 54 MHs) can be relayed to one of its three neighboring cells Cj.
As can be seen from the results shown in Figure 8, initially,
new calls can be supported via primary and secondary relay
only, and cascaded relay is not needed. The maximal number of new calls that can be supported via primary relay only,
however, is about 225, which is reached when the average
number of additional ARSs in cell A is about 19. Note that,
this number is lower than that shown in Figure 6 for scenario
1, even though in both cases, about 375 MHs involved in
the 500 new calls are covered by a A R S in cell A. This is because in this scenario, some of the MHs (about 150) have
relaying routes to a congested Bi only. Our simulation results have also shown that, with about 19 additional ARSs

in cell A, only about 50 out of about 94 existing MHs in the

cell can find a relaying route to a non-congested B,, although
75 out of 100 existing MHs has a relaying route to a B, as
shown in Figure 6 for Scenario 1 (where B,s will never be
congested). Finally, the results show that cascaded relay becomes useful as more and more new calls are supported by
B,s (i.e., their loads increase). In particular, when B,s are
more or less congested, cascaded relay can support about 140
additional new calls (because that number of existing MHs
in BZscan be relayed to C,s). Overall, the number of new
calls which can be supported in cell A will increase to about
415.

Related Work

In this section, we describe a few related works. [lo] presented a hierarchical structure for wireless mobile systems
with a fixed backbone. In order to access the backbone, all
MHs have to go through a Mobile Base Station (which can
be thought of as a cluster head). It is similar to the integrated
cellular and Ad-hoc relay system proposed here in that the
cellular infrastructure we have considered is fixed, and the
ARSs can be mobile and used to relay between MHs and
the fixed BTSs. However, in the proposed architecture, the
MHs have two air (or radio) interfaces so that they may communicate with BTSs directly without going through ARSs.
In addition, each ARS is under the control of a MSC, and has
limited mobility. Such a feature is important to ensure that a
relaying route can be set up fast and maintained with a high
degree of stability. Routing in the proposed next generation
wireless system is similar to that of having a hybrid (both hierarchical and flat) structure in [ l l ] for efficient routing and
handoffs in mobile ATM networks. The difference between
the two is that in the latter, path extension (or relay) is between two (fixed) BTSs through direct wired links.
In multihop cellular systems [12], relaying is performed
by MHs, and thus that approach shares many disadvantages
in terms of security (authentication, privacy), billing, and
mobility management (of the MHs) as discussed earlier in
Sec. 2. In addition, the main goal of the multihop cellular

160

systems is to reduce the number of BTSs or the transmission power of each BTS, but it can no longer guarantee a full
coverage of the area. In fact, even in the ideal case where every MH in an area uncovered by any BTS can find a relaying
route (through other MHs), the multihop approach will neither increase the system capacity nor decrease the call blockingldropping rate, unless a large percentage of the calls are
intra-cell calls (i.e., calls whose source and destination are in
the same cell), which usually is not the case in practice.

[9] C.Qiao, H. Wu, and 0.Tonguz, Load balancing via relay in next generation wireless systems, in Proceeding of IEEE Mobile Ad Hoc Networking & Computing,
2000. To appear.
[lo] LEAkyildiz, W. Yen, and B. Yener, A new hierarchical routing protocol for dynamic multihop wireless networks, in IEEE INFOCOM97,pp. 1422-1429,1997.
[ 111 S. Maloo and C. Qiao, Efficient routing and fast hand-

off in a mobile atm network with a hybrid topology, in

5th Intl Conf on Info Systems Analysis and Synthesis
(ISAS),vol. 4, pp. 614-621, 1999.

6 Conclusion
We have presented a novel architecture for next generation
wireless systems, which integrates the traditional cellular
and modern relay technologies. The basic idea is to place a
number of Ad-hoc relaying stations (ARS) in a cellular system to divert traffic in one (possibly congested) cell to another cell. We have also discussed the number and placement
of ARSs required to cover the entire system, and presented
a seed-growing model to show, via both analysis and simulation, that only a limited number of ARSs is needed for
the proposed relaying scheme to be effective. More specifically, the new architecture can efficiently balance the traffic between cells in the cellular system dynamically, thus reducing the new call blocking probability and increasing the
number of calls that can be supported concurrently in a cell.

[I21 Y.D.Lin and Y.C.Hsu, Multihop cellular: A new architecture for wireless communication, in IEEE INFOCOM2000, pp. 1273-1282,2000.

A The Proof for Theorem 1

Theorem 1 Assume that the total trafJic in a n-cell system is
T Erlangs, then the (system wide) call blocking probability
is minimized when the trafJic in each cell is T / n Erlangs.

Proof : Let the number of DCHs in each cell be M and

assume that the traffic intensity is Ti in each cell i where T =
Cy.lTi. The probability of all the channels in cell i being

References

busy is given by the following Erlang B formula.

[ 11 T. Rappaport, Wireless Communications Principle and

B , ( M ; T z )=

Practice. Prentice Hall, 1996.

[2] V.Garg and J. Wilkes, Wireless and Personal Communications Systems. Prentice Hall, 1996.

TzM/M!
T,Z/i!

For the n-cell system, the average blocking probability for

the entire system is

[3] C. Toh, Wireless ATM and Ad-Hoc Networks: Protocols and Architectures. Kluwer Academic Publishers,
1996.

T,, we may write T, = T - E:; T,.

Since T =
In other words, there are only n - 1 independent Tzs. In
order to compute the minimum value of B , we compute all
the partial derivatives of B and set them to be 0, that is,

[4] C. Perkins and E. Royer, Ad-hoc on demand distance

vector routing, in Proceedings of IEEE WMCSA99,

pp. 90-100,1999.

[5] D. Johnson and D. Maltz, Dynamic source routing in

ad hoc wireless networks, Mobile Computing, vol. 5 ,

(7)

pp. 153-181,1996.
[6] Z. Haas and B. Liang, Ad hoc mobility management
with uniform quorum systems, IEEE/ACM Transactions on Networking, vol. 7, no. 2, pp. 228-240,1999.

We omit the details but we can obtain the critical points (or
the solutions to the above equations) as

[7] L. Chen, H. Murata, S. Yoshida, and S. Hirose, Wireless dynamic channel assignment performance under
packet data traffic, IEEE Journal on Selected Areas in
Communications,vol. 17, no. 7, pp. 1257-1269,1999.

Ti = Tz = ... = T, = - = T

T
n

(8)

By computing the second order partial derivatives of B

which forms a matrix, and verifying that its determinant is
larger than 0 at the above critical points, we have shown that
the blocking probability reaches its minimum value when the
traffic is evenly distributed.

[8] J.-L. Pan, S. Rappaport, and P. Djuric, Multibeam cellular communication systems with dynamic channel assignment across multiple sectors, Wireless Networks,
vol. 5 , no. 4, pp. 231-243, 1999.

161