A dynamic channel assignment simulation system for large scale cellular

telecommunications
P.M.Papazoglou1, D.A.Karras2, R.C.Papademetriou3

Abstract - The rapid evolution of cellular technology and the augmentative user demand for advanced mobile services leads
the industry to develop more efficient network structures. The increasing number of cellular users and the demand for
broadband mobile communications (3rd and 4th generation) drives to the research of new methodologies for the design of
cellular networks and services. Simulation environments more than ever offer the opportunity to develop and study with low
cost new structures and methods for the implementation of new services. This paper thoroughly reviews centralized and
distributed DCA schemes in terms of simulation characteristics and needs and proposes a comprehensive simulation system
incorporating their principles, entities and concepts involved. The proposed system is implemented in Java, in order to
create a high performance generic simulation environment with the future capability of internetworking. The system is
designed with the goal to be efficient in simulating large scale generic cellular telecommunication systems. Moreover,
another goal of the proposed system is to serve as a test bed for the evaluation and development of DCA schemes
(especially for educational purposes) involved in such large scale cellular systems towards their planning as effective
cellular mobile radio networks.

Keywords - cellular network, simulation model, channel allocation scheme, simulation environment

Introduction Many approaches to simulation systems have been made

The concept of cellular networks using various simulation languages. Languages such as
Simula, Parsec[10] and other of special purpose offer high
The cellular principle divides the covered geographical area flexibility for simulation systems development but
into a set of smaller service areas called cells. Each cell has introduce lack of portability and lag in terms of general
a base station and a number of mobile terminals (e.g. purpose optimizations, etc.
mobile phone, palms, laptops, or other mobile devices). The Other software tools like Matlab, have a plethora of
base station is equipped with radio transmission and libraries and capabilities for flexible script code
reception equipment. The mobile terminal within a cell development but suffer from low performance execution.
communicates through wireless links with the base station A Large scale simulation system requires a high
associated with the cell. A number of base stations are performance environment with advanced characteristics
connected to the Base Station Controller (BSC) via such as portability, internetworking, etc.
microwave links or dedicated leased lines. The BSC The invention of new simulation languages introduce
contains logic for radio resource management of the base important drawbacks which are:
stations under its control. It is also responsible for
transferring an ongoing call from one base station to • New languages are domain-specific and are rarely
another as a mobile user moves from cell to cell. In order to adapted by scientific community
establish a communication with a base station, a mobile • The corresponding libraries impose the designers to adapt
terminal must first obtain a channel from the base station. A their applications in specific requirements
channel consists of a pair of frequencies: one frequency (the • Designers can not achieve high adaptability of the
forward link/downlink) for transmission from the base simulation kernel in their applications
station to the mobile terminal, and another frequency (the
reverse link/uplink) for the transmission in the reverse Java language is the most suitable for building flexible,
direction. An allocated channel is released under two portable and high performance network applications and is
scenarios: the user completes the call or the mobile user adapted by the majority of scientific community.
moves to another cell before the call is completed.
With this simulation program, we attempt to build a generic
The need of Simulation systems in cellular networks simulation system for cellular communication systems of
3rd towards 4th generation in order to perform simulations
The performance and the behavior of a real cellular network not in ad-hoc systems as most software tools do. With a
can be evaluated using simulation systems without the need generic simulation system, clear and general conclusions
to perform field experiments and develop prototypes. can be extracted compared to ad-hoc wireless networks.
The simulation solutions give us the opportunity to develop
custom protocols, channel allocation schemes, network
structures, etc, towards to a desired cellular network.
Channel Allocation & GoS consideration
Due to the complexity of real cellular networks the
The capacity of a cellular system can be described in terms
simulation software development strategy becomes a very
of the number of available channels, or the number of users
important factor that influences the resulted network model.
the system can support. The total number of channels made
The structure of the simulation environment affects the
available to a system depends on the allocated spectrum and
performance of simulated cellular network and for that
the bandwidth of each channel. The available frequency
reason we try to study the design and the development of
spectrum is limited and the number of mobile users is
such system.
increasing day by day, hence the channels must be reused as
The authors are with the (1) Technological Educational Institute of Lamia,
Greece and ECE Dept., University of Portsmouth, UK and (2) Chalkis much as possible to increase the system capacity. The
Institute of Technology, Greece and Hellenic Aerospace Industry, Greece assignment of channels to cells or mobile is one of the
and (3) ECE Dept., University of Portsmouth, UK (e_mails: fundamental resource management issues in a mobile
[email protected], [email protected])

1
communication system. The role of a channel assignment When the cell channels are divided in two groups –
scheme is to allocate channels to cells or mobiles in such a standard channels and borrowable channels – we have the
way as to minimize: a) the probability that the incoming Simple Complex Channel Borrowing Scheme (SCCB). In
calls are dropped, b) the probability that ongoing calls are Borrowing with Channel Ordering (BCO) we use channel
dropped, and c) the probability that the carrier-to- priorities to define the borrowing order. Other FCA
interference ratio of any call falls below a prespecified variations are Sharing with Bias (SHB), channel
value. In literature, many channel assignment schemes have Assignment with Borrowing and Reassignment (CABR),
been widely investigated with a goal to maximize the and Ordered Dynamic Channel Assignment with
frequency reuse. The channel assignment schemes in Rearrangement (ODCA) [8].
general can be classified into three strategies: Fixed Non borrowing schemes are mostly based on hand-of
Channel Assignment (FCA) [1,2], Dynamic Channel strategies where an available channel is difficult to locate.
Assignment (DCA) [1,3], and the Hybrid Channel
Assignment (HCA) [1]. In FCA, a set of channels are DCA variations
permanently allocated to each cell based on a pre-estimated
traffic intensity. In DCA, there is no permanent allocation As we mentioned before, centralized and distributed
of channels to cells. Rather, the entire set of available versions are the main variations of the DCA scheme.
channels is accessible to all the cells, and the channels are
assigned on a call-by-call basis in a dynamic manner. One Centralized DCA schemes
of the objectives in DCA is to develop a channel
assignment strategy, which minimizes the total number of The First Available (FA) is the simplest method where the
blocked calls. The FCA scheme is simple but does not adapt first available channel within the reuse distance encountered
to changing traffic conditions and user distribution. during a channel search is assigned to the new call. Future
Moreover, the frequency planning becomes more difficult blocking probability is a very important factor for the
in a microcellular environment as it is based on the accurate channel selection in a new call. Locally Optimized
knowledge of traffic and interference conditions. These Dynamic Assignment (LODA) examines this probability in
deficiencies are overcome by DCA but FCA outperforms order to assign a new channel. The Mean Square (MSQ)
most known DCA schemes under heavy load conditions[2]. scheme selects an available channel that minimizes the
To overcome the drawbacks of FCA and DCA, HCA was MSQ between cells that use the same channel.
proposed by Kahwa et al.[1], which combines the features
of both FCA and DCA techniques. DCA schemes can be Distributed DCA schemes
implemented as centralized or distributed. In the centralized
approach[1] all requests for channel allocation are The distributed DCA schemes are based on signal quality
forwarded to a channel controller that has access to system computations around the initiated calls. More precisely, are
wide channel usage information. The central controller then based on co-channel distance, signal strength measurement
assigns the channel by maintaining the required signal and signal to noise interference ratio. Locally Optimized
quality. I distributed DCA [3], the decision regarding the Least Interference DCA algorithm (LOLIA) [6] is a known
channel acquisition and release is taken by the concerned variation and computes the interference among users that
base station on the basis of the information from the use the same channel and examines the signal quality in
surrounding cells. As the decision is not based on the global comparison with a defined threshold.
status of the network, it can achieve only suboptimal
allocation compared to the centralized DCA and may cause One of the most important characteristic of a cellular
forced termination of ongoing calls. network is that a certain number of calls will fail to
establish an initial connection. We can state two measures
Many channel allocation schemes have been proposed. for dealing with unsuccessful call attempts: the never
There are many variations of the FCA, DCA and HCA serviced new calls, and the delayed successful calls (waiting
schemes. for initial connection).

FCA variations The service quality in the busy period of the day may be
expressed as the grade of service [5] or. Two requirements
Channel borrowing and non borrowing schemes can be set in order to measure GoS level or Quality of
Service [9]:
In channel borrowing schemes [7], a cell can borrow a
channel from an adjacent cell only if there is no free • GoS, loss system : probability of blocking < x%
channel in initial cell. After the use of the borrowed or congestion < x%
channel, the channel returns to its initial cell. Known
schemes in channel borrowing category [7] are the Simple • GoS, delay system : probability (waiting time > z
borrowing (SB), Borrow from the Richest (BFR) and sec) < x%
Borrow First Available (BFA) scheme. In SB, after the
initial channel assignment in each cell if there is no free Another quality metric can be the signal to noise
channel for a new call, the needed channel can be borrowed interference C/(N+I) ratio compared to a predefined
from a neighboring cell. In the case of the BFR scheme, the threshold.
available channel is borrowed from a neighboring cell that
holds the greatest number of free channels (for borrowing) The simulation system
compared to other neighboring cells. Without any
optimization when borrowing, BFA scheme is the simplest Our system is a generic simulation system for cellular
because the borrowing is based on the first available communications of 3rd and towards 4th generation in order
channel [7]. to perform simulations not in ad-hoc wireless networks.

2
The number of communication channels, the number of
users and other simulation parameters can be configured
using a special configuration text file. The supported
channel allocation schemes for a given cellular network are
distributed DCA and centralized DCA. As we mentioned in
previous section, the whole program was built in Java in
order to have the capability of future extensions in any
constrains.
In order to build a sufficient simulation system it is
necessary to take in consideration some basic aspects of the
operation of the cellular network which are:

User characteristics - user attributes including connection

status, current position, call holding time, allocated channel,
communication signal strength and interference (if Figure 2 - cell mesh (possible user positions in a cell)
necessary), etc.
Network parameters - Cells number, channels per cell, cell According to user conditions, the simulation program saves
positions, base station positions, etc. the relevant information in the user attribute table. These
Traffic and QoS - new call arrival schemes, user movement attributes summarize the user conditions and updated in
profile, call dropping conditions, hand-of conditions, etc. every user new state (Table 1).
Channel allocation schemes – channel allocation strategy
according to network conditions, congestions cases, User attribute Description
resources limitations, etc. X position Current X position in the
network
Our simulation system consists of one main module - the Y position Current Y position in the
simulation module, and nine additional modules. The network
additional modules are used for the necessary computations,
Connection status 0=no connected, 1=connected
the network model structure and simulation program
Call holding time Call termination time (sec)
execution.
Allocated channel Allocated channel (if
connected)
DCA schemes that are used in our simulation system
Path loss Path loss of the user (if
In distributed DCA, the call management is based on the connected)
Table 1 - user attributes
desired Carrier to noise interference C/(N+I) ratio. For the
centralized DCA scheme the first available channel is The simulation takes place during the simulation time. In
allocated to the new user in the cell only if the channel is this time, several procedures are executed:
not used in any other adjacent cell.
call termination check
The network structure reallocation check
new call arrival
The cellular network consists of 19 cells and each of the user movement
cells includes a base station in the center position (fig.1).
Around the 19 cells there are some copied cells (from real There are also additional procedures for the normal
cells) in order to generate the expected behavior when users operation of the simulation program, like configuration file
exist in boundary cells (1 to 19). processing, initialization, etc. The basic parameters of
simulation exist in a text configuration file that is read by
the program before simulation begins. The containing
information of the configuration file includes:

CNR on cell edge (dB), CNIR threshold (dB), average call

arrival rate (times/hour,lambda), average call holding time
(second),
Simulation time step, end simulation time, number of
channels per cell, path loss factor(alpha), standard deviation
of shadowing( sigma), cellmesh fineness, max users per
cell, Display options, network dimensions information, user
movement type(random, Gaussian, etc), etc

Call termination. The program uses an exponential function

Figure 1 - structure of cellular network model that generates the call duration for each new user. The call
holding time is added to current simulation time for later
Each cell is divided in a mesh of spots where the users may examination. The associated procedure scans the network
exist from the beginning of the call or due to hand-of cells and searches for any connected users and examines the
situation. The number of spots per cell mesh can be defined progressive call time. If this time is expired the user
from the simulation configuration. The basic parameter for attribute table is updated with the new connection status of
the mesh construction is the fineness which means the the particular user both in distributed and centralized DCA.
distance between the spots (fig.2)

3
In addition, in centralized DCA the new free channel after the new user and the base station. Channel allocation is
the call termination returns in the channel pool. successful if a free channel is found in the cell and the
Figure 3, shows the logic diagram of that procedure. C/(N+I) constraint is fulfilled (fig.4).

After a new call arrival, several actions take place in turn

(fig. 4a) :

a) Search for unconnected users in each cell

b) Calculate a random user position in the mesh
c) Place the new user according to cell’s base station
position and mesh spot
d) Calculate the signal strength between base station and
new user
σ ⋅n
Firstly we obtain the shadow attenuation, sh = 10 10 ,
where σ is the standard deviation of shadowing and n is a
number from the normal distribution. Using the shadow
attenuation and distance between user and base station, we
can derive the distance attenuation dw. Finally, we
calculate the CNIR (Carrier to Noise Interference Ratio)
cnedge
between user and base station cn = 10 10 ⋅ dw , where
cnedge is the CNR on cell edge (dB).
e) Scan the other cells and calculate interference among
new user and other users that use the same channel
Figure 3a Figure 3b f) Check if C/(N+I) ratio is acceptable according to
call termination procedure Reallocation check predefined threshold
(distributed DCA)
g) If C/(N+I) is acceptable, establish the new call and
update the table of user attributes.
Reallocation check (Distributed DCA )
New call arrival and channel allocation (centralized DCA)
The reallocation procedure searches all the cells for
connected users and performs check for each user. The
A new call arrival occurs under the same conditions in
computations are based on signal strength and how is
distributed DCA. After the calculation of new user position,
affected from other connected users in neighbor cells. If a
the algorithm searches for an available channel that is not
user signal does not fulfill the C/(N+I) threshold (Carrier to
used in current cell or in any adjacent cell. The search is
Noise and Interference ratio) the procedure tries to find
performed using the channel pool that keeps information for
another appropriate channel. The Carrier to Noise and
channel usage all over the cellular network. The channel
Interference ratio can be derived using the type
allocation criterion is based only in the channel usage in
ξ0 adjacent cells. This method requires less processing power
A P0 d 0− α 1 0 1 0 than the method in distributed DCA because is not based to
R cni = n ξi mathematical calculations.
N + ∑
i −1
A Pi d i− α 1 0 1 0
User movement (distributed & Centralized DCA)
Where n is the number of base stations and users, di is the
The simulation program supports two types of user
distance between user and base station, ξi is the distortion movement. The first type of user movement is based on
due to shadowing from user to base station, A is a
random number generation like new call arrival and the
proportional coefficient and P is the transmitted power. second on Gaussian distribution. The algorithm locates the
Firstly, the algorithm calculates the signal strength between
connected users and changes their current positions. The
user and base station and in later time calculates any reallocation procedure in distributed DCA checks the new
interference from around connected users. If an accepted
user positions and performs new channel assignment if
channel is found, is allocated to the new user, otherwise the needed. In centralized DCA, after user movement, the user
call is dropped. Figure 3b, illustrates the logical diagram of
is disconnected and the algorithm searches for a proper free
this algorithm. channel from the central channel pool.
New call arrival and channel allocation (distributed DCA) The Gaussian distribution density function
1
e −( x − µ ) / 2σ , where σ is the standard
2 2
New calls are resulted from a random distribution in
is f ( x) =
combination with the allowed number of calls per
simulation time step period. The algorithm searches all the
σ 2π
cells for not connected users (according to user attribute deviation and µ is the mean value. The Gaussian
table). If proper conditions exist for a new call arrival, a distribution in our program has mean value zero and
random position in the cell mesh is calculated. The user standard deviation one. Hence, the density function
position attribute is updated with the mesh spot position in becomes
combination with the position of the base station in the
center of the cell. The signal strength is calculated between

4
 1 − x2 / 2
f ( x) = e . In the user movement procedure,
2π
firstly we generate a Gaussian number e.g. x1 and the
corresponding f(x1) when x1 ∈ [ −0.5, 0.5] . If another
Gaussian number e,g x2, where x2 ≥ f ( x1 ) a user
move is generated.

Simulation results

The program produces the simulation results in text files

(.m files) for later process in matlab. Each result file
contains the simulation data and the appropriate matlab
commands for graph viewing and statistical analysis.
The simulation results include data for calculating
important measures such as blocking probability, dropping
probability, channel reuse, cell congestion, number of
reallocations, channel reuse per cell, etc.
The generated matlab files include additional command for
calculating some statistical metrics of simulation results like
Standard deviation, Covariance and Average value.

Example results

As we mentioned before, the simulation program produces

a variety of graphs. Assume that we simulate a cellular
network. The network uses the distributed DCA scheme.
Current network consists of 19 cells with capacity 50 users Figure 4a Figure 4b
per cell and 5 channels per cell. New call arrival New call arrival
Figure 5 shows the channel reuse for distributed DCA with (distributed DCA) (centralized DCA)
random call arrival and random movement. This DCA
algorithm assigns the first channel that fulfills the
CHANNEL REUSE,USERS=50
conditions. 1600

The blocking probability is one of the most important 1400

characteristics for the performance of a cellular network. 1200

When a new call arrival occurs and the network can not
allocate a channel then we say that this call is blocked. The
reuse num

1000

blocking probability Pbl is calculated from the

800
blocked calls
ratio Pbl = . Figure 6 illustrates the 600
call number
400
blocking probability graph.
200

0
1 2 3 4 5
channel
Figure 5 – channel reuse

5
 BLOCKING,USERS=50
0.7
Channel RE-USE per CELL
100
0.6
Blocking probability 90

0.5 80

Channel Reusability
70
0.4
60
0.3
50

0.2 40

30
0.1
20

0 10
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Sim time 0
0 2 4 6 8 10 12 14 16 18 20
Figure 6 - blocking probability
Cell Number
Figure 8 - channel reusability per cell
If the received power of each user is high enough, we can
make the assumption that the interference from other users
Number of Cell Congestions
can be ignored. Thus, we can compare the simulated 200
blocking probability with the theoretical which is 180

⎛ n − 1⎞
⎟ ( vh )
s 160

⎜
Num of Congestions
= s⎝
s ⎠ 140

Pblthe , where n is the number of

⎛ n − 1⎞
120

∑ ⎟ ( vh )
i
⎜ 100

i =0 ⎝ i ⎠ 80

users, s is the number of channels, v is the average call 60

arrival rate (for no connected user) and h is the average call 40
holding time.
20

The dropping probability is also an additional and very 0

0 2 4 6 8 10 12 14 16 18 20
important characteristic of the cellular network. When a call Cell Number
is in progress and the required quality conditions are not Figure 9 - number of congestions per cell
met then this call is obligatory driven to termination. The
dropping probability Pfc is calculated from the ratio DCA comparative results
forced call number
Pfc = We have simulated a cellular network of 19 cells capacity
call number − blocked call numbers with 10, 50, 80,100,200,400,750 users per cell and 5, 32
Next figure (fig. 7) shows this probability from the given channels per cell. The channel allocation schemes that have
example network. been used are Centralized DCA (random call arrival,
DROPPING,USERS=50 random user movement) and distributed DCA (random call
0.16
arrival and random/Gaussian use movement).
0.14 BLOCKING,channels per cell=5
1
Dropping probability

0.12 0.9

0.1 0.8
Blocking probability

0.08 0.7
DCA-crr,ch5,nm
DCA-drr,ch5,nm
0.6
0.06 DCA-drg,ch5,nm
DCA-crr,ch5,m
0.5
DCA-drr,ch5,m
0.04
DCA-drg,ch5,m
0.4
0.02
0.3
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 0.2
Sim time
0.1
Figure 7 - Dropping probability 0 100 200 300 400 500 600 700 800
USERS
Other graphs include the number of reallocation, forced Figure 10. Blocking probability of various channel assignment schemes (5
channels per cell)
calls and channel reusability during the simulation time.
Finally, apart from the statistical matlab commands that the
simulation program produces there are two more graphs Figure 10, shows that centralized DCA (DCA crr - random
that show the channel reusability per cell and the number of call arrival, random move) with user movement or not have
congestions per cell respectively (fig. 8 & 9). higher blocking probability than distributed DCA (DCA
drr/drg random/Gaussian movement) which means that

6
distributed DCA is more suitable for network with limited Conclusions & Future work
number of cells.
There is no doubt that cellular network simulation is a very
1
BLOCKING,channels per cell=32 complicated procedure. From the point of view of network
designer the development and research of various channel
0.9
allocation schemes must be faced from the lowest starting
0.8 point. The type and the number of parameters that the
Blocking probability

0.7 designer should take in consideration depend on the type of

0.6
the network and the focused procedure inside the network.
The divergence between simulation results and real network
0.5
behavior is affected by the structure of the implementation
0.4 DCA-crr,ch32,nm algorithms and the corresponding research of real network
DCA-drr,ch32,nm
0.3
DCA-drg,ch32,nm
conditions like new call arrival schemes, user movement,
0.2
DCA-crr,ch32,m etc. A deeper research requires full customizable software
DCA-drr,ch32,m
DCA-drg,ch32,m with high level of adaptation in real network specifications
0.1
and behavior.
0
0 100 200 300 400 500 600 700 800 In this paper we have presented a generic simulation
USERS environment for GSM cellular networks. The whole
program was developed from scratch using the Java
Figure 11. Blocking probability of various channel assignment schemes (32 Language.
channels per cell)
Our future goal is to develop a software tool for simulation
of large scale generic cellular networks in order to extract
Figure 11, confirms the previous note about the clear and reliable conclusions. With the internetworking
performance of Centralized DCA in small networks and capability of Java language, our simulation environment
shows that even with more available channels per cell the will be used for educational purposes too.
Centralized DCA is not satisfactory for small networks. The In order to develop an efficient and useful simulation
above figures also show that the user movement and user environment we have to study the evolution, the
movement distribution plays important role for the network specifications and the constrains of real cellular networks
behavior. and also the algorithm schemes that must used like multi-
threading for the user and network model, etc.
In Centralized DCA the number of cell congestions is
smaller in contrast with Distributed DCA (fig.12 & 13) and
References
that means that the network management is easiest is
centralized DCA.
[1] M.Zhang, and T. S. Yum, “Comparisons of
CELL CONGESTION,channels per cell=5
500 Channel Assignment Strategies in Cellular Mobile
450 Telephone Systems”, IEEE Transactions on Vehicular
Technology, vol.38,no.4,pp211-215,1989.
400
[2] W.K. Lai and G.C. Coghill, “Channel Assignment
350
Cell Congestion

DCA-crr,ch5,nm through Evolutionary Optimization”, IEEE Transactions on

DCA-drr,ch5,nm
300 DCA-drg,ch5,nm Vehicular Technology, vol.45,no.1,pp91-96,1996.
DCA-crr,ch5,m
250 DCA-drr,ch5,m
[3] L.J. Cimini and G.J. Foschini, “Distributed
200
DCA-drg,ch5,m Algorithms for Dynamic Channel Allocation in
Microcellular Systems”, IEEE Vehicular Technology
150
Conference, pp.641-644, 1992.
100 [4] S.Hurley, “Planning Effective Cellular Mobile
50 Radio Networks”, IEEE Transactions on Vehicular
0
Technology, vol.51,no.2,pp.254-253,2002.
0 100 200 300 400 500 600 700 800
[5] T.S. Rapaport, “Wireless Communications
USERS
Principles and Practice”, Prentice Hall, 2002
Figure 12 - cell congestions (5 channels)
[6] P.Cherriman, F.Romiti and L.Hanzo, “Channel
Allocation for Third-generation Mobile Radio Systems”,
CELL CONGESTION,channels per cell=32
500 ACTS’98, vol.1,pp.255-261,1998
450 [7] I.Katzela and M.Naghshineh, “Channel
400
assignment schemes for cellular mobile telecommunication
systems: A comprehensive survey”, IEEE Personal
350
Cell Congestion

Communications, pp.10-31,1996
300 [8] S.Jordan, “Resource Allocation in Wireless
250 Networks”, Journal of High Speed Networks,
200 DCA-crr,ch32,nm vol.5,no.1,pp.23-24,1996
DCA-drr,ch32,nm [9] M.Salamah and H.Lababidi, “Dynamically
150 DCA-drg,ch32,nm
DCA-crr,ch32,m adaptive channel reservation scheme for cellular networks”,
DCA-drr,ch32,m
100
DCA-drg,ch32,m
Elsevier Computer networks, article in press, 2005
50 [10] R.L.Banrodia, Et al, “Parsec: A parallel simulation
0
environment for complex systems”, IEEE Computer,
0 100 200 300 400 500 600 700 800
21(10):77-85, Oct.1998
USERS
Figure 13 - cell congestions (32 channels)