Hierarchical Genetic Algorithm For Dynamic Channel Units Allocation in TD-CDMA/TDD System

International Journal of Wireless & Mobile Networks (IJWMN), Vol 1, No 2, November 2009
103
HIERARCHICAL GENETIC ALGORITHM FOR
DYNAMIC CHANNEL UNITS ALLOCATION IN TD-
CDMA/TDD SYSTEM
Kuo-Ming Hung12
, Yao-Tien Wang1
and Ching-Tang Hsieh2
1
Department of Information Management, Kainan University, Taoyuan, Taiwan
2
Department of Electrical Engineering, TamKang University,Tamsui Taiwan
hkming@mail.knu.edu.tw
ABSTRACT
Hierarchical Genetic Algorithms (HGA) as a tool for search and optimizing methodology have now
reached a mature stage. The minimum resource facility to carry user traffic, termed a channel unit (CU),
is composed of a one time-slot and one code in the TD-CDMA/TDD system. The control of the number of
CUs depends on the traffic load solves varied and asymmetrical traffic problems in the 3G system. In a
cellular network, the call arrival rate, call duration and the communication overhead between the base
stations and the control center are vague and uncertain, regardless of whether the criteria of concern are
nonlinear, constrained, discrete or NP hard. In this paper, the HGA is used to tackle the neural network
(NN) topology as well as the fuzzy logic controller for the dynamic CU allocation scheme in wireless
cellular networks. Therefore, we propose a new efficient HGA CUs Allocation (HGACA) in cellular
networks. It aims to efficiently satisfy the diverse quality-of-service (QoS) requirements of multimedia
traffic. The results show our algorithm has a lower blocking rate, lower dropping rate, less update
overhead, and shorter channel-acquisition delay than previous methods.
KEYWORDS
Dynamic channel unit allocation, load balancing, hierarchical genetic algorithms, neural-fuzzy
controllers, TDD-CDMA/TDD, wireless networks, radio resource management.
1. INTRODUCTION
Code Division Multiple Access (CDMA) can be categorized as frequency division duplex
(FDD) and time division duplex (TDD) modes. CDMA in the TDD (TDD-CDMA) mode will
be based on the harmonization between UMTS Terrestrial Radio Access in the TDD mode
(UTRA TDD) and Time Division-Synchronous CDMA (TD-SCDMA). There have been several
proposals for supporting real-time and multimedia application services and so on [1-3]. TDD-
CDMA uses a combined time division and code division multiple access schemes, therefore the
signals of different users separated in both the time and code domains TDD mode can flexibly
cope with the traffic asymmetry by changing the number of time-slots allocated to the downlink
(DL) and uplink (UL). The minimum resource facility to carry user traffic, termed a channel
unit (CU), is composed of one time-slot and one code. Multiple CUs are allocated when the
traffic load exceeds one CU capability. The control of the number of CUs depending on the
traffic load solves varied and asymmetrical traffic problems in the TD-CDMA system [3,4].
This is one of the fundamental problems in the wireless cellular network. In fact, increasing the
bandwidth of a cell can increase the system capacity but not increase the efficiency in dealing
with the time varying imbalance in traffic.
There are two strategies for allocation of channels to cells [5-10]: Fixed Channel Allocation
(FCA) [10] and Dynamic Channel Allocation (DCA) [9,11-14]. The advantage of FCA is its
simplicity. It does not, however, reflect real scenarios where the load may vary from cell to cell.
DCA schemes can dynamically assign or reassign channels and are thus more flexible. In the

104
centralized DCA schemes [11-12,15], all channels are placed in a pool and assigned to the new
calls as required, and all the allocation jobs are performed by the control center. In the
distributed DCA schemes, BSs must be involved [16].
The channel allocation for load balancing usually uses some fixed threshold values to
distinguish the status of each cell [11-12]. A cell load is marked as “hot”, if the ratio of the
number of available channels to the total number of channels allocated to that cell is less than or
equal to some threshold value. Otherwise it is “cold”. The drawback is the threshold values are
fixed. Since the load state may display sharp distinction state levels, series fluctuation like the
ping-pong effect may occur when loads are around the threshold. This results in wasting a
significant amount of effort in transferring channels back and forth [11,12]. This is achieved by
efficiently transferring channels from lightly loaded cells (cold status) to heavily loaded ones (hot
status). While the great advantage of GAs is they find a solution through evolution [17-18], this
is also the biggest disadvantage. Evolution is inductive in nature. Life does not evolve towards a
good solution, rather it evolves away from bad circumstances, and search and optimizing are
very slow. This can cause a species to evolve into an evolutionary dead end. Likewise, GAs risk
finding a suboptimal solution, such as not always finding the exact solution but always finding
the best solution. [Editor: I don’t know if the previous sentence keeps your intended meaning.]
The load information collection can not only estimate the time-varying traffic load for the
cellular networks, but also provide useful information in making the channel-reallocation
decisions.
Traditional channel allocation approaches can be classified into update and search [19]. The
fundamental idea is a cell must consult all the interference cells (IN(C)) within the minimum
reuse distance before it can acquire a channel. Both approaches have advantages and
disadvantages. The update approach has a short acquisition delay but a higher message
complexity, while the search approach has a lower message complexity but a longer acquisition
delay. Due to this nature, using HGACA is the best way to approach the problem. The concept
of the fuzzy number plays a fundamental role in formulating quantitative fuzzy variables. The
fuzzy numbers represent the linguistic concepts, such as very hot, hot, and moderate etc [20-21].
The fuzzy expert system approach has also been applied to forecasting where the advantage of
an operator's expert knowledge is used. We adopt the number of available channels and cell
traffic load as the input variables for fuzzy sets and define a set of membership functions. In
addition, our scheme allows a requesting cell to borrow multiple CUs at a time, based on the
traffic loads of the cells and channel availability, thereby further reducing the channel allocation
overhead. Fig. 1 shows the block diagram of our HGACA.
Our HGACA consist of six modules: a fuzzy rule base, a fuzzy inference engine,
fuzzification, defuzzification modules, genetic algorithm, and neural networks. The HGACA
consists of cell load decision making, cell involved negotiation, and multi-CUs migration
phases. The structure of a dynamic channel borrowing for a wireless cellular network is
composed of three design phases by applying HGACA to them. The main purpose of a HGACA
is to obtain an optimal neural network topology, and neural network learning techniques to find
and tune the parameters. In this parameter learning phase, the possible parameters to be tuned
include those associated with membership functions such as the center, widths, and slope; the
parameters of the parameterized fuzzy connectives; and the weights of the fuzzy logic control
rules. The performance of our HGACA is compared with the simple borrowing [10,22], directed
retry [23], CBWL [24], and LBSB [11]. The experimental results reveal our proposed scheme
performs better than conventional dynamic channel allocation schemes. Our HGACA for CUs
algorithm not only effectively reduces the blocking rate and the dropping rate but also provides
significant improvement in overall performance, such as fewer update messages and shorter
CUs acquisition delay. The rest of this paper is organized as follows. In Section 2, we provide
the structure of the cellular system model and channel allocation strategy. The design issues of
our proposed HGACA wireless cellular system are in Section 3. The experimental results are
given in Section 4, and concluding remarks are given in Section 5.

105
Fig. 1: Block diagram of HGACA.
2. System Model and Channel Allocation Strategy
The universal mobile telecommunication system (UMTS) consists of the radio network
controller (RNC) that owns and controls the radio resources in its domain and the base stations
(BSs) connected to it. The RNC is the service access point for all services. The UMTS
Terrestrial RAN provides the core network (CN) and management of the connection to the user
equipment (UE). The concept also applies to the radio network controller in the next generation
of wireless cellular systems, and a BS directly communicating with all mobile stations (MSs) or
mobile equipment (ME) within its wireless transmission radius. The cellular system model in
this paper is assumed as follows. A given geographical area consists of a number of hexagonal
cells, each served by the base station (BS). The base station and the mobile stations
communicate through the wireless links using the channel. Each cell is allocated a fixed set of
CUs and the same set of channels are reused by those identical cells, which channels are
sufficiently far from one another to avoid interference, such as inter-cell and intra-cell
interference [4,19].
In the simple allocation strategy [10], this variant of the fixed assignment scheme proposes to
borrow a channel from neighboring cells provided it does not interfere with the existing calls
and locks in the co-channel cells of the lending one. In the directed retry with the load sharing
scheme [23], it is assumed the neighboring cells and the users overlap the region. The main
drawback of this scheme includes the increased number of hand-offs and co-channel
interference, as well as the load sharing depending on the number of users in the overlap region.
The channel borrowing without locking (CBWL) scheme [24] proposes channel borrowing
when the set of channels in a cell is exhausted; but it uses the borrowed channels under reduced
transmission power to avoid co-channel interference. Also, only a fraction of the channels in all
neighboring cells are available for borrowing. In the load balancing with selective borrowing

106
(LBSB) [11], a cell is classified as “hot”, if its degree of coldness defined as the ratio of the
number of available channels to the total number of channels allocated to that cell is less than or
equal to some threshold value. Otherwise the cell is “cold”. Aided by a channel allocation
strategy within each cell, it has been presented in that the centralized LBSB achieves has almost
perfect load balancing and leads to a significant improvement over FCA, simple borrowing,
directories and CBWL schemes in case of an overloaded cellular system. The LBSB scheme has
two disadvantages: (1) Too much dependency on the central server maintenance of continuous
status information of the cells in an environment. The traffic load changes dynamically, leading
to an enormous amount of updating traffic, consumption of bandwidth and message delays. (2)
The strategy of the channel allocation for load balancing usually uses fixed threshold values to
distinguish the status of each cell. However, the [insert noun] are fixed and cannot indicate the
degree of the load. Since the load status may display a sharp distinction state level, the channel
borrowing or lending action will be made frequently around the threshold, possibly resulting in
ping-pong series fluctuation. This results in wasting a significant amount of effort in
transferring channels back and forth. In this paper, the performance of a DCA strategy will
depend on how the state information has been decided at the BSs. Achieving this estimation,
however, is difficult and time-consuming. The relationship between the communication
resources is too complex to define a good rule for estimating the cell load.
3. HGACA Wireless Cellular System
Some of the techniques used to load balancing in heuristic techniques involve a threshold used
to determine where the load is cold or hot. This binary-state makes the system load state
fluctuate between the hot and cold load when the cell load is near the threshold value. It will
cause frequent channel reallocation because of the load change. Simulation techniques have
been widely used by researchers. Although it provides more flexibility and freedom, it has its
own limitations and drawbacks. For example, the load is usually artificial and predetermined.
Some methods use a simple queuing model of a mobile cellular system [10-13,15,19,23-24].
Those proposed schemes completely ignore resources other than the traffic load. Therefore,
while it may be reasonable to detect the performance of purely available channels, the utility of
this is questionable for channels that use the other resources of contention. We recognize it is
difficult, perhaps impossible, to find the cell load information that satisfies all the above
requirements. Moreover, they may be contradictory. But the cell load information may be
judged by the degree to which it meets the above criteria. The problem with such methods is
many unrealistic assumptions must be made to make the study feasible. For example, most
models use exponential distributions for the arrival and service times.
The typical architecture of fuzzy logic control includes four principal components: fuzzifier,
fuzzy rule base, inference engine and defuzzifer. The fuzzifier has the effect of transforming
crisp measured data into suitable linguistic values. The fuzzy rule base stores the empirical
knowledge of the operation of the process of the domain experts. The inference engine is the
kernel of fuzzy logic control: it also has the ability to simulate human decision making by
performing approximated reasoning to achieve a desired control strategy. Finally, the
defuzzifier is utilized to yield a non-fuzzy decision of control action from an inferred fuzzy
control action by the inference engine [19,21,25-26].
HGACA is based on the optimum size of a neural network to reduce the enormous search
spaces in learning and using mathematical methods to determine the architecture and parameters
of the neural network [27]. The advantage of this approach is genes of the chromosome are
classified into categories in a hierarchical form. The HGA differs from the standard GA with a
structure where each chromosome consists of multiple levels of genes. Each consists of two
types of genes, and the control genes and connection genes. The control genes in the form of a
bit are the genes for layers and neurons for activation. The connection genes, a real value
representation, are the genes for connection weightings and neuron bias. A neuron consists of an

107
activity level, a set of input and output connections with a basic value associated to each
connection.
Fig. 2 shows a typical neuron with n-input connections and a single output connection. The
output of the neuron is determined as: ∑ +
=
=
n
1
i
i
i )
x
f(
y η
ω , where n
2
1 x
,...,
x
,
x are input signals,
n
2
1 ,...,
, ω
ω
ω are connection weightings, η is the basic value, and f is output function.
The GA processes imitate natural evolution, and therefore include biomimetic operation such
as reproduction, crossover, and mutation.
(1) Population: The choice of an appropriate population size is a fundamental decision to be
taken in all GA implementations. If the population sizing is too small, the GA will usually
converge too quickly. If the population size is too large, a population will take a very long time
to evaluate. In the study, the population of at k-th generation, (k)
p , is divided into several
connection subgroups, j
i
,
G
G
and
P
...
G
G (k)
j
(k)
i
(k)
(k)
M
(k)
2
(k)
1 ≠
∀
=
∩
=
∪
∪ φ , where M is the maximum
number of possible connections represented by HGACA, and (k)
i
G is the subgroup of
chromosomes representing those networks with i active connection at k-th generation.
(2)Objective functions: The objective of training the network is to minimize two different
parameters, the accuracy of the network (f1) and the complexity of the network (f2). The
accuracy of the network (f1) is defined as: ∑
=
=
N
1
i
2
i
T
i
1 )
y
-
(y
N
1
f , where N is the size of the testing
vector, T
i
y and i
y are the network output and desired output for the i-th pattern of the test vector
respectively.
(3) Selection process: Parent selection is a routine to emulate the survival of the fittest
mechanism of nature. Chromosomes in the population are selected for the generation of new
chromosomes using certain selection schemes. There are two different objective functions, (f1)
and (f2) of the problem optimization process.
(4) Control and Connection, Genes Crossover: Crossover operates on two solution strings
and results in other two strings. A typical crossover operator exchanges the
segments of selected strings across a crossover point with probability. There are two steps
producing two offspring by crossover operator.(1) A given number of crossing sites are
uniformly selected along the parent strings at random. (2) Two new exchanging alternate pairs
of sections between the secreted forms string crossing sites.
(5) Mutation: The mutation operator prevents irreversible loss of certain patterns by
introducing small random changes into chromosomes. Change each bit value with the
probability.
(6) Fitness function: The intended insect uses the GA evolutionary process, and the feature of
this particular chromosome must be specified. The programmed will proceed and each of the
generated chromosomes will be checked according to this ideally specified chromosome. The
measure of this checking mechanism represents the fitness function. This can be a combination
of the genes, and the genetic algorithm is only able to optimize the characteristics explicit in the
fitness function. Fig. 3 shows the hybrid structure parameter learning of the HGACA.
∑ +
= η
ω i
i x
s
1
ω
n
ω
i
x
i
ω
η
f(s)
y =
]
,
,...,
,
[ 3
2
1 η
ω
ω
ω
Fig. 2: A single neuron.

108
Hot
Rule 1
Moderate
Very cold
Cold
Moderate
Hot
Very hot
Cold
Rule i
Rule n
.
.
.
.
.
.
Available
CUs
Traffic load
i
w
m
w
j
w
o
coa
Y
d
Y
Layer 1 Layer 4
Layer 3
Layer 2
Connection Gene
( 1, 1) (00100100), (1...0...1) n
w
w
w
w η
η ,
,
,
, 2
1
1
2
1 Κ
ChromosomeStructure=
)
( 1
η
1
w
2
w
Neuron Control Gene +
Layer Control Gene +
: Active genes (turn on)
: Inactive genes (turn off)
Fig. 3: Hybrid structure parameter learning of the HGACA.
0
coa
Y
*
e
e
0
coa
Y
*
e
e
Fig.4: Membership functions of the fuzzy input and output.
The system has a total of four layers. The nodes in layer 1 are linguistic nodes representing
input linguistic variables; layer 4 is the output layer. There are two linguistic nodes for each
output variable. One is for desired output to feed into the network; the other is for actual output

109
to be pumped out of the network. The nodes in layers 2 and 3 are term nodes, which act as
membership functions representing the terms of the respective linguistic variables. Actually,
layer 2 nodes can be either a single node performing a triangle-shaped membership function or
one performing a complex membership function. Each node in layer 3 is a rule node
representing one fuzzy rule. Also, the links between the rule nodes and the output term nodes
are initially fully connected. Only a suitable term in each output linguistic variable's term set
will be chosen after the learning process, where 0
coa
Y represents the number of migrating
channels, and d
y is our desired output.
3.1 Cell Load Decision-Making
This section addresses our strategy of estimating load status in a wireless cellular network.
This measure is vital for us to determine the most suitable site for migrating channels to share
the load in the system. We can construct different available channels membership functions, and
traffic load membership functions. The distributed channel assignment schemes have received
considerable attention because of their reliability and solvability. Many researchers use
available channels as the single load index for BS in cellular systems [11,14]. Although the
number of available channels is the obvious factor affecting system load, other factors are also
influential, including system load, call arrival rate and call duration. For the accuracy of
evaluating the load state of a cell, we employ the used available channel and traffic load as the
input variables for the fuzzy sets.
The fuzzification function is introduced for each input variable to express the associated
measurement uncertainty. We consider an interval of the real number and the
notation ( )
∫
= u i
i
e a
a
u
e /
*
, and ( )
∫
= u i
i
e b
b
u
e / , where e is denoted as the available channel and
*
e is denoted as the traffic load, i
a and i
b are actual input values, respectively. Let i
a represent
the center value of the linguistic labels of available channel membership function for 2
0 ≤
≤ i ,
and let i
b stand for the center value of the linguistic labels of traffic load membership function
for 4
0 ≤
≤ i . The status of *
e may be very cold (VC), cold (C), moderate (M), hot (H) or very
hot (VH) for different value of traffic load and the status of e may be low (L), moderate (M) or
high (H) for different values of available channels. The fuzzified information is then passed on to
the fuzzy inference engine. Fig. 4 shows the membership function for the number of available
channels and the system parameter traffic load. These functions are defined on the interval ],
[ 4
,
0 a
a
]
,
[ 2
0 b
b .
3.2 Cell Involved Negotiation
After the cell load level of each BS has been decided by the load information, the objective of
the cell negotiation is to select the cell to or from which channels will be borrowed when the
cell load reallocation event takes place. The traditional channel allocation algorithm in
negotiation can be classified into update and search methods [19]. In the search approach, a cell
does not inform its neighbors of its channel acquisitions or releases. When a cell requires a
channel, it searches all neighboring cells to calculate the set of currently available channels, and
then acquires one according to the underlying DCA strategy. In the update approach, a cell
always informs its neighbors whenever it acquires or releases a channel, so each cell knows the
set of channels available for its use and the underlying DCA strategy.
Both approaches have advantages and disadvantages. The update approach has short
acquisition delay and good channel reuse, but it also has higher message complexity. In other
words, the search approach has lower message complexity, but it has longer acquisition delay
and ineffective channel reuse [19]. Our research took advantage of HGACA and presented an
enhanced version of the negotiation scheme, termed cell involved negotiation. When the load
state is hot, it plays the role of the borrowing channel action; in contrast, it plays the role of the
lending channel action when its load state is cold. The moderate cells are not allowed to borrow
any channels from any other cell nor lend any channels to any other cell. At each BS, an

110
augmented load state table is maintained. The entries of the table are the current load status of
every cluster cell as well as the co-channel cells. The cell operation types of load state
information exchanges among cells, and each BSs keeps the state information of the cells and runs
the channel borrowing algorithm to update the load state.
The knowledge pertaining to the given control problem is formulated in terms of a set of fuzzy
inference rules. We use five load actions; very cold, cold, moderate (stabilized-state), hot, and
very hot. The BS keeps the load-state information of the cells and runs the fuzzy based channel-
borrowing algorithm to borrow free channels from the very cold or cold cells for the very hot
or hot cells whenever it finds any very hot cells or hot cells. The moderate cells are neither
allowed reallocation to or from any channels, nor to any other cells or update interfering
neighborhood cells. [Editor: I don’t know if the previous sentence keeps your intended
meaning.] Further, assume the following seven linguistic states are selected for migrating
channels of the variables: Negative large (NL), negative medium (NM), negative small (NS),
approximately zero (AZ), positive large (PL), positive medium (PM), and positive small (PS).
This paper has 15 rules, as shown in Table1.
Table 1 .IF-THEN rule.
Cold
Very Cold
Moderate
Hot
Very Hot
High
Moderate
Low
(Lending)
NM (4)
(Lending)
NS
(Lending)
NM
(Lending)
NS
(Stable)
AZ
(Lending)
NS
(Stable)
AZ
(Borrowing)
PS
(Stable)
AZ
(Borrowing)
PS
(Borrowing)
PM
(Borrowing)
PS
(Borrowing)
PM
(Borrowing)
PL
(Lending )
NL
e
e*
3.3 Multi-CUs Migration
The new channel allocation with multi-CUs transferring, can reallocate CUs well, especially
in an unpredictable variation of cell load. Our mechanism for multi-CUs transfer calculates the
amount of transferred channels by these two values. The number of available CUs and traffic
load are the values, representing the average during the recent minutes. The HGACA we
discussed in the previous section has a common property; when a requesting cell and a probed
cell are decided, the number of reallocated channels is just one channel in each iteration. It is
very inefficient if the cell load of these two cells greatly differ. We propose the idea of
allocating several CUs instead of only one between two cells whose BS load greatly differ. For
example, in the next generation multimedia mobile network, a call may require multiple CUs at
one time. Under our proposal, the cell load between two cells could be made more balanced. To

111
accomplish this, we use five load values, which are very hot, hot, moderate, cold and very cold
to distinguish the difference in cell load on two cells. If one cell is in the “very hot” state (PL), it
will borrow several channels from the cell in the “very cold” state (NL). Defuzzification is a
mapping from a space of fuzzy control actions defined over an output universe of discourse into
a space of non-fuzzy (crisp) control actions. This process is necessary because in many practical
applications, a crisp control action is required for the actual control. Fig. 5 shows the
membership function for channel borrowing or lending a quantity control number of the
channels range [-C, +C] of the fuzzy output. The function is defined in the interval [0,+c] for
the borrowing action, and in the interval [0, c] for the lending action.
We used the center of area (COA) method because it supports real-time software fuzzy
controls to distinguish the difference in the load of two cells. This value is calculated by the
formula:
( )
c
IN
w
B
w
Y n
1
i i
n
1
i i
i
o
coa −
















∑
∑ ×
=
=
= (1)
Where 0
coa
Y represents the number of migrate channels,
i
W = The antecedent degree of the ith control rule
i
B = The consequent center value of the ith control rule
Thus, the defuzzified value 0
coa
Y obtained by the formula can be interpreted as an expected value
of variable. Finally, we obtain
Migrate Channels =Min [Borrowing cell ( 0
coa
Y ), Lending cell ( 0
coa
Y )].
After multi-CUs are reallocated, we use GA to tune the output fuzzy membership function.
The genetic operation should be used in a way to rapidly achieve high fitness individuals in the
population without leading to total convergence. In this paper, we used the part random method
to achieve high fitness for a short time interval. There are two main factors to be considered in
the GA, one is the fitness function and the other is the encoding scheme. The fitness function F
uses the following formula for the load index. Note, the parameter gene remained unaltered and
merely changes the interpretation of its form. In this way, the complexity of tuning the fuzzy
memberships and rules can thus be optimized and the overall structure can be greatly reduced.
Each string included in a population is evaluated by the fitness function using the following
formula:
MES
ACQ
BP
DP
1
F
×
+
×
+
×
+
×
=
δ
γ
β
α
(2)
Where α : the weight for parameters of DP,
β : the weight for the parameters of BP,
:
γ the weight for the parameters of ACQ,
:
δ the weight for the parameters of MES,
DP: the handoff call-dropping rate,
BP: the new call-blocking rate,
ACQ: the channel acquisition delays and
MES: the average number of update messages overhead.
4. Experimental Results
The problem domain naturally lends itself to simulating multiple threads since there are a lot
of concurrences and global resource management issues in the system. The simulated model
consists of 14 clusters with 7 homogeneous cells each. This experiment used the number of
channels CH = 100 in a cell, total of N = 98 cells in the system. The amount of requested CU
[specified of minimum basic a channel unit (CU)] is 30Kbps of multi-channels migration.]
[Editor: I don’t understand this part of the previous sentence.] We assume o
λ = 100 calls/per

112
hour~ 2000 calls/per hour is the call originating rate per cell, 01
.
0
( ×
= o
h λ
λ ~ )
1
×
o
λ is the
hand-off traffic density per cell, d = 1 sec the communication delay between cells, and each
handoff and new call request delay constraint (DC=5) is five seconds. So, from the simulation
result, the value of the traffic load is chosen randomly and non-linearly. The maximum numbers
of hand-off calls are queued at 10 for the first-class priority, with new calls queued at 10 for the
second-class priority, respectively. [Editor: Does the previous sentence keep your intended
meaning.] Let the density of simulation be 500 people/per cell. We define the time of the
sample interval as 3 minutes and the sampling time will influence the previous one. The CU
acquires messages transmitted between the hot cell i and cold cell j . In our simulation, three
types of traffic services are assumed: voice service, videophone and video on demand. These
types are defined on the CUs requirement 30 Kbps, 256 Kbps and interval 1 Mbps to 3 Mbps,
respectively. The assumptions of four performance metrics for our simulation study are as
follows. (1) Blocking calls: If all the servers are busy, the cell does not succeed to borrow a CU
from its cluster cells and its waiting time (delay constraint) is over then the calls must be
blocked, otherwise they receive a service. (2) Dropping calls: When an MS moves into a
neighboring cell, the call must be transferred to the neighboring BS. This procedure is a hand-
off. If a channel cannot be assigned at the new BS and the particular cell does not borrow a
channel from its cluster cells, then the call generated at this particular cell are stored in the
queue. Its waiting time (delay constraint) will be over and the calls must be dropped, otherwise
they receive service. (3) Update-message complexity: Each cell needs to communicate with the
co-channel and cluster cells to exchange the set of load state information. (4) CUs-acquisition
delays: For the values it acquires before the selected channels, the cell must simultaneously
ensure the selected channels will not be acquired by any of its cluster cells and interference
cells. When a cell receives a channel request from an MS, it assigns a free channel, if any, to the
request. Otherwise, the cell will need to acquire a new channel from its cluster cells and then
assign channels to the request. The performance of our HGACA is compared with the simple
borrowing (SB), and existing strategies like channel borrowing directed retry (DR), CBWL, and
LBSB. The hand-off call dropping probabilities for HGACA and other methods are plotted in
Fig. 5 against the hand-off dropping probability at different traffic loads. In every case, when
the hand-off dropping probability is fixed, the HGACA has a lower hand-off call dropping
probability than other methods. The improvement in the performance of the HGACA over other
methods, however, decreases as the traffic load goes up. Fig. 6 compares the channel-
assignment algorithms according to the new call-blocking probability of channel request for the
multimedia services. Fig. 7 shows the hand-off call-dropping probability for various schemes at
various multimedia services. The number of multimedia requirements on the horizontal axis has
different meanings for voice service, videophone and video on demand. The HGACA scheme
always has a lower hand-off dropping rate than the existing channel-assignment schemes with
the same number of channels required. It also indicates the HGACA scheme can improve
performance over the other methods with the number of reserved channels by further reducing
the hand-off dropping probability. Fig. 8, which depicts the messages of different CU allocation
schemes, shows our proposed DCA scheme has the fewest updated messages. Our proposed
scheme performs especially well when the numbers of hot cells are large. The channel
acquisition delays are also discussed in our experiment. Fig. 9 shows our proposed scheme has
the shortest channel acquisition delays. This results in a channel-allocation scheme with
efficient channel use in all traffic conditions.

113
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Call
dropping
probability
100 500 1000 1500 1800 2000
Traffic arrival rate (calls/h)
DR
SB
CBWL
LBSB
NFC
GA
HGACA
Fig. 5: Compare dropping probability and traffic arrival rate.
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Call
blocking
probability
30Kbps 256Kbps 1Mbps 2Mbps 3Mbps
Multi-CUs requirement of various multimedia service
DR
SB
CBWL
LBSB
NFC
GA
HGACA
Fig. 6: Blocking probability and multi-CUs requirement of multimedia service.
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
Call
dropping
probability
30Kbps 256Kbps 1Mbps 2Mbps 3Mbps
Multi-CUs requirement of various multimedia service
DR
SB
CBWL
LBSB
NFC
GA
HGACA
Fig. 7: Dropping probability and multi CUs requirement of multimedia service.
0
1000
2000
3000
4000
5000
6000
7000
8000
9000
A
verage
number
of
message
complexity
10 20 30 40 50 60 70 80
Number of hot cells
Search
Update
NFC (Update)
GA (Update)
HGACA (Update)
Fig. 8: Average number of update message overhead in our scheme and others.

114
0
1000
2000
3000
4000
5000
6000
7000
8000
A
verage
number
of
CU
s
acquisition
delay
(sec)
10 20 30 40 50 60 70 80
Number of hot cells
Search
Update
NFC (Update)
GA (Update)
HGACA (Update)
Fig. 9: The CUs acquisition delay of various schemes
5. Conclusion
HGACA are complementary technologies in designing an intelligent wireless cellular
network. Neural networks are essentially low level computational structures and algorithms
offering good performance in dealing with sensory nonlinear input data, while fuzzy logic
techniques deal with reasoning on a higher level than networks. This is the first attempt to
formulate the dynamic CUs allocation problem with HGACA and with simulation for various
traffic loads and numbers of hot-cell nodes. Since fuzzy logic controls are constructed using
linguistic variables, intuitive knowledge is easily integrated into the control system. We believe
a HGACA for the control and managing cellular networks is more appropriate than the
conventional probabilistic models. It also can efficiently determine the suitable cell for
allocation CUs. The performance of the proposed scheme is better than conventional schemes in
the blocking rate, dropping rate, message complexity and channel acquisition delay.
REFERENCES
[1] J.J Won, H.W Lee, C.H Cho, Mobile Clustering Based Resource Reservation Schemes in Wireless
Mobile Networks, ICION KOREA, Vol. 1, pp. 900-909, 2003.
[2] J.J Won, E.S Hwang, H.W Lee, C.H Cho, Mobile Cluster Based Call Admission Control in
Wireless Mobile Networks, IEEE Vehicular Technology Conference, Vol. 3, pp. 1527–1531, 2003.
[3] Goeckel, D.L.Stark, W.E, Optimal diversity allocation in multi-user communication systems, IEEE
Transactions on Communications, Vol. 47, pp. 1828 – 1836, Dec, 1999 .
[4] Holma, H. Heikkinen, S. Lehtinen, Otto-A leksanteri Lehtinen, and Antii Toskala, Interference
considerations for the time division duplex mode of the UMTS Terrestrial Radio Access, IEEE on
Selected Areas in Communications, Vol. 18, pp. 1386 - 1393 Aug. 2000.
[5] “http://www.3gpp.org. 2002”.
[6] H. Holma and A. Toskala (eds.), WCDMA for UMTS. Wiley, 2000.
[7] 3rd Generation Partnership Project Technical Specification Group Radio Access Network. Working
Group 1, “Physical Layer – Measurements.” TS25.225 v4.0.0. 2001.
[8] 3rd Generation Partnership Project. Technical Specification Group. Radio Access Network “Radio
Inter-face Protocol Architecture.” TS25.301 v4.2.0. 2002.
[9] 3rd Generation Partnership Project. Technical Specification Group. Radio Access Network “Radio
Re-source Control (RRC); Protocol Specification.” TS25.331” 4.4.0, 2002.
[10] T. S. Yum M. Zhang, Comparisons of channel-assignment strategies in cellular mobile telephone
systems, IEEE Transactions on Vehicular Technology, Vol.38, pp. 211–215, 1989.

115
[11] S. K. Das, S. K. Sen and R. Jayaram, A structured channel borrowing scheme for dynamic load
balancing in cellular networks, IEEE Distributed Computing Systems Conference, Vol 3, pp. 116-
123, 1997.
[12] J. Kim, T. Lee and C. S. Hwang, A dynamic channel assignment scheme with two thresholds for
load balancing in cellular networks, IEEE Radio and Wireless Conference, Vol. 1, pp. 141-145,
1999.
[13] T. Lee J. Kim and C. S. Hwang, A dynamic channel assignment scheme with two thresholds for
load balancing in cellular networks, IEEE Radio and Wireless Conference, Vol. 1, pp. 141–145,
1999.
[14] S. Kim and P. K. Varshney, Adaptive Load Balancing with Preemption for Multimedia Cellular
Network, IEEE Wireless Communications and Networking Conference, Vol 3, pp. 1680–1684,
2003.
[15] S. Mitra and S. DasBit, A load balancing strategy using dynamic channel assignment and channel
borrowing in cellular mobile environment, IEEE Personal Wireless Communications Conference,
Vol. 1, pp 278-282, 2000.
[16] H. Haas and S. McLaughlin, A novel decentralized DCA concept for a TDD network applicable for
UMTS. IEEE Transactions on Vehicular Technology, Vol. 2, pp 881-885, 2001.
[17] S. S. M. Patra, K.Roy, S. Banerjee and D.P.Vidyarthi, Improved Generic Algorithm for Channel
Allocation with Channel Borrowing in Mobile Computing, IEEE Trans. on mobile computing, Vol.
5, No. 7, pp.884-892, Jul. 2006.
[18] Yao-Tien Wang and Kuo-Ming Hung, “A Genetic-Fuzzy Controller for Load Balancing in Wireless
Cellular.” International Journal of Information and Management Sciences (IJIMS), Vol. 18, pp.
467-494, December 2007.
[19] X. Dong and T. H. Lai, Distributed dynamic carrier allocations in mobile cellular networks: search
vs. update, IEEE Distributed Computing Systems Conference, Vol. 1, pp. 108–115, 1997.
[20] Y. -T. Wang and J.-P. Sheu, A Dynamic Channel Borrowing Approach with Fuzzy Logic Control
in Distributed Cellular Networks, the International journal of Simulation Modeling Practice and
Theory, Vol. 12, pp. 287 – 303, July 2004.
[21] Yao-Tien Wang and Kuo-Ming Hung “Fuzzy Logic Based Neural Network Model for Load
Balancing in Wireless Networks. “ IEEE Communications Society, International Journal of
Communications and Networks (IJCN), Vol. 10, pp.38- 43, March 2008.
[22] J. S. Engel and M. Peritsky, Statistically optimum dynamic sever assignment in systems with
interfering severs, IEEE Vehicular Technology Conference, Vol. 1, pp 1287-1293, 1973.
[23] J. Karlsson and B. Eklundh, A cellular mobile telephone system with load sharing-an enhancement
of directed retry, IEEE Transactions on Communications, Vol. 37, pp 530-535, 1989.
[24] H. Jiang and S. S. Rappaport, CBWL: a new channel assignment and sharing method for cellular
communication systems, IEEE Transactions on Vehicular Technology, Vol.43, pp. 313 –322, 1994.
[25] L. A. Zadeh, Fuzzy Algorithm. Information and Control, Vol.1, pp 94–102, 1968.
[26] Y.-T. Wang and J.-P. Sheu , Adaptive Channel Borrowing for Quality of Service in Wireless
Cellular Networks, International Journal of Communication Systems. Vol. 19, pp. 205-224,
March , 2006
[27] K.F.Man, KS. Tang and S. Kwong, Genetic Algorithms, Springer, 1999.

116
Authors
Kuo-Ming Hung is currently a lecturer at the Department of Information
Management, Kainan University, Taiwan. He received the B.S. and M.S.
degrees in electrical engineering from TamKang University, Taiwan, in
1990 and 2001. He is currently pursuing his Ph.D degree in Department of
Electrical Engineering,TamKang University, Taiwan. His research interests
include image processing, image inpainting,security, wireless
communications and mobile computing.
Yao-Tien Wang is currently an Assistant Professor at the Department of
Information Management, Kainan University, Taiwan. He received his
PhD. degree in Computer Science and Information Engineering from
National Central University ,Taiwan. His current research interests
include wireless communications, mobile computing, fuzzy logic control,
neural network and genetic algorithm.
Ching-Tang Hsieh is a professor of E.E. at Tamkang University, Taiwan,
R.O.C. . He received the B.S. degree in E.E. in 1976 from TKU and the
M.S. and Ph.D. degree in 1985 and 1988, respectively, from Tokyo
Institute of Technology, Japan. From 1990 to 1996, He acted as the
Chairman of the Dept. of E.E. His current research interests include
speaker recognition, fingerprint identification, image inpainting, image
processing, and fuzzy system.

Hierarchical Genetic Algorithm For Dynamic Channel Units Allocation in TD-CDMA/TDD System

More Related Content

What's hot (18)

Similar to Hierarchical Genetic Algorithm For Dynamic Channel Units Allocation in TD-CDMA/TDD System (20)

Recently uploaded (20)

Hierarchical Genetic Algorithm For Dynamic Channel Units Allocation in TD-CDMA/TDD System