Вы находитесь на странице: 1из 20

ADAPTIVE GUARD CHANNEL RESERVATION SCHEME

IN CELLULAR NETWORKS

PREPARED BY

ACHAL AGARWAL
2015B4A70436P

IN PARTIAL FULFILMENT OF

MATH F366 - LABORATORY PROJECT

SUBMITTED TO

DR. RAKHEE

BIRLA INSTITUTE OF TECHNOLOGY & SCIENCE, PILANI

APRIL, 2017
TABLE OF CONTENTS

1. ACKNOWLEDGEMENTS 3

2. ABSTRACT 4

3. INTRODUCTION 5

4. QUEUING THEORY IN CELLULAR NETWORKS 6

4.1. MARKOVIAN MODEL 6

5. SIMULATION 8

5.1 MOBILITY MODEL 8

5.1.1. Simulation Parameters 8

5.2 SINGLE CLASS TRAFFIC MODEL 10

5.2.1. Simulation Parameters 11

5.3 MULTI CLASS TRAFFIC MODEL 13

5.3.1. Simulation Parameters 13

5.3.1. Algorithms 15

5.3.2. Results 16

6. CONCLUSION 19

7. REFERENCES 20
3
1. ACKNOWLEDGEMENTS

The success and final outcome of this project required a lot of guidance and
assistance. All that I have done till now is only due to such supervision and assistance
provided by Dr. Rakhee. I thank Dr. Rakhee for all this.

I would also like to thank the Department of Mathematics, BITS Pilani, for giving me
the opportunity to learn.

Moreover I would like to express my gratitude towards the college and all the driving
forces for the same, including the Director, Vice Chancellor and many others.

Last but not the least, I would like to thank my parents and friends for primarily their
emotional support and many other things to enable me to work on this project.
4
2. ABSTRACT

This project is based on queuing theory for wireless networks and studies the handoff
and new connection blocking rates. Various network schemes are studied and
implemented using approximate simulation models to know the reduction in blocking
rates achieved by them. Reservation schemes are examined over a wide range of
offered load. Then multi class traffic is studied and various schemes are analysed.
After studying the recent works, a new scheme is proposed and thoroughly analysed.
5
3. INTRODUCTION

A wireless network is as good as it’s call admission control scheme. A call admission
control scheme allows new connections to hosts for various services. Any call
admission control scheme must reduce blocking and aim to judiciously utilise
available resources for maximum welfare. At the same time, it should primarily
prioritise containing the handoff blocking rate within acceptable limits. Handoff
connections are those which get transferred or handed off to new base stations whilst
the user moves from one cell to another. A cell is a region wherein a single base
station provides connection to a network.

Queuing theory is extensively used in network control where channels in a base


station act as servers and hosts in the corresponding cell as customers. This model is
scaled to multiple servers as we have a hexagonal mesh-like arrangement of wireless
networks.

In this project, a guard channel policy is implemented to reduce handoff blocking


rates while keeping the new call blocking rates minimum. A guard channel policy
reserves a subset of channels (bandwidth) in a given cell for handoff calls. This
policy is implemented and simulated using Java. Without loss of generality, only the
central cell is studied.

Pati’s guard channel policy is used and the results are compared. Further more,
network and simulation parameters are varied to test policy in different scenarios.
Handoff rate and Cell dwell time are important parameters to get a somewhat realistic
view and hence, a separate simulation was carried out to get these parameters using
the Random Waypoint Mobility model.

After analysing other such schemes, multi class traffic is incorporated in the model
and thoroughly studied. Various schemes are analysed, specifically the one provided
by Md. Sulamah. Thereafter another guard channel reservation scheme for multi class
traffic ism provided with detailed results and its convergence with respect to the
blocking and dropping probabilities are shown.

The analysis was carried out with different parameters spread over a wide range and
acceptable results were obtained. The random waypoint model designed before was
also incorporated for carrying out the thorough testing.
6

4. QUEUING THEORY IN CELLULAR NETWORKS

We model this problem to that of a typical Queuing Theory model. In each base
station controller, the channels are servers and the users in and around the nearby
cells constitutes the population.

As mentioned before, the call admission control scheme is a crucial aspect of any
wireless network. It controls the provision of services to the various users in the
network.

In this project, a distributed and adaptive, guard channel reservation scheme proposed
by Dr. Pati is used. This scheme is built upon the concept of guard channels, and it
imposes a hard constraint on the handoff call blocking probability. Guard channels
are channels reserved specifically for handoff calls and is not used for provisioning
new calls in the cell. It achieves optimal performance (i.e. maximum resource
utilisation with guaranteed Quality of Service) and has the ability to adapt
automatically to changes in network traffic.

4.1. MARKOVIAN MODEL

The Markovian model for this scheme is as shown in Figure 1. This is provided to
estimate the number of guard channels in this scheme.

FIGURE 1. Markovian Model

The complete model has been derived from Pati’s paper, please refer to the same for
intricate details of this model.

The service rate is defined as follows:

In the above and subsequent relations, the symbols carry their usual meanings.
7

Hence, probability in steady state will be given by:

Thus summing over all possible values of i, we can get P0 , given as follows:

Thereafter, our required parameter of interest, handoff call blocking probability can
be derived. This analytically solved expression can be contained within the required
limit. The probability is given as follows:

Now simulation is carried out with appropriate parameters to test this scheme.
8

5. SIMULATION

Simulation for the above described model was carried out. A few parameters were
taken from Pati’s publications but mobility parameters, the choice of which was not
reasoned clearly was taken after a separate simulation for the same.

5.1 MOBILITY MODEL

Cell dwelling time and rate of handoff are two crucial parameters to simulate a call
admission control scheme. Incorrect choice of parameters may give false results and
hence nullifying the role of the simulation.

Mobility models help get us a rough estimate of these two parameters to directly use
in the main simulation. In this project, a Random Waypoint Mobility model is chosen.

Random Waypoint model is a commonly used synthetic model for mobility, e.g., in
Ad Hoc networks. It is an elementary model which describes the movement pattern
of independent nodes by simple terms as state below.

• Each node moves along a line from one waypoint to the next.

• The waypoints are uniformly distributed over the given convex area. In this case,
disks centred at the Base Station Controller with radius as the range of service
provision.

• At the start of each leg a random velocity, a random direction and a random
distance are each drawn from their respective distributions.

• Optionally, the nodes may have so-called "thinking times" when they reach each
waypoint before continuing on the next leg, where durations are independent and
identically distributed random variables.

• Thereafter the simulation proceeds iteratively.

5.1.1. Simulation Parameters

• The simulation model was a representation of the hexagonal mesh-like structure of


the cells.
• Each cell was a disk centred at the cell and radius 50,000 units. The total time for
simulation was 20,000 units.
• Uniform distribution was used for all parameters viz. Velocity, Thinking Time,
Distance and Direction.
• 2000 hosts were distributed across 19 cells with a uniform distribution.
9
• Time-driven simulation was used as all the hosts’ location needed to be updated in
real time.

After carrying out the simulation the results were used to compute the Cell Dwelling
time and Handoff rate, which will be used in the subsequent policy simulation.

Given below are sample outputs plotted on an X-Y connected plot to visualise the
randomness generated.

Random Waypoint Mobility Model


800

600

400

Movement of Host
200
Y Axis

0
-600 -400 -200 0 200 400 600

-200

-400

-600
X Axis

FIGURE 2. Sample Output 1 from Mobility Simulation

Random Waypoint Mobility Model


800

600

400
Movement of Host

200
Y Axis

0
-800 -600 -400 -200 Y
0 200 400 600 800 1000

-200

-400

-600

-800
X Axis

FIGURE 2. Sample Output 2 from Mobility Simulation


10
5.2 SINGLE CLASS TRAFFIC MODEL

The physical coverage area is partitioned into a hexagonal mesh structure without any
overlapping meshes as shown in Figure 3. There are a total of 19 cells and the cell
marked 0 is considered for performance evaluation.

8 9 10

7 1 2 11

18 6 0 3 12

17 5 4 13

16 15 14

FIGURE 3: Hexagonal Cell Model used for Simulation

Discrete Event Simulation is used this time as it suits the purpose better. The events
are:

• New Call Arrival


• Call Termination
• Call Handover
11
The process flowchart is as shown in Figure 4.

FIGURE 4: Simulation Process Flow Chart

5.2.1. Simulation Parameters

The parameter values used in the simulation experiment are replicated from Pati’s
model for comparison.

The total simulation period is chosen to be 5,000,000 units so as to obtain accurate


results. As the algorithm designed was very efficient, the duration of the simulation
was not much. The simulator built from scratch is modular, tuneable and multi
threaded. Hence given a powerful computer, better results can be obtained.

5.2.2. Results

After carrying out the simulation, the essential parameters were recorded and plotted
to estimate the required probabilities. They are shown in Figures 5 and 6.
12

The handoff blocking rate is contained below 0.02%

FIGURE 5: Handoff Call Blocking Results

The new call blocking rate is contained below 0.5%

FIGURE 6: New Call Blocking Results


13
These figures show how the handoff call drop probability becomes stable and remains
below the targeted QoS as the model adapts to the traffic. This model also bounds the
new call blocking probability.

5.3 MULTI CLASS TRAFFIC MODEL

The physical coverage area is partitioned just as in the previous model (as shown in
Figure 3). There are a total of 19 cells and the cell marked 0 is considered for
performance evaluation.

Discrete Event Simulation is used this time as it suits the purpose better. The events
are:

• New Call Arrival (of any class C)


• Call Termination (of any class C)
• Call Handover (of any class C)
The process flowchart is as shown in Figure 4.

5.3.1. Simulation Parameters

In our multi class model (K classes), we require threshold values for each class of
traffic along with K*K probabilities for the relation of each class with the others
which will give us the probability with which a new call of given class can use a
reserve channel. Finally we require the fixed probability which will determine
whether a handoff call of a higher priority can utilise guard channels of a lower
priority

The total simulation period and the reason are similar as the previous case. The
simulator built from scratch is modular, tuneable and multi threaded. Hence given a
powerful computer, better results can be obtained.

The results were collected for each class of traffic under various conditions and with
different parameters. The results were then analysed and compared with other
schemes for multi class traffic.

These figures show how the handoff call drop probability becomes stable and remains
below the targeted QoS as the model adapts to the traffic. This model also bounds the
new call blocking probability.

To keep our analysis simple, we discuss the case with 3 classes of traffic. For most
practical purposes, three classes of traffic is suitable as in a broad view we can
classify traffic into Real Time, Streaming and Background calls. Similar results are
expected for more classes due to symmetry in our model.

The classes in decreasing order of priorities are 1,2 and 3.


14
For the results, the parameters are chosen as below:

Handoff Threshold

This measure denotes the approximate percentage of handoff drops acceptable. There
is a proportional (not exact) correlation between this value and the the acquired call
drop percentage. The lower the threshold, the better (lower) are the acquired results
depending on the total channels.

The handoff threshold values for the three classes are:

Class 1 - 0.001
Class 2 - 0.02
Class 3 - 0.1

Call Arrival Rate

This measure denotes the call arrival rate (lambda). Here lambda is the mean value of
call arrival rate. This lambda is used as the parameter in the exponential distribution
to generate calls in a cell. As expected, the higher this value, the higher is the number
of new calls generating in the particular cell.

The call arrival rates for the three classes are:

Class 1 - 0.8
Class 2 - 0.6

Class 3 - 0.2

Call Handoff Rate

This measure denotes the call handoff rate (lambda). The mean value of handoffs per
unit time distributed exponentially. As handoffs are sent to neighbouring cells, a
hand-in occurs in Cell 0 only when a handoff is generated in Cells 1,2,3,4,5,6 (refer
to Figure 3). This effect has a slow recursive flow along both in and out directions. As
expected, the higher this value in the neighbouring cells, the higher is the number of
handoffs generating in the particular cell (Cell 0 in our analysis).

The handoff rates for the three classes are:

Class 1 - 0.08
Class 2 - 0.005
Class 3 - 0.02

Call Termination Rate


15
This measure denotes the call termination rate. The mean termination rate will be
inversely related to the mean duration of the call. Thus if the arrival rate of real time
calls is x and the mean duration of a call is 100 (units of time), then the termination
rate of real time calls is x/100.

The termination rates for the three classes of calls are:

Class 1 - 0.008
Class 2 - 0.0055
Class 3 - 0.0014

5.3.1. Algorithms

There were newly devised algorithms for the multi class scheme to ensure
convergence for all classes of calls.

Provided below are the algorithms for all the basic events that take place in the
model. These use the thresholds as discussed in the previous section. The classes of
traffic are C1, C2, … CN. Where CN is of lowest priority

Algorithm 1 New Call Arrival


Input: New call of Class Ci
1: If there is a vacant unreserved guard channel,
then, we grant admission to the call. Return.
2: Generate a random number r.
3: Initialise P <= CN
4: While all classes are not covered
5: If r < threshold value for the two classes (P and Ci),
then, we grant admission to the new call. Return.
(In one of the vacant reserved guard channels of P)
6: Admission not granted. Return.

Algorithm 2 New Hand-in Arrival


Input: Handoff of Class Ci
1: Randomly decide a neighbour to HandIn.
2: If there exists a vacant channel in the reserved slots for class Ci,
then, grant admission to the call. Return.
1: If there is a vacant unreserved guard channel,
then, we grant admission to the call. Return.
2: Generate a random number r.
3: Initialise P to Class Ci+1
4: If r < threshold value for the two classes (P and Ci),
then, we grant admission to the new call. Return.
(In one of the vacant reserved guard channels of P)
6: Admission not granted. Return.
16

5.3.2. Results

After carrying out the simulation, the essential parameters were recorded and plotted
to estimate the required probabilities. They are shown in the figures below.

The data is shown for all the various classes of traffic (3). As the classes were
separated based on priority and our algorithms also utilise the priority hierarchy, it is
natural for the traffic of higher priority to attain lower blocking and dropping rates as
compared to other classes.
These figures show how the handoff call drop probability becomes stable and remains
below the targeted QoS as the model adapts to the traffic. This model also bounds the
new call blocking probability.

FIGURE 8: Handoff Call Dropping for Class 1


17

FIGURE 9: Handoff Call Dropping for Class 2

FIGURE 10: Handoff Call Dropping for Class 3


18

FIGURE 11: New Call Blocking for Class 1

Based on the provided graphs, we see that the probabilities converge as the
simulation runs. The parameters involved high probability and high arrival rates
traffic of Class 1, hence, the dropping probability is not as good as expected but on
the whole the results are acceptable.

Briefly, the convergence rates for handoff call blocks are as follows:

Class 1 - 2.2%
Class 2 - 1.8%
Class 3 - 2.2%
19
6. CONCLUSION

In this project, a distributed adaptive guard channel reservation (DAGCR) scheme


proposed by Pati is implemented with modifications in mobility.

This DAGCR scheme reserves channels to give priority to handoff calls and to
control new call admission into the network. Given that handoff calls have higher
priority than new calls to ensure customer satisfaction and retention, this proposed
adaptive algorithm can search automatically the optimal number of guard channels to
be reserved at a base station, and it can also adapt to changes in traffic conditions
such as changes in the call arrival rate. Thus, fixed CAC policies such as fixed guard
channel policies are less preferable in applications.

Simulation results showed that a simple adaptation algorithm can be used to


automatically determine the optimal number of guard channels to be reserved at a
base station under changing traffic conditions. Thus, this scheme can be used to
guarantee the targeted QoS and control the blocking performances in wireless
networks.

Having built an accurate and a modular simulator, this model can now be tested with
various mobility models and varying parameters. Thereafter, it can be compared with
other admission control schemes and improved upon.

In the conclusion of the interim report, there were a few points mentioned with
regards to the future scope of the project.

Amongst those points, a majority have been carried out -

• Simulating for a longer duration.


• Using more realistic and established distributions for mobility models.
• Classifying services into Real Time and Non Real Time types.
• Incorporating heterogenous controllers.

There is still more scope in this project. Some of them include:

• Analysis of heterogenous controllers.


• Incorporating an ecosystem with various mobilities in a single simulation.

The model we have proposed has almost negligible overhead and makes no
assumption on the network, hence this can be applied on almost all the networks
which face the problem we are addressing.
20

7. REFERENCES
1. Kunz, G., Stoffers, M., Gross, J., & Wehrle, K. (2011). Runtime efficient event
scheduling in multi-threaded network simulation. In Proceedings of the 4th
International ICST Conference on Simulation Tools and Techniques, pp. 359-366.

2. Hyytiä, E. and Virtamo, J. (2006). Random waypoint mobility model in cellular


networks. Wireless Networks, 13(2), pp.177-188.

3. Lin, X., Ganti, R., Fleming, P. and Andrews, J. (2013). Towards Understanding
the Fundamentals of Mobility in Cellular Networks. IEEE Transactions on
Wireless Communications, 12(4), pp.1686-1698.

4. Pati, H. (2007). A distributed adaptive guard channel reservation scheme for


cellular networks. International Journal of Communication Systems, 20(9), pp.
1037-1058.

5. Pati, H. (2013). A control-period-based distributed adaptive guard channel


reservation scheme for cellular networks. Wireless Networks, 19(7), pp.
1739-1753.

6. Salamah, M. (2006). An Adaptive Multi-Guard Channel Scheme for Multi-Class


Traffic in Cellular Networks. IEEE International Conference on Computer
Systems and Applications.

7. Papazoglou, P.m., et al. (2014). On Improved Modeling and Simulation of


Wireless Communication Systems Using Multiagent and Multi-Threading
Systems Architectures for Efficient Spectrum Reuse. 2014 22nd
Telecommunications Forum, pp. 324-327.

8. Kim, K., & Choi, H. (2009). A Mobility Model and Performance Analysis in
Wireless Cellular Network with General Distribution and Multi-Cell
Model. Wireless Personal Communications, 53(2), pp 179-195.

9. Rejeb, S. B., Nasser, N., & Tabbane, S. (2014). A novel resource allocation
scheme for LTE network in the presence of mobility. Journal of Network and
Computer Applications, 46(1), pp 352-361.

10. Somagari, S. R., & Pati, H. K. (2012). An Analytical Model for Adaptive Multi-
Guard Channel Scheme for Multi-Class Traffic in Cellular Networks with
Reduced Handoff Drop Probabilities. Procedia Technology.

Вам также может понравиться