Wireless Pers Commun

DOI 10.1007/s11277-015-2333-2

An Intelligent Hand-off Algorithm to Enhance Quality

of Service in High Altitude Platforms Using Neural
Network
S. H. Alsamhi N. S. Rajput

Springer Science+Business Media New York 2015

Abstract Efficient hand-off algorithm enhances the capacity and quality of service (QoS) of
cellular systems. Hand-off algorithm is used in wireless cellular systems to decide when and
to which base station (BS) will receive the handoff call, without any service interruption. High
altitude platforms (HAPs) is considered as a complementary BS to mobiles in an obstacle
position. HAPs can supply services to uncovered areas of terrestrial systems, thus with the
goodness of HAPs total capacity in a service-limited area will be improved. Recently, artificial
neural network (ANN) has been utilized to improve hand-off algorithms due to its ability
to handle large data. As a revolutionary wireless system, ANN helps in taking the hand-off
decision based on receive signal strength, speed, traffic intensity, and directivity. Radial based
function network is used for making a hand-off decision to the chosen neighbor BS. This
paper presents novel approaches of combining HAPs and terrestrial system in a particular
coverage area for the design of high performance hand-off algorithm. It is found that hand-off
rate and blocking rate are greatly improved using ANN for handoff decision.
Keywords High altitude platforms (HAPs) Radial base function network (RBFN)
Hand-off algorithm Artificial neural network (ANN)

1 Introduction
Cellular communications provides communication facility to mobile subscribers (MSs). A
service area is divided into a number of cells [1]. Several such cells constitute a cluster. The

S. H. Alsamhi (B) N. S. Rajput

Department of Electronics Engineering, Indian Institute of Technology (Banaras Hindu University),
Varanasi, UP, India
e-mail: s.alsamhi.rs.ece@iitbhu.ac.in
N. S. Rajput
e-mail: nsrajput.ece@iitbhu.ac.in
S. H. Alsamhi
IBB University, Ibb, Yemen

123

S. H. Alsamhi, N. S. Rajput

available frequency spectrum is used in each cluster. Each cell in a cluster uses a fraction
of the available channels in the spectrum allocated according to a channel assignment strategy and is served by base station (BS). Hand-off is a common technique employed by all
cellular systems, both terrestrial and satellite, which has been proven vital both for ensuring
uninterrupted connections and increasing system capacity [2,3].
A hand-off is the process of transferring a mobile stations serving BS from one to another
when the mobile station moves across the cell boundary. A properly designed hand-off
algorithm is essential in reducing the switching load of the system while maintaining the
desired QoS of the call in progress and a low probability of blocking new calls. The handoff process determines the spectral efficiency and the quality perceived by users [4]. For
enhancing the capacity an efficient hand-off algorithms is required.
HAPs is airplane or airship that operates at altitude 1721 km [5]. It provides line of sight,
better channel condition as well as high coverage area [6]. The speed of wind is sufficient low
in HAPs position. Coexistence of HAPs and terrestrial systems using spectrum etiquettes is
investigated [7]. The coverage area of HAPs is divided into three zone that are urban area
coverage (UAC), suburban area coverage (SAC) and rural area coverage (RAC) [6,8].
In this paper we are assuming that the platform of HAPs will be moved in vertical and
horizontal which will effect on the coverage area and Hand-off process. It is possible to employ
a combination of hand-off techniques and a steering mechanism, to avoid interruptions on
the link between the user and the platform. To do so, either the customer premises equipment
(CPE) should be keep track of the platform and / or the HAPs itself should employ an antenna
steering mechanism to maintain a constant coverage. HAPs is proposed as a complementary
BS to mobiles in an obstacle position as shown in Fig. 1. HAPs can supply services to the
mobile having weak signal from the serving terrestrial BS influenced by shadowing, turning
corner as well as being outside the terrestrial coverage.
Recently, ANN have been applied to many diverse problems. Neural network is trained to
predict a users transfer probabilities [9]. To achieve an efficient handoff, ANN is explored.
ANN helps in taking the handoff decision based on receive signal strength (RSS), bandwidth,
delay etc. Combination of these parameters, then carry on training. After training ANN is
capable of taking appreciate and efficient hand-off decision.
The rest of this paper is organized as follows. In Sects. 2 and 3, classification of hand-off
and desirable features of hand-off have been described, respectively. The HAPs movement
has been described in Sect. 4. In Sect. 5, neural network algorithm has been carried out and
the results have been shown in Sect. 6.

2 Classification of Hand-off
The hand-off process determines the maximum number of calls that can be served in a given
area [10]. Figure 2 shows a simple hand-off scenario in which an MSs travels from BS-A
to BS-B. Initially, the MSs are connected to BS-A. The overlap between the two cells is the
hand-off region in which the mobile may be connected to either BS-A or BS-B. At a certain
time during the travel, the mobile is handed-off from BS-A to BS-B, When the MS is close
to BS-B.
Hand-off classification can be classified in several ways [11] depend on type of type of
network, number of connection and entity as shown in Fig. 3. In first type, type of network represents as horizontal and vertical handoff, hard and soft handoff, mobile-controlled,
mobile-assisted, and network-controlled handoff.

123

An Intelligent Hand-off Algorithm

Fig. 1 Concept of HAP cellar

Fig. 2 Hand-off cellular system

123

S. H. Alsamhi, N. S. Rajput

Fig. 3 Hand-off classification

In Horizontal occurs when the MSs move between different BS of the same network. On
the other hand, vertical handoff occurs when handoff is required between different wireless
networks. Second type is number of connection which represents by hard and soft. Hard
handoff, the MS must break its connection from the current access network before it can
connect to a new one. But the MS can communicate and connect with more than one access
network during the handoff process in case of a soft handoff. Third type is depends on
entity and represents by mobile-controlled, mobile-assisted, and network-controlled handoff,
mobile-assisted handoff is the hybrid of mobile-controlled and network-controlled handoff
where the MS makes the handoff decisions in cooperation with the access network.

3 Desirable Features of Hand-off

A seamless hand-off is typically characterized by two performance requirements [12]:
a. The hand-off latency should be no more than a few hundreds of milliseconds.
b. The QoS provided by the source and target access networks should be nearly identical
in order to sustain the same communication experience.
Figure 4 describes several desirable features of hand-off algorithms as mentioned in the
literature [13]. Some of these features are described below:
1. Hand-off should be fast enough to avoid service degradation.
2. Hand-off should be reliable such that the MSs will be able to maintain the required QoS
after hand-off.
3. Successful handoffs to total attempted handoffs should be maximized.
4. Number of Hand-off: the number of hand-off must be minimized.
5. The effect of handoff on QoS should be minimal.
6. The handoff latency should be low.

123

An Intelligent Hand-off Algorithm

Fig. 4 Desirable hand-off features

4 HAPs Movement
Really, the position of HAPs is not fixed and will vary with time dependent on the prevailing
winding conditions in the stratosphere. Investigation of the potential use of phased array
technology was done in [14] to cope with platform movement. When the platform is moving,
it would also be necessary to compensate motion by electronic or mechanical means in order
to keep the cells stationary, or to hand-off connections between cells as is done in cellular
telephony.
4.1 Vertical Shifting
HAPs comprises of individual antennas for each cells on the ground which is fixed in relation
to each other as shown in Fig. 5. Thus the coverage area on the ground has a fixed subtended
angle. The coverage area of HAPs can be calculated by the following formula [15]:

[(h + h1 ) tan ]2 (h tan )2 for upward vertical shift
A =
(1)
[(h h2 ) tan ]2 (h tan )2 for downward vertical shift
where, A is the coverage area, h is the altitude, h1 and h2 is the change in upward and
downward height respectively, is subtended angle which is fixed.
4.2 Horizontal Shifting
HAPs movement can change position or distort the shape of the individual cells. In the case of
the HAPs drifts from the center of the coverage area, cells move from their intended position.
The coverage will be increase in the direction of the platform and the user in opposite direction
will lose coverage as shown in Fig. 6. The approximate coverage as the following:

123

S. H. Alsamhi, N. S. Rajput

Fig. 5 Vertical shifting up and down of HAPs

Fig. 6 Horizontal shifting of HAPs

123

An Intelligent Hand-off Algorithm

Fig. 7 Steerable antenna solution for hand-off

A disp(disp2 4r)0.5

(2)

where r is the coverage radius and disp is the horizontal shifting.

4.3 Hand off and Steerable Antenna
Steerable antennas can be used to cope with the movement of the HAPs. HAPs movements
have been addressed in the past such as in [1,15] and in [16] where various techniques have
been proposed to cope with various movements. Mechanism of antenna is proposed in order
to counterbalance the horizontal displacement with the ideal position of the HAPs and the
relevant correction required being specified using a Global Positioning System (GPS) [16].
Steerable antenna correction mechanism was proposed, which needs to be applied on every
antenna individually [15]. However, this would require a complex mechanical system with a
large number of motors and therefore it would add significant weight to the payload.
It was preferable that HAPs system would employ some sort of mechanically steerable
mechanism but for a group of antennas instead. As shown in Fig. 7, when HAPs moves
upward, the antennas will be pushed inward, and the center will move little upward. In the
other hand, when HAPs moves downward, the antennas will be pushed outward, and the
antenna at the center will move a little downward.
In case of horizontal movement, the steerable antenna of the centre cell is always pointing
to the centre of the HAPs coverage area, and all the antennas are interconnected with each
other. Rotation adjustment can make all the antennas pointing to the original position on the
ground, but elevation angle is different from the original angle.

123

S. H. Alsamhi, N. S. Rajput

Fig. 8 Model of artificial neuron

5 Neural Network Algorithm

Recently, ANN has been applied to many diverse problems. ANN is one tool of artificial
intelligence (AI). An ANN is a massively parallel distributed processor that stores experimental knowledge; this knowledge is acquired by a learning process and is stored in the form
of parameters of the ANN [17].
The ANN consists of a number of neurons arranged in a particular fashion. The three basic
elements of a neuron are the synaptic weights (or weights), the summing junction, and the
activation function. In Fig. 8 explains the fundamental component of the ANN, an artificial
neuron. Different activation functions include hard limit, linear, log-sig. threshold k can be
considered as one of the weight. The ANN consists of more than one neuron. The output of
a neuron k is given by:
uk =

n


Wkj Xj

(3)

Yk = f (uk k )

(4)

j=1

where Xj ( j = 1, 2, . . . . . . . . . . . . . . . , p) are the input, Wkj are weights, k is the threshold,

F(..) is the activate function, and Yk is the output of neuron.
ANN characteristics are massively parallel distributed architecture, ability to learn and
generalize, fault tolerance, nonlinearity, and adaptively. The learning in ANN can be unsupervised or supervised.
5.1 Radial Based Function Network
The RBFN consists of three different layers, an input layer, a hidden layer, and an output
layer as shown in Fig. 9. The input layer acts as an entry point for the input vector; no
processing takes place in the input layer. The hidden layer consists of several Gaussian
functions that constitute arbitrary basis functions (called radial basis functions); these basis
functions expand the input pattern onto the hidden layer space. This transformation from the
input space to the hidden layer space is nonlinear due to nonlinear radial-basis functions.
Two distinct phases of learning in the RBFN are selection of enters of the radial basis
functions and determination of linear weights. Some of the methods for the selection of
RBFN centers are random selection (based on the training patterns), unsupervised selection,
and supervised selection. Some of the methods for linear weight determination are pseudo-

123

An Intelligent Hand-off Algorithm

Fig. 9 RBF neural network

inverse memory and LMS algorithm. These weight determination methods and a mapping
between the hidden unit space and the output layer.
The output layer linearly combines the hidden layer responses to produce an output pattern.
The rationale behind the working of the RBFN, a pattern-classification problem expressed in
a high-dimensional space is more likely to be linearly separable than in a lower-dimensional
space. The parameters of the RBFN weights (in the output layer) and the positions and spreads
of the Gaussian functions. A complete learning procedure can be found in [17].
Input nodes are RSS of MS and BS, traffic intensity of MS and BS, steerable antenna,
elevation angle of HAPs, delay, bandwidth, HAPs position and distance between MS and
next BS. The output equals the summation of hidden layer. The output decides whether the
system needs hand-off or not. When Y 1 and Y 2 are equal to 0 that mean no hand-off will be
performance. If Y 1 and Y 2 are equal to 1, the system will hand-off the mobile to chosen the
BS.
Wk1 (n) = [Wk (n) , . . . . . . . . . . . . , Wk20 (n)]
(5)
Initialize all the following, the center value ji (0), the span value j (0), weight vector
WK (0), expect W11 (0) = W21 (0) = 1. Calculate the output of hidden layer and output
layer are given respectively by:
Zj = exp(((xi ij (n)))2
Yk = R

M


2 (n)

(6)

Wkj (n)ZJ , k = 1, 2; M = 20

(7)

J=I

The error calculates by:

ek = dk yk

(8)

where dk {0, 1] desired pattern and update the weight given by:
Wkj (n + 1) = Wkj (n) w ek zj

(9)

123

S. H. Alsamhi, N. S. Rajput

where w and represent the learning rate of weight and center respectively, update the
center and span momentum:
ij (n + 1) = ij (n) +

zj 


xi ji (n)
ek wkj (n)

j

2 zj
ek wkj (n)
lnzj
j (n + 1) = j (n)
j (n)

(10)
(11)

where learning rate of span. Repeat the steps until the mean square error convergence less
than small number.
5.2 Hand-off Algorithm
After every small time interval, the simulator checked whether the position of the HAPs had
changed as shown in Fig. 10 and then initiated hand-off if required.
If the position of the platform has changed, then all that users have been affected. Then the
users must be added back into the system and connected to the new cell. This is to eliminate
the case where users are being dropped from a cell that is waiting for some of its current users
to be connected to another cell. In this case the cell will have a number of channels available
as soon as its hand-off users release the channels they occupy. The point is to ensure that
these channels are available for the new hand-off users coming to the cell. The affected users
are only a small proportion of the total number of users within a cell.
Since the capacity is allocated on a case by case basis, the overhead will not be significantly
high. There are major feature of hand-off algorithm and several desirable feature of handoff algorithm should be fast, successful, the effect of hand-off on the equality of services
should be minimum, should be maintain the planning cellular borders to avoid congestion,
the number of hand-off should be minimized, target cell should be chosen correctly minimal
effect on new cell blacking, procedure should be minimize the number of connecting call
drop outs by providing desired QoS.
Traffic intensity is the average number of calls simultaneously in progress during a particular period of time. It measured in units of Erlangs. Thus 1 Erlang equals 1*3,600 call
seconds. Traffic intensity is equal to the summation of circuit holding time divided by the
duration of monitoring period.
I = Nc t/T
(12)
Where, I is traffic intensity, T is duration of monitoring period is average holding time. Nc
is total number of calls in monitoring period.
There are two type of traffic which either infinite or finite. Infinite traffic implies number
of call arrivals, each with a small holding time. In other hand,when the number of sources
offering traffic to group of trunks or circuits is comparatively small in comparison to the
number of circuits, this call finite traffic.
5.3 Comparison of Hand-off Approaches
The decision phase is the most important one in hand-off, the network performance, satisfactions, efficiency, flexibility, and complexity and reliability of the overall algorithm. The
different combinations of these criteria can be used to perform hand-off decisions: Bandwidth
(BW), Signal to Interference Ratio (SIR), delay, response time, network coverage area, Bite
Error Rate (BER), RSS, traffic load, and number of user (Tables 1, 2).

123

An Intelligent Hand-off Algorithm

Fig. 10 Hand off algorithm

123

S. H. Alsamhi, N. S. Rajput
Table 1 Comparison of RSS and
ANN based network performance

Hand-off feature

RSS

ANN

Multi criteria

No

Yes

User performance

No

Medium

Flexibility

Low

Medium

Complexity

Low

High

Efficiency

Low

High

Table 2 Compaction of hand-off algorithm methods

Hand-offs features

RSS

RSS with threshold

ANN

Resource management

Signal strength

Signal strength

Ping pong effect

Yes

Avoided

RSS, SIR, velocity, available

power, user performance,
BW, etc.
Avoided

Hand-off latency

Low

Low

Reduced

Number of hand-off

High

Reduced

Reduced

Number of unnecessary hand-off

High

Reduced

Low

NLOS

Possible

Possible

Can be avoided

0.25
RBF neural network

0.24

Backprop neural network

hand off rate

0.23
0.22
0.21
0.2
0.19
0.18
0.17
3

3.2

3.4

3.6

3.8

4.2

4.4

4.6

mean arrive time

Fig. 11 Hand-off rate versus mean arrival time

6 Result
The RBFN is used for making a hand-off decision for chosen neighbor BS. The input to
neurons consist combination of parameters which are required for taking hand-off decision.
Steerable antennas are used in HAPs, therefore movement of HAPs (vertically or horizontally) had not any effect in hand-off decision. Positioning MSs is obtained by apply the timing
advance concept. When mean arrival time increases the hand-off rate decreases smoothly as
shown in Fig. 11. On the other hand, hand-off rate increases when traffic intensity increases
as shown in Fig. 12.
The important of RBFN is shown in Figs. 11 and 12 for taking an efficient hand-offs.
Using number of parameters help RBFN to take appropriate and efficient hand-off decision
and the unnecessary hand-off reduces.

123

An Intelligent Hand-off Algorithm

0.29
0.28

hand off rate

0.27
0.26
0.25
0.24
0.23
0.22

RBF neural network

0.21

Backprop neural network

0.2
0.65

0.66

0.67

0.68

0.69

0.7

0.71

0.72

0.73

0.74

0.75

traffic intensity

Fig. 12 Hand-off rate versus traffic intensity

Handoff probability with diff speed

0.722

Probality

0.7215

0.721

0.7205

0.72

0.7195

10

20

30

40

50

60

70

80

speed(Km/h)

Fig. 13 Hand-off probability rate

Figure 13 clarifies the relationship between speed of user and hand-off rate. So that, if the
speed increases the hand off probability will increase.

7 Conclusion
A high performance hand-off algorithm provides many desirable features by making appropriate hand-off. The advantage of HAPs is that it can provide services to the users either they
are getting weak signals from the terrestrial systems or they are at the covered area influenced by shadowing. The RSS, direction of MSs, HAPs position, Traffic intensity, steerable
antenna, elevation angle of HAPs and delay are input of the neural networks. Effective handoff algorithm is done based on RBFN for combination of HAPs and terrestrial systems. As a
result, hand-off rate and dropping rate decrease as compared with other traditional methods.
Therefore, the hand-off rate increases when traffic intensity increases. As well as hand-off
rate decreases when mean arrival time increases.

123

S. H. Alsamhi, N. S. Rajput

References
1. Alsamhi, S. H., & Rajput, N. S. (2014). HAP antenna radiation pattern for providing coverage and service
characteristics. In Advances in computing, communications and informatics (ICACCI, 2014 International
Conference on, 14341439).
2. kumar, A., & Purohit, H. (2013). A comparative study of different type of handoff strategies in cellular system. International Journal of Advanced Research in Computer and Communication Engineering
(IJARCCE), 2, 42784287.
3. Alsamhi, S. H., & Rajput, N. S. (2014). Performance and analysis of propagation models for efficient
handoff in high altitude platform system to sustain QoS. In Electrical, Electronics and computer science
(SCEECS), 2014 IEEE Students Conference on, 16.
4. Stavroulakis, P. (2004). Fuzzy-neural applications in handoff. In P. Stavroulakis (Ed.), Neuro-fuzzy and
fuzzy-neural applications in telecommunications (pp. 149234). Berlin Heidelberg: Springer.
5. Alsamhi, S. H. A., & Rajput, N. S. (2012). Methodology for coexistence of high altitude platform ground
stations and radio relay stations with reduced interference. International Journal of Scientific & Engineering Research, 3, 17.
6. Alsamhi, S. H., & Rajput, N. S. (2015). An intelligent HAP for broadband wireless communications:
Developments, QoS and applications. International Journal of Electronics and Electrical Engineering,
3, 134144.
7. Alsamhi, S. H., & Rajput, N. S. (2014). Neural network in a joint HAPS and terrestrial fixed broadband
system. International Journal of Technological Exploration and Learning (IJTEL), 3, 344348.
8. Alsamhi, S. H. A., & Rajput, N. S. (2012). Interference environment between high altitude platform station
and fixed wireless access stations. International Journal of Engineering Research and Applications, 2,
15081513.
9. Akoush, S., & Sameh, A. (2007). Mobile user movement prediction using bayesian learning for neural networks. Proceedings of the International Conference on Wireless Communications and Mobile Computing,
191196.
10. Kim, J.-S., Serpedin, E., Shin, D.-R., & Qaraq, K. (2008). Handoff triggering and network selection
algorithms for load-balancing handoff in CDMAWLAN integrated networks. EURASIP Journal on
Wireless Communications and Networking, 2008, 114.
11. Saini, M., & Mann, S. (2010). Handoff schemes for vehicular ad-hoc networks: A survey. International
Journal of Innovations in Engineering and Technology, 8691.
12. Taaghol, P., Salkintzis, A. K., & Iyer, J. (2008). Seamless integration of mobile WiMAX in 3GPP networks.
Communications Magazine, IEEE, 46, 7485.
13. Nasser, N., Hasswa, A., & Hassanein, H. (2006). Handoffs in fourth generation heterogeneous networks.
Communications Magazine, IEEE, 44, 96103.
14. Park, J.-M., Ku, B.-J., Kim, Y.-S., & Ahn, D.-S (2002). Technology development for wireless communications system using stratospheric platform in Korea. In Personal, Indoor and Mobile Radio Communications, 2002. The 13th IEEE International Symposium. Vol. 4, pp. 15771581.
15. Capstick, M. H., & Grace, D. (2005). High altitude platform mm-wave aperture antenna steering solutions.
Wireless Personal Communications, 32, 215236.
16. Axiotis, D. I., Theologou, M. E., & Sykas, E. D. (2004). The effect of platform instability on the system
level performance of HAPS UMTS. Communications Letters, IEEE, 8, 111113.
17. Haykin, S. (1999). Neural networks: A comprehensive foundation. (2nd ed.). USA: Tom Robbins.
S. H. Alsamhi received the B. E. from Department of Electronic Engineering (Communication Division), IBB University, Yemen, in 2009.
In 2009, He worked as lecturer assistant in faculty of Engineering,
IBB University. He received M. Tech degree in Communication Systems, Electronics Engineering, Indian Institute of Technology (Banaras
Hindu University), IIT (BHU), Varanasi, India in 2012. He is currently pursuing Ph.D. degree program in same department. His area
of interest is in the field of wireless communication, Satellite Communication, WiMAX, Communication via HAPS and Tethered Balloon
Technology.

123

An Intelligent Hand-off Algorithm

N. S. Rajput received Ph.D. degree in the area of Intelligent Data
Analysis and Pattern Recognition in 2011, from Indian Institute of
Technology (BHU), Varanasi. He received the M. Eng. degree in communication systems in 1997. He is presently working as an Assistant
Professor (Stage-III) in the Department of Electronics Engineering, IIT
(BHU). His research interests include Intelligent Techniques on Networked Communication and Computation.

123