A Proactive Horizontal Handover Algorithm Based on RSSI Supported by

a New Gradient Predictor

Kholoud Atalah, Elsa Macías, Alvaro Suárez

Concurrency and Architecture Group (GAC)
Department of Telematics Engineering
University of Las Palmas de G. C.
Spain
[email protected]
{emacias, asuarez}@dit.ulpgc.es

ABSTRACT
Two important high bandwidth wireless certifications in our days are
Worldwide Interoperability for Microwave Access (WiMAX) and Wireless Fidelity
(WiFi). They can interoperate to form a wide geographical network in which
Mobile Client (MC) can move. We suppose a combination of two WiFi Access
Points (AP) and a WiMAX Base Station (BS) that interoperate putting a WiMAX
Wireless Network Interface Card (WNIC) in the WiFi AP. The BS can request real
time video or Video on Demand (VoD) from a server allocated in the fixed network
(to which the BS is connected). In order to model the movement of the MC, we
build a diagram state to represent the areas in which MC can stay, and from what
area to which one the MC could pass. Classifying the areas in connected and
disconnected ones, we can discover the different actions that are achieved by
different entities of a simple protocol, that are allocated in the BS and the AP.
These entities are in charge of controlling the buffers to efficiently support
disconnections due to handover and out of coverage situations (disconnected
states). Our algorithm is based on Received Signal Strength Indicator (RSSI)
measurements. These RSSI values are used to change the state in the diagram state.
But to predict the next state of MC, we use a new prediction technique that we call
RSSI Gradient Predictor. We test our predictor using some synthetic proposed
RSSI values for different MC movement shapes. These movements include
Regular and Irregular movements. The results show a good behaviour of our
algorithm.

Keywords: Buffer Management, Disconnections Prediction, Multimedia, RSSI

Gradient, Streaming, WiFi, WiMAX.

1 INTRODUCTION conditions deteriorate, data are lost, and the

medium saturates with retransmitted failed data,
Physical properties of WiMAX and WiFi which cause poor multimedia playing quality at the
channels make difficult to maintain client MC [2]. As a result, sporadic disruptions could
connections and guarantee the necessary Quality of occur.
Service (QoS) [1] for multimedia services. In our days, a MC is provided from fabric with
The RSSI is used to define different areas of several WNIC (multi-homed systems that has
coverage. In outdoors, in an open area with no multiple network interfaces [3]) to allow them to
obstacles, it starts at 100% powerful surrounding access to different wireless technologies [4] to
the WiFi AP or WiMAX BS and decreases as MC extend the area in which they can be connected (but
moves far from the AP (BS), until it arrives fewer there is no one with WiFi and WiMAX at the same
than 10% where is the end of the coverage area. In time), but with IEEE 802.21 standard, MC will be
this last part of the coverage area, channel supported with one common interface to access all

Ubiquitous Computing and Communication Journal

technologies [5]. algorithm, RSSI is used to keep video flow
When MC starts to loss its connection to the transferred while handover.
associated AP and stops receiving data for several In [9] it is introduced a new concept named
seconds [2], it searches for another AP or BS to Takeover in which the Neighbor Node (NN) takes
associate with. This process is named roaming [6] over Mobile Node (MN) handover operation. They
and handover [7], [8], [9], [10], [11], [12] that is have defined signaling messages and protocol for
required to maintain channel connection and frame Takeover operations and applied it to the proposed
transmission while MC movement to be seamless VH scheme. Authors found that Takeover fails
handover [1], [4], [8], [9], [13], [14]. Handover is when connection between the nodes is broken or
classified, from the BS point of view [1], [13], [15] Takeover process maximum time is finished before
as Vertical Handover (VH) that occurs when MC MC receives the completion message. In our
moves between different networks that use different opinion, one disadvantage is the ability of NN to
technologies [16] and Horizontal Handover (HH) deny the request message of Takeover, or MN can
when these networks use the same technology [17]. not connect to any MC for any reason e.g. if it was
Depending on its procedures it is classified as in unreachable area. In the present paper, we do not
Reactive Handover (RH) [17], [18] that delays use any other MC; we depend mainly on the BS or
Handover as much as possible i.e. Handover starts AP which MC is associated to.
only when MC completely losses its current AP In [10] authors divided the handover process
signal, and Proactive Handover (PH) which into two steps: the discovery of new AP by passive
triggers Handover before the complete loss of or active scanning, and the re-authentication that
original cell signal [17]. Two strategies under this involves authentication and re-association to new
type are available: Hard Proactive (HP) and Soft AP. They have demonstrated clearly that delay
Proactive (SP) [7], [19]. variability is due to scanning phase while re-
Many continuous multimedia services like VoD, authentication delay is constant, and the delay can
Voice over Internet Protocol (VoIP), and Real Time be reduced by the proper choice of the AP position.
Chatting communicate a continuous flow of packets. The system can support VoIP that is sensitive to the
Handover influences many parameters of delay and jitter. Up to our knowledge this technique
multimedia services such as: bandwidth, security, is not appropriated for video streaming. Scanning is
traffic shape, RSSI, packet loss and service a hard process that wastes time, and it is not needed
connection recovering. If the handover process in our proposed approach. Measuring RSSI is
takes a long time, then a lot of packets will be lost enough to decide handover necessity.
and the user of MC will not follow the video or The handover mechanism proposed in [22] is
audio adequately [6]. Efficient proactive HH design based on a priori knowledge of other networks
considering multimedia communication is a big depending on combined handover probabilities. It
challenge. increases handover success rate of up to 22%
In this paper, we present a proactive HH compared to classical blind handover. But in our
theoretical mechanism designed for WiFi networks opinion, these probabilities change rapidly because
of which AP are connected to a WiMAX BS MC changes its position frequently. In our proposed
considering a Random Walk Mobility Model [20] algorithm, our RSSI Gradient Predictor measures
(we do not include special profile, e.g. streets or RSSI frequently and the handover protocol just
tunnels). To complete the study of possible detects MC position changes which reduces time
disconnections, we also analyze the case the MC and memory needed to store these data.
losses the connection because it goes out of Anticipating, as much as possible, the
coverage. Different possibilities of disconnections information about MC movement is crucial for
are also considered: for example the MC is doing reactive handover. In [17], a migration process of
handover but it goes out of coverage and returns to mobile proxies in advance to the wireless cells
an area of coverage of another AP. where MC is going to reconnect, it is proposed (it
The mobility management system proposed in increases the size only when a client handover is
[1] may keep connections based on the end-to-end going to occur, and decreases the size when
principle by incorporating an intelligent network handover is not near, which reduces memory and
status detection mechanism. Although it in general bandwidth usage). These authors also propose
considers QoS it does not focus on multimedia mobility prediction solutions depending on the
applications such as VoD. Grey Model [23] and a way to exploit handover
The Handover algorithm proposed in [21] prediction to optimize client side pre-fetching
depends on the estimated mobility using RSSI. This buffers usage for streaming data depending on
algorithm is able to avoid ping-pong effect and to client side RSSI from the visible AP [24] (their
enhance overall Handover process performance. solution depends on reducing the buffer size needed
However they have not shown how it affects on to keep continuous streaming and to impose a very
packet lost and multimedia applications. In our limited overhead by exploiting RSSI data already

Ubiquitous Computing and Communication Journal

available at client). The performance of our RSSI in the environment of MC as 3GPP, IEEE 802.11 or
Gradient Predictor is better than Grey model 802.16, and IETF, and optimize VH between them
performance because it shows good behavior with by using multiple services [16]. IEEE 802.21
all movement shapes whereas the Grey model can enables co-operative handover decision between
not predict in the case of RSSI is constant. network (provider) and MC (user). The main
Reference [25] shows that Forward Handover using protocol, in IEEE 802.21, that controls handover
Partial Re-establishment is better than using the and provides information is called Media
backward one (using “Full Re-establishment, Independent Handover Function (MIHF) [26].
Multicasting, or Connection Extension”). It In IEEE 802.11 standard the driver of the
proposes a packet discarding mechanism where the WNIC is in charge of measuring the RSSI as an
information carried by high priority packets, is integer ranging from 0 to 255 (1 byte of size). Real
considered by the buffers to be more important than vendors choose a set of these 256 values ranging
information carried by low priority packets. In our from 0 to a maximum value (RSSI_Max) defined
proposed algorithm, we store the video when by them. For example Cisco chose RSSI_Max=100,
Handover is expected to happen, in Mobile Client while Atheros choose RSSI_Max=60. It does not
Buffer (MCB) or in Base Station Memory (BSM) in represent a power value in decibels (dB) or dB
order not to loss frames. Also our RSSI Gradient referenced to one mili Watt (dBm). The RSSI does
Predictor determines exactly where the MC will be not represent the full current power received by the
after any interval of time. WNIC what led many vendors to implement
The rest of the paper is organized as follows. In mapping tables to convert RSSI to dBm. IEEE
section 2 we outline the network architecture and 802.11 indicates that RSSI can be defined using
mobility pattern of MC. Section 3 is devoted to relative values to RSSI_Max as
present our mechanism that contemplates handover RSSI % = (RSSI × 100) / RSSI _ Max (1)
and also areas with no coverage. We present our RSSI% is used to assert if MC is inside
RSSI Gradient Predictor in section 4, and support it coverage area of a particular WiFi AP. Moreover,
with numerical results to show its performance. categorizing its values we can discriminate the part
Finally, we summarize our conclusions and present of the coverage area in which MC is. Let us note
directions for further research. that a Global Position System (GPS) set in the MC
it is not necessary because we are not interested in
2 THE NETWORK ARCHITECTURE AND the exact position of the MC. We also do not need
MOBILITY PATTERNS an accurate localization method.
It is appropriated to categorize the coverage
We consider a network that combine area in three different parts:
infrastructure WiFi cells of which AP are connected • Area 1 ( 100 ≥ RSSI % > 60 ), is the most near
to a WiMAX BS. The AP has two WNIC: one for part to AP, where coverage is very strong
the WiFi cell, and the other for communicating with
(ranging from excellent to very good level) and
the BS. The BS is connected to a wired backbone
MC is in the best state of connection.
(wired Internet). In Internet there is a VoD server
that streams multimedia packets to MC connected • Area 2 ( 60 ≥ RSSI % > 40 ), is the middle part,
to the WiFi AP. The Server and the Client use where the coverage is not so strong (ranging
Transmission Control Protocol (TCP) and User from very good to good level) and MC is in a
Datagram Protocol (UDP) for signaling and good state of connection.
multimedia data communication respectively. Over • Area 3 ( 40 ≥ RSSI % > 20 ), is the farthest part,
these protocols other multimedia protocols like where the coverage is weak (ranging from less
Real Time Streaming Protocol (RTSP) for video than good and very close to no connection
communications can be used. level) and MC is near to a bad state of
We suppose all the MC outdoor and no connection.
important obstacles are present in the surrounding Before transmitting, the MC WNIC checks if
area. We do not take into account the presence of current measured RSSI is less than Clear Channel
complex buildings, cars or other elements that Threshold (CCT). If it is greater or equal than CCT
provoke strong interferences in wireless channels. then wireless channel is not clear to transmit. When
Under these assumptions, the wireless channel the MC is ready to receive, its WNIC must test if
behavior is not strongly chaotic, because the the RSSI (transformed in dBm) received is greater
interferences and path loss conditions are than Reception Sensitivity Threshold (RST) that is
moderated. measured in dBm (a value very close to 0 but not 0).
In Fig. 1 we show a graphic representing the If RSSI is equal to RST, the WNIC can not
above assumptions. differentiate between noise and signal [27]. In any
Recently, in IEEE 802.21 standard, multiple part of the classified coverage areas, these
wireless technologies are considered to be available conditions must be assured.

Ubiquitous Computing and Communication Journal

 which is called the Roaming Threshold (RT). In that
case a roaming or handover process must start. We
call this part as the Handover Area (AH). Taking
this into account, we can discover that when MC is
in Area 3, and a handover process starts then MC is
in Area H. It could cross to Area H of the other AP
and then to Area 3. Let us note that with this
consideration, MC can not cross from Area 3 of one
AP to Area 3 of the other AP directly.
In the case the MC goes out of WiFi coverage
and also is not in AH, it means that it is under
WiMAX coverage area. We name this part as
WiMAX Area (AW). In Fig. 2 we show the different
parts of coverage areas and a particular movement
pattern of the MC.
Regarding to the movement of the MC, we
Figure 1: Physical range of wireless networks consider the following assumptions:
interconnection − The movement pattern is the most
conservative: in any point the MC can proceed
to any direction with constant speed.
− The MC defines a smooth movement, that is, it
can pass from Area i to Area i+1 or i-1
( 1 ≤ i ≤ 3 , i=H or i=W) but it can not jump
between non consecutive areas. MC can be in
the same area along the time or simply change
to another consecutive area.
− As a consequence of the above assumptions,
the MC can not cross from the Area 3 of an AP
to the Area 3 of another AP directly: it must
cross from Area 3 to the AW or AH before to
pass to Area 3 of the other AP.

3 THE HANDOVER MECHANISM AND

BUFFER MANAGEMENT

Figure 2: Coverage area parts and MC mobility While a particular MC is crossing from WiFi
cell 1 to WiFi cell 2, it can experiment
disconnections due to different situations:
a) MC is in AH, so connection can be lost during
any interval of time, although MC has wireless
connection, but it is trying to reconnect to the
appropriate AP.
b) MC is in AW and it losses the WiFi connection.
In this case it can reconnect to the AP it was
connected or to a new one.
In both cases, the result is a loss of video
reception during sometime.
The challenge is to anticipate predicted
information to the MC in order not to loss video
during the above disconnections. This challenge
exhibits different solution for real time video and
for VoD. To do this, we consider a mechanism that
Figure 3: States and transitions of the cooperative every ∆t discovers the state of MC, and anticipates
protocol among MCBM, APP and BSBM MC state immediately after this time. Also it must
inform the state to different entities allocated in the
In the case the MC crosses between Area 3 of AP, the BS and MC in order them to do some
different AP, the MC that is associated to an AP protocol actions. The first entity, named AP Proxy
can move its connection to another one, if the signal (APP) is a simple proxy that forwards signaling
level received from this AP drops to a low value information from MC to BS and data from BS to

Ubiquitous Computing and Communication Journal

MC. The second one is the BS Buffer Manager clearly the situation in which the MC is in A3 of an
(BSBM) and the third one is the MC Buffer AP, the event Cross_A3-AH occurs, and then the
Manager (MCBM). event Cross_AH-A3 occurs again. In this case we
These three entities cooperate using a state have to differentiate the case in which the MC
protocol. In Fig. 3, it is shown the states and the returns to connect to the original AP or handover to
transitions between states. In this state diagram is the other AP to be then in state A3 for that AP. The
captured the idea that MC only can cross to same occurs for Cross_A3-AW (A3 for an AP) and
consecutive parts of the coverage area or continue Cross_AW-A3 (A3 for the other AP).
in the same part. Annotating the amount of time a A VoD server can increase or decrease its
MC is in any of these states gives us an idea of the speed of video transmission till to reach the
space and time trajectory of the MC. Following the maximum bandwidth allowed for that connection in
transitions we easily discover that the MC starts in the wired network. In the same way, MC can
Area 1 (it is supposed that it has associated to an request the BS to increase or decrease transmission
AP with excellent coverage), then it can proceed to speed in order to store bigger amount of video
Area 2 and then Area 3, be disconnected (AW) frames in its buffer. In this way it can support a
during an interval of time, return to AH or Area 3 wireless disconnection or handover taking frames
of the same or another AP. In this simple way, we from its buffer.
capture the handover and wireless disconnection of Let us note, using the RSSI Gradient Predictor,
the MC, also we can memorize the trajectory of the we can predict the speed and trajectory of MC
MC. Doing this, a profile of movement can be when it is going to reach the AW state. So we can
obtained, that can be used to infer future possible ascertain if it will return to a connection state early,
movements e.g. an employee probably defines or on the contrary it will leave the wireless
every day the same trajectory from home to the coverage and will not return in an interval of time.
work place. In the first case we can arrange a buffer in the BS
that can be used to support this interval of
3.1 Previous Considerations, Protocol disconnection. When the MC returns to the wireless
Primitives, and Transitions Actions connection it can receive from this buffer and video
MCBM is in charge of measuring the different frames will not be lost. In the second case, simply
parameters of coverage consulting the WNIC the APP can send a message to the BSBM telling
driver: RSSI_Max, CCT and RST. Calculating that not to buffer video frames, and stop the VoD
RSSI%, it can discover the current state, and using server. To do this, the MCBM must communicate a
RST it can discover if it is in the AH state. message to the APP telling that before it will be out
Prediction of transition inter-states is achieved of coverage.
using handover prediction techniques (in section 4, The entities of the protocol exchange a number
our RSSI Gradient Predictor is proposed). In some of messages that regulates their actions. These
cases, it is very simple, e.g. a linear trajectory from messages are exchanged using a compact set of
Area 1 to Area 2 is easily identified if RSSI% array primitives explained in Table 1.
has the values: 100, 95, 90, 92, 80, 70, 65 and 62 We suppose that the APP entity simply
(probably, next value will be less than 55). This forwards the messages associated to the above
means that MCBM is in state A1 and will be in primitives. For that reason we do not include
state A2 activating transition Cross_A1-A2. This explicitly their actions. When the MCBM detects a
produces an event that will be processed by MCBM. transition, it sends a message to the BSBM. The
All other transitions generate the corresponding main actions associated to transitions are explained
anticipated event. It is trivial that no relevant in Table 2.
actions are associated to the Still_in transitions:
simply the entities remain doing the same actions. 3.2 Additional Considerations for Real Time
The states AW and AH are “disconnected states”, Video
which means MC can experiment loss of video Real time video is a more complex task to
frames. We also must note that when the MC is out support because MC or BS can not request its peer
of coverage area, it can return to a disconnected to increase or to decrease the video frame
state (AH), so it is important to register in APP and transmission speed. The future frames that have not
BSBM this situation, because it could connect to been produced at any time can not be consumed
another AP when it returns to coverage. This before.
situation corresponds with the following events We apply the ideas expressed in [28] to support
(transitions): Cross_A3-AW, Cross_AW-AH wireless disconnections for real time video, but we
(indeed a succession of Cross_AW-AH and base our solution taking into accounts the diagram
Cross_AH-AW can occur: ping-pong effect). state of Fig. 3.
Another interesting point is to differentiate

Ubiquitous Computing and Communication Journal

Table 1: Primitives of the protocol
Message Abbre. Description Syntax
Increase Speed of Video ISVF The entity sends this primitive to entity:ISVF(server)1,2
Frames Transmission the server
Decrease Speed of Video DSVF The contrary of ISVF
Frames Transmission
Start Buffering Additional StBAF entity starts buffering forward entity:StBAF()1
Frames video frames
Stop Buffering Additional SpBAF entity stops seeking and buffering entity:SpBAF()1
Frames forward video frames to be
buffered
Consume Video Frames CVF The MCBM simply consumes MCBM:CVF()
frames of video allocated in its
buffer
Send Send MCBM:Send(BSBM,par)3
Predict Disconnection Time PDT dt=MCBM:PDT()4
1. entity: [BSBM | MCBM]
2. server: [VoD | BSBM]
3. par: [hc | hd | rc] where:
a) hc (handover cancelled): In this case the MCBM consumes its buffer and in parallel it receives video frames from the BSBM.
b) hd (handover done): In this case the MC realizes the disassociation from the original AP and association to the new AP and
then the same actions as for hc.
c) rc (return to coverage): When the MC is in AW and returns to A3 it will be able to send this message.
4. dt: It is an output parameter that represents the amount of future disconnection time for MC.

Table 2: Actions of the protocol associated to transitions

Transition Description Action
Cross_A1-A2 The BSBM buffers forward video frames BSBM:ISVF(VoD)
BSBM:StBAF()
Cross_A2-A1 The BSBM stops buffering forward video frames BSBM:DSVF (VoD)
BSBM: SpBAF()
Cross_A2-A3 The MCBM buffers forward video frames MCBM:ISVF(BSBM)
MCBM:StBAF()
Cross_A3-A2 The MCBM stops buffering forward video frames MCBM:DSVF (BSBM)
MCBM: SpBAF()
Cross_A3-AH MCBM:CVF()
Cross_AH-A3 a) A3 of original AP a) MCBM:Send(BSBM, hc)
b) A3 of new AP b) MCBM:Send(BSBM, hd)
Cross_A3-AW MCBM predicts the amount of buffer (depending on dt), MCBM:PDT()
in order to support the disconnection (it is inefficient to MCBM:BSBM:StBAF()
allocate more buffer if the MC will not return to A3) MCBM:CVF()
Cross_AW-A3 Let us note that it is sent independently to which AP it MCBM:Send(BSBM, rc)
will be associated again
Cross_AH-AW It must also predict the amount of time for it will be dt=MCBM:PDT()
disconnected MCBM:BSBM:StBAF()
MCBM:Send(BSBM,hc)
MCBM:CVF()
Cross_AW-AH MCBM:CVF()

We divide the buffers BSM and MCB into buffers simultaneously. When the transition
several parts defining a series of limits: LimAi, i=1, Cross_Ai+1-Ai occurs, the last LimAi video frames
2, 3 and W or H, With LimAi < LimAi+1. LimA2, are stored. We can support a wireless disconnection
LimA3 and LimAW are the limits associated to or handover because in the BSM is stored a set of
transitions: Cross_A1-A2, Cross_A2-A3, and the last video frames sent by the real time video
Cross_A3-AW (the same is applied for AH). LimA1 server. When the MC is in AW or in AH, it can
is associated to transition Still_in_A1. When the repeat the consumption of the last video frames
transition Cross_Ai-Ai+1 occurs, the last LimAi+1 stored in the MCB (from 0 - LimA3). When the MC
video frames (from 0 to LimAi+1) are stored in these returns to A3, it can request the BSBM the last

Ubiquitous Computing and Communication Journal

LimAW video frames. Then it can receive these next state allows the handover algorithm to
frames and the real time frames. generate the corresponding transition before the
In this very simple way, the MC is always MC changes connection state, permitting it to take
showing video frames to the user, but also no video actions in advance especially if the time of the
frames are lost if the disconnection period is not handover or disrupting could be predicted.
high (probably in this case the video session must If MC is moving with RM and we measure n
be finished because it is not practical to show the values of RSSI% (as a sample set) in interval of
last video frames sent a long time ago). time t then the
vector x = [ x 0 , x1 ,......., x n ] represents these
4 HANDOVER PREDICTION BY RSSI
GRADIENT values, where: x0 is measured at time t0 , and xn is
measured at time t . The Instant Gradient of RSSI
Referring to vector calculus definitions about at any time is:
the gradient, Gradient of RSSI can be defined to
calculate next area in which MC could be because ∇xk = xk − xk −1 (2)
of its relation with time and location. Then we have the gradient vector:
∇x = [∇x1 , ∇x 2 ,........, ∇x n ] (3)
Definition 1: Coverage Area Scalar Field is a
2-Dimentional space with a real RSSI value And the Instant Gradient with respect to time at
attached to each point in the space. We define any time:
different areas inside the scalar field that ∇x k
corresponds with the ones in Fig. 2.
∇( x k ) t = (4)
∆t k
Definition 2: Coverage Area Vector Field is a ∆t k = t k − t k −1 (5)
scalar field with a vector associated to each point in The Average Gradient with respect to time
the space; this vector field defines the Gradient of after an interval of time equal t is:
these scalar values (RSSI values). This gradient has
∇xn
the direction to the greatest value of RSSI in the ∇( xn )t = (6)
vector field, which is the center where the AP is ∆t
positioned. It means, if the gradient has the opposite ∇x n = x n − x 0 (7)
direction, it will be a negative value (Fig. 4). The
gradient can be a function of other variables like the ∆t = t − t 0 (8)
velocity (it is a function of time). This leads us to Then, if x n was unknown, it can be calculated
define the gradient of RSSI with respect to time
(because time is needed to predict the next from the following equation if all the other
connection state of MC defined in Fig. 3). variables are known:
x n = (∇( x n ) t × ∆t ) + x0 (9)
Definition 3: Regular Movement (RM) is
To calculate x at any time tk , the same
defined if the absolute value of RSSI% gradient is
constant along the time of this movement, and the equation can be used changing the interval of time
speed of MC is constant. Global contribution of a and we will call this value of x the predicted
RM can be done following the sense of the value:
Coverage Area Vector Field (forward to the AP xˆk = (∇( xn )t × ∆t ) + x0 (10)
position) or on the contrary (backward to the AP
position). There are different classes of RM (Fig. ∆t = tk − t0 (11)
5): Straight line (A, B), Sine (SIN) or Cosine (COS), But if we take in mind that in RM where MC
Zigzag, Spiral (D, E, F) ... in each class the MC can moves with constant speed and direction, it is
cross one or more times the limits of a particular correct that Average Gradient equals the average of
Area. the summation of Instants Gradient in the same
interval of time:
Definition 4: Irregular Movement (IM) is
1 n
defined if the absolute value of RSSI% gradient is
not constant, this means that the speed and direction
∇( x n ) t = ∑ ∇( x k ) t
n k −1
(12)

of MC is not constant, and its movement has not a Then the predicted value of x will be:
known regular shape (shape G, Fig. 5).
xˆk = (∇( xn )t × ∆t ) + x0 (13)
4.1 The RSSI% Gradient Predictor ∆t n

Predicting the next MC connection state is xˆk = x0 + ∑ ∇( xk )t

n k =1
(14)
crucial for any handover mechanism and disrupting
multimedia services control in general, because the

Ubiquitous Computing and Communication Journal

 (t k − t0 ) n xk − xk −1
xˆ k = x0 +
n
∑ (
k =1 t k − t k −1
) (15) 4.2 RSSI% Gradient Predictor Performance
In this section, verification is showed by
proposed synthetic values of RSSI% to explain
differences between movement shapes available in
Any RSSI value can be estimated using this Fig. 5. These values depend on the distribution of
formula, exactly at the time we want to know it. RSSI values on the Coverage Area Vector Field,
Using Table 3, it can be easily estimated the state and MC moves with average constant speed in RM
where MC will be. (shape A, B, C, D, E, and F). RSSI% values are
measured each 1 s in all RM, and we will take 15
values as sample set in all movements except in
shape G we will increase samples to be 30. Fig. 6
demonstrates the movement in straight line with
increasing gradient results positive slope, which
means MC moves towards AP. Whereas Fig. 7
shows decreasing gradient with negative slope, it
means MC moves backward AP. By these two
graphs we can get the value of RSSI after any
interval of time.
Figure 4: Coverage area scalar field In Fig. 8, MC moves with constant speed and
constant RSSI, this means it moves in the circular
line (shape C in Fig. 5) around the AP, the slop in
Table 3: Predicted values and MC states this figure equal zero as the gradient.
Fig. 9 describes different motions where the
Instant Gradient is not equal the Average Gradient
x̂ k state because MC moves with zigzag motion. Here the
prediction does not give precise results as MC does
100% ≥ xˆk > 60% A1 not move in constant direction. The prediction in
this case depends on the number of values that are
60% ≥ xˆ k > 40% A2 taken as samples, i.e. as the samples increased as
the prediction of the final state is more precise.
40% ≥ xˆ k > 20% A3 The last IM is shown in Fig. 10; in this case we
take more values of RSSI as samples to increase
prediction accuracy.
20% ≥ xˆ k > 0% AH
It is important to notice the behaviour of our
Predictor in Figures 9 and 10. In these two figures,
xˆ k = 0 AW the Predictor can only detect the global contribution
of the gradient of RSSI%, the Predictor does not
detect the instant variation because it is a linear
predictor. So in the case of zigzag motion (as shape
E), if MC moves from A1 to A2 and goes back
many times (A1, A2, A1, A2 …), the predictor can
not detect all these instant transition, It predicts the
final state of MC. One advantage of this case is we
can avoid a lot of entire actions related to these
instant transitions.

100
90
80
70
60
RSSI %

RSSI
50
Gradient
40
30
20
10
0
0 2 4 6 8 10 12 14 16
Time (s)

Figure 5: Coverage area vector field and RM Figure 6: Gradient if MC moves as shape A
classes

Ubiquitous Computing and Communication Journal

 100 will change its state.
90
80 A simple graph definition shows us in a simple
70
way the different areas in which the MC can be and
RSSI (%)
60
RSSI
50
40
Gradient also the possible solutions using a simple protocol
30
20
based on very simple primitives.
10
0
We plan to simulate this algorithm using a well
0 2 4 6 8 10 12 14 16 known simulator. For example the speed of MC can
be simulated in order to obtain accurately the limits
Time (s)

of the buffers in MC and BS.

Figure 7: Gradient if MC moves as shape B We have supposed a simple scenario in which
terminals can experiment a gradual transition of
100 coverage (no jumps). Another study can be done for
90
80 strong chaotic wireless channels in which although
70
60 MC is static, it could experiment disconnections.
RSSI %

RSSI

We also will study more complex situations in

50
Gradient
40
30
20 which more than one AP overlap their coverage
10
0 areas. We must define new kind of parts of area in
0 2 4 6 8
Time (s)
10 12 14 16
which three or more APs overlap their coverage
areas. Also the automaton must be extended to
consider the new areas; this implies that the
Figure 8: Gradient if MC moves as shape C protocol also must be extended.
We presented a RSSI% Gradient Predictor
50 technique from the RSSI gradient calculated from a
45
40
sample set. This Predictor detects the global
35 contribution of RSSI% gradient which means it
30
predicts the final value of RSSI% that can decide
RSSI %

RSSI
25
20
Gradient
the exact time to store the video. We are now
working to develop a new Filter named RSSI%
15
10
5 Gradient Filter. This filter will solve many points:
0
0 2 4 6 8 10 12 14 16 1. It will detect all instant states that can not be
Tim e (s) detected by the RSSI Gradient Predictor as
zigzag, SIN, COS, spiral or IM cases.
Figure 9: Gradient if MC moves as shape E 2. These instant states can be buffered and stored
for future needs, for example to study the route
of MC.
3. All these states are very important to trace
100
90
80 transitions between non consecutive states.
70
60 Finally, we find that, the RSSI% Gradient
RSSI %

RSSI
50
40
Gradient Predictor can be used to know the exact time to
30 store the video and the RSSI% Gradient Filter can
20
10
be used to avoid all the entire actions that are not
0 necessary.
0 5 10 15 20 25 30 35
Time (s)

ACKNOWLEDGEMENT
Figure 10: Gradient if MC moves as shape G
This work was supported in part by the Spanish
CICYT (MEC) and European Research
Development Fund (FEDER) under Grant
5 CONCLUSIONS AND FUTURE WORK
TSI2005-07764-C02-01, and by The
In this paper, we presented an approach for Canaries Regional Education and FEDER under
efficient management of VoD and real time video contract PI042004/164.
in wireless networks that combine WiFi and
WiMAX technology. While in recent papers several 6 REFERENCES
authors centre their work in efficient handover
mechanism, we study a more complex problem that [1] Ch. Guo, Z. Guo, Q. Zhang, and W. Zhu: A
seamless and proactive end-to-end mobility
arises when disconnection areas are also studied in
combination with handover areas. We base our solution for roaming across heterogeneous
solution on an anticipated calculation of the RSSI wireless networks, IEEE Journal on Selected
values that can be used to discover when the MC Areas in Communications, issue 5, Vol. 22, pp.
834-848 (2004).

Ubiquitous Computing and Communication Journal

[2] M. Vilas, X. G. Pañeda, D. Melendi, R. García, S. N. Moiseev, and M. S. Kondakov: QoS-
and V. G. García: Signaling management to oriented intersystem handover between IEEE
reduce roaming effects over streaming services, 802.11b and overlay networks, IEEE
IEEE Proceedings of the 32nd EUROMICRO Transactions on Vehicular Technology, issue
Conference on Software Engineering and 99 (2007).
Advanced Applications (EUROMICRO- [15] S.c. Oh, C.y. Lee, Y-s Shin, and J.u. Kim: A
SEAA'06) (2006). vertical handover based on software
[3] L. Taylor, R. Titmuss, and C. Lebre: The communications architectures, IEEE 60th
challenges of seamless handover in future Vehicular Technology Conference (VTC2004),
mobile multimedia networks, IEEE Personal Vol. 3, pp. 1973-1977 (2004).
Communications, issue 2, Vol. 6, pp. 32-37 [16] J. J. J. Roy, V. Vaidehi, and S. Srikanth:
(1999). Always best-connected QoS integration model
[4] W. Wu, N. Banerjee, K. Basu, and S. K. Das: for the WLAN, WiMAX heterogeneous
SIP-based vertical handoff between WWANS network, First International Conference on
and WLANS, IEEE Wireless Communications, Industrial and Information Systems, pp. 361-
issue 3, Vol. 12, pp. 66-72 (2005). 366 (2006).
[5] J.M. Chung, J.-H. Seol, and S.-H. Choi: Media [17] P. Bellavista, A. Corradi, and C. Giannelli:
independent handover in broadband mobile Mobile proxies for proactive buffering in
networks, Proceedings of the 6th WSEAS Int. wireless internet multimedia streaming,
Conf. on Electronics, Hardware, Wireless and International Workshop on Services and
Optical Communications (2007). Infrastructure for the Ubiquitous and Mobile
[6] CDMA Development Group, CDMA2000 Internet (SIUMI), (2005).
international roaming: a status report (2005). [18] L. Dimopoulou, G. Leoleis, and L.S. Venieris:
[7] H.D. Cho, J.K. Park, W.J. Ko, K. Lim, Fast handover support in a WLAN
and W. Kim: A study on the MCHO method in environment: challenges and perspectives,
hard handover and soft handover between IEEE Network, issue 3, Vol. 19, pp. 14–20
WLAN and CDMA, International Conference (2005).
on Consumer Electronics ICCE 2005, Digest of [19] S. Choi, G.H. Hwang, T. Kwon, A.R. Lim, and
Technical Papers, pp. 391- 392 (2005). D.-H. Cho: Fast handover scheme for real-time
[8] C.K. Toh: The design & implementation of a downlink services in IEEE 802.16e BWA
hybrid handover protocol for multimedia system, IEEE 61st Vehicular Technology
wireless LANs, Proceedings of the First Annual Conference (VTC 2005), Vol. 3, pp. 2028-2032
ACM International Conference on Mobile (2005).
Computing and Networking (MOBICOM [20] Ch. Schindelhauer: Mobility in wireless
1995), pp. 49-61 (1995). networks, 32nd Conference on Current Trends
[9] H.H. Choi, and D.H. Cho: Takeover: a new in Theory and Practice of Computer Science
vertical handover concept for next-generation (SOFSEM 2006), pp. 110–116 (2006).
heterogeneous networks, IEEE 61st Vehicular [21] S.K. Kim, C.G. Kang, and K.S. Kim: A
Technology Conference (VTC2005), Vol. 4, pp. adaptive handover decision algorithm based on
2225-2229 (2005). the estimating mobility from signal strength
[10] R. Corvaja, A. Zanella, M. Dossi, A. Tontoli, measurements, IEEE 60th Vehicular
and P. Zennaro: Experimental performance of Technology Conference VTC2004-Fall, Vol. 2,
the handover procedure in a WiFi network, pp. 1004-1008 (2004).
Proceedings of WPMC04 (2004). [22] M. Pischella, F. Lebeugle, and S. B. Jamaa:
[11] Mo Li, K. Sandrasegaran, and T. Tung: A UMTS to WLAN handover based on priori
multi-interface proposal for IEEE 802.21 media knowledge of the networks, IEEE International
independent handover, International Conference on Communications, Vol. 5, pp.
Conference on the Management of Mobile 2009-2013 (2006).
Business (ICMB 2007), pp. 7 (2007). [23] P. Bellavista, A. Corradi, and C. Giannelli:
[12] G. Dong, and J. Dai: An improved handover Mobility prediction for mobile agent-based
algorithm for scheduling services in service continuity in the wireless internet,
IEEE8O2.16e, IEEE Mobile WiMAX Lecture Notes in Computer Science, Vol.
Symposium, pp. 38-42 (2007). 3284/2004, pp. 1-12 (2004).
[13] J. H. Li, S. Luo, S. Das, T. McAuley, M. Patel, [24] P. Bellavista, A. Corradi, and C. Giannelli:
A. Staikos, and M. Gerla: An integrated Adaptive buffering-based on handoff prediction
framework for seamless soft handoff in ad hoc for wireless internet continuous services,
networks, Military Communications Lecture Notes on Computer Science, Vol. 3726,
Conference (MILCOM 2006), pp. 1-7 (2006). pp. 1021-1032 (2005).
[14] A. V. Garmonov, S. H. Cheon, D. H. Yim, K. [25] R. Risueño, P. Cuenca, F. Delicado, L. Orozco-
T. Han, Y. S. Park, A. Y. Savinkov, S. A. Filin, Barbosa, and A. Garrido: On the effect of the

Ubiquitous Computing and Communication Journal

 handover mechanisms in QoS performance in [27] J. Bardwell: Converting signal strength
wireless multimedia networks, Proceedings of percentage to dBm values, WildPackets' White
the 24th International Conference on Distributed Paper, 2002 (on line available on June 2008)
Computing Systems, pp. 138–143 (2004). [28] A. Suárez, M. La Menza, E. Macías, and V.
[26] V. Gupta, M. G Williams, DJ Johnston, S. Sunderam: Automatic resumption of streaming
McCann, Ph. Barber, and Y. Ohba: IEEE sessions over WiFi using JADE, IAENG
802.21: overview of standard for media International Journal of Computer Science, Vol.
independent handover services, IEEE 802 33, pp. 92-100, ISSN 1819-656X (2007).
Plenary (2006).

Ubiquitous Computing and Communication Journal