Guard Channel CAC Algorithm for High Altitude Platform
Networks
Dung D. Luong, Anh P. M. Tran and Tien V. Do1
Department of Telecommunications and Media Informatics
Budapest University of Technology and Economics
Magyar tudósok körútja 2, Budapest, H-1117 Hungary
{luong,tran,do}@hit.bme.hu
Abstract
This paper proposes the guard-channel call admission control algorithm
for High Altitude Platform (HAP) networks. An approximation method
is proposed to calculate the channel guards which keep the dropping
probability of hand-off calls under predefined threshold while
approaching the maximum bandwidth reservation. The algorithm takes
into account the general operating scenario of HAP networks.
1 Introduction
Today’s users of broadband are generally dependent upon either cable or satellite technologies for
the connection to their home or office. Cable, particularly fiber, can provide excellent capability but is only
commercially viable in high usage areas. Satellite provides widespread geographic coverage but has limited
capability and capacity.
CAPANINA (Communications from Aerial Platform Networks delivering Broadband
Communications for all) is an international cooperation supported by the European Union under
Framework 6 Research Programme. The aim of the project is to deliver cost effective solutions providing a
viable alternative to cable and satellite, with the potential to reach rural, urban and traveling users.
Examples of HAPs include airships and solar powered aeroplanes operating at altitudes of around 20km,
well above any air traffic. CAPANINA will gives solutions to deliver low cost broadband communications
services to small offices and home users, or those traveling inside high-speed public transport vehicles (e.g.
trains), at data rates up to 120Mbit/s - a staggering 2000 faster than today’s dial-up modems and more than
200 times faster than a typical "wired" broadband facility [1].
The most general operating scenario with platform interconnection via inter-platform links (IPLs)
provides extended system coverage with significantly reduced terrestrial infrastructure (Figure 1). To
support communication between adjacent platforms without any ground network elements each HAP
payload includes a switching device and one or more inter-platform link terminals. Depending on the link
budget analysis we can choose between optical and radio frequency terminals. In this scenario ground
stations are used mainly as gateways to other public and/or private networks, while providing also a backup
interconnection between platforms in the case of IPL failure.
1
This work was supported by the European Commission within the framework of the EU IST CAPANINA
project.
Figure 1: The general operating scenario in HAP systems.
In the general operating scenario, fixed users and mobile users select the most suitable platform
(possibly the one with strongest signal level) and initiate the authentication and admission procedure in
order to gain access to the network. Arrivals or departures of fixed users can be considered as single events
while the arrivals or departures of hand-offs occur in batches since the mobile users travel in vehicles and
connected to HAP via on-board collective terminals.
As in other mobile networks, Call Admission Control in HAP networks should give higher priority
for hand-off calls. The aim of CAC algorithms should keep the dropping probability under a predefined
threshold while minimalizing the blocking probability, therefore, maximalizing the bandwidth reservation
at the same time. Dropping probability is the probability that a hand-off call is dropped by CAC while
blocking probability is the probability that a new call is blocked.
The guard-channel policy is the most widely-used CAC scheme for mobile networks with handoff calls. Channel guards are defined for each traffic class to prevent the acceptance of new calls of that
class and leave the remaining available resource for hand-off calls when the resource usage becomes high.
Mobile systems with multiple class traffic makes the process of finding optimal channel guards very
complicated and CPU-time-consuming due to multi-dimensional Markov chains [6]. Additionally, the
consideration of batch hand-offs makes the finding of good CAC algorithms for HAP networks more
challenging.
This paper propose the use of the guard-channel CAC algorithm for the general operating scenario
of HAP systems. An approximation method is proposed for the calculation of channel guards to keep the
resource reservation as high as possible while guaranteeing a certain degree of Quality of Service (QoS),
i.e. guaranteeing minimum bandwidth for each connection (in this paper, we use the term of connection and
call interchangeably with the same meaning), while keeping the dropping probability of hand-off
connections under a predefined threshold.
The rest of this paper is organized as follows. Section 2 describes the problem, objective for CAC
of HAP networks. Section 3 provides the solution and Section 4 gives simulation results. Section 5
concludes the paper.
2 Problem formulation
Modern data communication networks should be designed and dimensioned to serve real-time and
non real-time traffic classes. For instance, IEEE 802.16 standard2 specifies 4 different QoS classes for the
up-link:
UGS Unsolicited Grant Service. For real-time service flows which generate fixed size packets on
a periodic basis, e.g., VoIP without silence suppression.
rtPS Real-Time Polling Service. For real-time service flows which generate variable size packets
on a periodic basis, e.g., MPEG video.
nrtPS Non-Real-Time Polling Service. For non-real-time service flows with guaranteed minimum
rate.
BE Best Effort service. Without any QoS guarantee.
Real-time traffic classes are sensitive on delay and/or delay variation while the non-real time
connection are not sensitive on delay factors. CAC and bandwidth allocation support the implementation of
QoS at the flow level and the packet scheduling is one of the tools for QoS implementation at the packet
level. At the flow level, the delay factors usually are not considered directly, or they are considered with
certain constraints, assumptions about underlying packet scheduling algorithms. At this level, the
bandwidth requirements are the primary factors under consideration. Connections may require a fixed
amount of bandwidth, e.g. for voice without silence suppression, or require a minimum bandwidth
guarantee, e.g. for video streaming and non real-time traffic. Best effort traffic can be considered as a
special non real-time class without the minimum bandwidth guarantee.
In the HAP networks, user terminals are connected to the Customer Premise Equipment (CPE)
which performs the processing of signal from/to an antenna. We are talking about the fixed or mobile users
depending on whether the antenna hooks on fixed or moving objects. Fixed users are home or office users
having the antenna hooked on the roof of the house or building when mobile users are travelers on trains or
buses with antenna hooked on the vehicles.
For a mobile wireless system with the arriving of new and hand-off calls and QoS consideration,
the most widely-used CAC is the guard-channel scheme, proposed in the mid 80s [8]. However, on one
hand, fast calculation method for optimal channel guard for each traffic class is still an open problem. On
the other hand, mobile users as collective travelers locating on vehicles make HAP networks being
different from other mobile system in the sense of batch hand-off. That is, if a bus or a train moves at the
boundary of cells, then all users located on the vehicle have to perform hand-off at the same time. The main
objective of this work is a solution which requires small CPU-time consumption and addresses the specific
problem of batch arrivals of hand-off calls.
The following model is general for HAP networks: considering a system with different traffic
classes, with bi as the minimum bandwidth guarantee for class i (i=1, 2, …, N). Denote Gi is the channel
guard for class i. At the CAC decision time epoch t, we have ni(t) connections of class i admitted into the
network. Given that the link capacity is C, the guard channel admission policy is shown in Algorithm 1.
A call arrives from class i at time t.
if the incomming call is a new call then
if bi  Gi 

j
n j (t )b j then
The new call is admitted.
else
The new call is blocked.
end if
else
2
At the present CAPANINA has identified IEEE 802.16 technology as a candidate for the provision of
broadband communications services.
if bi  C 

j
n j (t )b j then
The hand-off call is admitted.
else
The hand-off call is dropped.
end if
end if
Algorithm 1 Guard channel algorithms
Denote with B (t ) 

j
n j (t )b j the total bandwidth reservation at time t, then B(t) can be
d
considered as a random variable. Let Pi be the dropping probability of class i, then we have to find optimal
G={G ,G ,,G } to maximalize the expected bandwidth consumption while guarantee the dropping
1
2
N
probabilities below a predefined thresholds:
P:
maximize E(B),
d
subject to Pi <d ,
i
over G :0<G C, i=1, 2, …, N
i
i
Assume that the arrival of individual new calls and batch hand-offs are Poisson processes. Denote
H
by  the arrival rate of new calls of class i and by  the arrival rate of hand-off batches (a batch can
i
contains multiple calls of different classes). The expected value of the number of connections of class i
inside a batch is m . Furthermore, we assume that the holding time of connections of class i is exponentially
i
distributed with expected value of 1/ . We have the same assumption for the cell residence time for
i
H
batches with expected value of 1/ .
If we separately consider the admission of calls within a batch, then the order of calls in the
admission process is crucial. If a call is processed earlier than another call, its chance to be admitted is
obviously higher than that of the later one. That is, the order of admission consideration affects the
dropping probability of each traffic class. Similarly, if we consider the admission of calls in a batch
together, then in case there is not enough available bandwidth for the whole batch, the dropping policy of
calls to decrease the bandwidth requirements of the batch also affects the dropping probability of each
traffic class. In this paper, we assume the random order of call admission processing. Furthermore, we only
consider the common dropping probability P of all calls instead of the separate dropping probability of
d
each traffic class.
With this approach and the consideration of the common dropping probability for all traffic
classes, the optimalization problem P is rewritten as:
P*:
maximize E(B),
d
subject to P <d,
over G :0<G C, i=1,2, ..., N.
i
i
For traditional 2G-3G cellular mobile systems where most of the hand-offs do not occur in batch
mode, the optimal vector of G can be obtained by writing multi-dimensional Markov chain. The solving of
such Markov chains is very complicated and heavily CPU-time consuming. Furthermore, the overall space
of G is too large to execute the optimization numerically. The introduction of batch hand-offs makes the
situation more complicated: there are transitions not only between adjacent states in the multi-dimensional
Markov chain, but also between non-adjacent states.
3 An approximation solution
As mentioned in previous section, finding optimal G typically is very complicated process
requiring heavy CPU-time consuming due to the large multi-dimensional Markov chain and large domain
of G. In order to have simple method to approximate optimal solution, we use the a transformation
approach. A similar approach called splitting aproach has been already proposed in [7] to split the multidimensional Markov chain into one-dimensional Markov chains. However, the splitting solution in [7] does
not take into account the multiplexing nature of aggregate traffic of multi-class environment.
Parameters
Minimum bandwidth requirement
Holding time
From
bj
1/µj
Arrival rate for new calls
αj
Arrival rate for hand-off calls
αHmj
To
bi
1/µi
 ji 
 Hji 
bj
b j i
j
bi  j
i
bi  j   H
 Hmj
Table 1 Transformation rules for calls from class j to class i
The transformation approach consists of N steps: the channel guard G is approximated in step i
i
(0<iN),which is described as the follows. All calls of class j:ji are replaced by calls of class i with
minimum bandwidth of b instead of b , holding time of 1/ instead of 1/ (Table 1). By this replacement,
i
j
i
j
there is only one traffic class in the system and the multi-dimensional Markov chain becomes onedimensional. The transformed arrival rate  of new calls of class j is calculated as:
ji
bj
j
 ji
 bi
j
i
Equation 1
Therefore,
 ji 
b j i
j
bi  j
Equation 2
H
The arrival rate of  m of hand-off calls is transformed to:
j
 Hji 
i
 Hmj
H
bi  j  
bj
Equation 3
H
where  + is the termination rate of the hand-off calls. The termination time of a hand-off call is
j
the minimum of holding time and cell-residence time. The transformation of parameters can be found in
Table 1.
[Figure missing!!!]
Figure 2 One dimensional Markov chain after transformation
The one-dimensional Markov chain after transformation for step i is shown in Figure 2, where
 iTSN   j 1 ji
N
is the aggregate arrival rate of new calls and
 iTSH   j 1 Hji
N
is the aggregate arrival rate of hand-off calls.
Let
pi ( j, H ) denote the steady-state probability that there are j calls in the system of step i with
the channel guard of H

( iTSN   iTSH ) j
pi (0, H )
jH

j! i j

pi ( j , H )   TSN
TSH H
TSH j  H
 ( i   i ) ( i )
pi (0, H ) H  j  C / bi 

j! i j
The bigger the H, the higher the expected bandwidth reservation and the same time the higher the
dropping probability. We need to find the largest H with that the dropping probability is smaller than
threshold d.
H i  max H : Pi d ( H ) d
we suppose that the aggregate bandwidth requirement of a batch is exponentially distributed with
the expected value:
B  k 1 bk mk
N
Given that there are j calls in the system, then the expected dropping probability is:
pid ( j )  P (a call is dropped | there are j calls in the system) 
1
1  x
ri ( j ) S e S ( x  ri ( j ))dx  ri (Sj )

e
1
 1  x
S
0 S e xdx

where
ri ( j )  C  jbi
The dropping probability can be calculated as
Pi ( H ) 
d
C / bi 
 p ( j, H ) P
j 1
i
i
d
( j)
d
Note that Pi (j) is independent of H, therefore, it makes the calculation of G simpler.
i
Furthermore, to fasten the process of finding optimal G , we can use the binary search by calling
i
C
BinarySearchGuard(0,  ). This function is described in Algorithm 2.
b
i
Require: begin ≤ end
if begin == end then
if Pi ( H )  d then
d
return begin
else
return notfound
end if
else
 begin  end 
H 

2

d
if Pi ( H )  d then
return BinarySearchGuard(H, end)
else if Pi ( H )  d then
d
return H
else
return BinarySearchGuard(begin, H)
end if
end if
Algorithm 2 Function BinarySearchGuard(begin, end)
4 Simulation
The approach proposed in the previous section is an approximation method to calculate the
optimal channel guard. The aim of simulation experiments here is to investigate how good the
approximation is. A system with two traffic classes was considered: a real-time and a non-real-time one.
The arrival rate of new connections, holding time, minimum bandwidth guarantee, batch size of each class
is specified in Table Error! Reference source not found.. Since hand-off calls arrive in common batches,
then each traffic class has the same cell residence time and batch arrival rate. The number of hand-off calls
of each class is assumed to be geometric distributed with the expected value of m =6 for real-time class and
1
m =4 for non-real-time one.
2
Real-time class
Non-real-time class
-1
Arrival rate of new calls (s )
1  9
2  6
Minimum bandwidth (Kbps)
b1  64
1  10
m1  6
b2  96
2  20
m2  4
Holding time(s)
Batch size
Cell residence time (s)
 H  60
 H  0.3
-1
Arrival rate of batches (s )
Table 2 Parameters for two traffic classes
Figure 3 Link utility versus G1 and G2
With the dropping threshold selected from {[0.01,0.09]:0.01} (this means the dropping threshold
starts from 0.01, then increases its value with step of 0.01 until it reaches 0.09). In simulation, to
approximate the optimal G for each dropping threshold, G and G are selected from {[0.7,0.95]:0.05} and
1
2
the dropping probability and the reserved bandwidth are calculated for each combination of G={G ,G }.
1
2
The optimal G is that one that gives maximum reserved bandwidth while its dropping probability is under
the dropping threshold. Table Error! Reference source not found. gives the optimal G calculated by the
simulation and the model approximation. Even both of them are approximation, the approximation
calculated by simulation has estimation error within the increasing step (0.05 in this case). If we considered
the estimate of simulation is the actual value of the optimal channel guard, the comparison shows that most
of the model estimates has absolute error within 0.05 and the remaining part just exceeds the 0.05 but not
much.
Dropping threshold
Simulation
Model approx.
G
1
2
1
2
0.09
0.9
0.95
0.909091
0.907473
0.08
0.9
0.95
0.903743
0.900356
0.07
0.9
0.95
0.898396
0.893238
0.06
0.95
0.9
0.893048
0.88968
0.05
0.95
0.9
0.882353
0.879004
0.04
0.95
0.9
0.871658
0.868327
0.03
0.9
0.9
0.860963
0.857651
0.02
0.9
0.9
0.839572
0.839858
0.01
0.85
0.9
0.807487
0.807829
Table 3 Comparison of simulation and model approximation
G
G
G
5 Conclusions
This paper considered the guard-channel CAC algorithm for the HAP networks. A model is built
for the general operating scenario taking the batch hand-offs into account. An approximation method was
proposed for the simple calculation of the vector of channel guards. Simulation results suggest that the
proposed approximation method is fairly accurate.
This work is a part of QoS framework for HAP network containing admission control at call level
and packet scheduling [9] at the packet level. The future work would be the investigation of these two
mechanisms working simultaneously.
6 References
[1] IST CAPANINA project. http://www.capanina.org/.
[2] IEEE Std 802.16-2001, IEEE Standard for Local and Metropolitan Area Networks -Part 16: Air
Interface for Fixed Broadband Wireless Access Systems.
http://standards.ieee.org/getieee802/download/802.16-2001.pdf.
[3] The Network Simulator - ns-2. http://www.isi.edu/nsnam/ns/.
[4] James Yang, Amir K. Khandani, and Noel Tin, Adaptive Modulation and Coding in 3G Wireless
Systems. IEEE Vehicular Technology Conference, 2002. Proceedings. VTC 2002-Fall, vol. 1 , pp.
544-548, 2002.
[5] M. Nakamura, Y. Awad and S. Vadgama, Adaptive Control of Link Adaptation for High Speed
Downlink Packet Access (HSDPA) W-CDMA. IEEE WPMC02, Vol. 2, Page(s): 382-386, 2002.
[6] T. Janevski, Traffic Analysis and Design of Wireless IP Networks. IEEE WPMC02, Vol. 2,
Page(s): 382-386, 2002. ARTECH HOUSE, ISBN 1-58053-331-0, Chapter 8, Page(s): 239-268,
2003.
[7] J. Misic, S. T. Chanson, and F. S. Lai, Quality of Service Management for Wireless Networks
with Heterogeneous Traffic. IEEE Globecom’98, Sydney, Australia, November 1998, pp. 1406–
1412.
[8] D. Hong and S. S. Rappaport, Traffic model and performance analysis for cellular mobile radio
telephone systems with prioritized and nonprioritized handoff procedures. IEEE Transactions on
Vehicular Technology, vol. 35, no. 3, pp. 77–92, Aug. 1986.
[9] D. D. Luong, Nam H. Do, Gergo Buchholcz, and Tien V. Do, Packet Fair Scheduling for Wireless
Channels with Adaptive Burst Pro¯les. InternationalWorkshop on High Altitude Platform Systems
-WHAPS 05, Athens, Greece, September 2005.