TRUNKING THEORY

An Old Idea
•  Trunking theory tells the size of a population that can be
served by a limited number of servers with a specified
grade-of-service (GOS)
•  In the simple case, the GOS is the blocking probability
•  Developed in the late 1800s by Erlang
Request Model

•  Assume that customers request service at

random times at a certain cumulative average
rate, λcum
•  e.g. λcum =13 requests per hour
•  Times between consecutive requests are
independent exponential random variables
(RVs) with parameter λcum

Call Request times time

Server Model

•  The durations of service (i.e. the lengths of the calls or

holding times ) are independent exponential RVs
with expected value H"

Service time

User 1
User 2
User 3

Cumulative Call Request times time

Measuring Traffic Intensity

•  An Erlang is the average fraction of time that a

channel is occupied
•  One continuous call is an example of traffic intensity of
1 Erlang
•  A channel that carries traffic only half the time carries
0.5 Erlangs of traffic
•  For the request and server models in previous slides,
the traffic intensity is λH
Single User vs. Total

•  Suppose each user generates a traffic intensity

of Au Erlangs
•  Suppose there are U users
•  Then the total traffic intensity in Erlangs is
A = AuU
•  Let λ be the rate of call requests per user.
Then the traffic intensity per user can be
expressed
Au = λH
Channels
•  Each call requires a channel
•  One approach is to dedicate a channel to each user
•  A user s call request is never denied
•  A channel sits idle when its user is not making a call

Not an efficient use of resources!

Trunked System

•  Channels are pooled

•  No user has a fixed channel
•  A new user is assigned some channel from the
pool
•  When a call is finished, the channel is released
back into the pool
f4
f1
f5 f2
f3
f6
 Block Calls Cleared
•  In a Block Calls Cleared type of system, a call
request is simply denied if all channels in the pool
are in use
•  The blocked caller is free to make a new request
•  Mobile Cellular systems are Block Calls Cleared
systems

f4
f1
f6 f2
f5 f3
Probability of Blocking

•  The GOS measure for Block Calls Cleared

systems is the probability that a user s call
request is blocked
•  The Erlang B formula determines the blocking
probability, p, given a certain total offered
traffic intensity, A, and a certain number of
channels C in the pool
p = B( A, C )
Assumptions
•  There are memory-less arrivals of requests
•  call can be made anytime
•  Probability of user occupying a channel is exponentially
distributed
•  Longer calls less likely
•  Finite number of channels are available
Erlang B Formula

•  A is the totaloffered traffic

•  Because some calls are blocked, A is not the traffic
carried by the system

C
A
p = B ( A, C ) = C C! k
A
∑k = 0 k!
Trunking Gain
•  Trunking gain is the improvement in offered
traffic intensity that is obtained when sets of
channels are merged into trunk pools
•  In the next slide, the offered traffic intensities for
a 10% blocking probability are compared for a C-
channel trunked system and C fixed, single-
channel systems
Graphical Comparison

•  Sketched from [Hernando and Pérez-Fontán,`99]

Total Offered Traffic, Erlangs

8
p=0.1

6 C-channel trunked system

4
trunking gain
2
C single-channel systems

0
0 2 4 6 8 10
No. of Channels, C
Summary
•  In a trunked system, channels are pooled for
common use on an as-needed basis
•  In a Block Calls Cleared system, a new request is
simply denied if all channels are busy
•  The more channels in the pool, the higher the
offered traffic can be for a given probability of
blocking
BLOCKED CALLS
DELAYED
Blocked (or Lost) Calls Delayed
•  In a previous module, we considered Blocked
Calls Cleared, and the grade-of-service (GOS) was
probability of blocking
•  In a Blocked Calls Delayed system, if all channels
are busy, a new user goes into a queue to wait
for service
•  Private mobile radio (PMR) systems are often the
Blocked Calls Delayed type [Hernando and F.
Pérez-Fontán, 1999]
Erlang C Formula
•  The GOS is given by P(delay>t), where t is some
specified time limit, like 20 seconds
P(delay > t ) = P(delay > t | delay > 0) P(delay > 0)
•  P(delay>0) shall be denoted C(A,N), and is given
by the Erlang C formula:
( A N %( N %
&& ##& #
' N! $' N − A $
C ( A, N ) = N −1 k
A ( A N %( N %
∑
k = 0 k!
+ && ##& #
' N ! $' N − A $
Alternative Expression for Erlang C
•  C(A,N) can be expressed in terms of the blocking
probability B(A,N):

NB( A, N )
C ( A, N ) =
N − A[1 − B( A, N )]
Full GOS Formula
•  The final expression for the GOS is

& t #
P(delay > t ) = C ( A, N ) exp$− ( N − A) !
% H"
where H is the average call duration.
Summary
•  In Blocked Calls Cleared, if all channels are busy, the new
user is denied service
•  In Blocked Calls Delayed, if all channels are busy, the new
user waits for service
•  Erlang B is used for the former, Erlang C for the latter
OTHER ISSUES WITH
CELLS
Four Topics:
•  Hand-offs
•  Cell Splitting
•  Sectoring
•  Power Control
Hand-offs

•  As a subscriber moves away from its serving

base station, the SIR decreases
•  The subscriber compares received powers
from the surrounding base stations
•  Eventually, the subscriber changes to another
serving base station to get a better SIR
Must Change Channels

•  Because the adjacent cells use a different set

of channels, the carrier frequency must
change during a call
•  This frequency change is a hand-off
•  Because of multipath fading, there needs to be
a hysteresis to avoid excessive switching
Hand-offs Have Priority
•  Since it is bad for an existing call to be dropped, hand-offs
have priority over new callers for channel assignments
•  For this reason, the Erlang B formula is not quite
applicable, but it still provides insight
Sectoring

•  Sectoring is a way to increase the SIR

•  The omni antennas are replaced with 3 or 6
directional antennas
•  The channel pool in the hexagonal cell is split
into three disjoint pools
•  Hand-offs now occur between sectors

3
1
2
Interference Reduced

•  The significant interference 1

3
now comes from only 2 1
3 2
1
3
2
sources instead of 6 2
1
3
3 2 3
•  Cluster sizes can be 3
1
2 3
1
2
1 1
reduced (e.g. from 12 to 7), 1
3 2
1
3 2
1
3
increasing the number of 2
1
3 2
1
3 2
1
3
2
channels per hexagon 1
3 2
1
3 2
1
3
2 3 2 3 2 3
•  However, since the channels 1
2 3
1
2 3
1
2
1 1
of a hexagon are split into 3 1
3 2
1
3 2
sectors, there is an overall 2
1
3 2

loss in trunking efficiency 1

3 2
1
3
2 3 2
1
2
Power Control
•  Base stations monitor the power received by subscribers
•  The Base Station commands the subscriber to use the
lowest power necessary to maintain GOS
•  This minimizes interference from other subscribers and is
important for CDMA
Summary

•  A hand-off is a channel change as a subscriber

moves to a new serving base station or to a
new sector
•  Cell-splitting is a way for a part of a network to
accommodate growth
•  Sectoring reduces interference and trunking
efficiency
•  Power control is used to minimize interference
on the uplink (from subscriber to BS)
 Co-channel Interference

co-channel
base stations

D
D+R

serving D-R
base station
mobile subscriber

D-R
D+R
D

Worst case co-channel interference on the forward channel.

8
 Worst Case Co-Channel Interference

• There are six co-channel base-stations, two at distance D − R, two

at distance D, and two at distance of D + R.

• The worst case carrier-to-interference ratio is

1 R−β
Λ =
2 (D − R)−β + D−β + (D + R)−β
1 1
=
2 D − 1 −β + D −β + D + 1 −β
! " ! " ! "
R R R
1 1
= !√ "−β !√ "−β !√ "−β
2 3N − 1 + 3N + 3N + 1

• Hence, for β = 3.5


 14.3 dB for N = 7
Λ(dB) = 9.2 dB for N = 4
 6.3 dB for N = 3

– Shadows will introduce variations in the worst case C/I.

9
 Cell Sectoring

D+0.7R serving
base station

co-channel
base stations

mobile subscriber

Worst case co-channel interference on the forward channel with 120o cell
sectoring.

10
• 120o cell sectoring reduces the number of co-channel base stations
from six to two. The co-channel base stations are at distances D
and D + 0.7R.

• The carrier-to-interference ratio becomes

R−β
Λ =
D−β + (D + 0.7R)−β
1
= ! D "−β ! D "−β
R
+ R + 0.7
1
= !√ "−β !√ "−β
3N + 3N + 0.7

• Hence 
 21.1 dB for N = 7
Λ(dB) = 17.1 dB for N = 4
 15.0 dB for N = 3

• For N = 7, 120o cell sectoring yields a 6.8 dB C/I improvement over

omni-cells.

• The minimum allowable cluster size is determined by the minimum

C/I requirement of the radio receiver. For example, if the radio
receiver can operate at Λ = 15.0 dB, then a 3/9 reuse cluster can
be used (3/9 means 3 cells or 9 sectors per cluster).

11