You are on page 1of 7

c  


  c 

  c   
  c 

 c    

!
c
  

  

Written by Dr. Blimegger, "Capacity Optimization in a CDMA Network", is published in


the Chinese language version of Mobile Communications Magazine.

By Leroy G. Blimegger Jr.


President & CEO, CNRGI, Inc.
"! 
This article describes the problems and solutions related to optimizing the air-interface
capacity of a CDMA network. It is assumed that all hardware provisioning from the Base
Station through to the PSTN is sufficient to carry any load produced. This article is not
vendor equipment specific. Therefore, data derived in the article may be different from
data provided by a specific equipment vendor. The intent of this article is to provide a
guideline for capacity optimization for any CDMA network (800MHz, 1.9GHz, IS-95A/B,
IS2000).

The article discusses capacity planning, hardware constraints, RF optimization issues


that reduce capacity, and finally multi carrier traffic balancing.

c  "  
The first and most important step in capacity optimization begins with capacity planning.
Knowing where and how subscribers will use their CDMA phones is a major part of a
good CDMA Network design. The questions that need to be asked during the design
are; how many subscribers will access the network, what percentage of these
subscribers will be high mobility, what percentage will be low mobility or fixed, what
percentage of traffic is expected to be derived from indoor use, and finally, what
percentage will take advantage of data services?

The importance of these questions may not be apparent, so as an example, let¶s look at
the difference in capacity between a high mobility network and a low mobility or fixed
network. Typical capacity values for a high mobility network are approximately 25
erlangs/sector/carrier, or 35 users/sector/carrier, assuming a 1% GOS. For a low
mobility or fixed network, the capacity increases to approximately 38
erlangs/sector/carrier, or 50 users/sector/carrier. There is approximately a 50% increase
in erlangs capacity for a low mobility or fixed network. This increase in capacity is
related to the much lower SHO factor in low mobility or fixed networks.

If a design was based on 38 erlangs/sector/carrier, but the area covered was a high
mobility area, the network in this area would not have enough capacity to meet
subscriber demand. Conversely, if a design were based on 25 erlangs/sector/carrier,
but the area covered was a low mobility or fixed area, the network would be over-
provisioned and money would be wasted.

If the design produced an under-provisioned network in certain areas, there are only
three methods to add additional capacity to the area; adding channel element cards to
existing base stations, adding new base stations, and adding additional carriers to
existing base stations.

(  c
  
Air-Interface capacity in a CDMA system is constrained by two values; the number of
channel elements available and the amount of forward traffic channel power available.
While it is true, there are other capacity limiting factors, such as available Walsh codes,
most blocking occurs due to a lack of channel element resources or forward traffic
channel power. So, we will limit our discussion to these.

Assuming unlimited forward traffic channel power, the number of originations /


terminations that can be made on a base station is limited by the number of channel
elements and the SHO factor. If the SHO factor (which is the average number of
CE/per/subscriber/call) is 1, then the number of originations / terminations available will
be equal to the number of channel elements provisioned on the base station. As the
SHO factor increases, the number of originations / terminations available decreases
based on the same number of channel elements available.

Assuming unlimited channel element resources, the number of originations /


terminations that can be made on a sector/carrier is based on the amount of forward
traffic channel power required for each call. Assuming each origination / termination
requires the same amount of forward traffic channel power, the number of originations /
terminations available would be the forward traffic channel power available divided by
the forward traffic channel power required for each origination / termination.

Therefore, in order to optimize the air-interface capacity, we must find ways to reduce
the SHO factor and reduce the amount of forward traffic channel power required by
each subscriber.

#($ 

The SHO factor is defined as the average number of channel element resources used
by a subscriber during a call. For non-CDMA systems, the SHO factor would always
equal 1, since only one channel resource can be used at a time. For CDMA systems,
the ability to receive and decode multipath signals simultaneously adds to the
robustness of the CDMA network. Most equipment vendors recommend a SHO factor of
2.0 to 2.2. A SHO factor in this range will provide the benefit of soft handoff, without
wasting channel elements.

$
 % !!c "

The overall forward traffic channel power available is defined as the total usable PA
power ± overhead, where overhead includes pilot, paging, and sync channels. The
overall amount of forward traffic channel power can thus be increased by decreasing
the power used by the overhead channels.

The average amount of forward traffic channel power that a single subscriber uses is
based on the RF environment at the time of the call. If the RF environment at the time of
a call contains significant noise, the average amount of traffic channel power used will
increase. Therefore, to reduce the average amount of forward traffic channel power
used by a subscriber, we must optimize the RF environment with methods such as
neighbor list analysis and coverage control.

  c  


There are several optimization methods used to reduce the SHO factor and forward
traffic channel power per subscriber, which in turn, will increase the network capacity.
The first and most important is coverage control and neighbor list tuning. As an
example, let¶s look at the following picture, showing a reference sector and its neighbor
sectors. Since most vendor equipment and mobile phones support only 20 neighbors,
we¶ll assume that there are 20 neighbors in the reference sector¶s neighbor list. Let¶s
also assume that PSMMs (Pilot Strength Measurement Messages) were used to
determine the top 20 neighbors for the reference sector.

There are six high priority neighbors, defined as those sectors that share a natural
coverage boundary with the reference sector, six second tier neighbors that are
considered acceptable, two possible third tier neighbors, and six neighbors that are five
or more tiers away from the reference sector. This is a typical scenario for a network
that has had little or no coverage control optimization done, or where coverage control
is limited by a requirement to provide indoor coverage using outdoor base stations.

From experience, this scenario produces a SHO factor of 3.2 to 3.5, which is 45% to
75% higher than the recommended SHO factor. What this means is that 45% to 75%
additional channel element resources are being utilized just to handle soft handoff
traffic. If the SHO factor could be reduced to the recommended value, these channel
elements would be available for originations and terminations, thus increasing the
network capacity.

This scenario also produces a higher average forward traffic channel power per
subscriber. If a mobile in the reference sector coverage area originates or terminates a
call on one of the Bad neighbors (shown in red), the amount of traffic channel power
required to maintain the Eb/No and FER targets for this far-away mobile increases. The
additional noise generated by this far-away mobile in the bad neighbor¶s coverage area
also forces the active calls within the bad neighbor¶s coverage area to increase power,
thus reducing the available forward traffic channel power and capacity.

To combat both of these issues (high SHO factor and high forward traffic channel power
per subscriber), coverage control is essential. Confining the coverage of sectors to their
designed coverage areas will allow for reduction of neighbor lists, and thus the SHO
factor, as well as reduce the incidence of the far-away mobile scenario. The coverage
control used to optimize capacity also provides additional benefits, such as minimizing
pilot pollution, reducing dropped calls due to excessive search times for large neighbor
lists, and reducing access failures and page retries due to excessive Io.

Additionally, related to the reduction of the SHO factor, some vendors have
implemented parameters that help to reduce the SHO factor, by effectively altering the
handoff parameters based on the number of active set PNs. As an example, let¶s
assume that the typical T_ADD (-14dB) and T_DROP (-16dB) values are being used in
the network. When a neighbor PN exceeds T_ADD, it will be added to the active set,
and when an active PN falls below T_DROP, it will be removed from the active set. With
the use of SHO reduction parameters, the first additional PN to be added to the active
set would be based on the normal T_ADD and T_DROP values, as we don¶t want to
completely inhibit SHO. Remember, the target SHO factor is approximately 2.0 to 2.2.

For the second through fifth additional PNs that are added to the active set, an offset
value is added to the initial T_ADD and T_DROP values, making it increasingly difficult
for neighbors to be added to the active set. We don¶t want to set these offsets too high,
otherwise, good RF energy would be wasted. Considering the initial T_ADD and
T_DROP values of -14dB and -16dB, if we were to add an offset of 1dB per step, the
fifth additional PN to be added to the active set would be required to meet a T_ADD
value of -10dB. Once added, the first active PN to drop below a T_DROP value of -
12dB would be removed.

This method of filtering candidate PNs is far superior to modification of T_ADD and
T_DROP directly. If T_ADD and T_DROP were modified directly, let¶s say to -10dB and
-12dB, a large portion of useful RF energy would be wasted. By only making it more
difficult for two or more additional PNs, RF energy is not wasted, and only the highest
quality neighbors are added to the active set.

% !!   
Traffic balancing, in general, rarely adds capacity to a network. It does, however,
provide for more efficient use of the existing capacity. More efficient use of existing
capacity means more carried erlangs and more revenue per sector, with no
expenditures for new equipment. There are two types of traffic balancing. The first is
balancing traffic amongst sectors in a given area. The second is balancing traffic across
multiple carriers. First, we¶ll look at balancing traffic amongst different sectors in a given
area. The following picture shows three base stations with sectors that share a
coverage boundary. Let¶s assume that initially, all of these sectors are setup identically
(same antenna, same downtilt, and same overhead power).

Let¶s then assume that BTS3, gamma sector is experiencing high blocking due to a
traffic hotspot within the sectors coverage. Assuming BTS1, alpha sector and BTS2,
beta sector are not blocking, several steps can be taken to more evenly distribute the
traffic from the hotspot amongst the three sectors covering the area. First, reduce the
coverage of BTS3, gamma sector by doing one or both of the following; increase
antenna downtilt, decrease overhead power. Second, increase the coverage of BTS1,
alpha sector and BTS2, beta sector by doing one or both of the following: reduce
antenna downtilt, increase overhead power. Additionally, if coverage from other sectors
allows it, azimuth changes can be made to BTS1, alpha sector and BTS2, beta sector to
place the traffic hotspot in their antenna¶s main beams, thus reducing the downtilt
changes and/or the overhead power changes.

It is important to note that reducing the coverage of BTS3, gamma sector by decreasing
overhead power does affect the capacity of the sector, in two ways. First, by decreasing
the coverage area, the number of subscribers covered by the sector will be reduced.
Second, reducing the overhead power allows more PA power to be used for the forward
traffic channel.

However, for BTS1, alpha sector and BTS2, beta sector, increasing coverage by
increasing overhead power also has two affects on capacity for these sectors. First, by
increasing overhead power, the coverage of the sector is extended, thus covering more
subscribers. Second, by increasing overhead power, there will be a reduction in overall
forward traffic channel power available.

So, even though we increased the overall traffic capacity of BTS3, gamma sector, we
decreased the overall capacity of BTS1, alpha sector and BTS2, beta sector. But, since
the traffic hotspot is now covered by all three sectors, we can expect a reduction in or
elimination of blocking, and thus an increase in carried erlangs.

Care must be taken when adjusting the coverage of sectors to balance traffic. Several
weeks of busy hour data, as well as drive test data for the surrounding area should be
reviewed prior to deciding which sectors can afford a reduction in capacity to fix a
capacity deficiency on another sector and if changing the coverage footprint of any of
the sectors would create a non-capacity related problem, such as coverage holes.

&c % !!   


There are three methods of balancing traffic across multiple carriers; hashing, global
service redirection, and Intelligent or Adaptive traffic algorithms. All three methods of
traffic balancing, if setup correctly, help to improve network capacity. If multi-carrier
traffic is not balanced, and blocking occurs on a carrier, even though there is excess
capacity on other carriers, this excess capacity is wasted, and thus the network capacity
is diminished. For this discussion, we will talk about a network with a second carrier
deployed in certain areas. But, whether we are talking about a network with 2, 3, or 4
carriers, these traffic balancing methods remain the same. The following picture shows
an island of f1/f2 coverage, surrounded by f1 only coverage.

The IS-95 Hash function distributes idle mode traffic evenly amongst multiple carriers. It
is the easiest of all the traffic distribution methods to implement, as it is a system wide
function with no settable parameters. When a mobile handoff occurs in idle mode (in f1)
to an f1/f2 sector, the mobile will receive a Channel List Message on the new paging
channel containing two frequencies. The hash function makes decisions (an odd/even
distribution function) based on the mobile¶s IMSI number, which generates a 50/50
outcome for mobiles to either stay on f1 or move to f2. Therefore, if all mobiles were idle
when entering the f1/f2 region, or if the same carriers were deployed throughout the
entire network, this method of traffic distribution would generate an even loading
between f1 and f2.

However, since additional carriers are rarely deployed across the entire network (at
least initially), and there is no guarantee that all mobiles will be idle when switching from
an f1 only coverage area to a multiple carrier area, the traffic will not be distributed
evenly. If a mobile originates or terminates a call in an f1 only coverage area, and then
moves into the f1/f2 coverage area, the call will remain on f1. Thus, f1 will always carry
more traffic than f2. If the f1/f2 coverage area is quite large and call durations are short,
the affect of this imbalance may be negligible. But, if the f1/f2 coverage area is small
and call durations are longer, the affect of this imbalance can be extreme.

Global Service Redirection (GSR) also performs traffic distribution in idle mode only, but
gives the service provider much more flexibility related to balancing the traffic between
carriers. First, rather than using an odd/even distribution, GSR uses the Access
Overload Class of the mobile to determine the carrier. The Access Overload Class of
the phone is determined by the last digit of the MIN, thus producing an equal distribution
of classes from 0-9. There are six additional classes (A-F or 10-15), which are used for
special purpose mobiles, such as test phones. Second, GSR is setup on a per sector
basis, which allows the service provider to set different traffic distribution schemes
throughout the f1/f2 coverage area, based on actual traffic patterns.

To account for non-idle phones entering the f1/f2 area from an f1 only area, GSR can
be setup to distribute a higher percentage of idle traffic to f2. If previously, the Hash
function produced a 70%f1/30%f2 distribution of traffic, GSR can be setup to direct 80%
(or 8 overload classes) to f2, and only 20% (or 2 overload classes) to f1, thus
counteracting the affects of mobiles entering the f1/f2 coverage area already in
conversation mode. In this example, the actual percentage of traffic to be directed to f2
to counteract the 70%f1/30%f2 ratio is 84%, but since we can¶t assign 8.4 overload
classes to a particular carrier, we must settle for 80%.

Additionally, with GSR, if a mobile is directed to use a different carrier on a particular


sector, but there are no resources available for that carrier, or the carrier is temporarily
off the air, the service provider can set a parameter in GSR to allow the mobile to return
to the original carrier. With Hash functionality only, a mobile in the same situation would
continually attempt to re-sync on the directed carrier even though it may be off the air,
or would continually have the origination or termination blocked until a channel element
of forward power resources became available on the sector.

Intelligent or adaptive traffic distribution algorithms, rather than using a statistical


method (IMSI or ACCOLC) to distribute traffic, use the state of the network to determine
how the traffic should be distributed. This type of traffic distribution has been
implemented on most vendor equipment.
Typically, an adaptive traffic distribution algorithm will use the following values to
determine which carrier is best for a new origination or termination: Channel Element
Resources Available, Walsh Code Resources Available, and Forward Traffic Channel
Power Available. The algorithm uses these values to estimate the capacity of the carrier
prior to adding the new connection. The algorithm estimates the capacity on the
additional carriers, typically in the order specified by the service provider (Carrier
Priority). If one of the other carriers is estimated to have more capacity available, the
new connection will be directed to that carrier.

On various vendor equipment, there are different options available for the adaptive
traffic distribution algorithm that determines how the algorithm selects a carrier. Some
options available; the ability to select the order in which the carriers are tested by the
algorithm; the ability to force the algorithm to try to keep originations and terminations
on their initial carriers, even if loading becomes unbalanced (to reduce access failures
due to frequency changes); and the ability to set an artificial capacity threshold for a
carrier ± the capacity of a carrier, as seen by the algorithm would be the estimated
capacity minus the capacity threshold ± this causes the algorithm to select other carriers
first, until all carriers are estimated to have the same capacity, then the algorithm would
distribute traffic equally amongst the carriers.

You might also like