Value Automated Optimization

The Value of Automated Optimization Case Study: Using Capesso™ Blue to improve the Performance of a WCDMA Network SYMENA Software & Consulting GmbH Wiedner Hauptstrasse 24/15, A-1040 Vienna, AUSTRIA Phone: +43 / 1 / 5855101 – 0

Fax: +43 / 1 / 5855101 – 99

[email protected], www.symena.com

1

DOING MORE WITH LESS

This paper describes the benefits of using Capesso™ automated optimization to refine the rollout plan for a WCDMA network. The source data for the paper was a pilot project undertaken by Symena for a major European wireless operator. The pilot project was conducted on an operating network. It went live in late 2002 and customers were introduced in the second quarter of 2003. The pilot project was completed over the same period.

2

STARTING C ONDITIONS

The starting state for the project was a WCDMA network deployed in and around a city surrounded by mountains. The terrain made the project particularly interesting. There is a 2000m altitude difference between users on the valley floor and those in the surrounding areas. In GSM most of the altitude problems were solved by using frequency overlays. That solution is not available for WCDMA as only a single frequency band was available. The network was designed using standard, manual planning techniques on a widely used WCDMA planning tool with manufacturer recommended equipment settings. It was produced by engineers from a well respected wireless consulting firm and by the operator’s own engineers. The network was designed to provide a 95% probability of service across all customers in the market. Traffic maps were based on the operator’s GSM experience for voice combined with market estimates for data traffic. There was no priority of service for any given customer. There were 187 sectors or cells in the network design. They were spread across about 70 base station sites. All sectors were collocated with the operator’s GSM sites.

Page 1 of 5

3

OPTIMIZATION PARAMETERS

3.1

Objectives

The objective of the optimization was to demonstrate a capacity improvement in the deployed infrastructure. Any improvement will delay the deployment of additional infrastructure to meet rising traffic demand. The 95% service probability constraint was maintained.

3.2

Parameters Available

A limited set of parameters were available for adjustment. They are ranked here by cost from least expensive to most expensive. 1.

Pilot and Common channel power levels can all be set remotely for each sector;

2.

Electrical antenna tilt where available. This was further limited to down-tilt only as it is a design restriction of the hardware;

3.

Mechanical antenna tilt including changing mounting brackets to allow for up tilt;

4.

Azimuth direction but limited to +/- 20° on either side of the current direction and only where physically feasible. The arc restriction was imposed by hardware limitations. Changes beyond 20° are several times more expensive than those used.

Other parameters were available but they were not permitted in this project. They include: •

Selection of the most appropriate antenna pattern from a range available;

•

Deployment of new sites based on a finite set of those available; and

•

Several others such as soft-hand-over window and antenna height were available but not included in the consideration set.

There was no opportunity for deploying additional base stations to meet coverage and capacity requirements. This was primarily a cost consideration. New base stations are expensive.

The Value of Automated Optimization

4

BACKGROUND

4.2

Interference Limited Systems

calculations required to produce such a shift are relatively simple.

When considering optimization perhaps the key difference between GSM and UMTS air interface technologies is the capacity limiting factor. GSM is a Time Division Multiple Access (TDMA) system. Each user is allocated a frequency and time slot so that interference between users is minimised. Frequency use is geographically separated to minimise interference. This means that there is a fixed limit on the capacity of the system. Typically the limiting factor for GSM is the availability of frequencies. Once all frequencies and time slots are used then no more users can be admitted to the system. WCDMA is, as you would expect, a Code Division Multiple Access (CDMA) system1. All users have their signals spread across the same frequency band. So that each signal can be identified each user is allocated a code. Knowing the code means that you can pick the signal from the interference or noise in the system. Because of the design of the code all of the other signals in the system look like noise. In such a system the noise level in the system is the thing that limits capacity. Consequently CDMA is known as an interference limited system. In interference limited systems any change in the operating parameters of any transmitter will affect the reception of all of the receivers in the system. Since the usual physical laws apply, distant transmitters have a much lower impact than close transmitters, but all contribute to noise. There is no easily definable fixed limit on the number of users in a CDMA system. The actual number that can use the system at any one time will depend on a range of factors. It is in the clever setting of those factors that the benefits from optimization lie.

4.3

Manual Optimization

In WCDMA coverage and capacity are adjusted by antenna tilt and pilot power. Tilt the antenna down and reduce pilot power and you make its footprint smaller. This increases capacity close to a base station. Tilt up and increase pilot power and coverage is increased. If you consider a pair of base stations then you can shift capacity from one to the other by the appropriate combination of tilts and pilot power changes. This is shown in Figure 1. below. The

1

The “W” is “Wideband.” That means that the signal is spread over a wider band (5MHz) than the CDMA standard IS-95 or CDMAOne™ (1.25MHz).

Page 2 of 5

Figure 1: Shifting Coverage and Capacity With three cells then you must juggle those two parameters for each. As more cells are added to the network then the problem becomes very much more complex very quickly. An adjustment on one side of the network can ripple across to the other side upsetting all of the previous adjustments. Now consider the number of cells a network needs to cover a city. Assume that you have one parameter (with only “0/1” as valid settings) to adjust on each base station in a network of 75 cells. Then there are 275 possible arrangements for just that parameter2. Needless to say this is much more complexity than the human mind can deal with directly and so approximations are essential. Consequently manual “optimization” tends to be a case of making sure that the system works rather than trying to extract the maximum value from the system. Of course there is nothing to say that manual methods can’t produce a good result but it might take quite some time!

4.4

Realistic Results

It is possible to get spectacular optimization gains in network performance for a given instance of traffic. However when the mobiles move the optimization conditions collapse and the optimal performance degenerates very quickly. Although the performance gains for that particular case are good, the result is useless and misleading if it don’t work for the general traffic situation. To guard against this effect traffic is usually modelled in terms of a series of probability distribution functions. These distribution functions are then used to generate instances of traffic. Each instance is then statistically independent of every other instance. The optimal parameter settings are tested against a large number of these independent instances to

2

275 is a very large number. If each decision were 1mm then 275mm is about the distance that light will travel in 4000 years!

The Value of Automated Optimization

ensure that the gains achieved are statistically significant. If the results achieved for one instance of traffic are great but on average the result is poor then the average gain will be low. Consequently optimal settings that result in such situations are discarded.

These results were achieved with a standard deviation of about 5%. Table 1 shows that in half of the traffic instances the performance improvement was more than 30%!

To measure the stability of the optimization gain you can use the standard deviation of the average gain. Standard deviation is a measure of how spread the results are about the mean. The higher the standard deviation the more spread out the results and the less reliable the performance. Symena tested its optimization solution against 50 statistically independent traffic patterns to ensure that the performance gain achieved is useful.

5.3

5

PROJECT RESULTS

5.2

Performance Gain From Optimization

Service Improvement Due to Optimization

These gains due to optimization can also be expressed in an improvement in service probability. Service probability is the likelihood that a user will receive service when they are in the coverage area.

The gain due to optimization is expressed as additional capacity as a percentage users served at the desired data rate with a service probability of 95%. Because Azimuth changes are expensive, two recommended optimal configurations were considered: •

No azimuth changes; and

•

Azimuth varied with emphasis on +/-20°.

The optimization results are shown in Table 1 below. The percentage of cells requiring a costly change is shown so that the approximate cost of the optimization is known. There were power adjustments to most of the cells but these parameters can be changed remotely and hence direct cost is insignificant. Radio Configurations Original Configuration

No Azimuth Changes

With Azimuth Changes

Relative Gain

-

29.7%

35.8%

Manual Tilts

-

17%

18%

Electrical Tilts

-

79%

73%

Both Tilt Types

-

17%

14%

Azimuth Change

-

-

33.7%

Figure 2: Increase in Service Probability due to Optimization It can be seen in Figure 2 above that the original plan could serve about 1000 users with a 95% service probability. Optimization has improved this to about 1300 without the deployment of additional infrastructure. This improvement in service probability is extremely useful in low traffic networks as it improves the performance perception of early adopters. Of course from a marketing point of view that is very valuable. My shiny new phone works well!

5.4

Gains are Spread across All Users

One of the features of the recommendations is that, in general pilot power should be decreased. It might be expected that this would cause indoor users to lose service. In fact it does not. That result is shown in Figure 3 below. Gains are spread more or less evenly across all users regardless of their situation.

Table 1: Optimization Results

Page 3 of 5

The Value of Automated Optimization

6

COST

Some assumptions can be made about the cost of conducting the required changes. Of course these will vary from location to location and they are only intended as a guide. Using these prices then the cost of delivering the performance gains are about:

Figure 3: Comparison of Gains due to Optimization for Different User Situations

•

€50,000 for the “No Azimuth Change” result; or

•

€180,000 with the inclusion of Azimuth.

That is between 1% and 2% of the cost of establishing the network in the first place assuming that each site is built on an existing GSM site. The savings are even greater if a new site is required. For a relatively small investment in services a very large gain is available from the infrastructure.

Serial

Change

Cost

Notes

1.

Power variations

Negligible

Performed remotely from the OMC.

2.

Electrical Tilt

Negligible

If remote electrical tilt is available then the cost of making a change is negligible although there is a charge for the equipment.

3.

Manual Tilt

€1500 each

Requires a visit to the mast head. There will also be lost revenue when the site is shut down but this has not been calculated.

4.

Combined Manual Electrical Tilt

€1500 each

No separate visit is required.

5.

Azimuth Change

€2000 each

Within the +/- 20° range.

and

Table 2: Approximate Cost of Parameter Variation

7

CONCLUSIONS

A 30% improvement in system performance seems unrealistic. The immediate reaction is that either the gain does not really exist or the original planners are less than competent. Neither is the case. Symena have demonstrated the additional capacity in several well regarded WCDMA planning tools and with a large sample of traffic scenarios. The extra performance is available. Additionally there was nothing wrong with the skills of the original engineers! Rather the problem lies in the planning technology. Moving from manual optimization methods to automated optimization is the source of the gain. When the network was planned, automated optimization was not available. While a CDMA network is simple in concept it is very complex to model because of its interference limited nature. The engineers planning the network cannot hold hundreds or even thousands of variables in the

Page 4 of 5

heads simultaneously. They must simplify the problem to human scale. That ignores many of the sources of interference. A computer can track all of the variables simultaneously. With some smart mathematics a better solution can be found. This is the source of the benefit. Unfortunately there are costs associated with the optimization but they only run to a few percent of the value created by doing them. There are some things that can be done to minimise these costs: •

Keep comprehensive, reliable records of the direction, tilt and type of each fixed antenna in the system;

•

Use Remote Electrical Tilt antennas wherever possible. Knowledge of optimal tilt for a traffic pattern actually gives you a reason for wanting to change tilt on a regular basis; and

The Value of Automated Optimization

•

Allow for changes in azimuth so that the maximum gain can be achieved.

Automated optimization as delivered by Symena’s Capesso™ Blue product can deliver a 30% performance improvement even in a well designed WCDMA network. This is done at a cost of a few percent of the value of the gain. An operator not using automated optimization is overbuilding infrastructure. Only automated optimization can get the maximum out of the infrastructure investment!

Contact SYMENA Software & Consulting GmbH Wiedner Hauptstraße 24/15 A-1040 Vienna, Austria Tel: [+43-1] 585 51 01-0 Fax: [+43-1] 585 51 01-99 [email protected] www.symena.com

Page 5 of 5

The Value of Automated Optimization