by
John Berry
ATDI Ltd.
www.atdi.co.uk

Abstract
All too often in LMDS network design the marketers and planning engineers (representing business requirement and technical solution respectively) live apart, each operating with their own dedicated software tools. GIS tools incorporating population demographics are the preferred option for marketers with classic coverage tools used by their counterparts. Given that such detached working cannot be efficient, this paper postulates that, since subscribers can be uniquely specified in LMDS, a combined approach is essential.

Introduction
The objective of any radio network plan is to balance service demand and network resources deployed in order to meet a business plan.

What we need to achieve this is a methodology. This methodology must consider that we have a network roll out which commences with a pilot study, enters the main phase of design and then rolls on to satisfy the service demand over a number of years. During this time, we as planners, require to drive the deployment of resources and prove that business plan month by month, year by year.

The overall picture is complex - too complex to progress beyond pilot stage without tools with which to model the potential, and in time actual, subscriber behaviour, to model the environment of our expanding network, to model our limited resources as we deploy them and more especially to model the interplay between all of these elements considering that we have a fixed population who cannot re-orient their antennas to gain service from new local radio base stations in subsequent network enhancements.

The methodology proposed here comes from work in the France, Germany, Spain, and the USA on both LMDS and other fixed radio access networks during which we have integrated GIS and marketing analysis within our planning tools to give a single holistic modelling solution to what is a rather unique problem in radio communications. This model comprises the following aspects:

The input of marketing data describing customer behaviour
The prediction of ability to connect each prospective customer
The ability to develop both of the above together as the network grows

Customer Behaviour
Our holistic planning model begins here. At the outset of planning the concept is that the Marketing Manager rushes down to the planning engineers clutching a CD-ROM of market data - the population to be connected against time. This data contains the locations, number of discrete users at these locations and the service demands of all potential subscribers.   Sources of such data commonly available as text files include the local voters roll, the local Chamber of Commerce and various independent marketing organisations (Dun & Bradstreet). Postal codesand addresses, where available, give an additional information set and since these link directly to many other demography data sets and data forms it can be invaluable to import these vectors at an early stage.

Practically we need to import this data to the planning tool to form the subscriber database. Our next task is to sort this data to produce a considered target subscriber database for initial planning. From this initial target we can then develop prospective connections for subsequent phases throughout the life of the network thus linking radio base station deployment to anticipated customer payments and ultimately to the business plan.

At this stage the individuals in the subscriber database are un-parented or ‘orphans’. They have no assigned server (radio base station) and they exist only to describe a population to be served.

Where such population information is not available or where we are at the business planning or concept stage, virtual subscribers can be generated statistically within the planning tool from social data.

Whilst it would now be simple to dive into the planning task, it is worth pausing for a moment to discuss the nature of this market data and how we might expect to make use of it in months and years to come.

If we can indicate at the earliest stage of planning which subscribers will be connected week by week, we can drop out from the planning tool the necessary raw data to create personally addressed mail-shots for despatch to the target subscribers. We can target our sales force to close contracts for connection of these subscribers and we can brief installation teams prior to site visits. All of these bring the engineering function into the marketing process.

We should of course be aware that during the roll out of the network this subscriber database will change.   Subscribers will cancel contracts and new subscribers will appear as the service appeals to a wider population.   We need then to manage this subscriber database. It is not a once only input but lives as the fundamental core of the marketing, planning, roll out, sales and network maintenance.

A Model for Design and Deployment
LMDS BWA planning differs from mobility planning in quite specific ways and where there are parallels this is only in the very early planning phase. Our activity begins classically with the Requirement Specification leading to the search for radio base stations and this is shown at the top left of the overview model in Figure 2. This model shows the marketing data describing the customer behaviour as the central input.

Figure 2. Overview of the Holistic Model

Theoretically this site search can be done automatically, using the planning tool to suggest locations for given subscriber connections. Practically it is a manual activity linked to site acquisition and field work.   In BFWA we have typically a range limit of 2 to 10km to achieve, for typical equipment parameters, a path reliability equivalent to a copper connection (between 9.99% and 9.999%). This allows us to use a cellular planning approach with potential sites held in database each having an associated coverage prediction using an omni-directional antenna. These sites can be activated and deactivated speedily within the modelling activity. We can test population covered at any point using any combination of sites in any given area providing a coarse check of business plan viability for that network configuration.

Given some combination of viable radio sites we need then to conclude the design at this stage and to create the overall outline plan. You should note again the reference back to the marketing data as a check on how well the outline plan meets the requirement.

As we saw, these subscribers are, so far, ‘orphans’. They have no intended connection to any of the radio base stations so far suggested as part of the plan. The next task in our model is connection testing and parenting. Here the population information used ceases to be a simple list of locations - they become a database of unique individuals with names, addresses, potential telephone numbers and service demands. Connection testing commences by assigning a given directional antenna to each subscriber on the basis of a best server test.   Given this best server output subscribers are parented to base stations.

Subscribers can then be ringed using operator specified polygons to include or exclude each from traffic analysis as we probe the balance between resources and service using any of the popular traffic assessment methods. At this stage we can model segmented sites to give the required traffic capacity and we can interleave sites specifically to provide traffic handling without re-parenting existing installations. Subsequently subscriber parenting can be adjusted manually either individually or in groups to achieve optimal site loading and subscriber grade of service. Figure 4 shows this subscriber parenting.

Figure 4. Final Subscriber Parenting

In the model in Figure 2 we have proposed a linear flow with check back against the requirement expressed in the marketing data. This is of course all too simple. We can expect to iterate many times between the parenting/traffic analysis/resource deployment activities to achieve the right balance and it is here that operator skill comes in using additional sub-tools within the planning tool.

The output of this iteration is our plan for this phase of roll out.

The final activity then is practical installation, briefing field teams on azimuth to the best and second best server depending on the certainty of connection. The predicted signal levels can also be supplied along with data on the likely spatial signal variation at the subscriber premises and the actual measured values fed back to check the model and the planning process. Subsequent phases of deployment are shown as arrows on the model in Figure 2 effectively re-tracing the circular route clockwise. The activity is a continuum for as long as the network is being used and subscribers are being connected.

Practical Issues
There is one principal practical issue that needs to be understood. This is the cost associated with the planning tool digital terrain information and the associated accuracy of the propagation prediction model.

The modelling of the path loss to a given subscriber is well documented. It involves constructing a path profile within the planning tool and assessing the losses along the path according to a series of established ITU or other algorithms. Whilst these algorithms represent the best considered world radio engineering opinion, as with any prediction there will be an inherent error. Understanding and controlling this error is key to overall success.   There are two inputs determining accuracy. The first is the resolution and accuracy of the terrain and buildings data. Typically we would expect a resolution of between 1 metre and 10 metres for the equipment, frequency and path lengths involved in LMDS - lower resolutions in rural environments degrading to higher resolution in urban areas. Figure 5 shows a three dimensional representation of terrain and buildings from the planning tool. Great care should be taken in deciding on how to model the subscriber installation and in making assumptions over antenna heights available on installation. It is in the foreground of the subscriber terminal where the greatest error will occur. In selecting the correct buildings database be aware of how rooflines and other skyline features are described. The second input is the accuracy of the prediction for that given terrain and buildings environment.   Typical accuracy achievable in predicting a median signal level across the terrain step would be an average error of better than 1dB with a standard deviation of error of certainly better than 5dB in urban areas improving further in suburban and rural areas.

Figure 5. Three Dimensional View of Terrain & Buildings

Given knowledge of the accuracy and statistical distribution of the modelling process, we can calculate the certainty of being able to connect a given subscriber and in turn planning margins or second server overlaps can be used to improve the required result if necessary.   This data too can be fed back to the subscriber database to give a probability of achieving connection to a given RBS.

The debate over the resolution needed is contentious. We have seen organisations prepared to proceed with 500 metre resolution simply because the business plan would not permit the purchase of high-resolution data.   And conversely we have seen others spend millions of dollars to create highly accurate terrain and building models from aerial survey with a resolution of 20cm. Each technology has a different trade off and Figure 6 shows something of the balance needed with coarser resolutions at the top of the central ‘y’ axis.

Figure 6. Trade Off in Resolution to Achieve Reality

Figure 6 shows that when frequency increases so the resolution needed is higher (with lower terrain/building model step size). As we need reduced risk in connection (or conversely higher certainty of connection) the step size must reduce. Similarly as the cell size reduces so the step must reduce and with all this increase in data, the budget for digital map data can be expected to rise quite considerably. A good compromise for 3.5 GHz and 26-28GHz BFWA is 2-4 metres terrain/building step for urban areas reducing resolution to 10 metres in suburban and rural areas.

Conclusion
We have looked at three primary data elements to achieve a successful roll out. We need a model of the environment expressed as terrain, buildings and vegetation. We need an expression of potential (and later actual) subscriber behaviour contained here within marketing data and we need information on and quantity of resources available for use in balancing the three elements within the radio network planning tool to achieve the business requirement.

As with most complex problems which cannot be broken down to solve them, we must create a model - in our case here a holistic model. These three elements are modelled continuously and together in the same radio network planning software tool to provide a firm link between engineering and marketing to assure that the engineering design mirrors the marketing need. The use of a combined modelling tool with managed subscriber data firmly links the roll out with the business plan, controlling engineering risk and helping assure network success.