In the development of a telecommunications network the time lag between identification of the need to provide subscribers’ equipment, lines and exchange plant, and the ability to meet those needs may be quite considerable. This means that to augment the network we must accurately forecast these needs so that plant arrives and is installed before existing capacity is exhausted. In an ideal telecommunications network, with no restrictions, forecasting and planning would ensure that demand for services are accurately foreseen and satisfied as they arise. However, in practice this will not be the case, as planning is often constrained by availability of funds and policies of the administration or government.
There are two main requirements for producing accurate forecasts. Firstly, an adequate supply of accurate and relevant data should be available in the form of historical records, supplemented by appropriate more general background information on the various planning constraints. Secondly, a procedure is required for organizing and processing the data efficiently which can be used to produce the forecast. These forecasts should indicate the degree of uncertainty which is inherent in their production, and as far as possible include some indication of its magnitude so that planning and provisioning strategies may account of this uncertainty. If we fail to make statement concerning this uncertainty then we are making a prediction.
1. Demand Forecast
To forecast the number of subscribers in a well defined area. For access network planning this forecast may be done for each section first and then this forecast may be combined to form forecasts for bigger areas like blocks and exchange areas.
For the purpose of planning the junction network forecasts of subscribers are required in some of the methods. This forecast could be based on macro techniques discussed later.
2. Traffic Forecast
For access network planning forecast of total block traffics (originating and terminating) may be required. This would help in deciding the number of subscribers that can be planned per optical terminal when concentration is used. In case of design of ring structures where a number of optical terminals would be put in a ring, inter-block traffic may be of some importance.
For junction network planning, total exchange originating and terminating traffics, traffics for different traffic zones and traffic dispersion (traffic interest) between different zones and exchanges will have to be projected.
3.1 Period of Forecast
The demand forecast would usually be made annually in short term, say for 3-5 years and then at an interval of 5 years such that the last forecast is for 20 year period.
3.2 Method of Forecasting
While the forecast made at the head office level follows "Top-Down" methodology using techno-economic techniques or mathematical models to arrive at a country level forecast which is then appropriated to the regions, exchange areas etc. Forecast at the regional level or exchange area level (or below) will involve surveying and field studies.
3.2.1 Information required for forecasting
1. A city area map showing all the exchange location and physical features
2. Exchange area map showing the exchange boundary, blocks, cabinet locations, cabinet/block boundaries
3. A detailed map of the exchange area showing all the plots/buildings. This is usually in the scale of 1:1000 4. Existing demand for block/cabinet area
3.2.2 Zoning of the area
For ease of survey and correctness of forecasting a new area needs to be subdivided into manageable units. A city would have a number of exchange areas, an exchange area would have a number of blocks or cabinet areas and each block would have a number of sections or DP areas. To zone a new area, the area can be divided into survey units and teams of surveyor could then survey the assigned survey units and record information about the types of tenancies, residential or business that may be existing, under construction or planned. From this information, tenancies/plots are grouped into sections. Sections should be predominantly residential or business areas or they could be areas where residential and business tenancies are uniformly mixed. Size of a section in governed by the size of DPs normally used. This could be an area that can be served by a DP of size 10 or 20 pairs. Multistoryed buildings could make separate sections while in a low density area the size of section could be large. The section would then be grouped into Blocks in such a way that it should be possible to serve each block by a cabinet. In theory atleast the cabinet and block areas are synonymous but in practice there may be times when one block has more than one cabinet. Section becomes the smallest unit of area for which the forecast is made. Section forecasts can be combined to make block forecast and the block forecasts are grouped to make exchange forecast.
3.2.3 Classification of sections
Detailed survey would be carried out in each section to identify the types of tenancies. Since all kind of tenancies do not have the same growth potential, the tenancies would need to be classified on some basis. A common method of classification is based on types of tenancies.
Broad classification of tenancies used is
Residential
Business Sub-classification is then done in each of the above categories. R1 could be Detached houses, R2 could be Condominiums or luxury
apartments, R3 could be Low cost housing and so on. Similarly for business, B1 could be
big office complexes, B2 could be big shopping malls, B3 could be detached shops, B4 could be factories/workshops, B5 could be hospitals, government offices, schools etc., B6 could be restaurant, cinema, petrol station, parks, mosque/church/temple, museums etc.
3.2.4 Collecting development information
The forecasting personnel must obtain information about the developments that are likely to take place in future. This information would help in calculating growth in various types of tenancies and also the penetration factor. These sources of information include but are not limited to municipalities, town planning department, developers and builders, banks and financial institutions etc. The forecaster must use his experience and judgement to gauge the reliability of the information received.
3.2.5 Assessing growth of tenancies and penetration factor
From the information of the existing network and development information calculated above the forecaster will assess the number of different types of tenancies expected to come up in various years for which forecast is to be made. At this stage different types of residential R1, R2,..) and business tenancies(B1, B2,...) can be assessed separately as the penetration factor will be different in different cases. For example, grouping all types of residental tenancies into just one category and applying only one penetration factor to get the forecast will result in unreliable figures. Assessing tenancies may not pose much of a problem. Assessing PF for various forecast years is a more difficult proposition. Penetration factor is the ratio of demand and tenancies and therefore changes whenever demand or tenancies change. These changes could be caused by economic growth, social changes, increase in population, migration of population, increase of disposable income levels etc. A forecaster may have to study the past growth trends and development information to arrive at PF values for future years. A trend analyis based software may be of help.
3.2.6 Calculating forecast figures
Demand forecast for any year(y) and any category of tenancy(t) is given by
Forecast(y,t) = Tenancies(y,t) x PF(y,t)
3.3 Reliability of forecast
The method of forecasting demand described above is known as "Bottom-up" forecasting method. The forecaster starts by making a forecast for sections and then arrives at Block forecast by summing up the forecasts of all the sections in the block. Block forecasts are then totalled to make exchange level forecast. Exchange level forecasts can be used to make the next higher level forecast i.e. regional or country wide. To make sure that the forecast figures given by the above method are reasonably correct then they must be compared with forecast figures calculated by some other method. A method which is commonly used known as "Top-down" method of forecasting. Here the forecasting process begins by forecasting demand for an upper level unit like the whole country or a region or an exchange area rather than a lower level unit like section. This forecast is based on macro level parameter like population, income levels, GDP without concerning about the exact location of the subscriber. Correlation, regression and other statistical techniques are used to arrive at the forecast figures. Figures obtained at, say exchange level, by both the techniques should be compared. A variation of 10% to 15% may be acceptable depending on the policy of the administration. A variation of more than this would require a reassessment of bottom-up forecasts.
Those familiar with copper based access network planning would know that traffic consideration have rarely been important for dimensioning the access network. The reason for this has been exclusive rights of a subscriber on the pair allocated to him. The maximum traffic that a pair can carry is 1 Erlang. Also that is the maximum a subscriber can generate. [ simply put, a circuit continuously busy during the observation period is said to be carrying 1 Erlang traffic]. With the advent of new technologies, use of concentration in the access network (a la V5.2 interface), subscriber will no longer have exclusive right over a channel from his phone to the exchange. A subscriber does not use his phone all 24 hours a day and channels can be more efficiently utilized if shared among many subscribers. This would need measurement and forecating of traffic.
4.1 Traffic data
The production of traffic forecasts and the subsequent application of traffic theory to the dimensioning and administration of a telephone network depends on the availability and quantity of reliable reference data. This means that the data should be properly identified and related to the correct quantities of existing equipment for the relevant measurement period. These data must be systematically checked during collection and processing to ensure their integrity. Finally, for data to be in sufficient quantity, methods for collecting and processing must be readily available.
Traffic data for planning purposes are of three main types:
a). Exchange Data
These are the general statistics which specify the traffic generating capabilities of part or all of the exchange and include measured data as well as data supplied from other sources, together with information derived from these data.
For example this information includes:
b).Traffic Route Data These data usually include established routes which are dimensioned for a good grade of service and contain information about route size, identity, and the traffic carried by it. This would be required for dimensioning junction network
c). Dispersion Data
These data are held in the form of a set of row vectors for each originating exchange. They may contain call dispersion and/or traffic dispersion and associated mean holding time statistics. This would be required for dimensioning junction network
4.2 Traffic forecasting methods
There is a wide variety of different forecasting methods available. Many of them are already used in the transport network. The principal ones used may be categorized in the following way:
Intuitive forecasting is the systematic assessment of informed opinion and is often the basis of subscriber surveys which are developed to produce forecasts of subscriber development.
Trend methods assume that the future will have a predictable relationship with past performance. Their application depends on the existence of a database of past statistics which can be analysed to determine past trends. Trend projection is a frequently used traffic forecasting method for the short to medium term.
For trend methods, the main mathematical technique used is to adapt a function to historical data, applying the least squares curve fitting method and obtain the best fit to these data points. This mathematical function is then extrapolated to produce traffic values for future time points. As we shall see the curve fitting approach is also used in normative forecasting. If we plot traffic quantities on a graph we often find that they show a consistent pattern over time, and consequently it may be possible to forecast future values by constructing a line of "best fit" through the data points. The problem is to find an equation which describes the relationship between traffic and time, and also to define the criterion by which the parameters defining the curve of best fit may be calculated. One method is to ensure that the sum of the squares of the deviations from the fitted curve should be a minimum. This is known as least squares regression, and subject to various assumptions, estimates made using this method have optimal statistical properties.
The method of least squares can be applied to the calculation of numerical values of a given number of unknown parameters which are coefficients of a function in a mathematical model. The parameters of this model may be computed using the least squares method in the following way: Suppose that we have set of observations {y1, y2,..., ys} at time points {t1, t2,..., ts}, which we wish to model using a function of the type
Y = u1 + u2t
We require values of u1 and u2 which will give a straight line fitting the data as accurately as possible; this line is known as the regression line. We could chose a model with more than two parameters and try to fit a polynomial through these points. The accuracy of the fitted curve would obviously increase as the order of the polynomial increases to (s-1) the most accurate curve resulting when the number of parameters (r) is equal to the number of observations.
The purpose of performing regression analyses in this context is to find the coefficients of the chosen model which is then extrapolated into the future, and yields a forecast for time points t > ts. Now, if we use a high-order polynomial which gives an excellent fit to the observed data, we may often find that such an extrapolation gives results which are quite unacceptable from a practical viewpoint. Thus in forecasting work, we usually limit regression analyses to the low-order polynomials (linear and quadratic).
In many cases of traffic forecasting, a simple polynomial model may not be satisfactory as an explanation of the pattern of growth. Most of the other models which are in common use belong to the exponential family of curves, and the simplest of these is the exponential model.
Other models in this family may be treated in precisely the same way. They may all be broadly described as Modified Exponential models since they all contain the Exponential function. Two examples of such models are:
i)Logistic
ii)Gompertz
A number of errors can arise in the use of trend curves for forecasting. These errors may be due to one or more of the following:
It is likely that uncertainty arising from the first cause above will increase as the period of the forecast is increased, but it is difficult to quantify this uncertainty. The effects of uncertainty from the remaining two cause may be determined statistically. Cause ii. Depends on the variance of the estimated coefficients while the third cause depends on the variability of the data about the fitted curve.
A matter which cannot be analyzed statistically is whether the model which we choose will remain a true representation of the future values of the entity being forecast. In this case we may need to use some form of normative forecasting. The term "normative" suggests that our starting point for the forecast lies in the future itself; so that instead of looking forward from the present as we described for trend forecasting, we project backwards to the present from this future point.
The procedures which we may employ in this form of forecasting include those from trend forecasting; however, this time we make use of explaining factors other than time which, in our subjective judgement, will be related to the variable which we wish to forecast. Note that it is important for us to be able to forecast the explaining factors with known confidence.
This type of forecasting may be more adequately described as structural analysis since in its simplest form we separate different segments of the traffic which we wish to forecast, produce separate forecasts of the individual segments and then combine the results. The principles of the method will be illustrated by a discussion of originating and terminating rates: generally known as usage rates.
The subscriber usage rate has the dimensions of erlangs per subscriber. This rate varies from exchange to exchange, but in aggregate it has been found to change little with time for large groups of subscribers, (for example, subscribers of a metropolitan area). However, it can be assumed that usage rates of existing subscribers tend to increase in the long run for the following reasons:
i) The total number of subscribers is increasing and hence there are more people for each individual to call.
ii)Subscribers become more accustomed to using their telephone or other telecommunications device and find more uses for them.
On the other hand, new subscribers usually have lower calling rates than the older ones and tend to bring the average usage rate down. This factor must be taken into consideration when making usage rate forecasts. Other relevant factors are changes in the tariff structure and level of economic activity. All usage forecasts have a natural upper limit of 1 erlang per line (not per subscriber, as he may have more than one line).
The usage rate is always estimated at the time of a carried traffic measurement and forms part of the summary report of the measurement. Therefore it forms part of the available traffic information contained within the historical Database. Trend forecasting techniques may be used to estimate the coefficients of a suitable model. The estimates obtained from extrapolating this model may then be combined with subscriber forecast estimates to provide estimates of future traffics.
The above discussion has illustrated some of the considerations which need to be taken into account in the use of a simple normative forecasting technique. The process may be extended further by breaking the elements of a usage rate into the rate for individual types of subscribers, e.g. residential/business, single line/multiline, etc.
At this level of detail we may be able to make measurements which will aid in the answering of questions such as:
a). What causes changes in the usage rate? b). What are the effects of tariff changes on usage rates of long distance calls?
Exogenous variables can be used as possible alternatives to 'time' as regression variables. Studies in the Bell System using 20 exogenous variables have demonstrated that when complete and accurate historical data are available usage forecasts for an entity (e.g. originating traffic) can be improved by applying quantified estimates of changes in selected variables for models developed from historical data. The problems which arise are that variables which have the most impact are not consistent among the entities, and that gaps in historical data make this approach impractical for many entities. The resulting model is then used to forecast the entity by inserting forecast values of the exogenous variables into the model. The exogenous variables chosen should represent factors which have a demonstrated impact on the entity being forecast and they should be drawn from a set of data which is available in the form of an accurate time series and has a proven record of forecasting accuracy.
Some examples of exogenous variables of the use in calling rate forecasts
are:
Note also that not all exogenous variables affecting calling rates can be
used in normative forecasts; for example, the weather influences calling
rates in various ways but it is impossible to use this fact for any long term
forecasting of calling rates!
Comparison methods, where traffic patterns in a particular area are forecasted on the basis of known historical developments in another area.
