Software models telecomms capacity

Nortel has used Maple to develop a generic capacity model for itsradio products, specifically designed to model the effect of theintroduction of higher bandwidth services to traditional PSTNradio systems.

Nortel is one of the worlds leading telecommsproducts manufacturers and has its principal European researchand development facility at Harlow in England where over 1200research engineers work on the design and development oftelecomms systems.

Boris Sedacca talks to Ben Freeman at Nortel,about the development of the model and the choice of Maple as aplatform.

Currently there are many different systems, in theworld, which offer PSTN services via radio systems.

Some arefixed wireless access, which offer PSTN services to homes via aradio link; other are mobile, such as GSM, the European digitalmobile radio standard.

The vast majority of the establishedsystems were originally designed for voice only (PSTN) and at thetime of their design the demand for higher bandwidth services,such as Internet access, was not envisaged.

One such standard forhigher-datarate services is ISDN, and demand for such a servicehas driven designers of fixed wireless access and mobile personalcommunications systems to look for means of offering higherbandwidth services in addition to PSTN on existing radio systems.ISDN was designed by the United Nations with the ultimate goal ofallowing any communications equipment to plug into any phone jackanywhere in the world.

While this goal is still many years fromrealisation, ISDN service is proving to be an extremely reliableand fast means of transmitting digital voice and data overexisting copper wires, fibre optics, radio and satellitechannels.

Many radio systems, such as GSM, operate on a frequencyand time slot air interface structure.

Any one user making a callis assigned a timeslot and a frequency, which may be fixed forthe duration of the call.

The radio interface protocols weredesigned to have a flexible timeslot structure to be able to copewith the provision of different types of service (eg voice, dataetc) although, as the majority of calls were expected to be PSTN,the single timeslot capacity was optimised for voice traffic.Data traffic therefore needs to occupy multiple timeslots and/orfrequencies.

For example, if each individual time slot has enoughcapacity for a PSTN voice call, say 32Kbit/s, then a higherbandwidth services such as ISDN would require multiple time slotsto meet its capacity demand, eg 64Kbit/s for ISDN basic rateinterface.

Nortel needed an improved capacity model which couldbe used to analyse the effects that mixtures of traffic typeswould have on their systems, especially how higher bandwidthtraffic would affect the service offered to existing PSTN users.The study of traffic modelling is a well-established disciplinein telephone engineering.

The basic factors involved are thenature of the demand from users on the system - ie the callattempt rate, call holding time or duration, and various systemparameters such as the number of channels available in the systemand minimum acceptable grade of service (GOS).

GOS is oftendefined as the probability that a user attempting a call will beblocked or delayed due to the system running at full capacity.

Itis commonly expressed as the fraction of calls failing to receiveimmediate service (blocked calls), or the fraction of callswaiting longer than a given service time (delayed calls).

AsNortel designs and manufactures a wide range of radio systems,involving many different standards and protocols, a generic modelwas required; one which could be easily adapted to differentsystems offering different services.

This generic structure wouldallow the blocking, or GOS, to be broken down by type of servicerequest - eg PSTN voice, PSTN fax, ISDN data etc - as well asbeing adapted to take account of specific channel selectionalgorithms implemented in the radio hardware as well as thedifferent call demand statistics for each of the service types.Ben Freeman of Nortel has developed just such a universalanalytic blocking model that takes advantage of Maple'ssymbolic computational features.

The model builds symbolicmultidimensional Markov finite state models, which are used tomodel the traffic on the radio system.

As an example of some ofthe more subtle complications of offering higher bandwidthservices to a primarily PSTN only system, take the case of a sixfrequency by 10 time slot system, which can maintain up to 60simultaneous single time slot calls.

In this example each timeslot has a data capacity of 32Kbit/s, which is adequate fordigital PSTN voice.

An ISDN call, requiring 64Kbit/s, needs twotime slots to obtain this data capacity.

However, the choice oftimeslots has restrictions because of the design of the radiosystem.

Cost constraints on mobile-type equipment prevent eitherreceiving or transmitting on multiple frequencies at the sametime, so for example, if the first timeslot were (T2,F4) then thesecond timeslot could not be (T2,F3).

"If you need two timeslots for ISDN you cannot just pick any two free timeslotsbecause of the system constraints.

If somebody wants to make anISDN connection, it is not sufficient just to have two time-slotsfree.

They have to be at an appropriate separation", saysFreeman.

The type of blocking which occurs when there is freecapacity, but not in the right place, is called 'softblocking' to distinguish it from 'hard blocking'when there is no free capacity and the system is working at fullcapacity.

Soft blocking can only affect multiple-time-slot callsand the probability of its occurrence gets progressively higheras the system nears capacity, ie as the airside capacity'fills up'.

Designers of a radio system need toaccurately quantify this affect and ensure that the radio systemintelligently allocates capacity to minimise 'softblocking', which will be experienced by the user as arefused call attempt by their ISDN radio modem.

Anothercomplication in traffic modelling is that different types ofusers have different call profiles so any model must be adaptableto allow the population to be defined in terms of user profiles,each with their own call statistics.

For example, a home Internetuser will typically make longer calls than a home phone user."We wanted a model that would allow us to look at theperformance of these systems by specifying the total capacity ina cell and running a model which would indicate what the GOS islikely to be", says Freeman, "The model we havedeveloped allows the user to plan capacity before installing anyequipment and hence to optimise their revenue for their system,whilst ensuring that all users experience a good GOS".Traditional methods of calculating GOS relied on Monte-Carlosimulation techniques, but the problem with this technique ontraffic modelling problems is that over 100,000 iterations wouldbe needed to approach realistic results, because of the burstnature of the traffic blocking.

Therefore, Nortel's modelwas designed to be analytic, providing accurate results within afew seconds running on modern PC platforms.

Another advantage ofanalytic models is that the computation time is the same,regardless of the probabilities of the events being modelled."With an analytical model, you need to put in a little moreeffort into understanding the problem, but once you have workedout your method, it is easier to compute", adds Freeman.

Oneof the key advantages of using symbolic mathematics is that theprogrammer does not need to specify the solution exactly in termsof all parameters but by a set of equations, each relating onepart of the system to another.

There is no rigid structure forspecifying the problem.

There are also advantages of reducedcoding time over lower-level languages like C or Modula-2, suchas built in support for complex variable types like sets, tablesand matrices.

Debugging and testing also takes less time, withthe effort applied to checking algorithms rather than the codesyntax.

Ben Freeman argues that with Maple, any mathematicalconcept can be tackled directly without having to write tools todeal with them.

The price of an interpretive system's slowerrun times, compared with a compiled C program, is more thanoutweighed by the savings in development time.

Maple software issupplied and supported by Adept Scientific.