Unit 3

Circuit voice networks In most forms of retailing, the introduction of new products follows the S curve sequence rst recognised by Rogers [1]. Rogers stated that adopters of any innovation or idea could be categorised as innovators (2.5%), early adopters (13.5%), early majority (34%), late majority (34%) and laggards (16%), based on a bell curve. Adopting this model, initial sales would come from the early adopters, with volume sales from the early and late majorities. In conventional retail products, this means that all of the development and initial production costs have to be incurred before any revenue is earned; but it is likely that the costs of volume production can be deferred until there is some feedback on whether the product is likely to be a success. For cellular network operators, the investment prole is different; the service offered is connectivity over an entire country or region.

Considerable attention must, therefore, be paid to the way in which coverage is rolled out if the project payback period the period that must elapse before the initial investment has been recovered through annual cash ows is to be kept short. This payback period, or the more comprehensive Project Net Present Value (NPV),which reects the current value of a series of projected future cash ows at a specied discount rate, are keymetrics in assessing whether a particular project is viable.

There are at least four dimensions that can be exploited by the operator to minimise investment in the early phases of a project: (i) Network coverage, (ii) Network capacity, (iii) Network features, (iv) Handset subsidies.

Coverage The sequence and extent of network rollout can have amajor impact on project protability. Addressing regions of high population density rst (such as city and urban areas) can maximise potential revenue for the minimum infrastructure investment. This attractive result arises both because the area to be covered (and hence the number of BTS sites) is minimised and also because this same area is likely to contain a high percentage of businesses, which are typically early adopters.

Coverage of rural areas with small numbers of potential subscribers typically occurs last and may only happen because of legal obligations to provide such coverage, arising as part of the spectrum licence conditions. Initially, coverage in rural areas will be far from comprehensive because not all cells will overlap perhaps because of local topography (e.g., hills attenuating RF signals). At a later stage, when revenues can justify it, in-ll sites may be introduced to address these limitations.

One estimate of the path loss from the base station to the mobile, at some distance r, can be obtained by projecting the power transmitted by the BTS antenna onto the surface of a sphere of radius r (assuming that the antenna radiates power equally in all directions). This loss is due to the intrinsic reduction in power density with range and is known as the free space path loss. However, except at short ranges from the BTS and with a few other exceptions, this form of simple propagation loss is rarely encountered in cellular systems because of a number of additional effects. For the ground-reected case, it can be shown that the received power Pr at some distance d from the transmitter is approximately equal to-

Figure 3.2 also illustrates a number of other frequently encountered effects. Shadowing occurs when a building or other feature that attenuates radio signals is located at a distance from the base station or mobile such that it signicantly obscures the line-of-sight propagation path. Propagation does occur, despite the shadowing object, and is the sum of the attenuated direct ray ASU and diffraction around the object ADU.A further effect needs to be considered when the user is moving. For locations where there are several reecting surfaces (most deployments) it is apparent that movement of the user will alter the phase of signals from the various paths so that when the signals combine at the receiver, they can give rise to fading. The fading may exhibit Rician or Rayleigh characteristics [3] depending on the relative strength of the direct and multi-path rays. A real-world deployment will usually include impairments from all of the above sources,with large numbers of reecting and scattering surfaces togetherwith shadowing from objects of arbitrary size and location. The denition of models that may be used in planning wide-area networks is, therefore, carried out on the basis of curve tting to extensive empirical data. This approach was rst proposed by Hata [4], in his landmark paper Empirical formula for propagation loss in landmobile radio services but has since been developed by many contributors. In calculating the maximum cell range possible all of the above effects need to be considered in the link budget,a convenient expression representing the losses that may be incurred between the base station and mobile. So for a system with a BTS equivalent isotropic radiated power (EIRP) of Pt dBW the radiated power once transmit antenna gain is considered and mobile sensitivity Pr dBW (mobile antenna gain = 1) the downlink budget may be expressed as-

Capacity The central tenet of cellular systems is that they use the spectrum available many times over the planned coverage area, with a key gure of merit being the cellular reuse factor. If every tenth cell uses the same frequency as the rst cell in a deployment, the network is said to have a reuse factor of nine. This gure is important as, for a given amount of spectrum, it determines the maximum number of frequencies per cell and, hence, cell capacity. This, in turn, determines the number of sites for a given subscriber density. For instance, for a given amount of spectrum, moving from a reuse pattern of nine to three would enable three times as many carriers to be deployed in each cell. It is useful at this point to recall the way in which circuit networks, historically used to carry voice or video trafc, are dimensioned. Regardless of whether the network is upporting xed or mobile subscribers, a relationship needs to be established between the transmission capacity n (the number of circuits available), the grade of service B% (the acceptable blocking level) and the offered busy hour load E,in Erlangs. Then for a GSM single carrier, and recognising that one time slot is needed to support the BCCH channel, the offered load in Erlangs can be found, by reference to Erlang B tables for seven circuits and 2% blocking probability, to be:

In a green-eld network, the key requirement is to provide coverage over the planned region as quickly as possible and for the minimum nancial outlay. GSMcells with radii up to 20 km are easily practical, depending upon the propagation characteristics of the deployment area and so, for a single carrier, omnidirectional cell and 25mE demand, subscriber densities supported in such cells can be calculated as:

Figure 3.3 shows a capacity evolution path, which it is often convenient to follow. As more subscribers join the network, capacity can be increased by converting the omnidirectional conguration to a tri-sectored site, with one or more GSM carriers in each sector.More detailed discussions in Chapter 5 will show that frequency allocations found in typical GSMlicences will allow up to two carriers per sector and, exceptionally, four carriers per sector. In the latter case, capacities of up to 2.2 subscribers/km2 are possible (2% blocking, 20 km cell radius and 25mE demand), once the Erlang benets of larger numbers of channels, known as trunking efciency, are considered.

The working assumptions for macrocell planning are that the antenna at the centre of the cell site is higher than most of the surrounding obstacles, and that buildings and other obstructions are small compared with the cell area. These assumptions mean that there is usually at least partial line of sight to the subscriber, and when this is not the case reection or diffraction from one or two buildings is sufcient to regain close to line-of-sight communication. Under these conditions there is still a dominant path to the user, with perhaps only 5 dB to 10 dB excess loss over the direct ray. Microcells are characterised by deployments that deliberately site the cell antenna well below the prevailing building height. The concrete and steel construction of city centre buildings means that there is no signicant signal propagation through buildings; the buildings themselves thus dene the cell boundaries in these deployments. These cells can achieve very high subscriber densities, depending upon the particular street geometry, and propagation is typically close to line of sight. Microcells are never the only type of deployment in a coverage region; macrocells must also be present to provide coverage, as users move between microcell canyons.

Network features and handset subsidies So far, the discussion has concentrated on how to minimise the cost of rolling out the network infrastructure to support voice and basic telephony services, such as voicemail, call forwarding and other functionality hosted by the MSC that typically forms part of the basic feature bundle. However, rather like the early experience of xed telephony operators with video calls, service revenues only take off if people nd that the terminal cost is acceptable and are thus prepared to buy them. When rolling out services in developed countries, the practice has been to subsidise the cost of the handset and recover this over a xed term contract. At some later time, generally as the technology matures to the point where handset costs drop dramatically, pay-as-you-go tariffs become available with the end user bearing all or most of the device cost but no xed monthly tariff. With most of the green-eld deployments now taking place in countries where the annual income of the population is much lower than in the developed world, particular attention is being paid to this market. They are attractive from a revenue perspective because the governments concerned have often decided not to roll out copper to the home and rely on cellular coverage instead. However, operators need to be able to make a prot when the average revenue per user (ARPU) is in the region of $10 per month. Planning for such networks is quite distinct from the corresponding activity in developed countries Figure 3.5 shows the make-up of OPEX for a typical low revenue market (LRM). The cost of handsets (within the subscriber acquisition category) assumes key importance in these markets and has given rise to a much publicised initiative from Motorola, Nokia and the GSM Association (GSMA) to develop lower cost handsets targeted at these markets. Handsets that sell for <$30 are forecast for 2010 as a result of this work.

Planning for circuit multimedia services In GSM, the initial approach for data bearers was to retain the standard GSM GMSK modulation, and to optimise the channel coding and interleaving to offer a variety of circuit data bearers with rates from 2.4 kbits/s through to 14.4 kbits/s, all within the framework of the link budget for speech. This has clear advantages, in terms of planning simplicity. However, to provide a higher rate circuit bearer, it was decided to take advantage of the new8PSK(phase shift keyed)modulation proposed for EDGE and offer circuit EDGE bearers at rates up to 43.5 kbits/s alongside the packet services .Because EDGE delivers this performance by transmitting up to 3 bits/symbol (rather than 1 bit for GSM), the increased C/I needed for satisfactory BER means that the highest bearer rates are only available to subscribers away from the cell edge. Thus, for all GSM circuit data solutions, the support of multimedia applications, albeit at fairly low bearer rates, does not alter the cell size from that planned for voice. UMTS was conceived from the outset to be a multimedia-capable system. It offers a variety of bearers with rates between 15 kbits/s and 1920 kbits/s on the downlink and 7.5 kbits/s and 960 kbits/s on the uplink. However, unlike GSM, the maximum cell radius is directly

impacted by the choice of bearer rate. For practical systems with nite mobile power and equal uplink and downlink data rates, it is usually the uplink that limits the maximum cell size. The signal to noise plus interference power ratio (C/I) required at the base site for satisfactory operation is dictated by-

where Eb/N0 is the energy per bit to noise spectral density ratio needed for BPSK at the required BER in the particular multi-path environment and D is the data rate of the required bearer. Since the available transmit power from the mobile is nite, one can expect to see the maximum size of cells supporting applications requiring high-rate bearers to be signicantly smaller than the corresponding cell where voice is the only service deployed.

Planning for packet multimedia services It is expected that the current trafc mix, dominated by voice and other codec-driven trafc, such as video, will change to include an increasing proportion of trafc generated by applications such as VoIP, web browsing, gaming and, in the future, machine-to-machine communications. This trafc is characterised by aperiodic bursts of data separated by relatively long periods of inactivity, as shown in Figure 3.6. In a cellular system, dedicating one circuit bearer per user, matched to the peak data rate, to carry such trafc would be grossly inefcient. A far more attractive approach would be to convey such trafc over a packet bearer whose average throughput is well below the peak application rate, provided the delay and packet loss requirements are still met. The lower transmit power (associated with the reduced bearer rate) and corresponding reduced interference releases capacity in the cell for other users.

Before addressing the detail of the packet-planning process, it is useful to consider where such planning will be required in future systems. The list below is included by way of example and is by no means exhaustive: 1.Single-user point-to-point links in cellular systems, 2.BTS to BTS controller transport, 3.Aggregate trafc planning on point-to-point links, 4.Point-to-multipoint wireless access systems, 5.Dimensioning multimedia soft-switched trunks, 6.Signalling and transport dimensioning for IP multimedia systems (IMS).

Potential planning approaches Measurement test beds For some problems it is often quicker to build a test bed to measure performance rather than spend time developing potentially complex simulation models. Indeed, this was one of the initial approaches used by the authors. The test bed employed two computers inter-connected by an Ethernet link of dened bandwidth. The source computer selected bursts from the relevant statistical population and converted these to sequences of time-stamped packets of constant size, which were passed to a buffer of predetermined capacity. The second computer was connected to the rst by the Ethernet link and contained software to time-stamp packets on arrival. This enabled a distribution of delay times and packetloss rates to be established. However, it soon became obvious that thiswas not going to be a useful approach.

Simulation Simulation can sometimes be an acceptable solution, at least in some engineering investigations, and has again been used by the authors. To ensure the fastest possible run times, excellent general modelling frameworks, such as MATLAB,were discarded in favour of specically optimised C code. Depending upon the tail parameters, this too can sometimes take a day or two to run but more typically one to two hours. A bigger disadvantage is the level of engineering expertise necessary to congure and run the tool and ensure that the results are sensible. Nevertheless, run times of several hours are not amenable to the sort of what if analysis proposed earlier.

Modelling The limitations of the two conventional approaches discussed above in addressing the planning requirements for the new generation of applications mean that modelling is the preferred way forward. Unfortunately, classical queuing theory relies on a number of assumptions, which will not be valid in the majority of future planning scenarios. Most notably, the convenient assumptions of an exponential burst size distribution and innite buffer size are unlikely to be met. The authors have developed a proprietary tool, called ARC, specically for this purpose and used within Motorola. It relies upon transformations to map the problem to a space that is amenable to analytic approaches. The ARC tool can address the applications that are currently foreseen in future networks.

The buffer-pipe model The buffer-pipe model of Figure 3.7 is central to the management of packet trafc and it is important to optimise buffer size and bearer rates in order to minimise the cost of transport in future systems. So what drives these choices? A given application will only provide an acceptable user experience under certain conditions: 1.A maximum delay time the time taken for a burst leaving the users device to arrive at the application server some time Td later. 2.An acceptable packet loss ratio Lr the percentage of packets transmitted by the user that do not arrive at the server.

The delay time, Td, comprises two parts. The rst component, Tt , reects the time taken for a burst of data to travel across the bearer, as typically the whole burst must be received at the server before it can be processed i.e., it cannot be used on a byte-by-byte basis.

An indication of the expected transport delay can be calculated from:

where Bm is the mean burst size in bits and R is the bearer rate in bits/s. This component of the overall delay time can be managed through the choice of bearer rate relative to the mean burst size. The second part of the total end-to-end delay, tb, recognises that, at any point in time in a system with signicant load, the buffer is likely to contain some bursts waiting to be transmitted. If it can be assumed that the burst distribution and burst arrival rate exhibit Poisson behaviour, the steady state buffer contents, S bits, can be estimated using Markov chain techniques and Tb can be evaluated. Modelling of the expected packet loss ratio is conceptually straightforward. Assuming that the steady state buffer contents, S, are known, the probability of packet loss occurring is simply the probability that B is greater than (M S), where B is the size of the current burst being transported and is a direct function of the application statistics. Depending upon the choice of packet size and the extent to which the burst size exceeds the available space in the buffer, (M S), the number of lost packets from a given burst can be calculated. By evaluating this gure over an appropriate sample of the trafc population, a good estimate of the packet-loss ratio can be obtained. It is clear that, once again, the packet-loss ratio can be managed to a desired level by the designers choice of the mean trafc level (Bm Q) relative to the bearer rate R and the choice of buffer size.