2.2 Automatic Planning Strategies
2.2.1 Introduction
In Automatic Planning, human analysis is fully assisted by new mechanized strate- gies. This is an efficient way to keep track of the optimum plan as the network grows and evolves along time. Automatic Planning permits adapting the radio plan to long-term, seasonal changes. Opposite to Dynamic Planning, it refers to emi- nently static mechanisms that aim at obtaining an enhanced network configuration considering its average behavior.
Resolution methods are basically oriented to the optimization of a cost function Fcost that gathers the operator’s requirements and expresses the global value of a certain radio planning solution S. For that purpose several goals are defined and quantified by different sub-functions fcost. The final function to be optimized is usually a linear combination of those. Since different sub-functions are expressed in different units, a previous normalization is needed to ensure that the sum of values is meaningful. In addition, by changing the weight coefficients (ωi), the importance of each indicator can be stressed or reduced according to each particular case:
Fcost(S) = X
i
ωifcost,i(S) (2.1)
For example, a widely used function is the one that relates dropping and block- ing. In this case, the final cost is defined as the weighted sum of their probabilities
(Pblockand Pdrop respectively), in which dropping is usually adjusted with a higher
weight coefficient value since dropping an existing user is perceived as more detri- mental than blocking a new one.
Fcost(S) = Pblock(S) + ωPdrop(S) , ω > 1 (2.2) A second approach is to solve the different sub-objectives by optimizing the
different sub-functions sequentially. However, this indirectly leads to favoring some functions over others. Of course, a bad choice in the relative weights that define Fcost can also lead to inappropriate favoring or penalizing different sub-functions.
Another widely used approach is defining and optimizing a simpler cost function but subject to several constraints derived from the other sub-objectives. Continuing with the previous example, a possibility would be:
Minimize Pblock(S) (2.3)
Subject to Pdrop(S) < 0.02
Regarding sub-functions most are mainly defined according to coverage and capacity related criteria, but eventually economic aspects must be considered as well. Some examples are given next:
• Percentage of target area that is appropriately covered. This indicator can be defined in a service by service basis because requirements and Radio Access Bearers (RABs) are different, and so coverage areas differ as well. Furthermore, coverage must be guaranteed in three fronts:
1. Pilot coverage indicates those areas in which the UE is able to detect an appropriate level of the broadcast pilot channel. Further details on this are given in Section3.2.
2. UL coverage, which is determined by the UE maximum available power. Although link budget calculus aim at balancing UE capabilities with DL and pilot coverage, each cell shows its own particularities in terms of traffic and propagation and unbalance situations can appear.
3. Similar effects can be described for DL coverage, considering that op- erators limit the maximum power that can be devoted to one single link connection and also that the BS is limited in power. Further informa- tion and analysis on coverage variations with capacity and general cell breathing for both UL and DL can be found in [TYC05].
• The percentage of target area susceptible of suffering pilot pollution is also a coverage related parameter. This effect is observed in areas where UEs do not have enough RAKE fingers for processing all the received pilot signals. This excess of pilot signals does not contribute to improve coverage, on the contrary it can seriously contribute to degrade the received Signal-to-Interference-Ratio (SIR) due to increased interference.
• The percentage of areas in SHO is also an important metric to evaluate, so that smooth transitions among cells are guaranteed.
• Regarding capacity, traffic in UMTS Release 99 (Rel’99) systems is usually quantified in terms of number of users. Each user is assigned a Dedicated Channel (DCH). In this sense, maximum capacity can be defined as the sit- uation when a certain percentage of UEs cannot be served with the required
Eb/N0. This magnitude should be evaluated for both UL and DL since degra- dation might occur in different links of different connections.
• Another indicator directly derived from the previous one is the percentage of users consuming more than a certain percentage of its maximum power (in the UL or DL). This measurement permits detecting pre-congestion situations and can be obtained for the whole system or in a cell by cell basis. In this second case, a useful indicator is the number of cells with a certain percentage of users reaching the corresponding threshold.
• In HSDPA systems, since all user traffic is carried through a DL shared chan- nel, a different dimensioning approach is necessary. In particular the through- put arises as the important dimensioning indicator, for instance the average and peak throughput per user and per cell should be measured.
Due to the soft capacity feature of WCDMA networks, maximizing capacity requires careful load balancing among cells. That is why, network performance is very sensitive to parameters such as the pilot powers, antennas configuration (tilt and azimuth angles, beamwidth, use of null filling mechanisms, etc) and variables that feed to the SHO or the Admission Control (AC) algorithm, among others. All these parameters influence the network in a non-linear manner and finding their optimum combination is a problem of complex resolution.
In this sense, two different approaches exist in the literature to find the optimal combination of planning parameters. First, researchers from the operations research field are frequently focussed on strategies involving mathematical programming. The problem is modeled in terms of (usually) linear equations and thus simplifica- tions or focus on specific aspects are required, some smart examples can be found in [SY04; SVY07] and in several contributions from the MOMENTUM research project, which is an Information Society Technologies (IST) Project from the 5h Framework Program of the European Community, see for example [EFF+03]. Be- cause of non-linearities and dependencies among parameters, this approach is diffi- cult to be applied when all variables in the system are taken into account. Actually, the problem can be considered a combinatorial optimization one with a very high number of solutions. That is why most of electrical engineers and scientists from computer science tend to make use of metaheuristics, which is also the approach chosen in this thesis.