2.3 Dynamic Planning and Automatic Tuning
2.3.2 Automatic Tuning of RRM Parameters
The second group of automatic tuning algorithms is a more recent direction of research and deals with RRM parameters. General ideas on this topic can be found in [LWN06; NDA06] which validate its feasibility and demonstrate a significant increase in network capacity in comparison with default parameter settings. Most efforts have been focused on the parameters that govern AC and SHO algorithms.
It is worth mentioning that the integration of RRM parameters adjustment in the radio planning optimization process and the use of strategies that dynami- cally change the planning parameters imply that planning and RRM have partially merged and the line between both has become blurred.
Admission Control
Concerning AC, an automatic tuning of the maximum allowable load factor, or indirectly, of the measurements to calculate it, is proposed in several works. For instance, paper [HPVL03] controls the UL planned total received power target in a cell-by-cell basis, the proposed algorithm is very similar to the method presented in [LWN06]. The target value is auto-tuned in each cell based on the Block Error Rate (BLER) and blocking of Real Time, Circuit Switched (CS) calls and Non Real Time traffic queuing. A Look-Up-Table (LUT) is designed relating these three indicators with an adjustment decision (increase, decrease, etc). The metrics are calculated with straightforward functions considering that degrading a connection is 5 times “worse” than blocking a new one, and blocking is 4 times worse than queuing, which has a very small impact on the tuning. The same authors deal with the DL in [HV02] and extend the application of the algorithm to adjust the maximum power that can be devoted to a single DL link. Both papers are similar and just differ in the parameter to be tuned. The throughput increase is smaller (39%) in the DL than in the UL (50%) when compared to that obtained with fixed default parameter settings. Finally, paper [PDGA04] also deals with load target auto tuning in a very
similar way but taking dropping and blocking rates as inputs. The authors assume an autonomous control unit for each sector that is fed by quality indicators from the sector itself and its neighbors. The control is again rule-based but quantified and represented by a matrix, which in essence contains a similar information to that in the two previous works. Each element of the matrix contains a correction value for a couple of blocking and dropping rates. Similar matrices are constructed for each neighboring cell and then all of them are aggregated to give a global decision. However, it is not deeply explained how these neighboring matrices are generated and examples of the mapping between rates and decisions are not given.
Soft Handover
Automatic tuning of SHO parameters is usually done aiming at load balancing by means of cells shapes modifications, so the objective is the same as in most pilot and downtilt related works. However, instead of acting over the measurements them- selves, the different hysteresis thresholds that control the algorithm are varied. This way, the authors in [YGNT00;FN02;LCH05;LFYG05] propose automatic displace- ments of SHO areas so that loaded cells shed part of their UEs, which are naturally directed towards limiting cells. To achieve this, slightly different mechanisms are used but not all of them are applicable to UMTS networks.
For example, [LCH05;BHHPSS06] reduce SHO thresholds so that a cell with a hotspot is less appealing to nearby UEs and also it can expel some of its more exter- nal ones. This method, however, is only feasible for multi carrier CDMA systems, since the condition that triggers the reconfiguration is the percentage of occupied bandwidth. Similarly, the work in [YGNT00] designs an algorithm for CDMA sys- tems but soft capacity is not considered. Nevertheless, these cases could be adapted to realistic UMTS systems with minor changes, just measuring the load factor and comparing it with a maximum allowable threshold.
The work in [LFYG05] is fully developed in the context of UMTS networks, in this case there is a direct mapping between different load levels and specific han- dover thresholds. The work is quite generalist in the sense that it does not give the definition of the considered load factor. This parameter is constantly monitored and thresholds are updated every 100 ms if needed. However, mechanisms to avoid ping-pong effect and excessive reconfigurations are missing. Thus, in a system with active UEs using different services, variations in load levels could impair the algo- rithm performance. This issue is not a problem itself because complementing the proposal with a filtering process would be straightforward. Finally, the authors in [FN02] make use of a second order gradient method so that the minimum of a prede- fined cost function is constantly tracked. This function relates handover thresholds with the blocking ratio and DL transmission power.
Previous works perform the network tuning by modifying the thresholds that govern the addition and dropping of cells. However, in the case of UMTS, the stan- dard defines a parameter that can be easily used to make more or less attractive particular cells. This is the Cell Individual Offset (CIO), which is usually not con-
sidered for load balancing. More details on how this parameter affects on SHO is given Chapter7. A second comment to existent work is that load evaluations are done considering just one link. However, SHO parameters impact on each link in an opposite manner and thus should be considered and optimized at the same time. This effect is studied in Section 7.3 and indeed this is the framework of the cur- rent proposal. The opposite effect is investigated and conveniently used to detect whether the UL or the DL is the limiting link and favor it so that congestion control algorithms are delayed. Also, among the novelties on this topic, an important one is the investigation of UL and DL behavior in front of variations of SHO parameters and some extra light is provided in the lack of consensus about gains or losses in the DL.
Base Station Parameter
Adjustment
3.1
Introduction
Prior to the design of any Automatic Planning strategy it is mandatory to investi- gate the effects of the variables to be optimized. In addition, interactions with other parameters must also be included in the study.
In the previous chapter it was shown that two of the parameters that have been used for optimization more often are the CPICH and downtilt of antennas. Indeed, modifications in both of them allow a controlled way of modifying the cell shape. This is the effect that is mainly considered by existent approaches that aim at transferring UEs between cells. However there are secondary effects in WCDMA wireless access networks that are usually missed and which imply that reducing the load in a cell is not as straightforward as just reducing its cell area.
This chapter focuses on the impact of these two planning parameters on the radio access performance and in particular, the study stresses effects beyond well known ones.
The chapter is organized in two differentiated parts. The first one deals with CPICH power variations and the second with downtilting. In both cases an intro- duction is given on their role in the network and how they impact in the outcome of the radio planning. Interactions with other parameters are studied in both the UL and DL and some considerations on UEs spatial distribution are given too. Con- clusions from this chapter are a direct input to the Automatic Planning strategy proposed in the next one.