The functional architecture for the ATS consists of three main blocks, named Con- trol algorithm, Learning & Memory and Monitoring and two interfaces, one with UTRAN and the second with stored reference values. Whereas UTRAN provides access to the RRM parameters and measurements, the reference source provides the operator’s concept of QoS and network performance. A conceptual representation of this structure is presented in Figure7.1.
Control UTRAN RRM Parameters Learning Memory Control Algorithm
Automatic Tuning System
Reference Counters Monitoring Learning and Memory System
Figure 7.1: On-line Automated Tuning System. Functional Architecture.
The UMTS ATS creates a statistical feedback loop between network measure- ments and RRM parameters. The network is constantly monitored; selected pa- rameters are placed into memory for statistical analysis and compared with the reference source. When any of the cells does not meet the reference criteria, the tuning algorithm is started. Thus, the radio network tuning process becomes an automatic one. Each of the constituting blocks are explained next in more detail:
• Learning & Memory: This block can be seen as a data-base that accumu- lates statistical information concerned with the network performance (mem- ory). It is also responsible for finding out trends in the network behavior (learning). This entity will also be used to adjust the rules or the steps that govern the control algorithm.
The Learning process can be executed online or offline. Furthermore, if it is executed without human intervention, one would talk of self-learning. Several works have dealt with this specific topic and a combination of neural networks plus fuzzy logic rules is accepted to be a good combination for that purpose [Wer92]. The tuning of the rules that control the ATS should not be confused with the tuning of the UTRAN parameters themselves. In most cases there is no need to change these rules along time once they have been found.
• Monitoring: This block is responsible of obtaining information from the network and processing it adequately to obtain KPIs that give a better un- derstanding of the real state of the network and help the control block to take adequate decisions.
In must be noted that Automatic Planning algorithms can simulate and obtain statistics about global performance indicators which allow knowing the exact outcome of a given optimization action. For instance, the maximum capacity in the network. However, in the context of ATSs, performance must be evaluated in real time and so this kind of measurements are bound to be unavailable. In particular, in order to obtain a KPI the following steps are taken:
– Performance indicators measurement.
– Data filtering to overcome instantaneous fluctuations. – Combination of filtered data and calculation of final KPIs.
Next, KPI values are compared with the thresholds that define the operator’s QoS concept and an alarm is triggered when they are not met. Eventually, KPIs could be transferred to the O&M system in constant time intervals or in near real time for additional monitoring, however this action is out of the ATS processes.
Regarding the selection of measurements from UTRAN, each optimization case must be considered separately but the election should desirably take into account 3GPP definitions on performance related data. These are mainly grouped into radio-related measurements and protocol event counters [Kre06]. Whereas the second group information is somewhat reduced (mainly described in [3GPq]), a much better situation is found when looking for definitions of the first group. While [3GPk] defines measurement parameters themselves, [3GPh] is focused in how they must be reported. For example it contains reporting ranges, how the measurements are encoded in signalling messages, etc. In the context of UMTS Rel’99, the most important radio measurements in [3GPk] are summarized next [Kre06]:
– Measurements related to a cell and reported from Node-B:
∗ Received total wideband power and consequently the UL cell load factor.
∗ Transmitted carrier power and so the DL cell load factor.
∗ Preambles of RACH: number of connection request attempts. Also the number of rejected attempts.
– Measurements related to single connections:
∗ SINR and measurement difference with respect to SINR target. Even- tually UEs in degraded mode.
∗ Power transmitted by a single cell on a dedicated physical channel and therefore the difference with respect to the maximum allowable power.
∗ Round trip time on the radio interface (Uu interface). – Measurements from UE:
∗ CPICH Ec/I0.
∗ Power received on a DL dedicated physical channel. ∗ Received Signal Strength Indicator (RSSI).
∗ Reports on several events (see Section7.3).
• Control Algorithm: This last stage receives the alarm from the Monitoring block and, with the information provided by Learning & Memory, decides on the actions to take, which may compromise the change of RRM parameters. Regarding which network element holds intelligence and generates reconfigu- ration orders, three strategies can be defined:
– A centralized strategy, wherein a central element receives information regarding the whole network (or a great part of it). In this case, op- timization can take into account all existent interdependencies among cells. On the other hand, signalling between monitoring devices and this central control can be excessively high. Moreover, delays should be con- sidered if real time functioning is desired.
– Pseudo-centralized strategies, wherein intermediate elements in the net- work hierarchy, e.g. RNCs, take decisions over the group of Nodes-B that they control. Actually, information can be shared among these devices to achieve a better understanding of the whole network performance. – Finally, most of the existent works propose distributed topologies in
which decisions are taken at Nodes-B. Unless an information sharing mechanism is implemented, decisions hardly take into account the state of nearby cells and therefore this strategy is not the best option for cer- tain cases (as for example when trying to balance load among cells). On the other hand signalling is minimum.
In subsequent sections details will be given on the mapping that is proposed for the three ATS blocks in the specific problem of balancing UL and DL capacity.