Online QoS, Revenue Management for Third Generation Mobile Communication Networks

This thesis shows how online management of both quality of service (QoS) and provider revenue can be performed in third generation (3G) mobile networks by adaptive control of system parameters to changing traffic conditions. As a main result, this approach is based on a novel call admission control and bandwidth degradation scheme for real-time traffic. The admission controller considers real-time calls with two priority levels: calls of high priority have a guaranteed bit-rate, whereas calls of low priority can be temporarily degraded to a lower bit-rate in order to reduce forced termination of calls due to a handover failure. A second contribution constitutes the development of a Markov model for the admission controller that incorporates important features of 3G mobile networks, such as code division multiple access (CDMA) intraand inter-cell interference and soft handover. Online evaluation of the Markov model enables a periodical adjustment of the threshold for maximal call degradation according to the currently measured traffic in the radio access network and a predefined goal for optimization. Using distinct optimization goals, this allows optimization of both QoS and provider revenue. Performance studies illustrate the effectiveness of the proposed approach and show that QoS and provider revenue can be increased significantly with a moderate degradation of low-priority calls. Compared with existing admission control policies, the overall utilization of cell capacity is significantly improved using the proposed degradation scheme, which can be considered as an “on demand” reservation of cell capacity. To enable online QoS/revenue management of both real-time and non real-time services, accurate analytical traffic models for non real-time services are required. This thesis identifies the batch Markovian arrival process (BMAP) as the analytically tractable model of choice for the joint characterization of packet arrivals and packet lengths. As a key idea, the BMAP is customized such that different packet lengths are represented by batch sizes of arrivals. Thus, the BMAP enables the “two-dimensional”, i.e., joint, characterization of packet arrivals and packet lengths, and is able to capture correlations between the packet arrival process and the packet length process. A novel expectation maximization (EM) algorithm is developed, and it is shown how to utilize the randomization technique and a stable calculation of Poisson jump probabilities effectively for computing time-dependent conditional expectations of a continuous-time Markov chain required by the expectation step of the EM algorithm. This methodological work enables the EM algorithm to be both efficient and numerical robust and constitutes an important step towards effective, analytically/numerically tractable traffic models. Case studies of measured IP traffic with different degrees of traffic burstiness evidently demonstrate the advantages of the BMAP modeling approach over other widely used analytically tractable models and show that the joint characterization of packet arrivals and packet lengths is decisively for realistic traffic modeling at packet level.