Interference and Resource Management in Heterogeneous Cellular Netowrks

The task of future wireless cellular networks is to keep pace with the rapidly increasing demand for mobile data service. Statistics have shown that mobile data usage has been increasing by a factor of ten every four to five years. It is predicted that the global mobile data consumption will reach 20 billion gigabytes per month by the year 2020. To accommodate this gigantic amount of data demand, it is necessary to deploy many more base tranceiver stations (BTS's) of different coverage to form heterogeneous networks (HetNets)..;According to some early studies, the network capacity grows nearly linearly with the density of BTS's in the network due to the cell-splitting gain. With each cell serving fewer users, both the resources allocated to each user and the corresponding transmit rate increase in order to support high speed data services, such as file sharing, video streaming and real-time gaming. However, as the density of the nodes in a HetNets grows, the efficiency of each node (the throughput of each cell) also decreases. This drawback in efficiency mainly comes from significant inter-cell interference and potentially large traffic variations across time and space. The inter-cell interference mainly exists among nearby small cells. As a result of random deployment of small-cell BTS's (micro, pico, femto and relays), the cells of different sizes overlap with each other without clear boundaries. Hence traditional cell planning and static frequency reuse become inefficient. How to direct traffic and assign resources efficiently are important problems in HetNets.;This thesis addresses the interference and resource management problems from the viewpoint of two different timescales. The fast timescale is on the level of time slots (milliseconds), where single packet transmissions, user scheduling and channel estimation happen. In contrast, the slow timescale is on the scale of user session activities (seconds or minutes). On the fast timescale, accurate channel state information (CSI) may be available within a limited region. Operations such as scheduling, beamforming and power control can be performed autonomously based on the instantaneous CSI. On the slow timescale, long term average traffic and channel information can be exchanged over the entire network, which is more suitable for global optimization of network resources.;First, we considers a fast timescale method to deal with interference, i.e., interference alignment (IA) in cellular networks. By aligning intra- and inter-cell interference within a subspace of the spatial dimensions provided by the multi-input multi-output (MIMO) channels, a significant gain in degrees of freedom can be realized by jointly designing the precoders and receivers at different BTS's and user equipments (UEs) with full CSI. The performance of IA relies on accurate estimation of the CSI. Due to the overhead in training and pathloss effects, it is advisable to consider IA within a cluster of several cells.;The thesis then focuses on slow-timescale resource allocation in dense HetNets. A slow-timescale HetNet model is introduced to characterize the spatial traffic distribution and dynamic interference. A topology adaptation scheme is proposed to improve the energy efficiency of a HetNet. Topology adaptation is realized by adapting the UE-BTS association and the BTS on/off status, which is formulated as a mixed integer program. Substantial energy savings can be achieved in low and moderate traffic regimes without user quality of service (QoS) degradation.;The slow-timescale HetNet model is modified to consider spectrum allocation with traffic aware interference. We allow a group of ;Finally, the thesis generalizes the spectrum allocation problem to networks with k types of UEs (representing UEs from different location). To keep the analysis tractable, the interference is considered under the full buffer constraint. The joint user association and spectrum allocation problem is formulated as a delay minimization problem. Similar to the conservative allocation, the optimal solution uses no more than k of the 2n available reuse patterns. Further delay and throughput gains can be achieved by jointly optimizing the user association and spectrum allocation instead of optimizing spectrum allocation alone. To extend the solution to large HetNets, the optimization is reformulated through considering only local interference and local reuse patterns, which reduces the number of variables from O(nk2 n) to O(nk). To guarantee consistency of the local reuse patterns from a global point of view, a heuristic coloring algorithm is used. To achieve a good approximate solution, we propose to iterate between solving the approximate optimization problem and matching the solution using the coloring algorithm. Numerical results show that the approximate solution achieves close to optimal solution in small networks, and significantly outperforms the full spectrum reuse and optimal orthogonal allocations in moderate size networks. The proposed approximate solution provides limited gain in large networks, due to a large number of interfering BTS's forming a cycle.