Today's wireless networks are facing crucial challenges, such as radio spectrum scarcity, and excessive energy consumption. The vision of next generation networks lies in overcoming these challenges, and providing seamless wireless communication. This thesis aims to contribute to shaping the vision of next generation networks. Specifically, the thesis investigates how to optimize various network functionalities to improve overall utilization of radio resources, and how to enhance practical significance of the radio resource allocation techniques proposed in academia.

The thesis first focuses on enhancing network energy efficiency (EE) for downlink transmission of macro-only cellular networks. It develops a novel joint design that incorporates resource blocks (RBs) and discrete power levels allocations. Next, the thesis focuses on enhancing EE fairness (balancing) among individual users for uplink transmission of macro-only cellular networks. Particularly, in such networks, it considers a joint design of RBs and power levels allocations to maximize minimum individual EE. Unfortunately, both the downlink and uplink optimization problems arising from the joint designs are non-convex, and hence difficult to solve. To overcome this difficulty, the thesis reformulates the design problems in a form that is amenable to the standard semidefinite relaxation (SDR) with Gaussian randomization approach. This approach has a polynomial-time complexity, and yields a close-to-optimal performance.

The thesis later shifts the focus to enhancing spectral efficiency in multi-tier heterogeneous cellular networks (HetNets). It investigates the problem of user-to-base station (BS) association in downlink transmission of HetNets. In its simplest form, this problem is nonconvex. Hence, the thesis uses the efficient two-stage SDR-based technique.

The last part of the thesis considers a joint design of user-to-BS association, and RBs and power levels allocations in HetNets, allowing RBs to be reused opportunistically. The thesis discusses two network instances: with and without time-sharing. In the time-sharing case, user-to-BS association, RBs and power allocations are time-shared, and hence, not fixed throughout the signalling interval. In contrast, in the no time-sharing case, the network functionalities, once determined, are fixed throughout the signalling interval. The optimization problems in both instances are non-convex. To circumvent this difficulty, the thesis generates bounds on the solution of the original problems.