The widespread popularity of wireless smart devices and proliferation of multimedia services have posed unprecedented challenges to the designers of wireless communication networks. To meet the fast growing data traffic demand, a great number of wireless access points (APs) should be deployed, which, nevertheless, would also cause severe interference if they are not properly placed. Therefore, it is of paramount importance to study how to optimize the AP placement of infrastructure-based wireless communication networks.

The difficulty of optimizing network topology mainly lies in the time-varying nature of users' positions. To address this issue, most of the existing works assumed that either the positions of users are fixed, or the spatial distributions of users are known. The problems were then solved with deterministic optimization techniques. In practice, however, it is challenging to characterize the distribution of users. Instead, sample positions of mobile users that are constantly measured by APs and stored in the central server are abundant. Therefore, stochastic optimization approaches, which rely upon a collection of samples to produce statistical estimates of the random variables, could be more suitable for solving network topology optimization problems.

This thesis is devoted to the AP placement optimization of infrastructure-based wireless communication networks, including distributed antenna systems (DASs), millimeter wave (mmWave) cellular networks and IEEE 802.11 networks. It begins from the base station (BS) antenna layout optimization of a multi-user DAS. The objective is formulated as the uplink ergodic sum capacity averaged over the distribution of users. A computationally efficient algorithm is proposed to solve it based on stochastic gradient descent. In order to calculate the gradient of the ergodic sum capacity, an accurate closed-form approximation of ergodic sum capacity is also derived. Simulation results corroborate that the proposed algorithm converges fast and significantly outperforms the existing representative BS antenna placement schemes.

The stochastic optimization framework is further extended to optimize the BS placement in mmWave cellular networks. The aim is to minimize the long-term outage probability, which is related to the BS placement and user association. The BS placement problem is formulated as maximizing the average number of physically accessible BSs of each user under an inaccessible probability constraint. As both the objective and constraint functions are in the form of expectation, a novel algorithm based on cooperative stochastic approximation (CSA) is proposed to effectively determine the optimal placement of BSs. For user association, a load-balancing user association scheme is proposed, which can achieve similar outage performance to the Hungarian-algorithm-based optimal user association scheme, but with much less running time. Combined with the proposed user association scheme, the proposed BS placement scheme can significantly improve the network outage probability in the long term, especially when the aggregation degree of users is large.

Finally, the optimal placement and coverage of APs for IEEE 802.11 networks are investigated under the stochastic optimization framework. The objective is to maximize the average network throughput under an outage probability constraint. A CSA-based algorithm is developed to solve the problem. Simulation results demonstrate that the proposed AP placement algorithm can quickly converge, and gains over the representative grid installation AP placement scheme increase as the outage constraint becomes loose or the users' spatial distribution becomes more clustered. The methodology in this thesis offers important insights into practical network topology design for infrastructure-based wireless communication networks.