The framework of cognitive wireless networks is expected to endow wireless devices with a cognition-intelligence ability with which they can efficiently learn and respond to the dynamic wireless environment. In this dissertation, we focus on the problem of developing cognitive network control mechanisms without knowing in advance an accurate network model. We study a series of cross-layer resource allocation problems in cognitive wireless networks. Based on model-free learning, optimization and game theory, we propose a framework of self-organized, adaptive strategy learning for wireless devices to (implicitly) build the understanding of the network dynamics through trial-and-error.

The work of this dissertation is divided into three parts. In the first part, we investigate a distributed, single-agent decision-making problem for real-time video streaming over a time-varying wireless channel between a single pair of transmitter and receiver. By modeling the joint source-channel resource allocation process for video streaming as a constrained Markov decision process, we propose a reinforcement learning scheme to search for the optimal transmission policy without the need to know in advance the details of network dynamics.

In the second part of this work, we extend our study from the single-agent to a multi-agent decision-making scenario, and study the energy-efficient power allocation problems in a two-tier, underlay heterogeneous network and in a self-sustainable green network. For the heterogeneous network, we propose a stochastic learning algorithm based on repeated games to allow individual macro- or femto-users to find a Stackelberg equilibrium without flooding the network with local action information. For the self-sustainable green network, we propose a combinatorial auction mechanism that allows mobile stations to adaptively choose the optimal base station and sub-carrier group for transmission only from local payoff and transmission strategy information.

In the third part of this work, we study a cross-layer routing problem in an interweaved Cognitive Radio Network (CRN), where an accurate network model is not available and the secondary users that are distributed within the CRN only have access to local action/utility information. In order to develop a spectrum-aware routing mechanism that is robust against potential insider attackers, we model the uncoordinated interaction between CRN nodes in the dynamic wireless environment as a stochastic game. Through decomposition of the stochastic routing game, we propose two stochastic learning algorithm based on a group of repeated stage games for the secondary users to learn the best-response strategies without the need of information flooding.

Library of Congress Subject Headings

Wireless communication systems; Machine learning

Publication Date


Document Type


Student Type


Degree Name

Computing and Information Sciences (Ph.D.)

Department, Program, or Center

Information Sciences and Technologies (GCCIS)


Andres Kwasinski

Advisor/Committee Member

Zhu Han

Advisor/Committee Member

Pengcheng Shi


Physical copy available from RIT's Wallace Library at TK5203.2 .W36 2016


RIT – Main Campus