From cognitive to docitive radios: the role of machine learning in intelligent wireless multimedia networks
In recent years, multimedia wireless transmissions have become a rapidly growing field and have received increasing attentions. Enabling multimedia communications over wireless networks to reach their full potential is a challenging task, due to the complex and time-varying features of wireless networks. This dissertation presents intelligent multimedia wireless transmission schemes that enable prioritized multimedia transmission over various wireless networks, using advanced wireless networking techniques and cutting-edge machine learning techniques. Particularly, crosslayer design for multimedia transmission, spectrum handoff for cognitive radio networks, and multichannel wireless mesh networks with multi-beam antennas are addressed for the improvement of multimedia wireless transmission. Non-linear optimization is utilized for cooperative design of cross-layer wireless transmission; manifold learning is explored for dimensional reduction and user similarity measurement. Mixed preemptive resume priority and non-preemptive resume priority M/G/1 queueing models are proposed to for modeling the spectrum usage behavior for prioritized multimedia applications in wireless networks. Reinforcement learning is adapted to enable users to learn from their experience and the environment and apprenticeship learning is adapted to enable users to learn from other experienced users in a similar wireless network environment. These proposed transmission schemes have one or multiple advantages as: (1) efficiently uses available wireless resources to achieve the optimum transmission performance by means of cooperative design between different wireless layers and/or different users; (2) explicitly considers complex wireless communication conditions; (3) enables prioritized multimedia applications through allocating more wireless resources to applications with a higher priority; (4) opportunistically optimizes spectrum usage through a hybrid queueing model that manages all the spectrum usage in the network; (5) enables users to conduct spectrum behavior intelligently through learning from experience of their own as well as of other experienced users.