The behavior of traffic in wireless networks is becoming more and more complex due to the diversity of sources as well as the associated access protocols and transmission policies, resulting in high degrees of randomness in the aggregate traffic load. This gives rise to imminent need for more accurate and efficient models to capture the random, bursty, and time-variant characteristics of wireless networks. Especially, it is of high importance to take into account the randomness in the occupancy of wireless channels towards modeling a fair and efficient coexistence for dissimilar radio access technologies operating in a common spectrum. Recognizing the inherent distinctions between a schedule-based and a random-access contention-based wireless network, and the inevitable need for them to share the scarce resource of spectrum, this dissertation takes WiFi and LTE-U as examples to address two important problems: modeling temporal characteristics of a random-access network (WiFi), and exploiting the temporal opportunities therein by a schedule-based network (LTE-U).

Towards this end, we first develop an analytical framework to model the temporal characteristics of WiFi channel statistics. In particular, of interest is the duration of white spaces resulting from idle periods in WiFi downlink and uplink traffic. We employ a bivariate Markov process that captures the underlying status and dynamics of WiFi network in formulating group Poisson arrivals corresponding to aggregated WiFi frames, and determine the parameters of this Markov modulated batch Poisson process by matching the associated first order traffic arrival statistics with those of the bivariate process. While this model is drastically simpler than others such as BMAP, nevertheless, it provides remarkably similar accuracy in modeling the duration and abundance of white spaces. The results from analytical models corroborate those of the simulation platform we developed in NS3 as well as MATLAB.

Next, we use this model to study the feasibility of exploiting WiFi white spaces for LTE-U transmissions under different traffic intensities and WiFi network scenarios. We further propose an opportunistic coexistence algorithm that enables LTE-U base station to dynamically estimate the duration of upcoming WiFi white spaces and determine LTE-U ON/OFF epochs. The proposed scheme demonstrates performance limits when latency imposed on WiFi activity is minimized, while LTE-U maximally utilizes the available spectral resources. A useful application of this model is when integrated with stochastic geometry to form a basis for optimal resource allocation in 5G small cells towards guaranteeing network quality of service.

Finally, recognizing the restrictions in implementation aspects of the above model, we propose an alternative machine learning-based resolution for the opportunistic coexistence of LTE-U with WiFi. In this approach, LTE-U uses carrier sensing along with the modelfree and goal-directed reinforcement learning method to minimize the latency imposed on WiFi network operation, while maximizing spectral utilization. This approach enables a robust and rather simple realization of online, as well as distributed learning for optimal decision making in autonomous LTE-U small cells, without requiring detailed knowledge of the coexisting WiFi network characteristics.