Abstract :

The ever-increasing demands for higher data rate, lower cost, reduced latency, and other quality-of-service (QoS) requirements has urged researchers to think out of the box. The enormous growth in hand-held devices and reduction in their size and cost has resulted in a tremendous rise in the load over a cellular network. For adhering such an immense and diversified QoS requirements, the integration of multiple radio access technologies (RATs) and data offloading to complementary networks are recognized as key players. Therefore, the central purpose of this thesis is to develop adaptive data offloading policies for a two-RAT wireless network. In line with this aim, the main contributions of this thesis are fourfold.
Firstly, a unified framework is developed for the analysis of a multi-RAT heterogeneous network (HetNet) by taking into account the effect of both contention-free and contention-based channel access schemes. Such a model helps in estimating the average signal-to-interference-noise ratio (SINR) and rate coverage experienced by a user in a given scenario. An special scenario by assuming a two-RAT network, including cellular and wireless fidelity (Wi-Fi) RATs with multiple tiers, has been thoroughly analyzed. Various results are discussed and through comparative analysis it is shown that the proposed framework better approximates the practical scenario as compared to the existing methods.
Secondly, a joint spatio-temporal model is derived for the analysis of a single-tier cellular RAT. By keeping in view the impact of variations in users behavior on the overall performance of the network, the use of ON/OFF traffic models is advocated for modeling the activity of users generating requests for different service. For performance evaluation of the proposed model, various performance metrics which include success probability, mean delay, and stable region probability are derived. By assuming homogeneous and heterogeneous traffic scenarios, various results are discussed and the proposed model is benchmarked against the state-of-the-art approach.
Thirdly, an intent-based offloading (IBO) framework is developed wherein a user expresses the intent and the UE translates that into an optimal data offloading policy whenever possible. The task of UE is to find optimal data offloading policy by minimizing the cost of data usage while satisfying the intent defined by the user. Due to some shortcomings and higher computational complexity of iterative learning algorithms, an analytical expression is derived by exploiting Markov process (MP) and the developed unified framework for rate coverage analysis of a multi-RAT wireless network. The proposed IBO framework is benchmarked against standard methods and state-of-the-art algorithm.
Lastly, an online data offloading policy is developed by using Q-learning algorithm.
The online methods require quite a large number of learning episodes for reaching an optimal solution. However, in practical scenarios, especially in the supposed context, such a large number of learning episodes is less likely to be experienced by a user. Therefore, a few simple methods are discussed for speeding up the learning process. With the help of several results, it is reported that though the approach is not able to find an optimal data offloading policy, its performance is comparable to on-the-spot-offload (OTSO) method.