Predictive delay-aware network selection in data offloading

To address the increasingly severe congestion problem in cellular networks, mobile operators are actively considering offloading the cellular traffic to other complementary networks. In this paper, we study the online network selection problem in operator-initiated data offloading with multiple mobile users, taking into account the operation cost, queueing delay, and traffic load in different access networks (e.g., cellular macrocell, femtocell, and Wi-Fi networks). We first design a Delay-Aware Network Selection (DNS) algorithm based on the Lyapunov optimization technique. The DNS algorithm yields an operation cost within O (1/V) bound of the optimal value, and guarantees an O (V) traffic delay for any control parameter V > 0. Next, we incorporate the prediction of users´ mobilities and traffic arrivals into the network selection. Specifically, we assume that the users´ locations and traffic arrivals in the next few time slots can be estimated accurately, and propose a Predictive Delay-Aware Network Selection (P-DNS) algorithm to utilize this information based on a novel frame-based design. We characterize the performance bounds of P-DNS in terms of cost-delay tradeoff theoretically. To further reduce the computational complexity, we propose a Greedy Predictive Delay-Aware Network Selection (GP-DNS) algorithm, where the operator solves the network selection problem approximately and iteratively. Numerical results show that GP-DNS improves the cost-delay performance over DNS, and reduces the queueing delay by roughly 40% with the same operation cost.